blob: 163c890f436d7f9bfccf565c08464d4814b641e7 [file] [log] [blame]
Paul Menageddbcc7e2007-10-18 23:39:30 -07001/*
Paul Menageddbcc7e2007-10-18 23:39:30 -07002 * Generic process-grouping system.
3 *
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
6 *
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08007 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
10 *
Paul Menageddbcc7e2007-10-18 23:39:30 -070011 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
15 *
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
18 *
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
23 *
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
27 */
28
29#include <linux/cgroup.h>
Paul Menagec6d57f32009-09-23 15:56:19 -070030#include <linux/ctype.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070031#include <linux/errno.h>
32#include <linux/fs.h>
33#include <linux/kernel.h>
34#include <linux/list.h>
35#include <linux/mm.h>
36#include <linux/mutex.h>
37#include <linux/mount.h>
38#include <linux/pagemap.h>
Paul Menagea4243162007-10-18 23:39:35 -070039#include <linux/proc_fs.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070040#include <linux/rcupdate.h>
41#include <linux/sched.h>
Paul Menage817929e2007-10-18 23:39:36 -070042#include <linux/backing-dev.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070043#include <linux/seq_file.h>
44#include <linux/slab.h>
45#include <linux/magic.h>
46#include <linux/spinlock.h>
47#include <linux/string.h>
Paul Menagebbcb81d2007-10-18 23:39:32 -070048#include <linux/sort.h>
Paul Menage81a6a5c2007-10-18 23:39:38 -070049#include <linux/kmod.h>
Ben Blume6a11052010-03-10 15:22:09 -080050#include <linux/module.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070051#include <linux/delayacct.h>
52#include <linux/cgroupstats.h>
Li Zefan472b1052008-04-29 01:00:11 -070053#include <linux/hash.h>
Al Viro3f8206d2008-07-26 03:46:43 -040054#include <linux/namei.h>
Li Zefan096b7fe2009-07-29 15:04:04 -070055#include <linux/pid_namespace.h>
Paul Menage2c6ab6d2009-09-23 15:56:23 -070056#include <linux/idr.h>
Ben Blumd1d9fd32009-09-23 15:56:28 -070057#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -080058#include <linux/eventfd.h>
59#include <linux/poll.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070060
Paul Menageddbcc7e2007-10-18 23:39:30 -070061#include <asm/atomic.h>
62
Paul Menage81a6a5c2007-10-18 23:39:38 -070063static DEFINE_MUTEX(cgroup_mutex);
64
Ben Blumaae8aab2010-03-10 15:22:07 -080065/*
66 * Generate an array of cgroup subsystem pointers. At boot time, this is
67 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
68 * registered after that. The mutable section of this array is protected by
69 * cgroup_mutex.
70 */
Paul Menageddbcc7e2007-10-18 23:39:30 -070071#define SUBSYS(_x) &_x ## _subsys,
Ben Blumaae8aab2010-03-10 15:22:07 -080072static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
Paul Menageddbcc7e2007-10-18 23:39:30 -070073#include <linux/cgroup_subsys.h>
74};
75
Paul Menagec6d57f32009-09-23 15:56:19 -070076#define MAX_CGROUP_ROOT_NAMELEN 64
77
Paul Menageddbcc7e2007-10-18 23:39:30 -070078/*
79 * A cgroupfs_root represents the root of a cgroup hierarchy,
80 * and may be associated with a superblock to form an active
81 * hierarchy
82 */
83struct cgroupfs_root {
84 struct super_block *sb;
85
86 /*
87 * The bitmask of subsystems intended to be attached to this
88 * hierarchy
89 */
90 unsigned long subsys_bits;
91
Paul Menage2c6ab6d2009-09-23 15:56:23 -070092 /* Unique id for this hierarchy. */
93 int hierarchy_id;
94
Paul Menageddbcc7e2007-10-18 23:39:30 -070095 /* The bitmask of subsystems currently attached to this hierarchy */
96 unsigned long actual_subsys_bits;
97
98 /* A list running through the attached subsystems */
99 struct list_head subsys_list;
100
101 /* The root cgroup for this hierarchy */
102 struct cgroup top_cgroup;
103
104 /* Tracks how many cgroups are currently defined in hierarchy.*/
105 int number_of_cgroups;
106
Li Zefane5f6a862009-01-07 18:07:41 -0800107 /* A list running through the active hierarchies */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700108 struct list_head root_list;
109
110 /* Hierarchy-specific flags */
111 unsigned long flags;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700112
Paul Menagee788e062008-07-25 01:46:59 -0700113 /* The path to use for release notifications. */
Paul Menage81a6a5c2007-10-18 23:39:38 -0700114 char release_agent_path[PATH_MAX];
Paul Menagec6d57f32009-09-23 15:56:19 -0700115
116 /* The name for this hierarchy - may be empty */
117 char name[MAX_CGROUP_ROOT_NAMELEN];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700118};
119
Paul Menageddbcc7e2007-10-18 23:39:30 -0700120/*
121 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
122 * subsystems that are otherwise unattached - it never has more than a
123 * single cgroup, and all tasks are part of that cgroup.
124 */
125static struct cgroupfs_root rootnode;
126
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700127/*
128 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
129 * cgroup_subsys->use_id != 0.
130 */
131#define CSS_ID_MAX (65535)
132struct css_id {
133 /*
134 * The css to which this ID points. This pointer is set to valid value
135 * after cgroup is populated. If cgroup is removed, this will be NULL.
136 * This pointer is expected to be RCU-safe because destroy()
137 * is called after synchronize_rcu(). But for safe use, css_is_removed()
138 * css_tryget() should be used for avoiding race.
139 */
Arnd Bergmann2c392b82010-02-24 19:41:39 +0100140 struct cgroup_subsys_state __rcu *css;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700141 /*
142 * ID of this css.
143 */
144 unsigned short id;
145 /*
146 * Depth in hierarchy which this ID belongs to.
147 */
148 unsigned short depth;
149 /*
150 * ID is freed by RCU. (and lookup routine is RCU safe.)
151 */
152 struct rcu_head rcu_head;
153 /*
154 * Hierarchy of CSS ID belongs to.
155 */
156 unsigned short stack[0]; /* Array of Length (depth+1) */
157};
158
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800159/*
160 * cgroup_event represents events which userspace want to recieve.
161 */
162struct cgroup_event {
163 /*
164 * Cgroup which the event belongs to.
165 */
166 struct cgroup *cgrp;
167 /*
168 * Control file which the event associated.
169 */
170 struct cftype *cft;
171 /*
172 * eventfd to signal userspace about the event.
173 */
174 struct eventfd_ctx *eventfd;
175 /*
176 * Each of these stored in a list by the cgroup.
177 */
178 struct list_head list;
179 /*
180 * All fields below needed to unregister event when
181 * userspace closes eventfd.
182 */
183 poll_table pt;
184 wait_queue_head_t *wqh;
185 wait_queue_t wait;
186 struct work_struct remove;
187};
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700188
Paul Menageddbcc7e2007-10-18 23:39:30 -0700189/* The list of hierarchy roots */
190
191static LIST_HEAD(roots);
Paul Menage817929e2007-10-18 23:39:36 -0700192static int root_count;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700193
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700194static DEFINE_IDA(hierarchy_ida);
195static int next_hierarchy_id;
196static DEFINE_SPINLOCK(hierarchy_id_lock);
197
Paul Menageddbcc7e2007-10-18 23:39:30 -0700198/* dummytop is a shorthand for the dummy hierarchy's top cgroup */
199#define dummytop (&rootnode.top_cgroup)
200
201/* This flag indicates whether tasks in the fork and exit paths should
Li Zefana043e3b2008-02-23 15:24:09 -0800202 * check for fork/exit handlers to call. This avoids us having to do
203 * extra work in the fork/exit path if none of the subsystems need to
204 * be called.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700205 */
Li Zefan8947f9d2008-07-25 01:46:56 -0700206static int need_forkexit_callback __read_mostly;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700207
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800208#ifdef CONFIG_PROVE_LOCKING
209int cgroup_lock_is_held(void)
210{
211 return lockdep_is_held(&cgroup_mutex);
212}
213#else /* #ifdef CONFIG_PROVE_LOCKING */
214int cgroup_lock_is_held(void)
215{
216 return mutex_is_locked(&cgroup_mutex);
217}
218#endif /* #else #ifdef CONFIG_PROVE_LOCKING */
219
220EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
221
Paul Menageddbcc7e2007-10-18 23:39:30 -0700222/* convenient tests for these bits */
Paul Menagebd89aab2007-10-18 23:40:44 -0700223inline int cgroup_is_removed(const struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700224{
Paul Menagebd89aab2007-10-18 23:40:44 -0700225 return test_bit(CGRP_REMOVED, &cgrp->flags);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700226}
227
228/* bits in struct cgroupfs_root flags field */
229enum {
230 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
231};
232
Adrian Bunke9685a02008-02-07 00:13:46 -0800233static int cgroup_is_releasable(const struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -0700234{
235 const int bits =
Paul Menagebd89aab2007-10-18 23:40:44 -0700236 (1 << CGRP_RELEASABLE) |
237 (1 << CGRP_NOTIFY_ON_RELEASE);
238 return (cgrp->flags & bits) == bits;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700239}
240
Adrian Bunke9685a02008-02-07 00:13:46 -0800241static int notify_on_release(const struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -0700242{
Paul Menagebd89aab2007-10-18 23:40:44 -0700243 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700244}
245
Daniel Lezcano97978e62010-10-27 15:33:35 -0700246static int clone_children(const struct cgroup *cgrp)
247{
248 return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
249}
250
Paul Menageddbcc7e2007-10-18 23:39:30 -0700251/*
252 * for_each_subsys() allows you to iterate on each subsystem attached to
253 * an active hierarchy
254 */
255#define for_each_subsys(_root, _ss) \
256list_for_each_entry(_ss, &_root->subsys_list, sibling)
257
Li Zefane5f6a862009-01-07 18:07:41 -0800258/* for_each_active_root() allows you to iterate across the active hierarchies */
259#define for_each_active_root(_root) \
Paul Menageddbcc7e2007-10-18 23:39:30 -0700260list_for_each_entry(_root, &roots, root_list)
261
Paul Menage81a6a5c2007-10-18 23:39:38 -0700262/* the list of cgroups eligible for automatic release. Protected by
263 * release_list_lock */
264static LIST_HEAD(release_list);
265static DEFINE_SPINLOCK(release_list_lock);
266static void cgroup_release_agent(struct work_struct *work);
267static DECLARE_WORK(release_agent_work, cgroup_release_agent);
Paul Menagebd89aab2007-10-18 23:40:44 -0700268static void check_for_release(struct cgroup *cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700269
Paul Menage817929e2007-10-18 23:39:36 -0700270/* Link structure for associating css_set objects with cgroups */
271struct cg_cgroup_link {
272 /*
273 * List running through cg_cgroup_links associated with a
274 * cgroup, anchored on cgroup->css_sets
275 */
Paul Menagebd89aab2007-10-18 23:40:44 -0700276 struct list_head cgrp_link_list;
Paul Menage7717f7b2009-09-23 15:56:22 -0700277 struct cgroup *cgrp;
Paul Menage817929e2007-10-18 23:39:36 -0700278 /*
279 * List running through cg_cgroup_links pointing at a
280 * single css_set object, anchored on css_set->cg_links
281 */
282 struct list_head cg_link_list;
283 struct css_set *cg;
284};
285
286/* The default css_set - used by init and its children prior to any
287 * hierarchies being mounted. It contains a pointer to the root state
288 * for each subsystem. Also used to anchor the list of css_sets. Not
289 * reference-counted, to improve performance when child cgroups
290 * haven't been created.
291 */
292
293static struct css_set init_css_set;
294static struct cg_cgroup_link init_css_set_link;
295
Ben Blume6a11052010-03-10 15:22:09 -0800296static int cgroup_init_idr(struct cgroup_subsys *ss,
297 struct cgroup_subsys_state *css);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700298
Paul Menage817929e2007-10-18 23:39:36 -0700299/* css_set_lock protects the list of css_set objects, and the
300 * chain of tasks off each css_set. Nests outside task->alloc_lock
301 * due to cgroup_iter_start() */
302static DEFINE_RWLOCK(css_set_lock);
303static int css_set_count;
304
Paul Menage7717f7b2009-09-23 15:56:22 -0700305/*
306 * hash table for cgroup groups. This improves the performance to find
307 * an existing css_set. This hash doesn't (currently) take into
308 * account cgroups in empty hierarchies.
309 */
Li Zefan472b1052008-04-29 01:00:11 -0700310#define CSS_SET_HASH_BITS 7
311#define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
312static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
313
314static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
315{
316 int i;
317 int index;
318 unsigned long tmp = 0UL;
319
320 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
321 tmp += (unsigned long)css[i];
322 tmp = (tmp >> 16) ^ tmp;
323
324 index = hash_long(tmp, CSS_SET_HASH_BITS);
325
326 return &css_set_table[index];
327}
328
Ben Blumc3783692009-09-23 15:56:29 -0700329static void free_css_set_rcu(struct rcu_head *obj)
330{
331 struct css_set *cg = container_of(obj, struct css_set, rcu_head);
332 kfree(cg);
333}
334
Paul Menage817929e2007-10-18 23:39:36 -0700335/* We don't maintain the lists running through each css_set to its
336 * task until after the first call to cgroup_iter_start(). This
337 * reduces the fork()/exit() overhead for people who have cgroups
338 * compiled into their kernel but not actually in use */
Li Zefan8947f9d2008-07-25 01:46:56 -0700339static int use_task_css_set_links __read_mostly;
Paul Menage817929e2007-10-18 23:39:36 -0700340
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700341static void __put_css_set(struct css_set *cg, int taskexit)
Paul Menageb4f48b62007-10-18 23:39:33 -0700342{
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -0700343 struct cg_cgroup_link *link;
344 struct cg_cgroup_link *saved_link;
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700345 /*
346 * Ensure that the refcount doesn't hit zero while any readers
347 * can see it. Similar to atomic_dec_and_lock(), but for an
348 * rwlock
349 */
350 if (atomic_add_unless(&cg->refcount, -1, 1))
351 return;
352 write_lock(&css_set_lock);
353 if (!atomic_dec_and_test(&cg->refcount)) {
354 write_unlock(&css_set_lock);
355 return;
356 }
Paul Menage81a6a5c2007-10-18 23:39:38 -0700357
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700358 /* This css_set is dead. unlink it and release cgroup refcounts */
359 hlist_del(&cg->hlist);
360 css_set_count--;
361
362 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
363 cg_link_list) {
364 struct cgroup *cgrp = link->cgrp;
365 list_del(&link->cg_link_list);
366 list_del(&link->cgrp_link_list);
Paul Menagebd89aab2007-10-18 23:40:44 -0700367 if (atomic_dec_and_test(&cgrp->count) &&
368 notify_on_release(cgrp)) {
Paul Menage81a6a5c2007-10-18 23:39:38 -0700369 if (taskexit)
Paul Menagebd89aab2007-10-18 23:40:44 -0700370 set_bit(CGRP_RELEASABLE, &cgrp->flags);
371 check_for_release(cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700372 }
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700373
374 kfree(link);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700375 }
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700376
377 write_unlock(&css_set_lock);
Ben Blumc3783692009-09-23 15:56:29 -0700378 call_rcu(&cg->rcu_head, free_css_set_rcu);
Paul Menage817929e2007-10-18 23:39:36 -0700379}
380
381/*
382 * refcounted get/put for css_set objects
383 */
384static inline void get_css_set(struct css_set *cg)
385{
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700386 atomic_inc(&cg->refcount);
Paul Menage817929e2007-10-18 23:39:36 -0700387}
388
389static inline void put_css_set(struct css_set *cg)
390{
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700391 __put_css_set(cg, 0);
Paul Menage817929e2007-10-18 23:39:36 -0700392}
393
Paul Menage81a6a5c2007-10-18 23:39:38 -0700394static inline void put_css_set_taskexit(struct css_set *cg)
395{
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700396 __put_css_set(cg, 1);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700397}
398
Paul Menage817929e2007-10-18 23:39:36 -0700399/*
Paul Menage7717f7b2009-09-23 15:56:22 -0700400 * compare_css_sets - helper function for find_existing_css_set().
401 * @cg: candidate css_set being tested
402 * @old_cg: existing css_set for a task
403 * @new_cgrp: cgroup that's being entered by the task
404 * @template: desired set of css pointers in css_set (pre-calculated)
405 *
406 * Returns true if "cg" matches "old_cg" except for the hierarchy
407 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
408 */
409static bool compare_css_sets(struct css_set *cg,
410 struct css_set *old_cg,
411 struct cgroup *new_cgrp,
412 struct cgroup_subsys_state *template[])
413{
414 struct list_head *l1, *l2;
415
416 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
417 /* Not all subsystems matched */
418 return false;
419 }
420
421 /*
422 * Compare cgroup pointers in order to distinguish between
423 * different cgroups in heirarchies with no subsystems. We
424 * could get by with just this check alone (and skip the
425 * memcmp above) but on most setups the memcmp check will
426 * avoid the need for this more expensive check on almost all
427 * candidates.
428 */
429
430 l1 = &cg->cg_links;
431 l2 = &old_cg->cg_links;
432 while (1) {
433 struct cg_cgroup_link *cgl1, *cgl2;
434 struct cgroup *cg1, *cg2;
435
436 l1 = l1->next;
437 l2 = l2->next;
438 /* See if we reached the end - both lists are equal length. */
439 if (l1 == &cg->cg_links) {
440 BUG_ON(l2 != &old_cg->cg_links);
441 break;
442 } else {
443 BUG_ON(l2 == &old_cg->cg_links);
444 }
445 /* Locate the cgroups associated with these links. */
446 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
447 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
448 cg1 = cgl1->cgrp;
449 cg2 = cgl2->cgrp;
450 /* Hierarchies should be linked in the same order. */
451 BUG_ON(cg1->root != cg2->root);
452
453 /*
454 * If this hierarchy is the hierarchy of the cgroup
455 * that's changing, then we need to check that this
456 * css_set points to the new cgroup; if it's any other
457 * hierarchy, then this css_set should point to the
458 * same cgroup as the old css_set.
459 */
460 if (cg1->root == new_cgrp->root) {
461 if (cg1 != new_cgrp)
462 return false;
463 } else {
464 if (cg1 != cg2)
465 return false;
466 }
467 }
468 return true;
469}
470
471/*
Paul Menage817929e2007-10-18 23:39:36 -0700472 * find_existing_css_set() is a helper for
473 * find_css_set(), and checks to see whether an existing
Li Zefan472b1052008-04-29 01:00:11 -0700474 * css_set is suitable.
Paul Menage817929e2007-10-18 23:39:36 -0700475 *
476 * oldcg: the cgroup group that we're using before the cgroup
477 * transition
478 *
Paul Menagebd89aab2007-10-18 23:40:44 -0700479 * cgrp: the cgroup that we're moving into
Paul Menage817929e2007-10-18 23:39:36 -0700480 *
481 * template: location in which to build the desired set of subsystem
482 * state objects for the new cgroup group
483 */
Paul Menage817929e2007-10-18 23:39:36 -0700484static struct css_set *find_existing_css_set(
485 struct css_set *oldcg,
Paul Menagebd89aab2007-10-18 23:40:44 -0700486 struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -0700487 struct cgroup_subsys_state *template[])
488{
489 int i;
Paul Menagebd89aab2007-10-18 23:40:44 -0700490 struct cgroupfs_root *root = cgrp->root;
Li Zefan472b1052008-04-29 01:00:11 -0700491 struct hlist_head *hhead;
492 struct hlist_node *node;
493 struct css_set *cg;
Paul Menage817929e2007-10-18 23:39:36 -0700494
Ben Blumaae8aab2010-03-10 15:22:07 -0800495 /*
496 * Build the set of subsystem state objects that we want to see in the
497 * new css_set. while subsystems can change globally, the entries here
498 * won't change, so no need for locking.
499 */
Paul Menage817929e2007-10-18 23:39:36 -0700500 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
Li Zefan8d53d552008-02-23 15:24:11 -0800501 if (root->subsys_bits & (1UL << i)) {
Paul Menage817929e2007-10-18 23:39:36 -0700502 /* Subsystem is in this hierarchy. So we want
503 * the subsystem state from the new
504 * cgroup */
Paul Menagebd89aab2007-10-18 23:40:44 -0700505 template[i] = cgrp->subsys[i];
Paul Menage817929e2007-10-18 23:39:36 -0700506 } else {
507 /* Subsystem is not in this hierarchy, so we
508 * don't want to change the subsystem state */
509 template[i] = oldcg->subsys[i];
510 }
511 }
512
Li Zefan472b1052008-04-29 01:00:11 -0700513 hhead = css_set_hash(template);
514 hlist_for_each_entry(cg, node, hhead, hlist) {
Paul Menage7717f7b2009-09-23 15:56:22 -0700515 if (!compare_css_sets(cg, oldcg, cgrp, template))
516 continue;
517
518 /* This css_set matches what we need */
519 return cg;
Li Zefan472b1052008-04-29 01:00:11 -0700520 }
Paul Menage817929e2007-10-18 23:39:36 -0700521
522 /* No existing cgroup group matched */
523 return NULL;
524}
525
Paul Menage817929e2007-10-18 23:39:36 -0700526static void free_cg_links(struct list_head *tmp)
527{
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -0700528 struct cg_cgroup_link *link;
529 struct cg_cgroup_link *saved_link;
530
531 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
Paul Menagebd89aab2007-10-18 23:40:44 -0700532 list_del(&link->cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -0700533 kfree(link);
534 }
535}
536
537/*
Li Zefan36553432008-07-29 22:33:19 -0700538 * allocate_cg_links() allocates "count" cg_cgroup_link structures
539 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
540 * success or a negative error
541 */
542static int allocate_cg_links(int count, struct list_head *tmp)
543{
544 struct cg_cgroup_link *link;
545 int i;
546 INIT_LIST_HEAD(tmp);
547 for (i = 0; i < count; i++) {
548 link = kmalloc(sizeof(*link), GFP_KERNEL);
549 if (!link) {
550 free_cg_links(tmp);
551 return -ENOMEM;
552 }
553 list_add(&link->cgrp_link_list, tmp);
554 }
555 return 0;
556}
557
Li Zefanc12f65d2009-01-07 18:07:42 -0800558/**
559 * link_css_set - a helper function to link a css_set to a cgroup
560 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
561 * @cg: the css_set to be linked
562 * @cgrp: the destination cgroup
563 */
564static void link_css_set(struct list_head *tmp_cg_links,
565 struct css_set *cg, struct cgroup *cgrp)
566{
567 struct cg_cgroup_link *link;
568
569 BUG_ON(list_empty(tmp_cg_links));
570 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
571 cgrp_link_list);
572 link->cg = cg;
Paul Menage7717f7b2009-09-23 15:56:22 -0700573 link->cgrp = cgrp;
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700574 atomic_inc(&cgrp->count);
Li Zefanc12f65d2009-01-07 18:07:42 -0800575 list_move(&link->cgrp_link_list, &cgrp->css_sets);
Paul Menage7717f7b2009-09-23 15:56:22 -0700576 /*
577 * Always add links to the tail of the list so that the list
578 * is sorted by order of hierarchy creation
579 */
580 list_add_tail(&link->cg_link_list, &cg->cg_links);
Li Zefanc12f65d2009-01-07 18:07:42 -0800581}
582
Li Zefan36553432008-07-29 22:33:19 -0700583/*
Paul Menage817929e2007-10-18 23:39:36 -0700584 * find_css_set() takes an existing cgroup group and a
585 * cgroup object, and returns a css_set object that's
586 * equivalent to the old group, but with the given cgroup
587 * substituted into the appropriate hierarchy. Must be called with
588 * cgroup_mutex held
589 */
Paul Menage817929e2007-10-18 23:39:36 -0700590static struct css_set *find_css_set(
Paul Menagebd89aab2007-10-18 23:40:44 -0700591 struct css_set *oldcg, struct cgroup *cgrp)
Paul Menage817929e2007-10-18 23:39:36 -0700592{
593 struct css_set *res;
594 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
Paul Menage817929e2007-10-18 23:39:36 -0700595
596 struct list_head tmp_cg_links;
Paul Menage817929e2007-10-18 23:39:36 -0700597
Li Zefan472b1052008-04-29 01:00:11 -0700598 struct hlist_head *hhead;
Paul Menage7717f7b2009-09-23 15:56:22 -0700599 struct cg_cgroup_link *link;
Li Zefan472b1052008-04-29 01:00:11 -0700600
Paul Menage817929e2007-10-18 23:39:36 -0700601 /* First see if we already have a cgroup group that matches
602 * the desired set */
Li Zefan7e9abd82008-07-25 01:46:54 -0700603 read_lock(&css_set_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -0700604 res = find_existing_css_set(oldcg, cgrp, template);
Paul Menage817929e2007-10-18 23:39:36 -0700605 if (res)
606 get_css_set(res);
Li Zefan7e9abd82008-07-25 01:46:54 -0700607 read_unlock(&css_set_lock);
Paul Menage817929e2007-10-18 23:39:36 -0700608
609 if (res)
610 return res;
611
612 res = kmalloc(sizeof(*res), GFP_KERNEL);
613 if (!res)
614 return NULL;
615
616 /* Allocate all the cg_cgroup_link objects that we'll need */
617 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
618 kfree(res);
619 return NULL;
620 }
621
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700622 atomic_set(&res->refcount, 1);
Paul Menage817929e2007-10-18 23:39:36 -0700623 INIT_LIST_HEAD(&res->cg_links);
624 INIT_LIST_HEAD(&res->tasks);
Li Zefan472b1052008-04-29 01:00:11 -0700625 INIT_HLIST_NODE(&res->hlist);
Paul Menage817929e2007-10-18 23:39:36 -0700626
627 /* Copy the set of subsystem state objects generated in
628 * find_existing_css_set() */
629 memcpy(res->subsys, template, sizeof(res->subsys));
630
631 write_lock(&css_set_lock);
632 /* Add reference counts and links from the new css_set. */
Paul Menage7717f7b2009-09-23 15:56:22 -0700633 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
634 struct cgroup *c = link->cgrp;
635 if (c->root == cgrp->root)
636 c = cgrp;
637 link_css_set(&tmp_cg_links, res, c);
638 }
Paul Menage817929e2007-10-18 23:39:36 -0700639
640 BUG_ON(!list_empty(&tmp_cg_links));
641
Paul Menage817929e2007-10-18 23:39:36 -0700642 css_set_count++;
Li Zefan472b1052008-04-29 01:00:11 -0700643
644 /* Add this cgroup group to the hash table */
645 hhead = css_set_hash(res->subsys);
646 hlist_add_head(&res->hlist, hhead);
647
Paul Menage817929e2007-10-18 23:39:36 -0700648 write_unlock(&css_set_lock);
649
650 return res;
Paul Menageb4f48b62007-10-18 23:39:33 -0700651}
652
Paul Menageddbcc7e2007-10-18 23:39:30 -0700653/*
Paul Menage7717f7b2009-09-23 15:56:22 -0700654 * Return the cgroup for "task" from the given hierarchy. Must be
655 * called with cgroup_mutex held.
656 */
657static struct cgroup *task_cgroup_from_root(struct task_struct *task,
658 struct cgroupfs_root *root)
659{
660 struct css_set *css;
661 struct cgroup *res = NULL;
662
663 BUG_ON(!mutex_is_locked(&cgroup_mutex));
664 read_lock(&css_set_lock);
665 /*
666 * No need to lock the task - since we hold cgroup_mutex the
667 * task can't change groups, so the only thing that can happen
668 * is that it exits and its css is set back to init_css_set.
669 */
670 css = task->cgroups;
671 if (css == &init_css_set) {
672 res = &root->top_cgroup;
673 } else {
674 struct cg_cgroup_link *link;
675 list_for_each_entry(link, &css->cg_links, cg_link_list) {
676 struct cgroup *c = link->cgrp;
677 if (c->root == root) {
678 res = c;
679 break;
680 }
681 }
682 }
683 read_unlock(&css_set_lock);
684 BUG_ON(!res);
685 return res;
686}
687
688/*
Paul Menageddbcc7e2007-10-18 23:39:30 -0700689 * There is one global cgroup mutex. We also require taking
690 * task_lock() when dereferencing a task's cgroup subsys pointers.
691 * See "The task_lock() exception", at the end of this comment.
692 *
693 * A task must hold cgroup_mutex to modify cgroups.
694 *
695 * Any task can increment and decrement the count field without lock.
696 * So in general, code holding cgroup_mutex can't rely on the count
697 * field not changing. However, if the count goes to zero, then only
Cliff Wickman956db3c2008-02-07 00:14:43 -0800698 * cgroup_attach_task() can increment it again. Because a count of zero
Paul Menageddbcc7e2007-10-18 23:39:30 -0700699 * means that no tasks are currently attached, therefore there is no
700 * way a task attached to that cgroup can fork (the other way to
701 * increment the count). So code holding cgroup_mutex can safely
702 * assume that if the count is zero, it will stay zero. Similarly, if
703 * a task holds cgroup_mutex on a cgroup with zero count, it
704 * knows that the cgroup won't be removed, as cgroup_rmdir()
705 * needs that mutex.
706 *
Paul Menageddbcc7e2007-10-18 23:39:30 -0700707 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
708 * (usually) take cgroup_mutex. These are the two most performance
709 * critical pieces of code here. The exception occurs on cgroup_exit(),
710 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
711 * is taken, and if the cgroup count is zero, a usermode call made
Li Zefana043e3b2008-02-23 15:24:09 -0800712 * to the release agent with the name of the cgroup (path relative to
713 * the root of cgroup file system) as the argument.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700714 *
715 * A cgroup can only be deleted if both its 'count' of using tasks
716 * is zero, and its list of 'children' cgroups is empty. Since all
717 * tasks in the system use _some_ cgroup, and since there is always at
718 * least one task in the system (init, pid == 1), therefore, top_cgroup
719 * always has either children cgroups and/or using tasks. So we don't
720 * need a special hack to ensure that top_cgroup cannot be deleted.
721 *
722 * The task_lock() exception
723 *
724 * The need for this exception arises from the action of
Cliff Wickman956db3c2008-02-07 00:14:43 -0800725 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
Li Zefana043e3b2008-02-23 15:24:09 -0800726 * another. It does so using cgroup_mutex, however there are
Paul Menageddbcc7e2007-10-18 23:39:30 -0700727 * several performance critical places that need to reference
728 * task->cgroup without the expense of grabbing a system global
729 * mutex. Therefore except as noted below, when dereferencing or, as
Cliff Wickman956db3c2008-02-07 00:14:43 -0800730 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
Paul Menageddbcc7e2007-10-18 23:39:30 -0700731 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
732 * the task_struct routinely used for such matters.
733 *
734 * P.S. One more locking exception. RCU is used to guard the
Cliff Wickman956db3c2008-02-07 00:14:43 -0800735 * update of a tasks cgroup pointer by cgroup_attach_task()
Paul Menageddbcc7e2007-10-18 23:39:30 -0700736 */
737
Paul Menageddbcc7e2007-10-18 23:39:30 -0700738/**
739 * cgroup_lock - lock out any changes to cgroup structures
740 *
741 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700742void cgroup_lock(void)
743{
744 mutex_lock(&cgroup_mutex);
745}
Ben Blum67523c42010-03-10 15:22:11 -0800746EXPORT_SYMBOL_GPL(cgroup_lock);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700747
748/**
749 * cgroup_unlock - release lock on cgroup changes
750 *
751 * Undo the lock taken in a previous cgroup_lock() call.
752 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700753void cgroup_unlock(void)
754{
755 mutex_unlock(&cgroup_mutex);
756}
Ben Blum67523c42010-03-10 15:22:11 -0800757EXPORT_SYMBOL_GPL(cgroup_unlock);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700758
759/*
760 * A couple of forward declarations required, due to cyclic reference loop:
761 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
762 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
763 * -> cgroup_mkdir.
764 */
765
Nick Piggin5adcee12011-01-07 17:49:20 +1100766static struct dentry *cgroup_lookup(struct inode *dir,
767 struct dentry *dentry, struct nameidata *nd);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700768static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
769static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
Paul Menagebd89aab2007-10-18 23:40:44 -0700770static int cgroup_populate_dir(struct cgroup *cgrp);
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700771static const struct inode_operations cgroup_dir_inode_operations;
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700772static const struct file_operations proc_cgroupstats_operations;
Paul Menagea4243162007-10-18 23:39:35 -0700773
774static struct backing_dev_info cgroup_backing_dev_info = {
Jens Axboed9938312009-06-12 14:45:52 +0200775 .name = "cgroup",
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700776 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Paul Menagea4243162007-10-18 23:39:35 -0700777};
Paul Menageddbcc7e2007-10-18 23:39:30 -0700778
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700779static int alloc_css_id(struct cgroup_subsys *ss,
780 struct cgroup *parent, struct cgroup *child);
781
Paul Menageddbcc7e2007-10-18 23:39:30 -0700782static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
783{
784 struct inode *inode = new_inode(sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700785
786 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400787 inode->i_ino = get_next_ino();
Paul Menageddbcc7e2007-10-18 23:39:30 -0700788 inode->i_mode = mode;
David Howells76aac0e2008-11-14 10:39:12 +1100789 inode->i_uid = current_fsuid();
790 inode->i_gid = current_fsgid();
Paul Menageddbcc7e2007-10-18 23:39:30 -0700791 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
792 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
793 }
794 return inode;
795}
796
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -0800797/*
798 * Call subsys's pre_destroy handler.
799 * This is called before css refcnt check.
800 */
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700801static int cgroup_call_pre_destroy(struct cgroup *cgrp)
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -0800802{
803 struct cgroup_subsys *ss;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700804 int ret = 0;
805
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -0800806 for_each_subsys(cgrp->root, ss)
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700807 if (ss->pre_destroy) {
808 ret = ss->pre_destroy(ss, cgrp);
809 if (ret)
Kirill A. Shutemov4ab78682010-03-10 15:22:34 -0800810 break;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700811 }
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800812
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700813 return ret;
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -0800814}
815
Paul Menagea47295e2009-01-07 18:07:44 -0800816static void free_cgroup_rcu(struct rcu_head *obj)
817{
818 struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head);
819
820 kfree(cgrp);
821}
822
Paul Menageddbcc7e2007-10-18 23:39:30 -0700823static void cgroup_diput(struct dentry *dentry, struct inode *inode)
824{
825 /* is dentry a directory ? if so, kfree() associated cgroup */
826 if (S_ISDIR(inode->i_mode)) {
Paul Menagebd89aab2007-10-18 23:40:44 -0700827 struct cgroup *cgrp = dentry->d_fsdata;
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800828 struct cgroup_subsys *ss;
Paul Menagebd89aab2007-10-18 23:40:44 -0700829 BUG_ON(!(cgroup_is_removed(cgrp)));
Paul Menage81a6a5c2007-10-18 23:39:38 -0700830 /* It's possible for external users to be holding css
831 * reference counts on a cgroup; css_put() needs to
832 * be able to access the cgroup after decrementing
833 * the reference count in order to know if it needs to
834 * queue the cgroup to be handled by the release
835 * agent */
836 synchronize_rcu();
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800837
838 mutex_lock(&cgroup_mutex);
839 /*
840 * Release the subsystem state objects.
841 */
Li Zefan75139b82009-01-07 18:07:33 -0800842 for_each_subsys(cgrp->root, ss)
843 ss->destroy(ss, cgrp);
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800844
845 cgrp->root->number_of_cgroups--;
846 mutex_unlock(&cgroup_mutex);
847
Paul Menagea47295e2009-01-07 18:07:44 -0800848 /*
849 * Drop the active superblock reference that we took when we
850 * created the cgroup
851 */
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800852 deactivate_super(cgrp->root->sb);
853
Ben Blum72a8cb32009-09-23 15:56:27 -0700854 /*
855 * if we're getting rid of the cgroup, refcount should ensure
856 * that there are no pidlists left.
857 */
858 BUG_ON(!list_empty(&cgrp->pidlists));
859
Paul Menagea47295e2009-01-07 18:07:44 -0800860 call_rcu(&cgrp->rcu_head, free_cgroup_rcu);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700861 }
862 iput(inode);
863}
864
865static void remove_dir(struct dentry *d)
866{
867 struct dentry *parent = dget(d->d_parent);
868
869 d_delete(d);
870 simple_rmdir(parent->d_inode, d);
871 dput(parent);
872}
873
874static void cgroup_clear_directory(struct dentry *dentry)
875{
876 struct list_head *node;
877
878 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
879 spin_lock(&dcache_lock);
880 node = dentry->d_subdirs.next;
881 while (node != &dentry->d_subdirs) {
882 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
883 list_del_init(node);
884 if (d->d_inode) {
885 /* This should never be called on a cgroup
886 * directory with child cgroups */
887 BUG_ON(d->d_inode->i_mode & S_IFDIR);
888 d = dget_locked(d);
889 spin_unlock(&dcache_lock);
890 d_delete(d);
891 simple_unlink(dentry->d_inode, d);
892 dput(d);
893 spin_lock(&dcache_lock);
894 }
895 node = dentry->d_subdirs.next;
896 }
897 spin_unlock(&dcache_lock);
898}
899
900/*
901 * NOTE : the dentry must have been dget()'ed
902 */
903static void cgroup_d_remove_dir(struct dentry *dentry)
904{
905 cgroup_clear_directory(dentry);
906
907 spin_lock(&dcache_lock);
908 list_del_init(&dentry->d_u.d_child);
909 spin_unlock(&dcache_lock);
910 remove_dir(dentry);
911}
912
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700913/*
914 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
915 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
916 * reference to css->refcnt. In general, this refcnt is expected to goes down
917 * to zero, soon.
918 *
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -0700919 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700920 */
921DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
922
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -0700923static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700924{
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -0700925 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700926 wake_up_all(&cgroup_rmdir_waitq);
927}
928
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -0700929void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
930{
931 css_get(css);
932}
933
934void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
935{
936 cgroup_wakeup_rmdir_waiter(css->cgroup);
937 css_put(css);
938}
939
Ben Blumaae8aab2010-03-10 15:22:07 -0800940/*
Ben Blumcf5d5942010-03-10 15:22:09 -0800941 * Call with cgroup_mutex held. Drops reference counts on modules, including
942 * any duplicate ones that parse_cgroupfs_options took. If this function
943 * returns an error, no reference counts are touched.
Ben Blumaae8aab2010-03-10 15:22:07 -0800944 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700945static int rebind_subsystems(struct cgroupfs_root *root,
946 unsigned long final_bits)
947{
948 unsigned long added_bits, removed_bits;
Paul Menagebd89aab2007-10-18 23:40:44 -0700949 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700950 int i;
951
Ben Blumaae8aab2010-03-10 15:22:07 -0800952 BUG_ON(!mutex_is_locked(&cgroup_mutex));
953
Paul Menageddbcc7e2007-10-18 23:39:30 -0700954 removed_bits = root->actual_subsys_bits & ~final_bits;
955 added_bits = final_bits & ~root->actual_subsys_bits;
956 /* Check that any added subsystems are currently free */
957 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
Li Zefan8d53d552008-02-23 15:24:11 -0800958 unsigned long bit = 1UL << i;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700959 struct cgroup_subsys *ss = subsys[i];
960 if (!(bit & added_bits))
961 continue;
Ben Blumaae8aab2010-03-10 15:22:07 -0800962 /*
963 * Nobody should tell us to do a subsys that doesn't exist:
964 * parse_cgroupfs_options should catch that case and refcounts
965 * ensure that subsystems won't disappear once selected.
966 */
967 BUG_ON(ss == NULL);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700968 if (ss->root != &rootnode) {
969 /* Subsystem isn't free */
970 return -EBUSY;
971 }
972 }
973
974 /* Currently we don't handle adding/removing subsystems when
975 * any child cgroups exist. This is theoretically supportable
976 * but involves complex error handling, so it's being left until
977 * later */
Paul Menage307257c2008-12-15 13:54:22 -0800978 if (root->number_of_cgroups > 1)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700979 return -EBUSY;
980
981 /* Process each subsystem */
982 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
983 struct cgroup_subsys *ss = subsys[i];
984 unsigned long bit = 1UL << i;
985 if (bit & added_bits) {
986 /* We're binding this subsystem to this hierarchy */
Ben Blumaae8aab2010-03-10 15:22:07 -0800987 BUG_ON(ss == NULL);
Paul Menagebd89aab2007-10-18 23:40:44 -0700988 BUG_ON(cgrp->subsys[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700989 BUG_ON(!dummytop->subsys[i]);
990 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
Paul Menage999cd8a2009-01-07 18:08:36 -0800991 mutex_lock(&ss->hierarchy_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -0700992 cgrp->subsys[i] = dummytop->subsys[i];
993 cgrp->subsys[i]->cgroup = cgrp;
Li Zefan33a68ac2009-01-07 18:07:42 -0800994 list_move(&ss->sibling, &root->subsys_list);
Lai Jiangshanb2aa30f2009-01-07 18:07:37 -0800995 ss->root = root;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700996 if (ss->bind)
Paul Menagebd89aab2007-10-18 23:40:44 -0700997 ss->bind(ss, cgrp);
Paul Menage999cd8a2009-01-07 18:08:36 -0800998 mutex_unlock(&ss->hierarchy_mutex);
Ben Blumcf5d5942010-03-10 15:22:09 -0800999 /* refcount was already taken, and we're keeping it */
Paul Menageddbcc7e2007-10-18 23:39:30 -07001000 } else if (bit & removed_bits) {
1001 /* We're removing this subsystem */
Ben Blumaae8aab2010-03-10 15:22:07 -08001002 BUG_ON(ss == NULL);
Paul Menagebd89aab2007-10-18 23:40:44 -07001003 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1004 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
Paul Menage999cd8a2009-01-07 18:08:36 -08001005 mutex_lock(&ss->hierarchy_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001006 if (ss->bind)
1007 ss->bind(ss, dummytop);
1008 dummytop->subsys[i]->cgroup = dummytop;
Paul Menagebd89aab2007-10-18 23:40:44 -07001009 cgrp->subsys[i] = NULL;
Lai Jiangshanb2aa30f2009-01-07 18:07:37 -08001010 subsys[i]->root = &rootnode;
Li Zefan33a68ac2009-01-07 18:07:42 -08001011 list_move(&ss->sibling, &rootnode.subsys_list);
Paul Menage999cd8a2009-01-07 18:08:36 -08001012 mutex_unlock(&ss->hierarchy_mutex);
Ben Blumcf5d5942010-03-10 15:22:09 -08001013 /* subsystem is now free - drop reference on module */
1014 module_put(ss->module);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001015 } else if (bit & final_bits) {
1016 /* Subsystem state should already exist */
Ben Blumaae8aab2010-03-10 15:22:07 -08001017 BUG_ON(ss == NULL);
Paul Menagebd89aab2007-10-18 23:40:44 -07001018 BUG_ON(!cgrp->subsys[i]);
Ben Blumcf5d5942010-03-10 15:22:09 -08001019 /*
1020 * a refcount was taken, but we already had one, so
1021 * drop the extra reference.
1022 */
1023 module_put(ss->module);
1024#ifdef CONFIG_MODULE_UNLOAD
1025 BUG_ON(ss->module && !module_refcount(ss->module));
1026#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -07001027 } else {
1028 /* Subsystem state shouldn't exist */
Paul Menagebd89aab2007-10-18 23:40:44 -07001029 BUG_ON(cgrp->subsys[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001030 }
1031 }
1032 root->subsys_bits = root->actual_subsys_bits = final_bits;
1033 synchronize_rcu();
1034
1035 return 0;
1036}
1037
1038static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
1039{
1040 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
1041 struct cgroup_subsys *ss;
1042
1043 mutex_lock(&cgroup_mutex);
1044 for_each_subsys(root, ss)
1045 seq_printf(seq, ",%s", ss->name);
1046 if (test_bit(ROOT_NOPREFIX, &root->flags))
1047 seq_puts(seq, ",noprefix");
Paul Menage81a6a5c2007-10-18 23:39:38 -07001048 if (strlen(root->release_agent_path))
1049 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
Daniel Lezcano97978e62010-10-27 15:33:35 -07001050 if (clone_children(&root->top_cgroup))
1051 seq_puts(seq, ",clone_children");
Paul Menagec6d57f32009-09-23 15:56:19 -07001052 if (strlen(root->name))
1053 seq_printf(seq, ",name=%s", root->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001054 mutex_unlock(&cgroup_mutex);
1055 return 0;
1056}
1057
1058struct cgroup_sb_opts {
1059 unsigned long subsys_bits;
1060 unsigned long flags;
Paul Menage81a6a5c2007-10-18 23:39:38 -07001061 char *release_agent;
Daniel Lezcano97978e62010-10-27 15:33:35 -07001062 bool clone_children;
Paul Menagec6d57f32009-09-23 15:56:19 -07001063 char *name;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001064 /* User explicitly requested empty subsystem */
1065 bool none;
Paul Menagec6d57f32009-09-23 15:56:19 -07001066
1067 struct cgroupfs_root *new_root;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001068
Paul Menageddbcc7e2007-10-18 23:39:30 -07001069};
1070
Ben Blumaae8aab2010-03-10 15:22:07 -08001071/*
1072 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
Ben Blumcf5d5942010-03-10 15:22:09 -08001073 * with cgroup_mutex held to protect the subsys[] array. This function takes
1074 * refcounts on subsystems to be used, unless it returns error, in which case
1075 * no refcounts are taken.
Ben Blumaae8aab2010-03-10 15:22:07 -08001076 */
Ben Blumcf5d5942010-03-10 15:22:09 -08001077static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001078{
Daniel Lezcano32a8cf22010-10-27 15:33:37 -07001079 char *token, *o = data;
1080 bool all_ss = false, one_ss = false;
Li Zefanf9ab5b52009-06-17 16:26:33 -07001081 unsigned long mask = (unsigned long)-1;
Ben Blumcf5d5942010-03-10 15:22:09 -08001082 int i;
1083 bool module_pin_failed = false;
Li Zefanf9ab5b52009-06-17 16:26:33 -07001084
Ben Blumaae8aab2010-03-10 15:22:07 -08001085 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1086
Li Zefanf9ab5b52009-06-17 16:26:33 -07001087#ifdef CONFIG_CPUSETS
1088 mask = ~(1UL << cpuset_subsys_id);
1089#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -07001090
Paul Menagec6d57f32009-09-23 15:56:19 -07001091 memset(opts, 0, sizeof(*opts));
Paul Menageddbcc7e2007-10-18 23:39:30 -07001092
1093 while ((token = strsep(&o, ",")) != NULL) {
1094 if (!*token)
1095 return -EINVAL;
Daniel Lezcano32a8cf22010-10-27 15:33:37 -07001096 if (!strcmp(token, "none")) {
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001097 /* Explicitly have no subsystems */
1098 opts->none = true;
Daniel Lezcano32a8cf22010-10-27 15:33:37 -07001099 continue;
1100 }
1101 if (!strcmp(token, "all")) {
1102 /* Mutually exclusive option 'all' + subsystem name */
1103 if (one_ss)
1104 return -EINVAL;
1105 all_ss = true;
1106 continue;
1107 }
1108 if (!strcmp(token, "noprefix")) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07001109 set_bit(ROOT_NOPREFIX, &opts->flags);
Daniel Lezcano32a8cf22010-10-27 15:33:37 -07001110 continue;
1111 }
1112 if (!strcmp(token, "clone_children")) {
Daniel Lezcano97978e62010-10-27 15:33:35 -07001113 opts->clone_children = true;
Daniel Lezcano32a8cf22010-10-27 15:33:37 -07001114 continue;
1115 }
1116 if (!strncmp(token, "release_agent=", 14)) {
Paul Menage81a6a5c2007-10-18 23:39:38 -07001117 /* Specifying two release agents is forbidden */
1118 if (opts->release_agent)
1119 return -EINVAL;
Paul Menagec6d57f32009-09-23 15:56:19 -07001120 opts->release_agent =
Dan Carpentere400c282010-08-10 18:02:54 -07001121 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001122 if (!opts->release_agent)
1123 return -ENOMEM;
Daniel Lezcano32a8cf22010-10-27 15:33:37 -07001124 continue;
1125 }
1126 if (!strncmp(token, "name=", 5)) {
Paul Menagec6d57f32009-09-23 15:56:19 -07001127 const char *name = token + 5;
1128 /* Can't specify an empty name */
1129 if (!strlen(name))
1130 return -EINVAL;
1131 /* Must match [\w.-]+ */
1132 for (i = 0; i < strlen(name); i++) {
1133 char c = name[i];
1134 if (isalnum(c))
1135 continue;
1136 if ((c == '.') || (c == '-') || (c == '_'))
1137 continue;
1138 return -EINVAL;
1139 }
1140 /* Specifying two names is forbidden */
1141 if (opts->name)
1142 return -EINVAL;
1143 opts->name = kstrndup(name,
Dan Carpentere400c282010-08-10 18:02:54 -07001144 MAX_CGROUP_ROOT_NAMELEN - 1,
Paul Menagec6d57f32009-09-23 15:56:19 -07001145 GFP_KERNEL);
1146 if (!opts->name)
1147 return -ENOMEM;
Daniel Lezcano32a8cf22010-10-27 15:33:37 -07001148
1149 continue;
1150 }
1151
1152 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1153 struct cgroup_subsys *ss = subsys[i];
1154 if (ss == NULL)
1155 continue;
1156 if (strcmp(token, ss->name))
1157 continue;
1158 if (ss->disabled)
1159 continue;
1160
1161 /* Mutually exclusive option 'all' + subsystem name */
1162 if (all_ss)
1163 return -EINVAL;
1164 set_bit(i, &opts->subsys_bits);
1165 one_ss = true;
1166
1167 break;
1168 }
1169 if (i == CGROUP_SUBSYS_COUNT)
1170 return -ENOENT;
1171 }
1172
1173 /*
1174 * If the 'all' option was specified select all the subsystems,
1175 * otherwise 'all, 'none' and a subsystem name options were not
1176 * specified, let's default to 'all'
1177 */
1178 if (all_ss || (!all_ss && !one_ss && !opts->none)) {
1179 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1180 struct cgroup_subsys *ss = subsys[i];
1181 if (ss == NULL)
1182 continue;
1183 if (ss->disabled)
1184 continue;
1185 set_bit(i, &opts->subsys_bits);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001186 }
1187 }
1188
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001189 /* Consistency checks */
1190
Li Zefanf9ab5b52009-06-17 16:26:33 -07001191 /*
1192 * Option noprefix was introduced just for backward compatibility
1193 * with the old cpuset, so we allow noprefix only if mounting just
1194 * the cpuset subsystem.
1195 */
1196 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1197 (opts->subsys_bits & mask))
1198 return -EINVAL;
1199
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001200
1201 /* Can't specify "none" and some subsystems */
1202 if (opts->subsys_bits && opts->none)
1203 return -EINVAL;
1204
1205 /*
1206 * We either have to specify by name or by subsystems. (So all
1207 * empty hierarchies must have a name).
1208 */
Paul Menagec6d57f32009-09-23 15:56:19 -07001209 if (!opts->subsys_bits && !opts->name)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001210 return -EINVAL;
1211
Ben Blumcf5d5942010-03-10 15:22:09 -08001212 /*
1213 * Grab references on all the modules we'll need, so the subsystems
1214 * don't dance around before rebind_subsystems attaches them. This may
1215 * take duplicate reference counts on a subsystem that's already used,
1216 * but rebind_subsystems handles this case.
1217 */
1218 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1219 unsigned long bit = 1UL << i;
1220
1221 if (!(bit & opts->subsys_bits))
1222 continue;
1223 if (!try_module_get(subsys[i]->module)) {
1224 module_pin_failed = true;
1225 break;
1226 }
1227 }
1228 if (module_pin_failed) {
1229 /*
1230 * oops, one of the modules was going away. this means that we
1231 * raced with a module_delete call, and to the user this is
1232 * essentially a "subsystem doesn't exist" case.
1233 */
1234 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1235 /* drop refcounts only on the ones we took */
1236 unsigned long bit = 1UL << i;
1237
1238 if (!(bit & opts->subsys_bits))
1239 continue;
1240 module_put(subsys[i]->module);
1241 }
1242 return -ENOENT;
1243 }
1244
Paul Menageddbcc7e2007-10-18 23:39:30 -07001245 return 0;
1246}
1247
Ben Blumcf5d5942010-03-10 15:22:09 -08001248static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1249{
1250 int i;
1251 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1252 unsigned long bit = 1UL << i;
1253
1254 if (!(bit & subsys_bits))
1255 continue;
1256 module_put(subsys[i]->module);
1257 }
1258}
1259
Paul Menageddbcc7e2007-10-18 23:39:30 -07001260static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1261{
1262 int ret = 0;
1263 struct cgroupfs_root *root = sb->s_fs_info;
Paul Menagebd89aab2007-10-18 23:40:44 -07001264 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001265 struct cgroup_sb_opts opts;
1266
Paul Menagebd89aab2007-10-18 23:40:44 -07001267 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001268 mutex_lock(&cgroup_mutex);
1269
1270 /* See what subsystems are wanted */
1271 ret = parse_cgroupfs_options(data, &opts);
1272 if (ret)
1273 goto out_unlock;
1274
Ben Blumcf5d5942010-03-10 15:22:09 -08001275 /* Don't allow flags or name to change at remount */
1276 if (opts.flags != root->flags ||
1277 (opts.name && strcmp(opts.name, root->name))) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07001278 ret = -EINVAL;
Ben Blumcf5d5942010-03-10 15:22:09 -08001279 drop_parsed_module_refcounts(opts.subsys_bits);
Paul Menagec6d57f32009-09-23 15:56:19 -07001280 goto out_unlock;
1281 }
1282
Paul Menageddbcc7e2007-10-18 23:39:30 -07001283 ret = rebind_subsystems(root, opts.subsys_bits);
Ben Blumcf5d5942010-03-10 15:22:09 -08001284 if (ret) {
1285 drop_parsed_module_refcounts(opts.subsys_bits);
Li Zefan0670e082009-04-02 16:57:30 -07001286 goto out_unlock;
Ben Blumcf5d5942010-03-10 15:22:09 -08001287 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001288
1289 /* (re)populate subsystem files */
Li Zefan0670e082009-04-02 16:57:30 -07001290 cgroup_populate_dir(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001291
Paul Menage81a6a5c2007-10-18 23:39:38 -07001292 if (opts.release_agent)
1293 strcpy(root->release_agent_path, opts.release_agent);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001294 out_unlock:
Jesper Juhl66bdc9c2009-04-02 16:57:27 -07001295 kfree(opts.release_agent);
Paul Menagec6d57f32009-09-23 15:56:19 -07001296 kfree(opts.name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001297 mutex_unlock(&cgroup_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -07001298 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001299 return ret;
1300}
1301
Alexey Dobriyanb87221d2009-09-21 17:01:09 -07001302static const struct super_operations cgroup_ops = {
Paul Menageddbcc7e2007-10-18 23:39:30 -07001303 .statfs = simple_statfs,
1304 .drop_inode = generic_delete_inode,
1305 .show_options = cgroup_show_options,
1306 .remount_fs = cgroup_remount,
1307};
1308
Paul Menagecc31edc2008-10-18 20:28:04 -07001309static void init_cgroup_housekeeping(struct cgroup *cgrp)
1310{
1311 INIT_LIST_HEAD(&cgrp->sibling);
1312 INIT_LIST_HEAD(&cgrp->children);
1313 INIT_LIST_HEAD(&cgrp->css_sets);
1314 INIT_LIST_HEAD(&cgrp->release_list);
Ben Blum72a8cb32009-09-23 15:56:27 -07001315 INIT_LIST_HEAD(&cgrp->pidlists);
1316 mutex_init(&cgrp->pidlist_mutex);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08001317 INIT_LIST_HEAD(&cgrp->event_list);
1318 spin_lock_init(&cgrp->event_list_lock);
Paul Menagecc31edc2008-10-18 20:28:04 -07001319}
Paul Menagec6d57f32009-09-23 15:56:19 -07001320
Paul Menageddbcc7e2007-10-18 23:39:30 -07001321static void init_cgroup_root(struct cgroupfs_root *root)
1322{
Paul Menagebd89aab2007-10-18 23:40:44 -07001323 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001324 INIT_LIST_HEAD(&root->subsys_list);
1325 INIT_LIST_HEAD(&root->root_list);
1326 root->number_of_cgroups = 1;
Paul Menagebd89aab2007-10-18 23:40:44 -07001327 cgrp->root = root;
1328 cgrp->top_cgroup = cgrp;
Paul Menagecc31edc2008-10-18 20:28:04 -07001329 init_cgroup_housekeeping(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001330}
1331
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001332static bool init_root_id(struct cgroupfs_root *root)
1333{
1334 int ret = 0;
1335
1336 do {
1337 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1338 return false;
1339 spin_lock(&hierarchy_id_lock);
1340 /* Try to allocate the next unused ID */
1341 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1342 &root->hierarchy_id);
1343 if (ret == -ENOSPC)
1344 /* Try again starting from 0 */
1345 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1346 if (!ret) {
1347 next_hierarchy_id = root->hierarchy_id + 1;
1348 } else if (ret != -EAGAIN) {
1349 /* Can only get here if the 31-bit IDR is full ... */
1350 BUG_ON(ret);
1351 }
1352 spin_unlock(&hierarchy_id_lock);
1353 } while (ret);
1354 return true;
1355}
1356
Paul Menageddbcc7e2007-10-18 23:39:30 -07001357static int cgroup_test_super(struct super_block *sb, void *data)
1358{
Paul Menagec6d57f32009-09-23 15:56:19 -07001359 struct cgroup_sb_opts *opts = data;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001360 struct cgroupfs_root *root = sb->s_fs_info;
1361
Paul Menagec6d57f32009-09-23 15:56:19 -07001362 /* If we asked for a name then it must match */
1363 if (opts->name && strcmp(opts->name, root->name))
1364 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001365
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001366 /*
1367 * If we asked for subsystems (or explicitly for no
1368 * subsystems) then they must match
1369 */
1370 if ((opts->subsys_bits || opts->none)
1371 && (opts->subsys_bits != root->subsys_bits))
Paul Menageddbcc7e2007-10-18 23:39:30 -07001372 return 0;
1373
1374 return 1;
1375}
1376
Paul Menagec6d57f32009-09-23 15:56:19 -07001377static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1378{
1379 struct cgroupfs_root *root;
1380
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001381 if (!opts->subsys_bits && !opts->none)
Paul Menagec6d57f32009-09-23 15:56:19 -07001382 return NULL;
1383
1384 root = kzalloc(sizeof(*root), GFP_KERNEL);
1385 if (!root)
1386 return ERR_PTR(-ENOMEM);
1387
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001388 if (!init_root_id(root)) {
1389 kfree(root);
1390 return ERR_PTR(-ENOMEM);
1391 }
Paul Menagec6d57f32009-09-23 15:56:19 -07001392 init_cgroup_root(root);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001393
Paul Menagec6d57f32009-09-23 15:56:19 -07001394 root->subsys_bits = opts->subsys_bits;
1395 root->flags = opts->flags;
1396 if (opts->release_agent)
1397 strcpy(root->release_agent_path, opts->release_agent);
1398 if (opts->name)
1399 strcpy(root->name, opts->name);
Daniel Lezcano97978e62010-10-27 15:33:35 -07001400 if (opts->clone_children)
1401 set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
Paul Menagec6d57f32009-09-23 15:56:19 -07001402 return root;
1403}
1404
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001405static void cgroup_drop_root(struct cgroupfs_root *root)
1406{
1407 if (!root)
1408 return;
1409
1410 BUG_ON(!root->hierarchy_id);
1411 spin_lock(&hierarchy_id_lock);
1412 ida_remove(&hierarchy_ida, root->hierarchy_id);
1413 spin_unlock(&hierarchy_id_lock);
1414 kfree(root);
1415}
1416
Paul Menageddbcc7e2007-10-18 23:39:30 -07001417static int cgroup_set_super(struct super_block *sb, void *data)
1418{
1419 int ret;
Paul Menagec6d57f32009-09-23 15:56:19 -07001420 struct cgroup_sb_opts *opts = data;
1421
1422 /* If we don't have a new root, we can't set up a new sb */
1423 if (!opts->new_root)
1424 return -EINVAL;
1425
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001426 BUG_ON(!opts->subsys_bits && !opts->none);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001427
1428 ret = set_anon_super(sb, NULL);
1429 if (ret)
1430 return ret;
1431
Paul Menagec6d57f32009-09-23 15:56:19 -07001432 sb->s_fs_info = opts->new_root;
1433 opts->new_root->sb = sb;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001434
1435 sb->s_blocksize = PAGE_CACHE_SIZE;
1436 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1437 sb->s_magic = CGROUP_SUPER_MAGIC;
1438 sb->s_op = &cgroup_ops;
1439
1440 return 0;
1441}
1442
1443static int cgroup_get_rootdir(struct super_block *sb)
1444{
1445 struct inode *inode =
1446 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1447 struct dentry *dentry;
1448
1449 if (!inode)
1450 return -ENOMEM;
1451
Paul Menageddbcc7e2007-10-18 23:39:30 -07001452 inode->i_fop = &simple_dir_operations;
1453 inode->i_op = &cgroup_dir_inode_operations;
1454 /* directories start off with i_nlink == 2 (for "." entry) */
1455 inc_nlink(inode);
1456 dentry = d_alloc_root(inode);
1457 if (!dentry) {
1458 iput(inode);
1459 return -ENOMEM;
1460 }
1461 sb->s_root = dentry;
1462 return 0;
1463}
1464
Al Virof7e83572010-07-26 13:23:11 +04001465static struct dentry *cgroup_mount(struct file_system_type *fs_type,
Paul Menageddbcc7e2007-10-18 23:39:30 -07001466 int flags, const char *unused_dev_name,
Al Virof7e83572010-07-26 13:23:11 +04001467 void *data)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001468{
1469 struct cgroup_sb_opts opts;
Paul Menagec6d57f32009-09-23 15:56:19 -07001470 struct cgroupfs_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001471 int ret = 0;
1472 struct super_block *sb;
Paul Menagec6d57f32009-09-23 15:56:19 -07001473 struct cgroupfs_root *new_root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001474
1475 /* First find the desired set of subsystems */
Ben Blumaae8aab2010-03-10 15:22:07 -08001476 mutex_lock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001477 ret = parse_cgroupfs_options(data, &opts);
Ben Blumaae8aab2010-03-10 15:22:07 -08001478 mutex_unlock(&cgroup_mutex);
Paul Menagec6d57f32009-09-23 15:56:19 -07001479 if (ret)
1480 goto out_err;
1481
1482 /*
1483 * Allocate a new cgroup root. We may not need it if we're
1484 * reusing an existing hierarchy.
1485 */
1486 new_root = cgroup_root_from_opts(&opts);
1487 if (IS_ERR(new_root)) {
1488 ret = PTR_ERR(new_root);
Ben Blumcf5d5942010-03-10 15:22:09 -08001489 goto drop_modules;
Paul Menage81a6a5c2007-10-18 23:39:38 -07001490 }
Paul Menagec6d57f32009-09-23 15:56:19 -07001491 opts.new_root = new_root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001492
Paul Menagec6d57f32009-09-23 15:56:19 -07001493 /* Locate an existing or new sb for this hierarchy */
1494 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001495 if (IS_ERR(sb)) {
Paul Menagec6d57f32009-09-23 15:56:19 -07001496 ret = PTR_ERR(sb);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001497 cgroup_drop_root(opts.new_root);
Ben Blumcf5d5942010-03-10 15:22:09 -08001498 goto drop_modules;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001499 }
1500
Paul Menagec6d57f32009-09-23 15:56:19 -07001501 root = sb->s_fs_info;
1502 BUG_ON(!root);
1503 if (root == opts.new_root) {
1504 /* We used the new root structure, so this is a new hierarchy */
1505 struct list_head tmp_cg_links;
Li Zefanc12f65d2009-01-07 18:07:42 -08001506 struct cgroup *root_cgrp = &root->top_cgroup;
Paul Menage817929e2007-10-18 23:39:36 -07001507 struct inode *inode;
Paul Menagec6d57f32009-09-23 15:56:19 -07001508 struct cgroupfs_root *existing_root;
Li Zefan28fd5df2008-04-29 01:00:13 -07001509 int i;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001510
1511 BUG_ON(sb->s_root != NULL);
1512
1513 ret = cgroup_get_rootdir(sb);
1514 if (ret)
1515 goto drop_new_super;
Paul Menage817929e2007-10-18 23:39:36 -07001516 inode = sb->s_root->d_inode;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001517
Paul Menage817929e2007-10-18 23:39:36 -07001518 mutex_lock(&inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001519 mutex_lock(&cgroup_mutex);
1520
Paul Menagec6d57f32009-09-23 15:56:19 -07001521 if (strlen(root->name)) {
1522 /* Check for name clashes with existing mounts */
1523 for_each_active_root(existing_root) {
1524 if (!strcmp(existing_root->name, root->name)) {
1525 ret = -EBUSY;
1526 mutex_unlock(&cgroup_mutex);
1527 mutex_unlock(&inode->i_mutex);
1528 goto drop_new_super;
1529 }
1530 }
1531 }
1532
Paul Menage817929e2007-10-18 23:39:36 -07001533 /*
1534 * We're accessing css_set_count without locking
1535 * css_set_lock here, but that's OK - it can only be
1536 * increased by someone holding cgroup_lock, and
1537 * that's us. The worst that can happen is that we
1538 * have some link structures left over
1539 */
1540 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1541 if (ret) {
1542 mutex_unlock(&cgroup_mutex);
1543 mutex_unlock(&inode->i_mutex);
1544 goto drop_new_super;
1545 }
1546
Paul Menageddbcc7e2007-10-18 23:39:30 -07001547 ret = rebind_subsystems(root, root->subsys_bits);
1548 if (ret == -EBUSY) {
1549 mutex_unlock(&cgroup_mutex);
Paul Menage817929e2007-10-18 23:39:36 -07001550 mutex_unlock(&inode->i_mutex);
Paul Menagec6d57f32009-09-23 15:56:19 -07001551 free_cg_links(&tmp_cg_links);
1552 goto drop_new_super;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001553 }
Ben Blumcf5d5942010-03-10 15:22:09 -08001554 /*
1555 * There must be no failure case after here, since rebinding
1556 * takes care of subsystems' refcounts, which are explicitly
1557 * dropped in the failure exit path.
1558 */
Paul Menageddbcc7e2007-10-18 23:39:30 -07001559
1560 /* EBUSY should be the only error here */
1561 BUG_ON(ret);
1562
1563 list_add(&root->root_list, &roots);
Paul Menage817929e2007-10-18 23:39:36 -07001564 root_count++;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001565
Li Zefanc12f65d2009-01-07 18:07:42 -08001566 sb->s_root->d_fsdata = root_cgrp;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001567 root->top_cgroup.dentry = sb->s_root;
1568
Paul Menage817929e2007-10-18 23:39:36 -07001569 /* Link the top cgroup in this hierarchy into all
1570 * the css_set objects */
1571 write_lock(&css_set_lock);
Li Zefan28fd5df2008-04-29 01:00:13 -07001572 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1573 struct hlist_head *hhead = &css_set_table[i];
1574 struct hlist_node *node;
Paul Menage817929e2007-10-18 23:39:36 -07001575 struct css_set *cg;
Li Zefan28fd5df2008-04-29 01:00:13 -07001576
Li Zefanc12f65d2009-01-07 18:07:42 -08001577 hlist_for_each_entry(cg, node, hhead, hlist)
1578 link_css_set(&tmp_cg_links, cg, root_cgrp);
Li Zefan28fd5df2008-04-29 01:00:13 -07001579 }
Paul Menage817929e2007-10-18 23:39:36 -07001580 write_unlock(&css_set_lock);
1581
1582 free_cg_links(&tmp_cg_links);
1583
Li Zefanc12f65d2009-01-07 18:07:42 -08001584 BUG_ON(!list_empty(&root_cgrp->sibling));
1585 BUG_ON(!list_empty(&root_cgrp->children));
Paul Menageddbcc7e2007-10-18 23:39:30 -07001586 BUG_ON(root->number_of_cgroups != 1);
1587
Li Zefanc12f65d2009-01-07 18:07:42 -08001588 cgroup_populate_dir(root_cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001589 mutex_unlock(&cgroup_mutex);
Xiaotian Feng34f77a92009-09-23 15:56:18 -07001590 mutex_unlock(&inode->i_mutex);
Paul Menagec6d57f32009-09-23 15:56:19 -07001591 } else {
1592 /*
1593 * We re-used an existing hierarchy - the new root (if
1594 * any) is not needed
1595 */
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001596 cgroup_drop_root(opts.new_root);
Ben Blumcf5d5942010-03-10 15:22:09 -08001597 /* no subsys rebinding, so refcounts don't change */
1598 drop_parsed_module_refcounts(opts.subsys_bits);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001599 }
1600
Paul Menagec6d57f32009-09-23 15:56:19 -07001601 kfree(opts.release_agent);
1602 kfree(opts.name);
Al Virof7e83572010-07-26 13:23:11 +04001603 return dget(sb->s_root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001604
1605 drop_new_super:
Al Viro6f5bbff2009-05-06 01:34:22 -04001606 deactivate_locked_super(sb);
Ben Blumcf5d5942010-03-10 15:22:09 -08001607 drop_modules:
1608 drop_parsed_module_refcounts(opts.subsys_bits);
Paul Menagec6d57f32009-09-23 15:56:19 -07001609 out_err:
1610 kfree(opts.release_agent);
1611 kfree(opts.name);
Al Virof7e83572010-07-26 13:23:11 +04001612 return ERR_PTR(ret);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001613}
1614
1615static void cgroup_kill_sb(struct super_block *sb) {
1616 struct cgroupfs_root *root = sb->s_fs_info;
Paul Menagebd89aab2007-10-18 23:40:44 -07001617 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001618 int ret;
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -07001619 struct cg_cgroup_link *link;
1620 struct cg_cgroup_link *saved_link;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001621
1622 BUG_ON(!root);
1623
1624 BUG_ON(root->number_of_cgroups != 1);
Paul Menagebd89aab2007-10-18 23:40:44 -07001625 BUG_ON(!list_empty(&cgrp->children));
1626 BUG_ON(!list_empty(&cgrp->sibling));
Paul Menageddbcc7e2007-10-18 23:39:30 -07001627
1628 mutex_lock(&cgroup_mutex);
1629
1630 /* Rebind all subsystems back to the default hierarchy */
1631 ret = rebind_subsystems(root, 0);
1632 /* Shouldn't be able to fail ... */
1633 BUG_ON(ret);
1634
Paul Menage817929e2007-10-18 23:39:36 -07001635 /*
1636 * Release all the links from css_sets to this hierarchy's
1637 * root cgroup
1638 */
1639 write_lock(&css_set_lock);
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -07001640
1641 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1642 cgrp_link_list) {
Paul Menage817929e2007-10-18 23:39:36 -07001643 list_del(&link->cg_link_list);
Paul Menagebd89aab2007-10-18 23:40:44 -07001644 list_del(&link->cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -07001645 kfree(link);
1646 }
1647 write_unlock(&css_set_lock);
1648
Paul Menage839ec542009-01-29 14:25:22 -08001649 if (!list_empty(&root->root_list)) {
1650 list_del(&root->root_list);
1651 root_count--;
1652 }
Li Zefane5f6a862009-01-07 18:07:41 -08001653
Paul Menageddbcc7e2007-10-18 23:39:30 -07001654 mutex_unlock(&cgroup_mutex);
1655
Paul Menageddbcc7e2007-10-18 23:39:30 -07001656 kill_litter_super(sb);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001657 cgroup_drop_root(root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001658}
1659
1660static struct file_system_type cgroup_fs_type = {
1661 .name = "cgroup",
Al Virof7e83572010-07-26 13:23:11 +04001662 .mount = cgroup_mount,
Paul Menageddbcc7e2007-10-18 23:39:30 -07001663 .kill_sb = cgroup_kill_sb,
1664};
1665
Greg KH676db4a2010-08-05 13:53:35 -07001666static struct kobject *cgroup_kobj;
1667
Paul Menagebd89aab2007-10-18 23:40:44 -07001668static inline struct cgroup *__d_cgrp(struct dentry *dentry)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001669{
1670 return dentry->d_fsdata;
1671}
1672
1673static inline struct cftype *__d_cft(struct dentry *dentry)
1674{
1675 return dentry->d_fsdata;
1676}
1677
Li Zefana043e3b2008-02-23 15:24:09 -08001678/**
1679 * cgroup_path - generate the path of a cgroup
1680 * @cgrp: the cgroup in question
1681 * @buf: the buffer to write the path into
1682 * @buflen: the length of the buffer
1683 *
Paul Menagea47295e2009-01-07 18:07:44 -08001684 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1685 * reference. Writes path of cgroup into buf. Returns 0 on success,
1686 * -errno on error.
Paul Menageddbcc7e2007-10-18 23:39:30 -07001687 */
Paul Menagebd89aab2007-10-18 23:40:44 -07001688int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001689{
1690 char *start;
Li Zefan9a9686b2010-04-22 17:29:24 +08001691 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1692 rcu_read_lock_held() ||
1693 cgroup_lock_is_held());
Paul Menageddbcc7e2007-10-18 23:39:30 -07001694
Paul Menagea47295e2009-01-07 18:07:44 -08001695 if (!dentry || cgrp == dummytop) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07001696 /*
1697 * Inactive subsystems have no dentry for their root
1698 * cgroup
1699 */
1700 strcpy(buf, "/");
1701 return 0;
1702 }
1703
1704 start = buf + buflen;
1705
1706 *--start = '\0';
1707 for (;;) {
Paul Menagea47295e2009-01-07 18:07:44 -08001708 int len = dentry->d_name.len;
Li Zefan9a9686b2010-04-22 17:29:24 +08001709
Paul Menageddbcc7e2007-10-18 23:39:30 -07001710 if ((start -= len) < buf)
1711 return -ENAMETOOLONG;
Li Zefan9a9686b2010-04-22 17:29:24 +08001712 memcpy(start, dentry->d_name.name, len);
Paul Menagebd89aab2007-10-18 23:40:44 -07001713 cgrp = cgrp->parent;
1714 if (!cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001715 break;
Li Zefan9a9686b2010-04-22 17:29:24 +08001716
1717 dentry = rcu_dereference_check(cgrp->dentry,
1718 rcu_read_lock_held() ||
1719 cgroup_lock_is_held());
Paul Menagebd89aab2007-10-18 23:40:44 -07001720 if (!cgrp->parent)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001721 continue;
1722 if (--start < buf)
1723 return -ENAMETOOLONG;
1724 *start = '/';
1725 }
1726 memmove(buf, start, buf + buflen - start);
1727 return 0;
1728}
Ben Blum67523c42010-03-10 15:22:11 -08001729EXPORT_SYMBOL_GPL(cgroup_path);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001730
Li Zefana043e3b2008-02-23 15:24:09 -08001731/**
1732 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1733 * @cgrp: the cgroup the task is attaching to
1734 * @tsk: the task to be attached
Paul Menagebbcb81d2007-10-18 23:39:32 -07001735 *
Li Zefana043e3b2008-02-23 15:24:09 -08001736 * Call holding cgroup_mutex. May take task_lock of
1737 * the task 'tsk' during call.
Paul Menagebbcb81d2007-10-18 23:39:32 -07001738 */
Cliff Wickman956db3c2008-02-07 00:14:43 -08001739int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
Paul Menagebbcb81d2007-10-18 23:39:32 -07001740{
1741 int retval = 0;
Daisuke Nishimura2468c722010-03-10 15:22:03 -08001742 struct cgroup_subsys *ss, *failed_ss = NULL;
Paul Menagebd89aab2007-10-18 23:40:44 -07001743 struct cgroup *oldcgrp;
Lai Jiangshan77efecd2009-01-07 18:07:39 -08001744 struct css_set *cg;
Paul Menage817929e2007-10-18 23:39:36 -07001745 struct css_set *newcg;
Paul Menagebd89aab2007-10-18 23:40:44 -07001746 struct cgroupfs_root *root = cgrp->root;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001747
1748 /* Nothing to do if the task is already in that cgroup */
Paul Menage7717f7b2009-09-23 15:56:22 -07001749 oldcgrp = task_cgroup_from_root(tsk, root);
Paul Menagebd89aab2007-10-18 23:40:44 -07001750 if (cgrp == oldcgrp)
Paul Menagebbcb81d2007-10-18 23:39:32 -07001751 return 0;
1752
1753 for_each_subsys(root, ss) {
1754 if (ss->can_attach) {
Ben Blumbe367d02009-09-23 15:56:31 -07001755 retval = ss->can_attach(ss, cgrp, tsk, false);
Daisuke Nishimura2468c722010-03-10 15:22:03 -08001756 if (retval) {
1757 /*
1758 * Remember on which subsystem the can_attach()
1759 * failed, so that we only call cancel_attach()
1760 * against the subsystems whose can_attach()
1761 * succeeded. (See below)
1762 */
1763 failed_ss = ss;
1764 goto out;
1765 }
Paul Menagebbcb81d2007-10-18 23:39:32 -07001766 }
1767 }
1768
Lai Jiangshan77efecd2009-01-07 18:07:39 -08001769 task_lock(tsk);
1770 cg = tsk->cgroups;
1771 get_css_set(cg);
1772 task_unlock(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07001773 /*
1774 * Locate or allocate a new css_set for this task,
1775 * based on its final set of cgroups
1776 */
Paul Menagebd89aab2007-10-18 23:40:44 -07001777 newcg = find_css_set(cg, cgrp);
Lai Jiangshan77efecd2009-01-07 18:07:39 -08001778 put_css_set(cg);
Daisuke Nishimura2468c722010-03-10 15:22:03 -08001779 if (!newcg) {
1780 retval = -ENOMEM;
1781 goto out;
1782 }
Paul Menage817929e2007-10-18 23:39:36 -07001783
Paul Menagebbcb81d2007-10-18 23:39:32 -07001784 task_lock(tsk);
1785 if (tsk->flags & PF_EXITING) {
1786 task_unlock(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07001787 put_css_set(newcg);
Daisuke Nishimura2468c722010-03-10 15:22:03 -08001788 retval = -ESRCH;
1789 goto out;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001790 }
Paul Menage817929e2007-10-18 23:39:36 -07001791 rcu_assign_pointer(tsk->cgroups, newcg);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001792 task_unlock(tsk);
1793
Paul Menage817929e2007-10-18 23:39:36 -07001794 /* Update the css_set linked lists if we're using them */
1795 write_lock(&css_set_lock);
1796 if (!list_empty(&tsk->cg_list)) {
1797 list_del(&tsk->cg_list);
1798 list_add(&tsk->cg_list, &newcg->tasks);
1799 }
1800 write_unlock(&css_set_lock);
1801
Paul Menagebbcb81d2007-10-18 23:39:32 -07001802 for_each_subsys(root, ss) {
Paul Jacksone18f6312008-02-07 00:13:44 -08001803 if (ss->attach)
Ben Blumbe367d02009-09-23 15:56:31 -07001804 ss->attach(ss, cgrp, oldcgrp, tsk, false);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001805 }
Paul Menagebd89aab2007-10-18 23:40:44 -07001806 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001807 synchronize_rcu();
Paul Menage817929e2007-10-18 23:39:36 -07001808 put_css_set(cg);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07001809
1810 /*
1811 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1812 * is no longer empty.
1813 */
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07001814 cgroup_wakeup_rmdir_waiter(cgrp);
Daisuke Nishimura2468c722010-03-10 15:22:03 -08001815out:
1816 if (retval) {
1817 for_each_subsys(root, ss) {
1818 if (ss == failed_ss)
1819 /*
1820 * This subsystem was the one that failed the
1821 * can_attach() check earlier, so we don't need
1822 * to call cancel_attach() against it or any
1823 * remaining subsystems.
1824 */
1825 break;
1826 if (ss->cancel_attach)
1827 ss->cancel_attach(ss, cgrp, tsk, false);
1828 }
1829 }
1830 return retval;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001831}
1832
Sridhar Samudralad7926ee2010-05-30 22:24:39 +02001833/**
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -07001834 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1835 * @from: attach to all cgroups of a given task
Sridhar Samudralad7926ee2010-05-30 22:24:39 +02001836 * @tsk: the task to be attached
1837 */
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -07001838int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
Sridhar Samudralad7926ee2010-05-30 22:24:39 +02001839{
1840 struct cgroupfs_root *root;
Sridhar Samudralad7926ee2010-05-30 22:24:39 +02001841 int retval = 0;
1842
1843 cgroup_lock();
1844 for_each_active_root(root) {
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -07001845 struct cgroup *from_cg = task_cgroup_from_root(from, root);
1846
1847 retval = cgroup_attach_task(from_cg, tsk);
Sridhar Samudralad7926ee2010-05-30 22:24:39 +02001848 if (retval)
1849 break;
1850 }
1851 cgroup_unlock();
1852
1853 return retval;
1854}
Michael S. Tsirkin31583bb2010-09-09 16:37:37 -07001855EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
Sridhar Samudralad7926ee2010-05-30 22:24:39 +02001856
Paul Menagebbcb81d2007-10-18 23:39:32 -07001857/*
Paul Menageaf351022008-07-25 01:47:01 -07001858 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1859 * held. May take task_lock of task
Paul Menagebbcb81d2007-10-18 23:39:32 -07001860 */
Paul Menageaf351022008-07-25 01:47:01 -07001861static int attach_task_by_pid(struct cgroup *cgrp, u64 pid)
Paul Menagebbcb81d2007-10-18 23:39:32 -07001862{
Paul Menagebbcb81d2007-10-18 23:39:32 -07001863 struct task_struct *tsk;
David Howellsc69e8d92008-11-14 10:39:19 +11001864 const struct cred *cred = current_cred(), *tcred;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001865 int ret;
1866
Paul Menagebbcb81d2007-10-18 23:39:32 -07001867 if (pid) {
1868 rcu_read_lock();
Pavel Emelyanov73507f32008-02-07 00:14:47 -08001869 tsk = find_task_by_vpid(pid);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001870 if (!tsk || tsk->flags & PF_EXITING) {
1871 rcu_read_unlock();
1872 return -ESRCH;
1873 }
Paul Menagebbcb81d2007-10-18 23:39:32 -07001874
David Howellsc69e8d92008-11-14 10:39:19 +11001875 tcred = __task_cred(tsk);
1876 if (cred->euid &&
1877 cred->euid != tcred->uid &&
1878 cred->euid != tcred->suid) {
1879 rcu_read_unlock();
Paul Menagebbcb81d2007-10-18 23:39:32 -07001880 return -EACCES;
1881 }
David Howellsc69e8d92008-11-14 10:39:19 +11001882 get_task_struct(tsk);
1883 rcu_read_unlock();
Paul Menagebbcb81d2007-10-18 23:39:32 -07001884 } else {
1885 tsk = current;
1886 get_task_struct(tsk);
1887 }
1888
Cliff Wickman956db3c2008-02-07 00:14:43 -08001889 ret = cgroup_attach_task(cgrp, tsk);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001890 put_task_struct(tsk);
1891 return ret;
1892}
1893
Paul Menageaf351022008-07-25 01:47:01 -07001894static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
1895{
1896 int ret;
1897 if (!cgroup_lock_live_group(cgrp))
1898 return -ENODEV;
1899 ret = attach_task_by_pid(cgrp, pid);
1900 cgroup_unlock();
1901 return ret;
1902}
1903
Paul Menagee788e062008-07-25 01:46:59 -07001904/**
1905 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1906 * @cgrp: the cgroup to be checked for liveness
1907 *
Paul Menage84eea842008-07-25 01:47:00 -07001908 * On success, returns true; the lock should be later released with
1909 * cgroup_unlock(). On failure returns false with no lock held.
Paul Menagee788e062008-07-25 01:46:59 -07001910 */
Paul Menage84eea842008-07-25 01:47:00 -07001911bool cgroup_lock_live_group(struct cgroup *cgrp)
Paul Menagee788e062008-07-25 01:46:59 -07001912{
1913 mutex_lock(&cgroup_mutex);
1914 if (cgroup_is_removed(cgrp)) {
1915 mutex_unlock(&cgroup_mutex);
1916 return false;
1917 }
1918 return true;
1919}
Ben Blum67523c42010-03-10 15:22:11 -08001920EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
Paul Menagee788e062008-07-25 01:46:59 -07001921
1922static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
1923 const char *buffer)
1924{
1925 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
Evgeny Kuznetsovf4a25892010-10-27 15:33:37 -07001926 if (strlen(buffer) >= PATH_MAX)
1927 return -EINVAL;
Paul Menagee788e062008-07-25 01:46:59 -07001928 if (!cgroup_lock_live_group(cgrp))
1929 return -ENODEV;
1930 strcpy(cgrp->root->release_agent_path, buffer);
Paul Menage84eea842008-07-25 01:47:00 -07001931 cgroup_unlock();
Paul Menagee788e062008-07-25 01:46:59 -07001932 return 0;
1933}
1934
1935static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
1936 struct seq_file *seq)
1937{
1938 if (!cgroup_lock_live_group(cgrp))
1939 return -ENODEV;
1940 seq_puts(seq, cgrp->root->release_agent_path);
1941 seq_putc(seq, '\n');
Paul Menage84eea842008-07-25 01:47:00 -07001942 cgroup_unlock();
Paul Menagee788e062008-07-25 01:46:59 -07001943 return 0;
1944}
1945
Paul Menage84eea842008-07-25 01:47:00 -07001946/* A buffer size big enough for numbers or short strings */
1947#define CGROUP_LOCAL_BUFFER_SIZE 64
1948
Paul Menagee73d2c62008-04-29 01:00:06 -07001949static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
Paul Menagef4c753b2008-04-29 00:59:56 -07001950 struct file *file,
1951 const char __user *userbuf,
1952 size_t nbytes, loff_t *unused_ppos)
Paul Menage355e0c42007-10-18 23:39:33 -07001953{
Paul Menage84eea842008-07-25 01:47:00 -07001954 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
Paul Menage355e0c42007-10-18 23:39:33 -07001955 int retval = 0;
Paul Menage355e0c42007-10-18 23:39:33 -07001956 char *end;
1957
1958 if (!nbytes)
1959 return -EINVAL;
1960 if (nbytes >= sizeof(buffer))
1961 return -E2BIG;
1962 if (copy_from_user(buffer, userbuf, nbytes))
1963 return -EFAULT;
1964
1965 buffer[nbytes] = 0; /* nul-terminate */
Paul Menagee73d2c62008-04-29 01:00:06 -07001966 if (cft->write_u64) {
KOSAKI Motohiro478988d2009-10-26 16:49:36 -07001967 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
Paul Menagee73d2c62008-04-29 01:00:06 -07001968 if (*end)
1969 return -EINVAL;
1970 retval = cft->write_u64(cgrp, cft, val);
1971 } else {
KOSAKI Motohiro478988d2009-10-26 16:49:36 -07001972 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
Paul Menagee73d2c62008-04-29 01:00:06 -07001973 if (*end)
1974 return -EINVAL;
1975 retval = cft->write_s64(cgrp, cft, val);
1976 }
Paul Menage355e0c42007-10-18 23:39:33 -07001977 if (!retval)
1978 retval = nbytes;
1979 return retval;
1980}
1981
Paul Menagedb3b1492008-07-25 01:46:58 -07001982static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
1983 struct file *file,
1984 const char __user *userbuf,
1985 size_t nbytes, loff_t *unused_ppos)
1986{
Paul Menage84eea842008-07-25 01:47:00 -07001987 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
Paul Menagedb3b1492008-07-25 01:46:58 -07001988 int retval = 0;
1989 size_t max_bytes = cft->max_write_len;
1990 char *buffer = local_buffer;
1991
1992 if (!max_bytes)
1993 max_bytes = sizeof(local_buffer) - 1;
1994 if (nbytes >= max_bytes)
1995 return -E2BIG;
1996 /* Allocate a dynamic buffer if we need one */
1997 if (nbytes >= sizeof(local_buffer)) {
1998 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1999 if (buffer == NULL)
2000 return -ENOMEM;
2001 }
Li Zefan5a3eb9f2008-07-29 22:33:18 -07002002 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2003 retval = -EFAULT;
2004 goto out;
2005 }
Paul Menagedb3b1492008-07-25 01:46:58 -07002006
2007 buffer[nbytes] = 0; /* nul-terminate */
KOSAKI Motohiro478988d2009-10-26 16:49:36 -07002008 retval = cft->write_string(cgrp, cft, strstrip(buffer));
Paul Menagedb3b1492008-07-25 01:46:58 -07002009 if (!retval)
2010 retval = nbytes;
Li Zefan5a3eb9f2008-07-29 22:33:18 -07002011out:
Paul Menagedb3b1492008-07-25 01:46:58 -07002012 if (buffer != local_buffer)
2013 kfree(buffer);
2014 return retval;
2015}
2016
Paul Menageddbcc7e2007-10-18 23:39:30 -07002017static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2018 size_t nbytes, loff_t *ppos)
2019{
2020 struct cftype *cft = __d_cft(file->f_dentry);
Paul Menagebd89aab2007-10-18 23:40:44 -07002021 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002022
Li Zefan75139b82009-01-07 18:07:33 -08002023 if (cgroup_is_removed(cgrp))
Paul Menageddbcc7e2007-10-18 23:39:30 -07002024 return -ENODEV;
Paul Menage355e0c42007-10-18 23:39:33 -07002025 if (cft->write)
Paul Menagebd89aab2007-10-18 23:40:44 -07002026 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
Paul Menagee73d2c62008-04-29 01:00:06 -07002027 if (cft->write_u64 || cft->write_s64)
2028 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
Paul Menagedb3b1492008-07-25 01:46:58 -07002029 if (cft->write_string)
2030 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
Pavel Emelyanovd447ea22008-04-29 01:00:08 -07002031 if (cft->trigger) {
2032 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2033 return ret ? ret : nbytes;
2034 }
Paul Menage355e0c42007-10-18 23:39:33 -07002035 return -EINVAL;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002036}
2037
Paul Menagef4c753b2008-04-29 00:59:56 -07002038static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2039 struct file *file,
2040 char __user *buf, size_t nbytes,
2041 loff_t *ppos)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002042{
Paul Menage84eea842008-07-25 01:47:00 -07002043 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
Paul Menagef4c753b2008-04-29 00:59:56 -07002044 u64 val = cft->read_u64(cgrp, cft);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002045 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2046
2047 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2048}
2049
Paul Menagee73d2c62008-04-29 01:00:06 -07002050static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2051 struct file *file,
2052 char __user *buf, size_t nbytes,
2053 loff_t *ppos)
2054{
Paul Menage84eea842008-07-25 01:47:00 -07002055 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
Paul Menagee73d2c62008-04-29 01:00:06 -07002056 s64 val = cft->read_s64(cgrp, cft);
2057 int len = sprintf(tmp, "%lld\n", (long long) val);
2058
2059 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2060}
2061
Paul Menageddbcc7e2007-10-18 23:39:30 -07002062static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2063 size_t nbytes, loff_t *ppos)
2064{
2065 struct cftype *cft = __d_cft(file->f_dentry);
Paul Menagebd89aab2007-10-18 23:40:44 -07002066 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002067
Li Zefan75139b82009-01-07 18:07:33 -08002068 if (cgroup_is_removed(cgrp))
Paul Menageddbcc7e2007-10-18 23:39:30 -07002069 return -ENODEV;
2070
2071 if (cft->read)
Paul Menagebd89aab2007-10-18 23:40:44 -07002072 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
Paul Menagef4c753b2008-04-29 00:59:56 -07002073 if (cft->read_u64)
2074 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
Paul Menagee73d2c62008-04-29 01:00:06 -07002075 if (cft->read_s64)
2076 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002077 return -EINVAL;
2078}
2079
Paul Menage91796562008-04-29 01:00:01 -07002080/*
2081 * seqfile ops/methods for returning structured data. Currently just
2082 * supports string->u64 maps, but can be extended in future.
2083 */
2084
2085struct cgroup_seqfile_state {
2086 struct cftype *cft;
2087 struct cgroup *cgroup;
2088};
2089
2090static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2091{
2092 struct seq_file *sf = cb->state;
2093 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2094}
2095
2096static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2097{
2098 struct cgroup_seqfile_state *state = m->private;
2099 struct cftype *cft = state->cft;
Serge E. Hallyn29486df2008-04-29 01:00:14 -07002100 if (cft->read_map) {
2101 struct cgroup_map_cb cb = {
2102 .fill = cgroup_map_add,
2103 .state = m,
2104 };
2105 return cft->read_map(state->cgroup, cft, &cb);
2106 }
2107 return cft->read_seq_string(state->cgroup, cft, m);
Paul Menage91796562008-04-29 01:00:01 -07002108}
2109
Adrian Bunk96930a62008-07-25 19:46:21 -07002110static int cgroup_seqfile_release(struct inode *inode, struct file *file)
Paul Menage91796562008-04-29 01:00:01 -07002111{
2112 struct seq_file *seq = file->private_data;
2113 kfree(seq->private);
2114 return single_release(inode, file);
2115}
2116
Alexey Dobriyan828c0952009-10-01 15:43:56 -07002117static const struct file_operations cgroup_seqfile_operations = {
Paul Menage91796562008-04-29 01:00:01 -07002118 .read = seq_read,
Paul Menagee788e062008-07-25 01:46:59 -07002119 .write = cgroup_file_write,
Paul Menage91796562008-04-29 01:00:01 -07002120 .llseek = seq_lseek,
2121 .release = cgroup_seqfile_release,
2122};
2123
Paul Menageddbcc7e2007-10-18 23:39:30 -07002124static int cgroup_file_open(struct inode *inode, struct file *file)
2125{
2126 int err;
2127 struct cftype *cft;
2128
2129 err = generic_file_open(inode, file);
2130 if (err)
2131 return err;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002132 cft = __d_cft(file->f_dentry);
Li Zefan75139b82009-01-07 18:07:33 -08002133
Serge E. Hallyn29486df2008-04-29 01:00:14 -07002134 if (cft->read_map || cft->read_seq_string) {
Paul Menage91796562008-04-29 01:00:01 -07002135 struct cgroup_seqfile_state *state =
2136 kzalloc(sizeof(*state), GFP_USER);
2137 if (!state)
2138 return -ENOMEM;
2139 state->cft = cft;
2140 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2141 file->f_op = &cgroup_seqfile_operations;
2142 err = single_open(file, cgroup_seqfile_show, state);
2143 if (err < 0)
2144 kfree(state);
2145 } else if (cft->open)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002146 err = cft->open(inode, file);
2147 else
2148 err = 0;
2149
2150 return err;
2151}
2152
2153static int cgroup_file_release(struct inode *inode, struct file *file)
2154{
2155 struct cftype *cft = __d_cft(file->f_dentry);
2156 if (cft->release)
2157 return cft->release(inode, file);
2158 return 0;
2159}
2160
2161/*
2162 * cgroup_rename - Only allow simple rename of directories in place.
2163 */
2164static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2165 struct inode *new_dir, struct dentry *new_dentry)
2166{
2167 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2168 return -ENOTDIR;
2169 if (new_dentry->d_inode)
2170 return -EEXIST;
2171 if (old_dir != new_dir)
2172 return -EIO;
2173 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2174}
2175
Alexey Dobriyan828c0952009-10-01 15:43:56 -07002176static const struct file_operations cgroup_file_operations = {
Paul Menageddbcc7e2007-10-18 23:39:30 -07002177 .read = cgroup_file_read,
2178 .write = cgroup_file_write,
2179 .llseek = generic_file_llseek,
2180 .open = cgroup_file_open,
2181 .release = cgroup_file_release,
2182};
2183
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07002184static const struct inode_operations cgroup_dir_inode_operations = {
Nick Piggin5adcee12011-01-07 17:49:20 +11002185 .lookup = cgroup_lookup,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002186 .mkdir = cgroup_mkdir,
2187 .rmdir = cgroup_rmdir,
2188 .rename = cgroup_rename,
2189};
2190
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08002191/*
2192 * Check if a file is a control file
2193 */
2194static inline struct cftype *__file_cft(struct file *file)
2195{
2196 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2197 return ERR_PTR(-EINVAL);
2198 return __d_cft(file->f_dentry);
2199}
2200
Nick Piggin5adcee12011-01-07 17:49:20 +11002201static int cgroup_delete_dentry(struct dentry *dentry)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002202{
Nick Piggin5adcee12011-01-07 17:49:20 +11002203 return 1;
2204}
2205
2206static struct dentry *cgroup_lookup(struct inode *dir,
2207 struct dentry *dentry, struct nameidata *nd)
2208{
2209 static const struct dentry_operations cgroup_dentry_operations = {
2210 .d_delete = cgroup_delete_dentry,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002211 .d_iput = cgroup_diput,
2212 };
2213
Nick Piggin5adcee12011-01-07 17:49:20 +11002214 if (dentry->d_name.len > NAME_MAX)
2215 return ERR_PTR(-ENAMETOOLONG);
2216 dentry->d_op = &cgroup_dentry_operations;
2217 d_add(dentry, NULL);
2218 return NULL;
2219}
2220
2221static int cgroup_create_file(struct dentry *dentry, mode_t mode,
2222 struct super_block *sb)
2223{
Paul Menageddbcc7e2007-10-18 23:39:30 -07002224 struct inode *inode;
2225
2226 if (!dentry)
2227 return -ENOENT;
2228 if (dentry->d_inode)
2229 return -EEXIST;
2230
2231 inode = cgroup_new_inode(mode, sb);
2232 if (!inode)
2233 return -ENOMEM;
2234
2235 if (S_ISDIR(mode)) {
2236 inode->i_op = &cgroup_dir_inode_operations;
2237 inode->i_fop = &simple_dir_operations;
2238
2239 /* start off with i_nlink == 2 (for "." entry) */
2240 inc_nlink(inode);
2241
2242 /* start with the directory inode held, so that we can
2243 * populate it without racing with another mkdir */
Paul Menage817929e2007-10-18 23:39:36 -07002244 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002245 } else if (S_ISREG(mode)) {
2246 inode->i_size = 0;
2247 inode->i_fop = &cgroup_file_operations;
2248 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07002249 d_instantiate(dentry, inode);
2250 dget(dentry); /* Extra count - pin the dentry in core */
2251 return 0;
2252}
2253
2254/*
Li Zefana043e3b2008-02-23 15:24:09 -08002255 * cgroup_create_dir - create a directory for an object.
2256 * @cgrp: the cgroup we create the directory for. It must have a valid
2257 * ->parent field. And we are going to fill its ->dentry field.
2258 * @dentry: dentry of the new cgroup
2259 * @mode: mode to set on new directory.
Paul Menageddbcc7e2007-10-18 23:39:30 -07002260 */
Paul Menagebd89aab2007-10-18 23:40:44 -07002261static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
Li Zefan099fca32009-04-02 16:57:29 -07002262 mode_t mode)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002263{
2264 struct dentry *parent;
2265 int error = 0;
2266
Paul Menagebd89aab2007-10-18 23:40:44 -07002267 parent = cgrp->parent->dentry;
2268 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002269 if (!error) {
Paul Menagebd89aab2007-10-18 23:40:44 -07002270 dentry->d_fsdata = cgrp;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002271 inc_nlink(parent->d_inode);
Paul Menagea47295e2009-01-07 18:07:44 -08002272 rcu_assign_pointer(cgrp->dentry, dentry);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002273 dget(dentry);
2274 }
2275 dput(dentry);
2276
2277 return error;
2278}
2279
Li Zefan099fca32009-04-02 16:57:29 -07002280/**
2281 * cgroup_file_mode - deduce file mode of a control file
2282 * @cft: the control file in question
2283 *
2284 * returns cft->mode if ->mode is not 0
2285 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2286 * returns S_IRUGO if it has only a read handler
2287 * returns S_IWUSR if it has only a write hander
2288 */
2289static mode_t cgroup_file_mode(const struct cftype *cft)
2290{
2291 mode_t mode = 0;
2292
2293 if (cft->mode)
2294 return cft->mode;
2295
2296 if (cft->read || cft->read_u64 || cft->read_s64 ||
2297 cft->read_map || cft->read_seq_string)
2298 mode |= S_IRUGO;
2299
2300 if (cft->write || cft->write_u64 || cft->write_s64 ||
2301 cft->write_string || cft->trigger)
2302 mode |= S_IWUSR;
2303
2304 return mode;
2305}
2306
Paul Menagebd89aab2007-10-18 23:40:44 -07002307int cgroup_add_file(struct cgroup *cgrp,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002308 struct cgroup_subsys *subsys,
2309 const struct cftype *cft)
2310{
Paul Menagebd89aab2007-10-18 23:40:44 -07002311 struct dentry *dir = cgrp->dentry;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002312 struct dentry *dentry;
2313 int error;
Li Zefan099fca32009-04-02 16:57:29 -07002314 mode_t mode;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002315
2316 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
Paul Menagebd89aab2007-10-18 23:40:44 -07002317 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07002318 strcpy(name, subsys->name);
2319 strcat(name, ".");
2320 }
2321 strcat(name, cft->name);
2322 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2323 dentry = lookup_one_len(name, dir, strlen(name));
2324 if (!IS_ERR(dentry)) {
Li Zefan099fca32009-04-02 16:57:29 -07002325 mode = cgroup_file_mode(cft);
2326 error = cgroup_create_file(dentry, mode | S_IFREG,
Paul Menagebd89aab2007-10-18 23:40:44 -07002327 cgrp->root->sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002328 if (!error)
2329 dentry->d_fsdata = (void *)cft;
2330 dput(dentry);
2331 } else
2332 error = PTR_ERR(dentry);
2333 return error;
2334}
Ben Blume6a11052010-03-10 15:22:09 -08002335EXPORT_SYMBOL_GPL(cgroup_add_file);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002336
Paul Menagebd89aab2007-10-18 23:40:44 -07002337int cgroup_add_files(struct cgroup *cgrp,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002338 struct cgroup_subsys *subsys,
2339 const struct cftype cft[],
2340 int count)
2341{
2342 int i, err;
2343 for (i = 0; i < count; i++) {
Paul Menagebd89aab2007-10-18 23:40:44 -07002344 err = cgroup_add_file(cgrp, subsys, &cft[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002345 if (err)
2346 return err;
2347 }
2348 return 0;
2349}
Ben Blume6a11052010-03-10 15:22:09 -08002350EXPORT_SYMBOL_GPL(cgroup_add_files);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002351
Li Zefana043e3b2008-02-23 15:24:09 -08002352/**
2353 * cgroup_task_count - count the number of tasks in a cgroup.
2354 * @cgrp: the cgroup in question
2355 *
2356 * Return the number of tasks in the cgroup.
2357 */
Paul Menagebd89aab2007-10-18 23:40:44 -07002358int cgroup_task_count(const struct cgroup *cgrp)
Paul Menagebbcb81d2007-10-18 23:39:32 -07002359{
2360 int count = 0;
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -07002361 struct cg_cgroup_link *link;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002362
Paul Menage817929e2007-10-18 23:39:36 -07002363 read_lock(&css_set_lock);
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -07002364 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
Lai Jiangshan146aa1b2008-10-18 20:28:03 -07002365 count += atomic_read(&link->cg->refcount);
Paul Menage817929e2007-10-18 23:39:36 -07002366 }
2367 read_unlock(&css_set_lock);
Paul Menagebbcb81d2007-10-18 23:39:32 -07002368 return count;
2369}
2370
2371/*
Paul Menage817929e2007-10-18 23:39:36 -07002372 * Advance a list_head iterator. The iterator should be positioned at
2373 * the start of a css_set
2374 */
Paul Menagebd89aab2007-10-18 23:40:44 -07002375static void cgroup_advance_iter(struct cgroup *cgrp,
Paul Menage7717f7b2009-09-23 15:56:22 -07002376 struct cgroup_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07002377{
2378 struct list_head *l = it->cg_link;
2379 struct cg_cgroup_link *link;
2380 struct css_set *cg;
2381
2382 /* Advance to the next non-empty css_set */
2383 do {
2384 l = l->next;
Paul Menagebd89aab2007-10-18 23:40:44 -07002385 if (l == &cgrp->css_sets) {
Paul Menage817929e2007-10-18 23:39:36 -07002386 it->cg_link = NULL;
2387 return;
2388 }
Paul Menagebd89aab2007-10-18 23:40:44 -07002389 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -07002390 cg = link->cg;
2391 } while (list_empty(&cg->tasks));
2392 it->cg_link = l;
2393 it->task = cg->tasks.next;
2394}
2395
Cliff Wickman31a7df02008-02-07 00:14:42 -08002396/*
2397 * To reduce the fork() overhead for systems that are not actually
2398 * using their cgroups capability, we don't maintain the lists running
2399 * through each css_set to its tasks until we see the list actually
2400 * used - in other words after the first call to cgroup_iter_start().
2401 *
2402 * The tasklist_lock is not held here, as do_each_thread() and
2403 * while_each_thread() are protected by RCU.
2404 */
Adrian Bunk3df91fe2008-04-29 00:59:54 -07002405static void cgroup_enable_task_cg_lists(void)
Cliff Wickman31a7df02008-02-07 00:14:42 -08002406{
2407 struct task_struct *p, *g;
2408 write_lock(&css_set_lock);
2409 use_task_css_set_links = 1;
2410 do_each_thread(g, p) {
2411 task_lock(p);
Li Zefan0e043882008-04-17 11:37:15 +08002412 /*
2413 * We should check if the process is exiting, otherwise
2414 * it will race with cgroup_exit() in that the list
2415 * entry won't be deleted though the process has exited.
2416 */
2417 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
Cliff Wickman31a7df02008-02-07 00:14:42 -08002418 list_add(&p->cg_list, &p->cgroups->tasks);
2419 task_unlock(p);
2420 } while_each_thread(g, p);
2421 write_unlock(&css_set_lock);
2422}
2423
Paul Menagebd89aab2007-10-18 23:40:44 -07002424void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07002425{
2426 /*
2427 * The first time anyone tries to iterate across a cgroup,
2428 * we need to enable the list linking each css_set to its
2429 * tasks, and fix up all existing tasks.
2430 */
Cliff Wickman31a7df02008-02-07 00:14:42 -08002431 if (!use_task_css_set_links)
2432 cgroup_enable_task_cg_lists();
2433
Paul Menage817929e2007-10-18 23:39:36 -07002434 read_lock(&css_set_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07002435 it->cg_link = &cgrp->css_sets;
2436 cgroup_advance_iter(cgrp, it);
Paul Menage817929e2007-10-18 23:39:36 -07002437}
2438
Paul Menagebd89aab2007-10-18 23:40:44 -07002439struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -07002440 struct cgroup_iter *it)
2441{
2442 struct task_struct *res;
2443 struct list_head *l = it->task;
Lai Jiangshan2019f632009-01-07 18:07:36 -08002444 struct cg_cgroup_link *link;
Paul Menage817929e2007-10-18 23:39:36 -07002445
2446 /* If the iterator cg is NULL, we have no tasks */
2447 if (!it->cg_link)
2448 return NULL;
2449 res = list_entry(l, struct task_struct, cg_list);
2450 /* Advance iterator to find next entry */
2451 l = l->next;
Lai Jiangshan2019f632009-01-07 18:07:36 -08002452 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2453 if (l == &link->cg->tasks) {
Paul Menage817929e2007-10-18 23:39:36 -07002454 /* We reached the end of this task list - move on to
2455 * the next cg_cgroup_link */
Paul Menagebd89aab2007-10-18 23:40:44 -07002456 cgroup_advance_iter(cgrp, it);
Paul Menage817929e2007-10-18 23:39:36 -07002457 } else {
2458 it->task = l;
2459 }
2460 return res;
2461}
2462
Paul Menagebd89aab2007-10-18 23:40:44 -07002463void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07002464{
2465 read_unlock(&css_set_lock);
2466}
2467
Cliff Wickman31a7df02008-02-07 00:14:42 -08002468static inline int started_after_time(struct task_struct *t1,
2469 struct timespec *time,
2470 struct task_struct *t2)
2471{
2472 int start_diff = timespec_compare(&t1->start_time, time);
2473 if (start_diff > 0) {
2474 return 1;
2475 } else if (start_diff < 0) {
2476 return 0;
2477 } else {
2478 /*
2479 * Arbitrarily, if two processes started at the same
2480 * time, we'll say that the lower pointer value
2481 * started first. Note that t2 may have exited by now
2482 * so this may not be a valid pointer any longer, but
2483 * that's fine - it still serves to distinguish
2484 * between two tasks started (effectively) simultaneously.
2485 */
2486 return t1 > t2;
2487 }
2488}
2489
2490/*
2491 * This function is a callback from heap_insert() and is used to order
2492 * the heap.
2493 * In this case we order the heap in descending task start time.
2494 */
2495static inline int started_after(void *p1, void *p2)
2496{
2497 struct task_struct *t1 = p1;
2498 struct task_struct *t2 = p2;
2499 return started_after_time(t1, &t2->start_time, t2);
2500}
2501
2502/**
2503 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2504 * @scan: struct cgroup_scanner containing arguments for the scan
2505 *
2506 * Arguments include pointers to callback functions test_task() and
2507 * process_task().
2508 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2509 * and if it returns true, call process_task() for it also.
2510 * The test_task pointer may be NULL, meaning always true (select all tasks).
2511 * Effectively duplicates cgroup_iter_{start,next,end}()
2512 * but does not lock css_set_lock for the call to process_task().
2513 * The struct cgroup_scanner may be embedded in any structure of the caller's
2514 * creation.
2515 * It is guaranteed that process_task() will act on every task that
2516 * is a member of the cgroup for the duration of this call. This
2517 * function may or may not call process_task() for tasks that exit
2518 * or move to a different cgroup during the call, or are forked or
2519 * move into the cgroup during the call.
2520 *
2521 * Note that test_task() may be called with locks held, and may in some
2522 * situations be called multiple times for the same task, so it should
2523 * be cheap.
2524 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2525 * pre-allocated and will be used for heap operations (and its "gt" member will
2526 * be overwritten), else a temporary heap will be used (allocation of which
2527 * may cause this function to fail).
2528 */
2529int cgroup_scan_tasks(struct cgroup_scanner *scan)
2530{
2531 int retval, i;
2532 struct cgroup_iter it;
2533 struct task_struct *p, *dropped;
2534 /* Never dereference latest_task, since it's not refcounted */
2535 struct task_struct *latest_task = NULL;
2536 struct ptr_heap tmp_heap;
2537 struct ptr_heap *heap;
2538 struct timespec latest_time = { 0, 0 };
2539
2540 if (scan->heap) {
2541 /* The caller supplied our heap and pre-allocated its memory */
2542 heap = scan->heap;
2543 heap->gt = &started_after;
2544 } else {
2545 /* We need to allocate our own heap memory */
2546 heap = &tmp_heap;
2547 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2548 if (retval)
2549 /* cannot allocate the heap */
2550 return retval;
2551 }
2552
2553 again:
2554 /*
2555 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2556 * to determine which are of interest, and using the scanner's
2557 * "process_task" callback to process any of them that need an update.
2558 * Since we don't want to hold any locks during the task updates,
2559 * gather tasks to be processed in a heap structure.
2560 * The heap is sorted by descending task start time.
2561 * If the statically-sized heap fills up, we overflow tasks that
2562 * started later, and in future iterations only consider tasks that
2563 * started after the latest task in the previous pass. This
2564 * guarantees forward progress and that we don't miss any tasks.
2565 */
2566 heap->size = 0;
2567 cgroup_iter_start(scan->cg, &it);
2568 while ((p = cgroup_iter_next(scan->cg, &it))) {
2569 /*
2570 * Only affect tasks that qualify per the caller's callback,
2571 * if he provided one
2572 */
2573 if (scan->test_task && !scan->test_task(p, scan))
2574 continue;
2575 /*
2576 * Only process tasks that started after the last task
2577 * we processed
2578 */
2579 if (!started_after_time(p, &latest_time, latest_task))
2580 continue;
2581 dropped = heap_insert(heap, p);
2582 if (dropped == NULL) {
2583 /*
2584 * The new task was inserted; the heap wasn't
2585 * previously full
2586 */
2587 get_task_struct(p);
2588 } else if (dropped != p) {
2589 /*
2590 * The new task was inserted, and pushed out a
2591 * different task
2592 */
2593 get_task_struct(p);
2594 put_task_struct(dropped);
2595 }
2596 /*
2597 * Else the new task was newer than anything already in
2598 * the heap and wasn't inserted
2599 */
2600 }
2601 cgroup_iter_end(scan->cg, &it);
2602
2603 if (heap->size) {
2604 for (i = 0; i < heap->size; i++) {
Paul Jackson4fe91d52008-04-29 00:59:55 -07002605 struct task_struct *q = heap->ptrs[i];
Cliff Wickman31a7df02008-02-07 00:14:42 -08002606 if (i == 0) {
Paul Jackson4fe91d52008-04-29 00:59:55 -07002607 latest_time = q->start_time;
2608 latest_task = q;
Cliff Wickman31a7df02008-02-07 00:14:42 -08002609 }
2610 /* Process the task per the caller's callback */
Paul Jackson4fe91d52008-04-29 00:59:55 -07002611 scan->process_task(q, scan);
2612 put_task_struct(q);
Cliff Wickman31a7df02008-02-07 00:14:42 -08002613 }
2614 /*
2615 * If we had to process any tasks at all, scan again
2616 * in case some of them were in the middle of forking
2617 * children that didn't get processed.
2618 * Not the most efficient way to do it, but it avoids
2619 * having to take callback_mutex in the fork path
2620 */
2621 goto again;
2622 }
2623 if (heap == &tmp_heap)
2624 heap_free(&tmp_heap);
2625 return 0;
2626}
2627
Paul Menage817929e2007-10-18 23:39:36 -07002628/*
Ben Blum102a7752009-09-23 15:56:26 -07002629 * Stuff for reading the 'tasks'/'procs' files.
Paul Menagebbcb81d2007-10-18 23:39:32 -07002630 *
2631 * Reading this file can return large amounts of data if a cgroup has
2632 * *lots* of attached tasks. So it may need several calls to read(),
2633 * but we cannot guarantee that the information we produce is correct
2634 * unless we produce it entirely atomically.
2635 *
Paul Menagebbcb81d2007-10-18 23:39:32 -07002636 */
Paul Menagebbcb81d2007-10-18 23:39:32 -07002637
2638/*
Ben Blumd1d9fd32009-09-23 15:56:28 -07002639 * The following two functions "fix" the issue where there are more pids
2640 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2641 * TODO: replace with a kernel-wide solution to this problem
2642 */
2643#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2644static void *pidlist_allocate(int count)
2645{
2646 if (PIDLIST_TOO_LARGE(count))
2647 return vmalloc(count * sizeof(pid_t));
2648 else
2649 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
2650}
2651static void pidlist_free(void *p)
2652{
2653 if (is_vmalloc_addr(p))
2654 vfree(p);
2655 else
2656 kfree(p);
2657}
2658static void *pidlist_resize(void *p, int newcount)
2659{
2660 void *newlist;
2661 /* note: if new alloc fails, old p will still be valid either way */
2662 if (is_vmalloc_addr(p)) {
2663 newlist = vmalloc(newcount * sizeof(pid_t));
2664 if (!newlist)
2665 return NULL;
2666 memcpy(newlist, p, newcount * sizeof(pid_t));
2667 vfree(p);
2668 } else {
2669 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
2670 }
2671 return newlist;
2672}
2673
2674/*
Ben Blum102a7752009-09-23 15:56:26 -07002675 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2676 * If the new stripped list is sufficiently smaller and there's enough memory
2677 * to allocate a new buffer, will let go of the unneeded memory. Returns the
2678 * number of unique elements.
Paul Menagebbcb81d2007-10-18 23:39:32 -07002679 */
Ben Blum102a7752009-09-23 15:56:26 -07002680/* is the size difference enough that we should re-allocate the array? */
2681#define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
2682static int pidlist_uniq(pid_t **p, int length)
Paul Menagebbcb81d2007-10-18 23:39:32 -07002683{
Ben Blum102a7752009-09-23 15:56:26 -07002684 int src, dest = 1;
2685 pid_t *list = *p;
2686 pid_t *newlist;
2687
2688 /*
2689 * we presume the 0th element is unique, so i starts at 1. trivial
2690 * edge cases first; no work needs to be done for either
2691 */
2692 if (length == 0 || length == 1)
2693 return length;
2694 /* src and dest walk down the list; dest counts unique elements */
2695 for (src = 1; src < length; src++) {
2696 /* find next unique element */
2697 while (list[src] == list[src-1]) {
2698 src++;
2699 if (src == length)
2700 goto after;
2701 }
2702 /* dest always points to where the next unique element goes */
2703 list[dest] = list[src];
2704 dest++;
2705 }
2706after:
2707 /*
2708 * if the length difference is large enough, we want to allocate a
2709 * smaller buffer to save memory. if this fails due to out of memory,
2710 * we'll just stay with what we've got.
2711 */
2712 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
Ben Blumd1d9fd32009-09-23 15:56:28 -07002713 newlist = pidlist_resize(list, dest);
Ben Blum102a7752009-09-23 15:56:26 -07002714 if (newlist)
2715 *p = newlist;
2716 }
2717 return dest;
2718}
2719
2720static int cmppid(const void *a, const void *b)
2721{
2722 return *(pid_t *)a - *(pid_t *)b;
2723}
2724
2725/*
Ben Blum72a8cb32009-09-23 15:56:27 -07002726 * find the appropriate pidlist for our purpose (given procs vs tasks)
2727 * returns with the lock on that pidlist already held, and takes care
2728 * of the use count, or returns NULL with no locks held if we're out of
2729 * memory.
2730 */
2731static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
2732 enum cgroup_filetype type)
2733{
2734 struct cgroup_pidlist *l;
2735 /* don't need task_nsproxy() if we're looking at ourself */
Li Zefanb70cc5f2010-03-10 15:22:12 -08002736 struct pid_namespace *ns = current->nsproxy->pid_ns;
2737
Ben Blum72a8cb32009-09-23 15:56:27 -07002738 /*
2739 * We can't drop the pidlist_mutex before taking the l->mutex in case
2740 * the last ref-holder is trying to remove l from the list at the same
2741 * time. Holding the pidlist_mutex precludes somebody taking whichever
2742 * list we find out from under us - compare release_pid_array().
2743 */
2744 mutex_lock(&cgrp->pidlist_mutex);
2745 list_for_each_entry(l, &cgrp->pidlists, links) {
2746 if (l->key.type == type && l->key.ns == ns) {
Ben Blum72a8cb32009-09-23 15:56:27 -07002747 /* make sure l doesn't vanish out from under us */
2748 down_write(&l->mutex);
2749 mutex_unlock(&cgrp->pidlist_mutex);
Ben Blum72a8cb32009-09-23 15:56:27 -07002750 return l;
2751 }
2752 }
2753 /* entry not found; create a new one */
2754 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
2755 if (!l) {
2756 mutex_unlock(&cgrp->pidlist_mutex);
Ben Blum72a8cb32009-09-23 15:56:27 -07002757 return l;
2758 }
2759 init_rwsem(&l->mutex);
2760 down_write(&l->mutex);
2761 l->key.type = type;
Li Zefanb70cc5f2010-03-10 15:22:12 -08002762 l->key.ns = get_pid_ns(ns);
Ben Blum72a8cb32009-09-23 15:56:27 -07002763 l->use_count = 0; /* don't increment here */
2764 l->list = NULL;
2765 l->owner = cgrp;
2766 list_add(&l->links, &cgrp->pidlists);
2767 mutex_unlock(&cgrp->pidlist_mutex);
2768 return l;
2769}
2770
2771/*
Ben Blum102a7752009-09-23 15:56:26 -07002772 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
2773 */
Ben Blum72a8cb32009-09-23 15:56:27 -07002774static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
2775 struct cgroup_pidlist **lp)
Ben Blum102a7752009-09-23 15:56:26 -07002776{
2777 pid_t *array;
2778 int length;
2779 int pid, n = 0; /* used for populating the array */
Paul Menage817929e2007-10-18 23:39:36 -07002780 struct cgroup_iter it;
2781 struct task_struct *tsk;
Ben Blum102a7752009-09-23 15:56:26 -07002782 struct cgroup_pidlist *l;
2783
2784 /*
2785 * If cgroup gets more users after we read count, we won't have
2786 * enough space - tough. This race is indistinguishable to the
2787 * caller from the case that the additional cgroup users didn't
2788 * show up until sometime later on.
2789 */
2790 length = cgroup_task_count(cgrp);
Ben Blumd1d9fd32009-09-23 15:56:28 -07002791 array = pidlist_allocate(length);
Ben Blum102a7752009-09-23 15:56:26 -07002792 if (!array)
2793 return -ENOMEM;
2794 /* now, populate the array */
Paul Menagebd89aab2007-10-18 23:40:44 -07002795 cgroup_iter_start(cgrp, &it);
2796 while ((tsk = cgroup_iter_next(cgrp, &it))) {
Ben Blum102a7752009-09-23 15:56:26 -07002797 if (unlikely(n == length))
Paul Menage817929e2007-10-18 23:39:36 -07002798 break;
Ben Blum102a7752009-09-23 15:56:26 -07002799 /* get tgid or pid for procs or tasks file respectively */
Ben Blum72a8cb32009-09-23 15:56:27 -07002800 if (type == CGROUP_FILE_PROCS)
2801 pid = task_tgid_vnr(tsk);
2802 else
2803 pid = task_pid_vnr(tsk);
Ben Blum102a7752009-09-23 15:56:26 -07002804 if (pid > 0) /* make sure to only use valid results */
2805 array[n++] = pid;
Paul Menage817929e2007-10-18 23:39:36 -07002806 }
Paul Menagebd89aab2007-10-18 23:40:44 -07002807 cgroup_iter_end(cgrp, &it);
Ben Blum102a7752009-09-23 15:56:26 -07002808 length = n;
2809 /* now sort & (if procs) strip out duplicates */
2810 sort(array, length, sizeof(pid_t), cmppid, NULL);
Ben Blum72a8cb32009-09-23 15:56:27 -07002811 if (type == CGROUP_FILE_PROCS)
Ben Blum102a7752009-09-23 15:56:26 -07002812 length = pidlist_uniq(&array, length);
Ben Blum72a8cb32009-09-23 15:56:27 -07002813 l = cgroup_pidlist_find(cgrp, type);
2814 if (!l) {
Ben Blumd1d9fd32009-09-23 15:56:28 -07002815 pidlist_free(array);
Ben Blum72a8cb32009-09-23 15:56:27 -07002816 return -ENOMEM;
Ben Blum102a7752009-09-23 15:56:26 -07002817 }
Ben Blum72a8cb32009-09-23 15:56:27 -07002818 /* store array, freeing old if necessary - lock already held */
Ben Blumd1d9fd32009-09-23 15:56:28 -07002819 pidlist_free(l->list);
Ben Blum102a7752009-09-23 15:56:26 -07002820 l->list = array;
2821 l->length = length;
2822 l->use_count++;
2823 up_write(&l->mutex);
Ben Blum72a8cb32009-09-23 15:56:27 -07002824 *lp = l;
Ben Blum102a7752009-09-23 15:56:26 -07002825 return 0;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002826}
2827
Balbir Singh846c7bb2007-10-18 23:39:44 -07002828/**
Li Zefana043e3b2008-02-23 15:24:09 -08002829 * cgroupstats_build - build and fill cgroupstats
Balbir Singh846c7bb2007-10-18 23:39:44 -07002830 * @stats: cgroupstats to fill information into
2831 * @dentry: A dentry entry belonging to the cgroup for which stats have
2832 * been requested.
Li Zefana043e3b2008-02-23 15:24:09 -08002833 *
2834 * Build and fill cgroupstats so that taskstats can export it to user
2835 * space.
Balbir Singh846c7bb2007-10-18 23:39:44 -07002836 */
2837int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2838{
2839 int ret = -EINVAL;
Paul Menagebd89aab2007-10-18 23:40:44 -07002840 struct cgroup *cgrp;
Balbir Singh846c7bb2007-10-18 23:39:44 -07002841 struct cgroup_iter it;
2842 struct task_struct *tsk;
Li Zefan33d283b2008-11-19 15:36:48 -08002843
Balbir Singh846c7bb2007-10-18 23:39:44 -07002844 /*
Li Zefan33d283b2008-11-19 15:36:48 -08002845 * Validate dentry by checking the superblock operations,
2846 * and make sure it's a directory.
Balbir Singh846c7bb2007-10-18 23:39:44 -07002847 */
Li Zefan33d283b2008-11-19 15:36:48 -08002848 if (dentry->d_sb->s_op != &cgroup_ops ||
2849 !S_ISDIR(dentry->d_inode->i_mode))
Balbir Singh846c7bb2007-10-18 23:39:44 -07002850 goto err;
2851
2852 ret = 0;
Paul Menagebd89aab2007-10-18 23:40:44 -07002853 cgrp = dentry->d_fsdata;
Balbir Singh846c7bb2007-10-18 23:39:44 -07002854
Paul Menagebd89aab2007-10-18 23:40:44 -07002855 cgroup_iter_start(cgrp, &it);
2856 while ((tsk = cgroup_iter_next(cgrp, &it))) {
Balbir Singh846c7bb2007-10-18 23:39:44 -07002857 switch (tsk->state) {
2858 case TASK_RUNNING:
2859 stats->nr_running++;
2860 break;
2861 case TASK_INTERRUPTIBLE:
2862 stats->nr_sleeping++;
2863 break;
2864 case TASK_UNINTERRUPTIBLE:
2865 stats->nr_uninterruptible++;
2866 break;
2867 case TASK_STOPPED:
2868 stats->nr_stopped++;
2869 break;
2870 default:
2871 if (delayacct_is_task_waiting_on_io(tsk))
2872 stats->nr_io_wait++;
2873 break;
2874 }
2875 }
Paul Menagebd89aab2007-10-18 23:40:44 -07002876 cgroup_iter_end(cgrp, &it);
Balbir Singh846c7bb2007-10-18 23:39:44 -07002877
Balbir Singh846c7bb2007-10-18 23:39:44 -07002878err:
2879 return ret;
2880}
2881
Paul Menage8f3ff202009-09-23 15:56:25 -07002882
Paul Menagecc31edc2008-10-18 20:28:04 -07002883/*
Ben Blum102a7752009-09-23 15:56:26 -07002884 * seq_file methods for the tasks/procs files. The seq_file position is the
Paul Menagecc31edc2008-10-18 20:28:04 -07002885 * next pid to display; the seq_file iterator is a pointer to the pid
Ben Blum102a7752009-09-23 15:56:26 -07002886 * in the cgroup->l->list array.
Paul Menagecc31edc2008-10-18 20:28:04 -07002887 */
2888
Ben Blum102a7752009-09-23 15:56:26 -07002889static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
Paul Menagecc31edc2008-10-18 20:28:04 -07002890{
2891 /*
2892 * Initially we receive a position value that corresponds to
2893 * one more than the last pid shown (or 0 on the first call or
2894 * after a seek to the start). Use a binary-search to find the
2895 * next pid to display, if any
2896 */
Ben Blum102a7752009-09-23 15:56:26 -07002897 struct cgroup_pidlist *l = s->private;
Paul Menagecc31edc2008-10-18 20:28:04 -07002898 int index = 0, pid = *pos;
2899 int *iter;
2900
Ben Blum102a7752009-09-23 15:56:26 -07002901 down_read(&l->mutex);
Paul Menagecc31edc2008-10-18 20:28:04 -07002902 if (pid) {
Ben Blum102a7752009-09-23 15:56:26 -07002903 int end = l->length;
Stephen Rothwell20777762008-10-21 16:11:20 +11002904
Paul Menagecc31edc2008-10-18 20:28:04 -07002905 while (index < end) {
2906 int mid = (index + end) / 2;
Ben Blum102a7752009-09-23 15:56:26 -07002907 if (l->list[mid] == pid) {
Paul Menagecc31edc2008-10-18 20:28:04 -07002908 index = mid;
2909 break;
Ben Blum102a7752009-09-23 15:56:26 -07002910 } else if (l->list[mid] <= pid)
Paul Menagecc31edc2008-10-18 20:28:04 -07002911 index = mid + 1;
2912 else
2913 end = mid;
2914 }
2915 }
2916 /* If we're off the end of the array, we're done */
Ben Blum102a7752009-09-23 15:56:26 -07002917 if (index >= l->length)
Paul Menagecc31edc2008-10-18 20:28:04 -07002918 return NULL;
2919 /* Update the abstract position to be the actual pid that we found */
Ben Blum102a7752009-09-23 15:56:26 -07002920 iter = l->list + index;
Paul Menagecc31edc2008-10-18 20:28:04 -07002921 *pos = *iter;
2922 return iter;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002923}
2924
Ben Blum102a7752009-09-23 15:56:26 -07002925static void cgroup_pidlist_stop(struct seq_file *s, void *v)
Paul Menagecc31edc2008-10-18 20:28:04 -07002926{
Ben Blum102a7752009-09-23 15:56:26 -07002927 struct cgroup_pidlist *l = s->private;
2928 up_read(&l->mutex);
Paul Menagecc31edc2008-10-18 20:28:04 -07002929}
2930
Ben Blum102a7752009-09-23 15:56:26 -07002931static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
Paul Menagecc31edc2008-10-18 20:28:04 -07002932{
Ben Blum102a7752009-09-23 15:56:26 -07002933 struct cgroup_pidlist *l = s->private;
2934 pid_t *p = v;
2935 pid_t *end = l->list + l->length;
Paul Menagecc31edc2008-10-18 20:28:04 -07002936 /*
2937 * Advance to the next pid in the array. If this goes off the
2938 * end, we're done
2939 */
2940 p++;
2941 if (p >= end) {
2942 return NULL;
2943 } else {
2944 *pos = *p;
2945 return p;
2946 }
2947}
2948
Ben Blum102a7752009-09-23 15:56:26 -07002949static int cgroup_pidlist_show(struct seq_file *s, void *v)
Paul Menagecc31edc2008-10-18 20:28:04 -07002950{
2951 return seq_printf(s, "%d\n", *(int *)v);
2952}
2953
Ben Blum102a7752009-09-23 15:56:26 -07002954/*
2955 * seq_operations functions for iterating on pidlists through seq_file -
2956 * independent of whether it's tasks or procs
2957 */
2958static const struct seq_operations cgroup_pidlist_seq_operations = {
2959 .start = cgroup_pidlist_start,
2960 .stop = cgroup_pidlist_stop,
2961 .next = cgroup_pidlist_next,
2962 .show = cgroup_pidlist_show,
Paul Menagecc31edc2008-10-18 20:28:04 -07002963};
2964
Ben Blum102a7752009-09-23 15:56:26 -07002965static void cgroup_release_pid_array(struct cgroup_pidlist *l)
Paul Menagecc31edc2008-10-18 20:28:04 -07002966{
Ben Blum72a8cb32009-09-23 15:56:27 -07002967 /*
2968 * the case where we're the last user of this particular pidlist will
2969 * have us remove it from the cgroup's list, which entails taking the
2970 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
2971 * pidlist_mutex, we have to take pidlist_mutex first.
2972 */
2973 mutex_lock(&l->owner->pidlist_mutex);
Ben Blum102a7752009-09-23 15:56:26 -07002974 down_write(&l->mutex);
2975 BUG_ON(!l->use_count);
2976 if (!--l->use_count) {
Ben Blum72a8cb32009-09-23 15:56:27 -07002977 /* we're the last user if refcount is 0; remove and free */
2978 list_del(&l->links);
2979 mutex_unlock(&l->owner->pidlist_mutex);
Ben Blumd1d9fd32009-09-23 15:56:28 -07002980 pidlist_free(l->list);
Ben Blum72a8cb32009-09-23 15:56:27 -07002981 put_pid_ns(l->key.ns);
2982 up_write(&l->mutex);
2983 kfree(l);
2984 return;
Paul Menagecc31edc2008-10-18 20:28:04 -07002985 }
Ben Blum72a8cb32009-09-23 15:56:27 -07002986 mutex_unlock(&l->owner->pidlist_mutex);
Ben Blum102a7752009-09-23 15:56:26 -07002987 up_write(&l->mutex);
Paul Menagecc31edc2008-10-18 20:28:04 -07002988}
2989
Ben Blum102a7752009-09-23 15:56:26 -07002990static int cgroup_pidlist_release(struct inode *inode, struct file *file)
Paul Menagebbcb81d2007-10-18 23:39:32 -07002991{
Ben Blum102a7752009-09-23 15:56:26 -07002992 struct cgroup_pidlist *l;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002993 if (!(file->f_mode & FMODE_READ))
2994 return 0;
Ben Blum102a7752009-09-23 15:56:26 -07002995 /*
2996 * the seq_file will only be initialized if the file was opened for
2997 * reading; hence we check if it's not null only in that case.
2998 */
2999 l = ((struct seq_file *)file->private_data)->private;
3000 cgroup_release_pid_array(l);
Paul Menagecc31edc2008-10-18 20:28:04 -07003001 return seq_release(inode, file);
3002}
3003
Ben Blum102a7752009-09-23 15:56:26 -07003004static const struct file_operations cgroup_pidlist_operations = {
Paul Menagecc31edc2008-10-18 20:28:04 -07003005 .read = seq_read,
3006 .llseek = seq_lseek,
3007 .write = cgroup_file_write,
Ben Blum102a7752009-09-23 15:56:26 -07003008 .release = cgroup_pidlist_release,
Paul Menagecc31edc2008-10-18 20:28:04 -07003009};
3010
3011/*
Ben Blum102a7752009-09-23 15:56:26 -07003012 * The following functions handle opens on a file that displays a pidlist
3013 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3014 * in the cgroup.
Paul Menagecc31edc2008-10-18 20:28:04 -07003015 */
Ben Blum102a7752009-09-23 15:56:26 -07003016/* helper function for the two below it */
Ben Blum72a8cb32009-09-23 15:56:27 -07003017static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
Paul Menagecc31edc2008-10-18 20:28:04 -07003018{
3019 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
Ben Blum72a8cb32009-09-23 15:56:27 -07003020 struct cgroup_pidlist *l;
Paul Menagecc31edc2008-10-18 20:28:04 -07003021 int retval;
3022
3023 /* Nothing to do for write-only files */
3024 if (!(file->f_mode & FMODE_READ))
3025 return 0;
Paul Menagebbcb81d2007-10-18 23:39:32 -07003026
Ben Blum102a7752009-09-23 15:56:26 -07003027 /* have the array populated */
Ben Blum72a8cb32009-09-23 15:56:27 -07003028 retval = pidlist_array_load(cgrp, type, &l);
Ben Blum102a7752009-09-23 15:56:26 -07003029 if (retval)
3030 return retval;
3031 /* configure file information */
3032 file->f_op = &cgroup_pidlist_operations;
Paul Menagebbcb81d2007-10-18 23:39:32 -07003033
Ben Blum102a7752009-09-23 15:56:26 -07003034 retval = seq_open(file, &cgroup_pidlist_seq_operations);
Paul Menagecc31edc2008-10-18 20:28:04 -07003035 if (retval) {
Ben Blum102a7752009-09-23 15:56:26 -07003036 cgroup_release_pid_array(l);
Paul Menagecc31edc2008-10-18 20:28:04 -07003037 return retval;
Paul Menagebbcb81d2007-10-18 23:39:32 -07003038 }
Ben Blum102a7752009-09-23 15:56:26 -07003039 ((struct seq_file *)file->private_data)->private = l;
Paul Menagebbcb81d2007-10-18 23:39:32 -07003040 return 0;
3041}
Ben Blum102a7752009-09-23 15:56:26 -07003042static int cgroup_tasks_open(struct inode *unused, struct file *file)
3043{
Ben Blum72a8cb32009-09-23 15:56:27 -07003044 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
Ben Blum102a7752009-09-23 15:56:26 -07003045}
3046static int cgroup_procs_open(struct inode *unused, struct file *file)
3047{
Ben Blum72a8cb32009-09-23 15:56:27 -07003048 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
Ben Blum102a7752009-09-23 15:56:26 -07003049}
Paul Menagebbcb81d2007-10-18 23:39:32 -07003050
Paul Menagebd89aab2007-10-18 23:40:44 -07003051static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
Paul Menage81a6a5c2007-10-18 23:39:38 -07003052 struct cftype *cft)
3053{
Paul Menagebd89aab2007-10-18 23:40:44 -07003054 return notify_on_release(cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -07003055}
3056
Paul Menage6379c102008-07-25 01:47:01 -07003057static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3058 struct cftype *cft,
3059 u64 val)
3060{
3061 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3062 if (val)
3063 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3064 else
3065 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3066 return 0;
3067}
3068
Paul Menagebbcb81d2007-10-18 23:39:32 -07003069/*
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003070 * Unregister event and free resources.
3071 *
3072 * Gets called from workqueue.
3073 */
3074static void cgroup_event_remove(struct work_struct *work)
3075{
3076 struct cgroup_event *event = container_of(work, struct cgroup_event,
3077 remove);
3078 struct cgroup *cgrp = event->cgrp;
3079
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003080 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3081
3082 eventfd_ctx_put(event->eventfd);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003083 kfree(event);
Kirill A. Shutemova0a4db52010-03-10 15:22:34 -08003084 dput(cgrp->dentry);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003085}
3086
3087/*
3088 * Gets called on POLLHUP on eventfd when user closes it.
3089 *
3090 * Called with wqh->lock held and interrupts disabled.
3091 */
3092static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3093 int sync, void *key)
3094{
3095 struct cgroup_event *event = container_of(wait,
3096 struct cgroup_event, wait);
3097 struct cgroup *cgrp = event->cgrp;
3098 unsigned long flags = (unsigned long)key;
3099
3100 if (flags & POLLHUP) {
Changli Gaoa93d2f12010-05-07 14:33:26 +08003101 __remove_wait_queue(event->wqh, &event->wait);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003102 spin_lock(&cgrp->event_list_lock);
3103 list_del(&event->list);
3104 spin_unlock(&cgrp->event_list_lock);
3105 /*
3106 * We are in atomic context, but cgroup_event_remove() may
3107 * sleep, so we have to call it in workqueue.
3108 */
3109 schedule_work(&event->remove);
3110 }
3111
3112 return 0;
3113}
3114
3115static void cgroup_event_ptable_queue_proc(struct file *file,
3116 wait_queue_head_t *wqh, poll_table *pt)
3117{
3118 struct cgroup_event *event = container_of(pt,
3119 struct cgroup_event, pt);
3120
3121 event->wqh = wqh;
3122 add_wait_queue(wqh, &event->wait);
3123}
3124
3125/*
3126 * Parse input and register new cgroup event handler.
3127 *
3128 * Input must be in format '<event_fd> <control_fd> <args>'.
3129 * Interpretation of args is defined by control file implementation.
3130 */
3131static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3132 const char *buffer)
3133{
3134 struct cgroup_event *event = NULL;
3135 unsigned int efd, cfd;
3136 struct file *efile = NULL;
3137 struct file *cfile = NULL;
3138 char *endp;
3139 int ret;
3140
3141 efd = simple_strtoul(buffer, &endp, 10);
3142 if (*endp != ' ')
3143 return -EINVAL;
3144 buffer = endp + 1;
3145
3146 cfd = simple_strtoul(buffer, &endp, 10);
3147 if ((*endp != ' ') && (*endp != '\0'))
3148 return -EINVAL;
3149 buffer = endp + 1;
3150
3151 event = kzalloc(sizeof(*event), GFP_KERNEL);
3152 if (!event)
3153 return -ENOMEM;
3154 event->cgrp = cgrp;
3155 INIT_LIST_HEAD(&event->list);
3156 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3157 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3158 INIT_WORK(&event->remove, cgroup_event_remove);
3159
3160 efile = eventfd_fget(efd);
3161 if (IS_ERR(efile)) {
3162 ret = PTR_ERR(efile);
3163 goto fail;
3164 }
3165
3166 event->eventfd = eventfd_ctx_fileget(efile);
3167 if (IS_ERR(event->eventfd)) {
3168 ret = PTR_ERR(event->eventfd);
3169 goto fail;
3170 }
3171
3172 cfile = fget(cfd);
3173 if (!cfile) {
3174 ret = -EBADF;
3175 goto fail;
3176 }
3177
3178 /* the process need read permission on control file */
3179 ret = file_permission(cfile, MAY_READ);
3180 if (ret < 0)
3181 goto fail;
3182
3183 event->cft = __file_cft(cfile);
3184 if (IS_ERR(event->cft)) {
3185 ret = PTR_ERR(event->cft);
3186 goto fail;
3187 }
3188
3189 if (!event->cft->register_event || !event->cft->unregister_event) {
3190 ret = -EINVAL;
3191 goto fail;
3192 }
3193
3194 ret = event->cft->register_event(cgrp, event->cft,
3195 event->eventfd, buffer);
3196 if (ret)
3197 goto fail;
3198
3199 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3200 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3201 ret = 0;
3202 goto fail;
3203 }
3204
Kirill A. Shutemova0a4db52010-03-10 15:22:34 -08003205 /*
3206 * Events should be removed after rmdir of cgroup directory, but before
3207 * destroying subsystem state objects. Let's take reference to cgroup
3208 * directory dentry to do that.
3209 */
3210 dget(cgrp->dentry);
3211
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003212 spin_lock(&cgrp->event_list_lock);
3213 list_add(&event->list, &cgrp->event_list);
3214 spin_unlock(&cgrp->event_list_lock);
3215
3216 fput(cfile);
3217 fput(efile);
3218
3219 return 0;
3220
3221fail:
3222 if (cfile)
3223 fput(cfile);
3224
3225 if (event && event->eventfd && !IS_ERR(event->eventfd))
3226 eventfd_ctx_put(event->eventfd);
3227
3228 if (!IS_ERR_OR_NULL(efile))
3229 fput(efile);
3230
3231 kfree(event);
3232
3233 return ret;
3234}
3235
Daniel Lezcano97978e62010-10-27 15:33:35 -07003236static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3237 struct cftype *cft)
3238{
3239 return clone_children(cgrp);
3240}
3241
3242static int cgroup_clone_children_write(struct cgroup *cgrp,
3243 struct cftype *cft,
3244 u64 val)
3245{
3246 if (val)
3247 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3248 else
3249 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3250 return 0;
3251}
3252
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003253/*
Paul Menagebbcb81d2007-10-18 23:39:32 -07003254 * for the common functions, 'private' gives the type of file
3255 */
Ben Blum102a7752009-09-23 15:56:26 -07003256/* for hysterical raisins, we can't put this on the older files */
3257#define CGROUP_FILE_GENERIC_PREFIX "cgroup."
Paul Menage81a6a5c2007-10-18 23:39:38 -07003258static struct cftype files[] = {
3259 {
3260 .name = "tasks",
3261 .open = cgroup_tasks_open,
Paul Menageaf351022008-07-25 01:47:01 -07003262 .write_u64 = cgroup_tasks_write,
Ben Blum102a7752009-09-23 15:56:26 -07003263 .release = cgroup_pidlist_release,
Li Zefan099fca32009-04-02 16:57:29 -07003264 .mode = S_IRUGO | S_IWUSR,
Paul Menage81a6a5c2007-10-18 23:39:38 -07003265 },
Ben Blum102a7752009-09-23 15:56:26 -07003266 {
3267 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3268 .open = cgroup_procs_open,
3269 /* .write_u64 = cgroup_procs_write, TODO */
3270 .release = cgroup_pidlist_release,
3271 .mode = S_IRUGO,
3272 },
Paul Menage81a6a5c2007-10-18 23:39:38 -07003273 {
3274 .name = "notify_on_release",
Paul Menagef4c753b2008-04-29 00:59:56 -07003275 .read_u64 = cgroup_read_notify_on_release,
Paul Menage6379c102008-07-25 01:47:01 -07003276 .write_u64 = cgroup_write_notify_on_release,
Paul Menage81a6a5c2007-10-18 23:39:38 -07003277 },
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003278 {
3279 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3280 .write_string = cgroup_write_event_control,
3281 .mode = S_IWUGO,
3282 },
Daniel Lezcano97978e62010-10-27 15:33:35 -07003283 {
3284 .name = "cgroup.clone_children",
3285 .read_u64 = cgroup_clone_children_read,
3286 .write_u64 = cgroup_clone_children_write,
3287 },
Paul Menage81a6a5c2007-10-18 23:39:38 -07003288};
3289
3290static struct cftype cft_release_agent = {
3291 .name = "release_agent",
Paul Menagee788e062008-07-25 01:46:59 -07003292 .read_seq_string = cgroup_release_agent_show,
3293 .write_string = cgroup_release_agent_write,
3294 .max_write_len = PATH_MAX,
Paul Menagebbcb81d2007-10-18 23:39:32 -07003295};
3296
Paul Menagebd89aab2007-10-18 23:40:44 -07003297static int cgroup_populate_dir(struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003298{
3299 int err;
3300 struct cgroup_subsys *ss;
3301
3302 /* First clear out any existing files */
Paul Menagebd89aab2007-10-18 23:40:44 -07003303 cgroup_clear_directory(cgrp->dentry);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003304
Paul Menagebd89aab2007-10-18 23:40:44 -07003305 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
Paul Menagebbcb81d2007-10-18 23:39:32 -07003306 if (err < 0)
3307 return err;
3308
Paul Menagebd89aab2007-10-18 23:40:44 -07003309 if (cgrp == cgrp->top_cgroup) {
3310 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
Paul Menage81a6a5c2007-10-18 23:39:38 -07003311 return err;
3312 }
3313
Paul Menagebd89aab2007-10-18 23:40:44 -07003314 for_each_subsys(cgrp->root, ss) {
3315 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003316 return err;
3317 }
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07003318 /* This cgroup is ready now */
3319 for_each_subsys(cgrp->root, ss) {
3320 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3321 /*
3322 * Update id->css pointer and make this css visible from
3323 * CSS ID functions. This pointer will be dereferened
3324 * from RCU-read-side without locks.
3325 */
3326 if (css->id)
3327 rcu_assign_pointer(css->id->css, css);
3328 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07003329
3330 return 0;
3331}
3332
3333static void init_cgroup_css(struct cgroup_subsys_state *css,
3334 struct cgroup_subsys *ss,
Paul Menagebd89aab2007-10-18 23:40:44 -07003335 struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003336{
Paul Menagebd89aab2007-10-18 23:40:44 -07003337 css->cgroup = cgrp;
Paul Menagee7c5ec92009-01-07 18:08:38 -08003338 atomic_set(&css->refcnt, 1);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003339 css->flags = 0;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07003340 css->id = NULL;
Paul Menagebd89aab2007-10-18 23:40:44 -07003341 if (cgrp == dummytop)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003342 set_bit(CSS_ROOT, &css->flags);
Paul Menagebd89aab2007-10-18 23:40:44 -07003343 BUG_ON(cgrp->subsys[ss->subsys_id]);
3344 cgrp->subsys[ss->subsys_id] = css;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003345}
3346
Paul Menage999cd8a2009-01-07 18:08:36 -08003347static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3348{
3349 /* We need to take each hierarchy_mutex in a consistent order */
3350 int i;
3351
Ben Blumaae8aab2010-03-10 15:22:07 -08003352 /*
3353 * No worry about a race with rebind_subsystems that might mess up the
3354 * locking order, since both parties are under cgroup_mutex.
3355 */
Paul Menage999cd8a2009-01-07 18:08:36 -08003356 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3357 struct cgroup_subsys *ss = subsys[i];
Ben Blumaae8aab2010-03-10 15:22:07 -08003358 if (ss == NULL)
3359 continue;
Paul Menage999cd8a2009-01-07 18:08:36 -08003360 if (ss->root == root)
Li Zefancfebe562009-02-11 13:04:36 -08003361 mutex_lock(&ss->hierarchy_mutex);
Paul Menage999cd8a2009-01-07 18:08:36 -08003362 }
3363}
3364
3365static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3366{
3367 int i;
3368
3369 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3370 struct cgroup_subsys *ss = subsys[i];
Ben Blumaae8aab2010-03-10 15:22:07 -08003371 if (ss == NULL)
3372 continue;
Paul Menage999cd8a2009-01-07 18:08:36 -08003373 if (ss->root == root)
3374 mutex_unlock(&ss->hierarchy_mutex);
3375 }
3376}
3377
Paul Menageddbcc7e2007-10-18 23:39:30 -07003378/*
Li Zefana043e3b2008-02-23 15:24:09 -08003379 * cgroup_create - create a cgroup
3380 * @parent: cgroup that will be parent of the new cgroup
3381 * @dentry: dentry of the new cgroup
3382 * @mode: mode to set on new inode
Paul Menageddbcc7e2007-10-18 23:39:30 -07003383 *
Li Zefana043e3b2008-02-23 15:24:09 -08003384 * Must be called with the mutex on the parent inode held
Paul Menageddbcc7e2007-10-18 23:39:30 -07003385 */
Paul Menageddbcc7e2007-10-18 23:39:30 -07003386static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
Li Zefan099fca32009-04-02 16:57:29 -07003387 mode_t mode)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003388{
Paul Menagebd89aab2007-10-18 23:40:44 -07003389 struct cgroup *cgrp;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003390 struct cgroupfs_root *root = parent->root;
3391 int err = 0;
3392 struct cgroup_subsys *ss;
3393 struct super_block *sb = root->sb;
3394
Paul Menagebd89aab2007-10-18 23:40:44 -07003395 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3396 if (!cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003397 return -ENOMEM;
3398
3399 /* Grab a reference on the superblock so the hierarchy doesn't
3400 * get deleted on unmount if there are child cgroups. This
3401 * can be done outside cgroup_mutex, since the sb can't
3402 * disappear while someone has an open control file on the
3403 * fs */
3404 atomic_inc(&sb->s_active);
3405
3406 mutex_lock(&cgroup_mutex);
3407
Paul Menagecc31edc2008-10-18 20:28:04 -07003408 init_cgroup_housekeeping(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003409
Paul Menagebd89aab2007-10-18 23:40:44 -07003410 cgrp->parent = parent;
3411 cgrp->root = parent->root;
3412 cgrp->top_cgroup = parent->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003413
Li Zefanb6abdb02008-03-04 14:28:19 -08003414 if (notify_on_release(parent))
3415 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3416
Daniel Lezcano97978e62010-10-27 15:33:35 -07003417 if (clone_children(parent))
3418 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3419
Paul Menageddbcc7e2007-10-18 23:39:30 -07003420 for_each_subsys(root, ss) {
Paul Menagebd89aab2007-10-18 23:40:44 -07003421 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
Li Zefan4528fd02010-02-02 13:44:10 -08003422
Paul Menageddbcc7e2007-10-18 23:39:30 -07003423 if (IS_ERR(css)) {
3424 err = PTR_ERR(css);
3425 goto err_destroy;
3426 }
Paul Menagebd89aab2007-10-18 23:40:44 -07003427 init_cgroup_css(css, ss, cgrp);
Li Zefan4528fd02010-02-02 13:44:10 -08003428 if (ss->use_id) {
3429 err = alloc_css_id(ss, parent, cgrp);
3430 if (err)
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07003431 goto err_destroy;
Li Zefan4528fd02010-02-02 13:44:10 -08003432 }
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07003433 /* At error, ->destroy() callback has to free assigned ID. */
Daniel Lezcano97978e62010-10-27 15:33:35 -07003434 if (clone_children(parent) && ss->post_clone)
3435 ss->post_clone(ss, cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003436 }
3437
Paul Menage999cd8a2009-01-07 18:08:36 -08003438 cgroup_lock_hierarchy(root);
Paul Menagebd89aab2007-10-18 23:40:44 -07003439 list_add(&cgrp->sibling, &cgrp->parent->children);
Paul Menage999cd8a2009-01-07 18:08:36 -08003440 cgroup_unlock_hierarchy(root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003441 root->number_of_cgroups++;
3442
Paul Menagebd89aab2007-10-18 23:40:44 -07003443 err = cgroup_create_dir(cgrp, dentry, mode);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003444 if (err < 0)
3445 goto err_remove;
3446
3447 /* The cgroup directory was pre-locked for us */
Paul Menagebd89aab2007-10-18 23:40:44 -07003448 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
Paul Menageddbcc7e2007-10-18 23:39:30 -07003449
Paul Menagebd89aab2007-10-18 23:40:44 -07003450 err = cgroup_populate_dir(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003451 /* If err < 0, we have a half-filled directory - oh well ;) */
3452
3453 mutex_unlock(&cgroup_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -07003454 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003455
3456 return 0;
3457
3458 err_remove:
3459
KAMEZAWA Hiroyukibaef99a2009-01-29 14:25:10 -08003460 cgroup_lock_hierarchy(root);
Paul Menagebd89aab2007-10-18 23:40:44 -07003461 list_del(&cgrp->sibling);
KAMEZAWA Hiroyukibaef99a2009-01-29 14:25:10 -08003462 cgroup_unlock_hierarchy(root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003463 root->number_of_cgroups--;
3464
3465 err_destroy:
3466
3467 for_each_subsys(root, ss) {
Paul Menagebd89aab2007-10-18 23:40:44 -07003468 if (cgrp->subsys[ss->subsys_id])
3469 ss->destroy(ss, cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003470 }
3471
3472 mutex_unlock(&cgroup_mutex);
3473
3474 /* Release the reference count that we took on the superblock */
3475 deactivate_super(sb);
3476
Paul Menagebd89aab2007-10-18 23:40:44 -07003477 kfree(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003478 return err;
3479}
3480
3481static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3482{
3483 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3484
3485 /* the vfs holds inode->i_mutex already */
3486 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3487}
3488
Li Zefan55b6fd02008-07-29 22:33:20 -07003489static int cgroup_has_css_refs(struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -07003490{
3491 /* Check the reference count on each subsystem. Since we
3492 * already established that there are no tasks in the
Paul Menagee7c5ec92009-01-07 18:08:38 -08003493 * cgroup, if the css refcount is also 1, then there should
Paul Menage81a6a5c2007-10-18 23:39:38 -07003494 * be no outstanding references, so the subsystem is safe to
3495 * destroy. We scan across all subsystems rather than using
3496 * the per-hierarchy linked list of mounted subsystems since
3497 * we can be called via check_for_release() with no
3498 * synchronization other than RCU, and the subsystem linked
3499 * list isn't RCU-safe */
3500 int i;
Ben Blumaae8aab2010-03-10 15:22:07 -08003501 /*
3502 * We won't need to lock the subsys array, because the subsystems
3503 * we're concerned about aren't going anywhere since our cgroup root
3504 * has a reference on them.
3505 */
Paul Menage81a6a5c2007-10-18 23:39:38 -07003506 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3507 struct cgroup_subsys *ss = subsys[i];
3508 struct cgroup_subsys_state *css;
Ben Blumaae8aab2010-03-10 15:22:07 -08003509 /* Skip subsystems not present or not in this hierarchy */
3510 if (ss == NULL || ss->root != cgrp->root)
Paul Menage81a6a5c2007-10-18 23:39:38 -07003511 continue;
Paul Menagebd89aab2007-10-18 23:40:44 -07003512 css = cgrp->subsys[ss->subsys_id];
Paul Menage81a6a5c2007-10-18 23:39:38 -07003513 /* When called from check_for_release() it's possible
3514 * that by this point the cgroup has been removed
3515 * and the css deleted. But a false-positive doesn't
3516 * matter, since it can only happen if the cgroup
3517 * has been deleted and hence no longer needs the
3518 * release agent to be called anyway. */
Paul Menagee7c5ec92009-01-07 18:08:38 -08003519 if (css && (atomic_read(&css->refcnt) > 1))
Paul Menage81a6a5c2007-10-18 23:39:38 -07003520 return 1;
Paul Menage81a6a5c2007-10-18 23:39:38 -07003521 }
3522 return 0;
3523}
3524
Paul Menagee7c5ec92009-01-07 18:08:38 -08003525/*
3526 * Atomically mark all (or else none) of the cgroup's CSS objects as
3527 * CSS_REMOVED. Return true on success, or false if the cgroup has
3528 * busy subsystems. Call with cgroup_mutex held
3529 */
3530
3531static int cgroup_clear_css_refs(struct cgroup *cgrp)
3532{
3533 struct cgroup_subsys *ss;
3534 unsigned long flags;
3535 bool failed = false;
3536 local_irq_save(flags);
3537 for_each_subsys(cgrp->root, ss) {
3538 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3539 int refcnt;
Paul Menage804b3c22009-01-29 14:25:21 -08003540 while (1) {
Paul Menagee7c5ec92009-01-07 18:08:38 -08003541 /* We can only remove a CSS with a refcnt==1 */
3542 refcnt = atomic_read(&css->refcnt);
3543 if (refcnt > 1) {
3544 failed = true;
3545 goto done;
3546 }
3547 BUG_ON(!refcnt);
3548 /*
3549 * Drop the refcnt to 0 while we check other
3550 * subsystems. This will cause any racing
3551 * css_tryget() to spin until we set the
3552 * CSS_REMOVED bits or abort
3553 */
Paul Menage804b3c22009-01-29 14:25:21 -08003554 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3555 break;
3556 cpu_relax();
3557 }
Paul Menagee7c5ec92009-01-07 18:08:38 -08003558 }
3559 done:
3560 for_each_subsys(cgrp->root, ss) {
3561 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3562 if (failed) {
3563 /*
3564 * Restore old refcnt if we previously managed
3565 * to clear it from 1 to 0
3566 */
3567 if (!atomic_read(&css->refcnt))
3568 atomic_set(&css->refcnt, 1);
3569 } else {
3570 /* Commit the fact that the CSS is removed */
3571 set_bit(CSS_REMOVED, &css->flags);
3572 }
3573 }
3574 local_irq_restore(flags);
3575 return !failed;
3576}
3577
Paul Menageddbcc7e2007-10-18 23:39:30 -07003578static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3579{
Paul Menagebd89aab2007-10-18 23:40:44 -07003580 struct cgroup *cgrp = dentry->d_fsdata;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003581 struct dentry *d;
3582 struct cgroup *parent;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003583 DEFINE_WAIT(wait);
Kirill A. Shutemov4ab78682010-03-10 15:22:34 -08003584 struct cgroup_event *event, *tmp;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003585 int ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003586
3587 /* the vfs holds both inode->i_mutex already */
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003588again:
Paul Menageddbcc7e2007-10-18 23:39:30 -07003589 mutex_lock(&cgroup_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -07003590 if (atomic_read(&cgrp->count) != 0) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07003591 mutex_unlock(&cgroup_mutex);
3592 return -EBUSY;
3593 }
Paul Menagebd89aab2007-10-18 23:40:44 -07003594 if (!list_empty(&cgrp->children)) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07003595 mutex_unlock(&cgroup_mutex);
3596 return -EBUSY;
3597 }
KAMEZAWA Hiroyuki3fa59df2008-11-19 15:36:34 -08003598 mutex_unlock(&cgroup_mutex);
Li Zefana043e3b2008-02-23 15:24:09 -08003599
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -08003600 /*
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003601 * In general, subsystem has no css->refcnt after pre_destroy(). But
3602 * in racy cases, subsystem may have to get css->refcnt after
3603 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3604 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3605 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3606 * and subsystem's reference count handling. Please see css_get/put
3607 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3608 */
3609 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3610
3611 /*
Li Zefana043e3b2008-02-23 15:24:09 -08003612 * Call pre_destroy handlers of subsys. Notify subsystems
3613 * that rmdir() request comes.
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -08003614 */
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003615 ret = cgroup_call_pre_destroy(cgrp);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003616 if (ret) {
3617 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003618 return ret;
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003619 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07003620
KAMEZAWA Hiroyuki3fa59df2008-11-19 15:36:34 -08003621 mutex_lock(&cgroup_mutex);
3622 parent = cgrp->parent;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003623 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003624 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003625 mutex_unlock(&cgroup_mutex);
3626 return -EBUSY;
3627 }
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003628 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003629 if (!cgroup_clear_css_refs(cgrp)) {
3630 mutex_unlock(&cgroup_mutex);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003631 /*
3632 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3633 * prepare_to_wait(), we need to check this flag.
3634 */
3635 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
3636 schedule();
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003637 finish_wait(&cgroup_rmdir_waitq, &wait);
3638 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3639 if (signal_pending(current))
3640 return -EINTR;
3641 goto again;
3642 }
3643 /* NO css_tryget() can success after here. */
3644 finish_wait(&cgroup_rmdir_waitq, &wait);
3645 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003646
Paul Menage81a6a5c2007-10-18 23:39:38 -07003647 spin_lock(&release_list_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07003648 set_bit(CGRP_REMOVED, &cgrp->flags);
3649 if (!list_empty(&cgrp->release_list))
3650 list_del(&cgrp->release_list);
Paul Menage81a6a5c2007-10-18 23:39:38 -07003651 spin_unlock(&release_list_lock);
Paul Menage999cd8a2009-01-07 18:08:36 -08003652
3653 cgroup_lock_hierarchy(cgrp->root);
3654 /* delete this cgroup from parent->children */
Paul Menagebd89aab2007-10-18 23:40:44 -07003655 list_del(&cgrp->sibling);
Paul Menage999cd8a2009-01-07 18:08:36 -08003656 cgroup_unlock_hierarchy(cgrp->root);
3657
Paul Menagebd89aab2007-10-18 23:40:44 -07003658 spin_lock(&cgrp->dentry->d_lock);
3659 d = dget(cgrp->dentry);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003660 spin_unlock(&d->d_lock);
3661
3662 cgroup_d_remove_dir(d);
3663 dput(d);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003664
Paul Menagebd89aab2007-10-18 23:40:44 -07003665 set_bit(CGRP_RELEASABLE, &parent->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -07003666 check_for_release(parent);
3667
Kirill A. Shutemov4ab78682010-03-10 15:22:34 -08003668 /*
3669 * Unregister events and notify userspace.
3670 * Notify userspace about cgroup removing only after rmdir of cgroup
3671 * directory to avoid race between userspace and kernelspace
3672 */
3673 spin_lock(&cgrp->event_list_lock);
3674 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
3675 list_del(&event->list);
3676 remove_wait_queue(event->wqh, &event->wait);
3677 eventfd_signal(event->eventfd, 1);
3678 schedule_work(&event->remove);
3679 }
3680 spin_unlock(&cgrp->event_list_lock);
3681
Paul Menageddbcc7e2007-10-18 23:39:30 -07003682 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003683 return 0;
3684}
3685
Li Zefan06a11922008-04-29 01:00:07 -07003686static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003687{
Paul Menageddbcc7e2007-10-18 23:39:30 -07003688 struct cgroup_subsys_state *css;
Diego Callejacfe36bd2007-11-14 16:58:54 -08003689
3690 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003691
3692 /* Create the top cgroup state for this subsystem */
Li Zefan33a68ac2009-01-07 18:07:42 -08003693 list_add(&ss->sibling, &rootnode.subsys_list);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003694 ss->root = &rootnode;
3695 css = ss->create(ss, dummytop);
3696 /* We don't handle early failures gracefully */
3697 BUG_ON(IS_ERR(css));
3698 init_cgroup_css(css, ss, dummytop);
3699
Li Zefane8d55fd2008-04-29 01:00:13 -07003700 /* Update the init_css_set to contain a subsys
Paul Menage817929e2007-10-18 23:39:36 -07003701 * pointer to this state - since the subsystem is
Li Zefane8d55fd2008-04-29 01:00:13 -07003702 * newly registered, all tasks and hence the
3703 * init_css_set is in the subsystem's top cgroup. */
3704 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
Paul Menageddbcc7e2007-10-18 23:39:30 -07003705
3706 need_forkexit_callback |= ss->fork || ss->exit;
3707
Li Zefane8d55fd2008-04-29 01:00:13 -07003708 /* At system boot, before all subsystems have been
3709 * registered, no tasks have been forked, so we don't
3710 * need to invoke fork callbacks here. */
3711 BUG_ON(!list_empty(&init_task.tasks));
3712
Paul Menage999cd8a2009-01-07 18:08:36 -08003713 mutex_init(&ss->hierarchy_mutex);
Li Zefancfebe562009-02-11 13:04:36 -08003714 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003715 ss->active = 1;
Ben Blume6a11052010-03-10 15:22:09 -08003716
3717 /* this function shouldn't be used with modular subsystems, since they
3718 * need to register a subsys_id, among other things */
3719 BUG_ON(ss->module);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003720}
3721
3722/**
Ben Blume6a11052010-03-10 15:22:09 -08003723 * cgroup_load_subsys: load and register a modular subsystem at runtime
3724 * @ss: the subsystem to load
3725 *
3726 * This function should be called in a modular subsystem's initcall. If the
Thomas Weber88393162010-03-16 11:47:56 +01003727 * subsystem is built as a module, it will be assigned a new subsys_id and set
Ben Blume6a11052010-03-10 15:22:09 -08003728 * up for use. If the subsystem is built-in anyway, work is delegated to the
3729 * simpler cgroup_init_subsys.
3730 */
3731int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
3732{
3733 int i;
3734 struct cgroup_subsys_state *css;
3735
3736 /* check name and function validity */
3737 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
3738 ss->create == NULL || ss->destroy == NULL)
3739 return -EINVAL;
3740
3741 /*
3742 * we don't support callbacks in modular subsystems. this check is
3743 * before the ss->module check for consistency; a subsystem that could
3744 * be a module should still have no callbacks even if the user isn't
3745 * compiling it as one.
3746 */
3747 if (ss->fork || ss->exit)
3748 return -EINVAL;
3749
3750 /*
3751 * an optionally modular subsystem is built-in: we want to do nothing,
3752 * since cgroup_init_subsys will have already taken care of it.
3753 */
3754 if (ss->module == NULL) {
3755 /* a few sanity checks */
3756 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
3757 BUG_ON(subsys[ss->subsys_id] != ss);
3758 return 0;
3759 }
3760
3761 /*
3762 * need to register a subsys id before anything else - for example,
3763 * init_cgroup_css needs it.
3764 */
3765 mutex_lock(&cgroup_mutex);
3766 /* find the first empty slot in the array */
3767 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
3768 if (subsys[i] == NULL)
3769 break;
3770 }
3771 if (i == CGROUP_SUBSYS_COUNT) {
3772 /* maximum number of subsystems already registered! */
3773 mutex_unlock(&cgroup_mutex);
3774 return -EBUSY;
3775 }
3776 /* assign ourselves the subsys_id */
3777 ss->subsys_id = i;
3778 subsys[i] = ss;
3779
3780 /*
3781 * no ss->create seems to need anything important in the ss struct, so
3782 * this can happen first (i.e. before the rootnode attachment).
3783 */
3784 css = ss->create(ss, dummytop);
3785 if (IS_ERR(css)) {
3786 /* failure case - need to deassign the subsys[] slot. */
3787 subsys[i] = NULL;
3788 mutex_unlock(&cgroup_mutex);
3789 return PTR_ERR(css);
3790 }
3791
3792 list_add(&ss->sibling, &rootnode.subsys_list);
3793 ss->root = &rootnode;
3794
3795 /* our new subsystem will be attached to the dummy hierarchy. */
3796 init_cgroup_css(css, ss, dummytop);
3797 /* init_idr must be after init_cgroup_css because it sets css->id. */
3798 if (ss->use_id) {
3799 int ret = cgroup_init_idr(ss, css);
3800 if (ret) {
3801 dummytop->subsys[ss->subsys_id] = NULL;
3802 ss->destroy(ss, dummytop);
3803 subsys[i] = NULL;
3804 mutex_unlock(&cgroup_mutex);
3805 return ret;
3806 }
3807 }
3808
3809 /*
3810 * Now we need to entangle the css into the existing css_sets. unlike
3811 * in cgroup_init_subsys, there are now multiple css_sets, so each one
3812 * will need a new pointer to it; done by iterating the css_set_table.
3813 * furthermore, modifying the existing css_sets will corrupt the hash
3814 * table state, so each changed css_set will need its hash recomputed.
3815 * this is all done under the css_set_lock.
3816 */
3817 write_lock(&css_set_lock);
3818 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
3819 struct css_set *cg;
3820 struct hlist_node *node, *tmp;
3821 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
3822
3823 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
3824 /* skip entries that we already rehashed */
3825 if (cg->subsys[ss->subsys_id])
3826 continue;
3827 /* remove existing entry */
3828 hlist_del(&cg->hlist);
3829 /* set new value */
3830 cg->subsys[ss->subsys_id] = css;
3831 /* recompute hash and restore entry */
3832 new_bucket = css_set_hash(cg->subsys);
3833 hlist_add_head(&cg->hlist, new_bucket);
3834 }
3835 }
3836 write_unlock(&css_set_lock);
3837
3838 mutex_init(&ss->hierarchy_mutex);
3839 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3840 ss->active = 1;
3841
Ben Blume6a11052010-03-10 15:22:09 -08003842 /* success! */
3843 mutex_unlock(&cgroup_mutex);
3844 return 0;
3845}
3846EXPORT_SYMBOL_GPL(cgroup_load_subsys);
3847
3848/**
Ben Blumcf5d5942010-03-10 15:22:09 -08003849 * cgroup_unload_subsys: unload a modular subsystem
3850 * @ss: the subsystem to unload
3851 *
3852 * This function should be called in a modular subsystem's exitcall. When this
3853 * function is invoked, the refcount on the subsystem's module will be 0, so
3854 * the subsystem will not be attached to any hierarchy.
3855 */
3856void cgroup_unload_subsys(struct cgroup_subsys *ss)
3857{
3858 struct cg_cgroup_link *link;
3859 struct hlist_head *hhead;
3860
3861 BUG_ON(ss->module == NULL);
3862
3863 /*
3864 * we shouldn't be called if the subsystem is in use, and the use of
3865 * try_module_get in parse_cgroupfs_options should ensure that it
3866 * doesn't start being used while we're killing it off.
3867 */
3868 BUG_ON(ss->root != &rootnode);
3869
3870 mutex_lock(&cgroup_mutex);
3871 /* deassign the subsys_id */
3872 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
3873 subsys[ss->subsys_id] = NULL;
3874
3875 /* remove subsystem from rootnode's list of subsystems */
3876 list_del(&ss->sibling);
3877
3878 /*
3879 * disentangle the css from all css_sets attached to the dummytop. as
3880 * in loading, we need to pay our respects to the hashtable gods.
3881 */
3882 write_lock(&css_set_lock);
3883 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
3884 struct css_set *cg = link->cg;
3885
3886 hlist_del(&cg->hlist);
3887 BUG_ON(!cg->subsys[ss->subsys_id]);
3888 cg->subsys[ss->subsys_id] = NULL;
3889 hhead = css_set_hash(cg->subsys);
3890 hlist_add_head(&cg->hlist, hhead);
3891 }
3892 write_unlock(&css_set_lock);
3893
3894 /*
3895 * remove subsystem's css from the dummytop and free it - need to free
3896 * before marking as null because ss->destroy needs the cgrp->subsys
3897 * pointer to find their state. note that this also takes care of
3898 * freeing the css_id.
3899 */
3900 ss->destroy(ss, dummytop);
3901 dummytop->subsys[ss->subsys_id] = NULL;
3902
3903 mutex_unlock(&cgroup_mutex);
3904}
3905EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
3906
3907/**
Li Zefana043e3b2008-02-23 15:24:09 -08003908 * cgroup_init_early - cgroup initialization at system boot
3909 *
3910 * Initialize cgroups at system boot, and initialize any
3911 * subsystems that request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07003912 */
3913int __init cgroup_init_early(void)
3914{
3915 int i;
Lai Jiangshan146aa1b2008-10-18 20:28:03 -07003916 atomic_set(&init_css_set.refcount, 1);
Paul Menage817929e2007-10-18 23:39:36 -07003917 INIT_LIST_HEAD(&init_css_set.cg_links);
3918 INIT_LIST_HEAD(&init_css_set.tasks);
Li Zefan472b1052008-04-29 01:00:11 -07003919 INIT_HLIST_NODE(&init_css_set.hlist);
Paul Menage817929e2007-10-18 23:39:36 -07003920 css_set_count = 1;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003921 init_cgroup_root(&rootnode);
Paul Menage817929e2007-10-18 23:39:36 -07003922 root_count = 1;
3923 init_task.cgroups = &init_css_set;
3924
3925 init_css_set_link.cg = &init_css_set;
Paul Menage7717f7b2009-09-23 15:56:22 -07003926 init_css_set_link.cgrp = dummytop;
Paul Menagebd89aab2007-10-18 23:40:44 -07003927 list_add(&init_css_set_link.cgrp_link_list,
Paul Menage817929e2007-10-18 23:39:36 -07003928 &rootnode.top_cgroup.css_sets);
3929 list_add(&init_css_set_link.cg_link_list,
3930 &init_css_set.cg_links);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003931
Li Zefan472b1052008-04-29 01:00:11 -07003932 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
3933 INIT_HLIST_HEAD(&css_set_table[i]);
3934
Ben Blumaae8aab2010-03-10 15:22:07 -08003935 /* at bootup time, we don't worry about modular subsystems */
3936 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07003937 struct cgroup_subsys *ss = subsys[i];
3938
3939 BUG_ON(!ss->name);
3940 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
3941 BUG_ON(!ss->create);
3942 BUG_ON(!ss->destroy);
3943 if (ss->subsys_id != i) {
Diego Callejacfe36bd2007-11-14 16:58:54 -08003944 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
Paul Menageddbcc7e2007-10-18 23:39:30 -07003945 ss->name, ss->subsys_id);
3946 BUG();
3947 }
3948
3949 if (ss->early_init)
3950 cgroup_init_subsys(ss);
3951 }
3952 return 0;
3953}
3954
3955/**
Li Zefana043e3b2008-02-23 15:24:09 -08003956 * cgroup_init - cgroup initialization
3957 *
3958 * Register cgroup filesystem and /proc file, and initialize
3959 * any subsystems that didn't request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07003960 */
3961int __init cgroup_init(void)
3962{
3963 int err;
3964 int i;
Li Zefan472b1052008-04-29 01:00:11 -07003965 struct hlist_head *hhead;
Paul Menagea4243162007-10-18 23:39:35 -07003966
3967 err = bdi_init(&cgroup_backing_dev_info);
3968 if (err)
3969 return err;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003970
Ben Blumaae8aab2010-03-10 15:22:07 -08003971 /* at bootup time, we don't worry about modular subsystems */
3972 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07003973 struct cgroup_subsys *ss = subsys[i];
3974 if (!ss->early_init)
3975 cgroup_init_subsys(ss);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07003976 if (ss->use_id)
Ben Blume6a11052010-03-10 15:22:09 -08003977 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003978 }
3979
Li Zefan472b1052008-04-29 01:00:11 -07003980 /* Add init_css_set to the hash table */
3981 hhead = css_set_hash(init_css_set.subsys);
3982 hlist_add_head(&init_css_set.hlist, hhead);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07003983 BUG_ON(!init_root_id(&rootnode));
Greg KH676db4a2010-08-05 13:53:35 -07003984
3985 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
3986 if (!cgroup_kobj) {
3987 err = -ENOMEM;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003988 goto out;
Greg KH676db4a2010-08-05 13:53:35 -07003989 }
3990
3991 err = register_filesystem(&cgroup_fs_type);
3992 if (err < 0) {
3993 kobject_put(cgroup_kobj);
3994 goto out;
3995 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07003996
Li Zefan46ae2202008-04-29 01:00:08 -07003997 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
Paul Menagea4243162007-10-18 23:39:35 -07003998
Paul Menageddbcc7e2007-10-18 23:39:30 -07003999out:
Paul Menagea4243162007-10-18 23:39:35 -07004000 if (err)
4001 bdi_destroy(&cgroup_backing_dev_info);
4002
Paul Menageddbcc7e2007-10-18 23:39:30 -07004003 return err;
4004}
Paul Menageb4f48b62007-10-18 23:39:33 -07004005
Paul Menagea4243162007-10-18 23:39:35 -07004006/*
4007 * proc_cgroup_show()
4008 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4009 * - Used for /proc/<pid>/cgroup.
4010 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4011 * doesn't really matter if tsk->cgroup changes after we read it,
Cliff Wickman956db3c2008-02-07 00:14:43 -08004012 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
Paul Menagea4243162007-10-18 23:39:35 -07004013 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4014 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4015 * cgroup to top_cgroup.
4016 */
4017
4018/* TODO: Use a proper seq_file iterator */
4019static int proc_cgroup_show(struct seq_file *m, void *v)
4020{
4021 struct pid *pid;
4022 struct task_struct *tsk;
4023 char *buf;
4024 int retval;
4025 struct cgroupfs_root *root;
4026
4027 retval = -ENOMEM;
4028 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4029 if (!buf)
4030 goto out;
4031
4032 retval = -ESRCH;
4033 pid = m->private;
4034 tsk = get_pid_task(pid, PIDTYPE_PID);
4035 if (!tsk)
4036 goto out_free;
4037
4038 retval = 0;
4039
4040 mutex_lock(&cgroup_mutex);
4041
Li Zefane5f6a862009-01-07 18:07:41 -08004042 for_each_active_root(root) {
Paul Menagea4243162007-10-18 23:39:35 -07004043 struct cgroup_subsys *ss;
Paul Menagebd89aab2007-10-18 23:40:44 -07004044 struct cgroup *cgrp;
Paul Menagea4243162007-10-18 23:39:35 -07004045 int count = 0;
4046
Paul Menage2c6ab6d2009-09-23 15:56:23 -07004047 seq_printf(m, "%d:", root->hierarchy_id);
Paul Menagea4243162007-10-18 23:39:35 -07004048 for_each_subsys(root, ss)
4049 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
Paul Menagec6d57f32009-09-23 15:56:19 -07004050 if (strlen(root->name))
4051 seq_printf(m, "%sname=%s", count ? "," : "",
4052 root->name);
Paul Menagea4243162007-10-18 23:39:35 -07004053 seq_putc(m, ':');
Paul Menage7717f7b2009-09-23 15:56:22 -07004054 cgrp = task_cgroup_from_root(tsk, root);
Paul Menagebd89aab2007-10-18 23:40:44 -07004055 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
Paul Menagea4243162007-10-18 23:39:35 -07004056 if (retval < 0)
4057 goto out_unlock;
4058 seq_puts(m, buf);
4059 seq_putc(m, '\n');
4060 }
4061
4062out_unlock:
4063 mutex_unlock(&cgroup_mutex);
4064 put_task_struct(tsk);
4065out_free:
4066 kfree(buf);
4067out:
4068 return retval;
4069}
4070
4071static int cgroup_open(struct inode *inode, struct file *file)
4072{
4073 struct pid *pid = PROC_I(inode)->pid;
4074 return single_open(file, proc_cgroup_show, pid);
4075}
4076
Alexey Dobriyan828c0952009-10-01 15:43:56 -07004077const struct file_operations proc_cgroup_operations = {
Paul Menagea4243162007-10-18 23:39:35 -07004078 .open = cgroup_open,
4079 .read = seq_read,
4080 .llseek = seq_lseek,
4081 .release = single_release,
4082};
4083
4084/* Display information about each subsystem and each hierarchy */
4085static int proc_cgroupstats_show(struct seq_file *m, void *v)
4086{
4087 int i;
Paul Menagea4243162007-10-18 23:39:35 -07004088
Paul Menage8bab8dd2008-04-04 14:29:57 -07004089 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
Ben Blumaae8aab2010-03-10 15:22:07 -08004090 /*
4091 * ideally we don't want subsystems moving around while we do this.
4092 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4093 * subsys/hierarchy state.
4094 */
Paul Menagea4243162007-10-18 23:39:35 -07004095 mutex_lock(&cgroup_mutex);
Paul Menagea4243162007-10-18 23:39:35 -07004096 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4097 struct cgroup_subsys *ss = subsys[i];
Ben Blumaae8aab2010-03-10 15:22:07 -08004098 if (ss == NULL)
4099 continue;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07004100 seq_printf(m, "%s\t%d\t%d\t%d\n",
4101 ss->name, ss->root->hierarchy_id,
Paul Menage8bab8dd2008-04-04 14:29:57 -07004102 ss->root->number_of_cgroups, !ss->disabled);
Paul Menagea4243162007-10-18 23:39:35 -07004103 }
4104 mutex_unlock(&cgroup_mutex);
4105 return 0;
4106}
4107
4108static int cgroupstats_open(struct inode *inode, struct file *file)
4109{
Al Viro9dce07f2008-03-29 03:07:28 +00004110 return single_open(file, proc_cgroupstats_show, NULL);
Paul Menagea4243162007-10-18 23:39:35 -07004111}
4112
Alexey Dobriyan828c0952009-10-01 15:43:56 -07004113static const struct file_operations proc_cgroupstats_operations = {
Paul Menagea4243162007-10-18 23:39:35 -07004114 .open = cgroupstats_open,
4115 .read = seq_read,
4116 .llseek = seq_lseek,
4117 .release = single_release,
4118};
4119
Paul Menageb4f48b62007-10-18 23:39:33 -07004120/**
4121 * cgroup_fork - attach newly forked task to its parents cgroup.
Li Zefana043e3b2008-02-23 15:24:09 -08004122 * @child: pointer to task_struct of forking parent process.
Paul Menageb4f48b62007-10-18 23:39:33 -07004123 *
4124 * Description: A task inherits its parent's cgroup at fork().
4125 *
4126 * A pointer to the shared css_set was automatically copied in
4127 * fork.c by dup_task_struct(). However, we ignore that copy, since
4128 * it was not made under the protection of RCU or cgroup_mutex, so
Cliff Wickman956db3c2008-02-07 00:14:43 -08004129 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
Paul Menage817929e2007-10-18 23:39:36 -07004130 * have already changed current->cgroups, allowing the previously
4131 * referenced cgroup group to be removed and freed.
Paul Menageb4f48b62007-10-18 23:39:33 -07004132 *
4133 * At the point that cgroup_fork() is called, 'current' is the parent
4134 * task, and the passed argument 'child' points to the child task.
4135 */
4136void cgroup_fork(struct task_struct *child)
4137{
Paul Menage817929e2007-10-18 23:39:36 -07004138 task_lock(current);
4139 child->cgroups = current->cgroups;
4140 get_css_set(child->cgroups);
4141 task_unlock(current);
4142 INIT_LIST_HEAD(&child->cg_list);
Paul Menageb4f48b62007-10-18 23:39:33 -07004143}
4144
4145/**
Li Zefana043e3b2008-02-23 15:24:09 -08004146 * cgroup_fork_callbacks - run fork callbacks
4147 * @child: the new task
4148 *
4149 * Called on a new task very soon before adding it to the
4150 * tasklist. No need to take any locks since no-one can
4151 * be operating on this task.
Paul Menageb4f48b62007-10-18 23:39:33 -07004152 */
4153void cgroup_fork_callbacks(struct task_struct *child)
4154{
4155 if (need_forkexit_callback) {
4156 int i;
Ben Blumaae8aab2010-03-10 15:22:07 -08004157 /*
4158 * forkexit callbacks are only supported for builtin
4159 * subsystems, and the builtin section of the subsys array is
4160 * immutable, so we don't need to lock the subsys array here.
4161 */
4162 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
Paul Menageb4f48b62007-10-18 23:39:33 -07004163 struct cgroup_subsys *ss = subsys[i];
4164 if (ss->fork)
4165 ss->fork(ss, child);
4166 }
4167 }
4168}
4169
4170/**
Li Zefana043e3b2008-02-23 15:24:09 -08004171 * cgroup_post_fork - called on a new task after adding it to the task list
4172 * @child: the task in question
4173 *
4174 * Adds the task to the list running through its css_set if necessary.
4175 * Has to be after the task is visible on the task list in case we race
4176 * with the first call to cgroup_iter_start() - to guarantee that the
4177 * new task ends up on its list.
4178 */
Paul Menage817929e2007-10-18 23:39:36 -07004179void cgroup_post_fork(struct task_struct *child)
4180{
4181 if (use_task_css_set_links) {
4182 write_lock(&css_set_lock);
Lai Jiangshanb12b5332009-01-07 18:07:36 -08004183 task_lock(child);
Paul Menage817929e2007-10-18 23:39:36 -07004184 if (list_empty(&child->cg_list))
4185 list_add(&child->cg_list, &child->cgroups->tasks);
Lai Jiangshanb12b5332009-01-07 18:07:36 -08004186 task_unlock(child);
Paul Menage817929e2007-10-18 23:39:36 -07004187 write_unlock(&css_set_lock);
4188 }
4189}
4190/**
Paul Menageb4f48b62007-10-18 23:39:33 -07004191 * cgroup_exit - detach cgroup from exiting task
4192 * @tsk: pointer to task_struct of exiting process
Li Zefana043e3b2008-02-23 15:24:09 -08004193 * @run_callback: run exit callbacks?
Paul Menageb4f48b62007-10-18 23:39:33 -07004194 *
4195 * Description: Detach cgroup from @tsk and release it.
4196 *
4197 * Note that cgroups marked notify_on_release force every task in
4198 * them to take the global cgroup_mutex mutex when exiting.
4199 * This could impact scaling on very large systems. Be reluctant to
4200 * use notify_on_release cgroups where very high task exit scaling
4201 * is required on large systems.
4202 *
4203 * the_top_cgroup_hack:
4204 *
4205 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4206 *
4207 * We call cgroup_exit() while the task is still competent to
4208 * handle notify_on_release(), then leave the task attached to the
4209 * root cgroup in each hierarchy for the remainder of its exit.
4210 *
4211 * To do this properly, we would increment the reference count on
4212 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4213 * code we would add a second cgroup function call, to drop that
4214 * reference. This would just create an unnecessary hot spot on
4215 * the top_cgroup reference count, to no avail.
4216 *
4217 * Normally, holding a reference to a cgroup without bumping its
4218 * count is unsafe. The cgroup could go away, or someone could
4219 * attach us to a different cgroup, decrementing the count on
4220 * the first cgroup that we never incremented. But in this case,
4221 * top_cgroup isn't going away, and either task has PF_EXITING set,
Cliff Wickman956db3c2008-02-07 00:14:43 -08004222 * which wards off any cgroup_attach_task() attempts, or task is a failed
4223 * fork, never visible to cgroup_attach_task.
Paul Menageb4f48b62007-10-18 23:39:33 -07004224 */
4225void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4226{
4227 int i;
Paul Menage817929e2007-10-18 23:39:36 -07004228 struct css_set *cg;
Paul Menageb4f48b62007-10-18 23:39:33 -07004229
4230 if (run_callbacks && need_forkexit_callback) {
Ben Blumaae8aab2010-03-10 15:22:07 -08004231 /*
4232 * modular subsystems can't use callbacks, so no need to lock
4233 * the subsys array
4234 */
4235 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
Paul Menageb4f48b62007-10-18 23:39:33 -07004236 struct cgroup_subsys *ss = subsys[i];
4237 if (ss->exit)
4238 ss->exit(ss, tsk);
4239 }
4240 }
Paul Menage817929e2007-10-18 23:39:36 -07004241
4242 /*
4243 * Unlink from the css_set task list if necessary.
4244 * Optimistically check cg_list before taking
4245 * css_set_lock
4246 */
4247 if (!list_empty(&tsk->cg_list)) {
4248 write_lock(&css_set_lock);
4249 if (!list_empty(&tsk->cg_list))
4250 list_del(&tsk->cg_list);
4251 write_unlock(&css_set_lock);
4252 }
4253
Paul Menageb4f48b62007-10-18 23:39:33 -07004254 /* Reassign the task to the init_css_set. */
4255 task_lock(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07004256 cg = tsk->cgroups;
4257 tsk->cgroups = &init_css_set;
Paul Menageb4f48b62007-10-18 23:39:33 -07004258 task_unlock(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07004259 if (cg)
Paul Menage81a6a5c2007-10-18 23:39:38 -07004260 put_css_set_taskexit(cg);
Paul Menageb4f48b62007-10-18 23:39:33 -07004261}
Paul Menage697f4162007-10-18 23:39:34 -07004262
4263/**
Li Zefana043e3b2008-02-23 15:24:09 -08004264 * cgroup_clone - clone the cgroup the given subsystem is attached to
4265 * @tsk: the task to be moved
4266 * @subsys: the given subsystem
Serge E. Hallyne885dcd2008-07-25 01:47:06 -07004267 * @nodename: the name for the new cgroup
Li Zefana043e3b2008-02-23 15:24:09 -08004268 *
4269 * Duplicate the current cgroup in the hierarchy that the given
4270 * subsystem is attached to, and move this task into the new
4271 * child.
Paul Menage697f4162007-10-18 23:39:34 -07004272 */
Serge E. Hallyne885dcd2008-07-25 01:47:06 -07004273int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys,
4274 char *nodename)
Paul Menage697f4162007-10-18 23:39:34 -07004275{
4276 struct dentry *dentry;
4277 int ret = 0;
Paul Menage697f4162007-10-18 23:39:34 -07004278 struct cgroup *parent, *child;
4279 struct inode *inode;
4280 struct css_set *cg;
4281 struct cgroupfs_root *root;
4282 struct cgroup_subsys *ss;
4283
4284 /* We shouldn't be called by an unregistered subsystem */
4285 BUG_ON(!subsys->active);
4286
4287 /* First figure out what hierarchy and cgroup we're dealing
4288 * with, and pin them so we can drop cgroup_mutex */
4289 mutex_lock(&cgroup_mutex);
4290 again:
4291 root = subsys->root;
4292 if (root == &rootnode) {
Paul Menage697f4162007-10-18 23:39:34 -07004293 mutex_unlock(&cgroup_mutex);
4294 return 0;
4295 }
Paul Menage697f4162007-10-18 23:39:34 -07004296
Paul Menage697f4162007-10-18 23:39:34 -07004297 /* Pin the hierarchy */
Li Zefan1404f062009-01-29 14:25:21 -08004298 if (!atomic_inc_not_zero(&root->sb->s_active)) {
Li Zefan7b574b72009-01-04 12:00:45 -08004299 /* We race with the final deactivate_super() */
4300 mutex_unlock(&cgroup_mutex);
4301 return 0;
4302 }
Paul Menage697f4162007-10-18 23:39:34 -07004303
Paul Menage817929e2007-10-18 23:39:36 -07004304 /* Keep the cgroup alive */
Li Zefan1404f062009-01-29 14:25:21 -08004305 task_lock(tsk);
4306 parent = task_cgroup(tsk, subsys->subsys_id);
4307 cg = tsk->cgroups;
Paul Menage817929e2007-10-18 23:39:36 -07004308 get_css_set(cg);
Lai Jiangshan104cbd52009-01-07 18:07:38 -08004309 task_unlock(tsk);
Li Zefan1404f062009-01-29 14:25:21 -08004310
Paul Menage697f4162007-10-18 23:39:34 -07004311 mutex_unlock(&cgroup_mutex);
4312
4313 /* Now do the VFS work to create a cgroup */
4314 inode = parent->dentry->d_inode;
4315
4316 /* Hold the parent directory mutex across this operation to
4317 * stop anyone else deleting the new cgroup */
4318 mutex_lock(&inode->i_mutex);
4319 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
4320 if (IS_ERR(dentry)) {
4321 printk(KERN_INFO
Diego Callejacfe36bd2007-11-14 16:58:54 -08004322 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
Paul Menage697f4162007-10-18 23:39:34 -07004323 PTR_ERR(dentry));
4324 ret = PTR_ERR(dentry);
4325 goto out_release;
4326 }
4327
4328 /* Create the cgroup directory, which also creates the cgroup */
Li Zefan75139b82009-01-07 18:07:33 -08004329 ret = vfs_mkdir(inode, dentry, 0755);
Paul Menagebd89aab2007-10-18 23:40:44 -07004330 child = __d_cgrp(dentry);
Paul Menage697f4162007-10-18 23:39:34 -07004331 dput(dentry);
4332 if (ret) {
4333 printk(KERN_INFO
4334 "Failed to create cgroup %s: %d\n", nodename,
4335 ret);
4336 goto out_release;
4337 }
4338
Paul Menage697f4162007-10-18 23:39:34 -07004339 /* The cgroup now exists. Retake cgroup_mutex and check
4340 * that we're still in the same state that we thought we
4341 * were. */
4342 mutex_lock(&cgroup_mutex);
4343 if ((root != subsys->root) ||
4344 (parent != task_cgroup(tsk, subsys->subsys_id))) {
4345 /* Aargh, we raced ... */
4346 mutex_unlock(&inode->i_mutex);
Paul Menage817929e2007-10-18 23:39:36 -07004347 put_css_set(cg);
Paul Menage697f4162007-10-18 23:39:34 -07004348
Li Zefan1404f062009-01-29 14:25:21 -08004349 deactivate_super(root->sb);
Paul Menage697f4162007-10-18 23:39:34 -07004350 /* The cgroup is still accessible in the VFS, but
4351 * we're not going to try to rmdir() it at this
4352 * point. */
4353 printk(KERN_INFO
4354 "Race in cgroup_clone() - leaking cgroup %s\n",
4355 nodename);
4356 goto again;
4357 }
4358
4359 /* do any required auto-setup */
4360 for_each_subsys(root, ss) {
4361 if (ss->post_clone)
4362 ss->post_clone(ss, child);
4363 }
4364
4365 /* All seems fine. Finish by moving the task into the new cgroup */
Cliff Wickman956db3c2008-02-07 00:14:43 -08004366 ret = cgroup_attach_task(child, tsk);
Paul Menage697f4162007-10-18 23:39:34 -07004367 mutex_unlock(&cgroup_mutex);
4368
4369 out_release:
4370 mutex_unlock(&inode->i_mutex);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004371
4372 mutex_lock(&cgroup_mutex);
Paul Menage817929e2007-10-18 23:39:36 -07004373 put_css_set(cg);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004374 mutex_unlock(&cgroup_mutex);
Li Zefan1404f062009-01-29 14:25:21 -08004375 deactivate_super(root->sb);
Paul Menage697f4162007-10-18 23:39:34 -07004376 return ret;
4377}
4378
Li Zefana043e3b2008-02-23 15:24:09 -08004379/**
Grzegorz Nosek313e9242009-04-02 16:57:23 -07004380 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
Li Zefana043e3b2008-02-23 15:24:09 -08004381 * @cgrp: the cgroup in question
Grzegorz Nosek313e9242009-04-02 16:57:23 -07004382 * @task: the task in question
Li Zefana043e3b2008-02-23 15:24:09 -08004383 *
Grzegorz Nosek313e9242009-04-02 16:57:23 -07004384 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4385 * hierarchy.
Paul Menage697f4162007-10-18 23:39:34 -07004386 *
4387 * If we are sending in dummytop, then presumably we are creating
4388 * the top cgroup in the subsystem.
4389 *
4390 * Called only by the ns (nsproxy) cgroup.
4391 */
Grzegorz Nosek313e9242009-04-02 16:57:23 -07004392int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
Paul Menage697f4162007-10-18 23:39:34 -07004393{
4394 int ret;
4395 struct cgroup *target;
Paul Menage697f4162007-10-18 23:39:34 -07004396
Paul Menagebd89aab2007-10-18 23:40:44 -07004397 if (cgrp == dummytop)
Paul Menage697f4162007-10-18 23:39:34 -07004398 return 1;
4399
Paul Menage7717f7b2009-09-23 15:56:22 -07004400 target = task_cgroup_from_root(task, cgrp->root);
Paul Menagebd89aab2007-10-18 23:40:44 -07004401 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4402 cgrp = cgrp->parent;
4403 ret = (cgrp == target);
Paul Menage697f4162007-10-18 23:39:34 -07004404 return ret;
4405}
Paul Menage81a6a5c2007-10-18 23:39:38 -07004406
Paul Menagebd89aab2007-10-18 23:40:44 -07004407static void check_for_release(struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -07004408{
4409 /* All of these checks rely on RCU to keep the cgroup
4410 * structure alive */
Paul Menagebd89aab2007-10-18 23:40:44 -07004411 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4412 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
Paul Menage81a6a5c2007-10-18 23:39:38 -07004413 /* Control Group is currently removeable. If it's not
4414 * already queued for a userspace notification, queue
4415 * it now */
4416 int need_schedule_work = 0;
4417 spin_lock(&release_list_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07004418 if (!cgroup_is_removed(cgrp) &&
4419 list_empty(&cgrp->release_list)) {
4420 list_add(&cgrp->release_list, &release_list);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004421 need_schedule_work = 1;
4422 }
4423 spin_unlock(&release_list_lock);
4424 if (need_schedule_work)
4425 schedule_work(&release_agent_work);
4426 }
4427}
4428
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -08004429/* Caller must verify that the css is not for root cgroup */
4430void __css_put(struct cgroup_subsys_state *css, int count)
Paul Menage81a6a5c2007-10-18 23:39:38 -07004431{
Paul Menagebd89aab2007-10-18 23:40:44 -07004432 struct cgroup *cgrp = css->cgroup;
KAMEZAWA Hiroyuki3dece832009-10-01 15:44:09 -07004433 int val;
Paul Menage81a6a5c2007-10-18 23:39:38 -07004434 rcu_read_lock();
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -08004435 val = atomic_sub_return(count, &css->refcnt);
KAMEZAWA Hiroyuki3dece832009-10-01 15:44:09 -07004436 if (val == 1) {
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004437 if (notify_on_release(cgrp)) {
4438 set_bit(CGRP_RELEASABLE, &cgrp->flags);
4439 check_for_release(cgrp);
4440 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07004441 cgroup_wakeup_rmdir_waiter(cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004442 }
4443 rcu_read_unlock();
KAMEZAWA Hiroyuki3dece832009-10-01 15:44:09 -07004444 WARN_ON_ONCE(val < 1);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004445}
Ben Blum67523c42010-03-10 15:22:11 -08004446EXPORT_SYMBOL_GPL(__css_put);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004447
4448/*
4449 * Notify userspace when a cgroup is released, by running the
4450 * configured release agent with the name of the cgroup (path
4451 * relative to the root of cgroup file system) as the argument.
4452 *
4453 * Most likely, this user command will try to rmdir this cgroup.
4454 *
4455 * This races with the possibility that some other task will be
4456 * attached to this cgroup before it is removed, or that some other
4457 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4458 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4459 * unused, and this cgroup will be reprieved from its death sentence,
4460 * to continue to serve a useful existence. Next time it's released,
4461 * we will get notified again, if it still has 'notify_on_release' set.
4462 *
4463 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4464 * means only wait until the task is successfully execve()'d. The
4465 * separate release agent task is forked by call_usermodehelper(),
4466 * then control in this thread returns here, without waiting for the
4467 * release agent task. We don't bother to wait because the caller of
4468 * this routine has no use for the exit status of the release agent
4469 * task, so no sense holding our caller up for that.
Paul Menage81a6a5c2007-10-18 23:39:38 -07004470 */
Paul Menage81a6a5c2007-10-18 23:39:38 -07004471static void cgroup_release_agent(struct work_struct *work)
4472{
4473 BUG_ON(work != &release_agent_work);
4474 mutex_lock(&cgroup_mutex);
4475 spin_lock(&release_list_lock);
4476 while (!list_empty(&release_list)) {
4477 char *argv[3], *envp[3];
4478 int i;
Paul Menagee788e062008-07-25 01:46:59 -07004479 char *pathbuf = NULL, *agentbuf = NULL;
Paul Menagebd89aab2007-10-18 23:40:44 -07004480 struct cgroup *cgrp = list_entry(release_list.next,
Paul Menage81a6a5c2007-10-18 23:39:38 -07004481 struct cgroup,
4482 release_list);
Paul Menagebd89aab2007-10-18 23:40:44 -07004483 list_del_init(&cgrp->release_list);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004484 spin_unlock(&release_list_lock);
4485 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
Paul Menagee788e062008-07-25 01:46:59 -07004486 if (!pathbuf)
4487 goto continue_free;
4488 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4489 goto continue_free;
4490 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4491 if (!agentbuf)
4492 goto continue_free;
Paul Menage81a6a5c2007-10-18 23:39:38 -07004493
4494 i = 0;
Paul Menagee788e062008-07-25 01:46:59 -07004495 argv[i++] = agentbuf;
4496 argv[i++] = pathbuf;
Paul Menage81a6a5c2007-10-18 23:39:38 -07004497 argv[i] = NULL;
4498
4499 i = 0;
4500 /* minimal command environment */
4501 envp[i++] = "HOME=/";
4502 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4503 envp[i] = NULL;
4504
4505 /* Drop the lock while we invoke the usermode helper,
4506 * since the exec could involve hitting disk and hence
4507 * be a slow process */
4508 mutex_unlock(&cgroup_mutex);
4509 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004510 mutex_lock(&cgroup_mutex);
Paul Menagee788e062008-07-25 01:46:59 -07004511 continue_free:
4512 kfree(pathbuf);
4513 kfree(agentbuf);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004514 spin_lock(&release_list_lock);
4515 }
4516 spin_unlock(&release_list_lock);
4517 mutex_unlock(&cgroup_mutex);
4518}
Paul Menage8bab8dd2008-04-04 14:29:57 -07004519
4520static int __init cgroup_disable(char *str)
4521{
4522 int i;
4523 char *token;
4524
4525 while ((token = strsep(&str, ",")) != NULL) {
4526 if (!*token)
4527 continue;
Ben Blumaae8aab2010-03-10 15:22:07 -08004528 /*
4529 * cgroup_disable, being at boot time, can't know about module
4530 * subsystems, so we don't worry about them.
4531 */
4532 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
Paul Menage8bab8dd2008-04-04 14:29:57 -07004533 struct cgroup_subsys *ss = subsys[i];
4534
4535 if (!strcmp(token, ss->name)) {
4536 ss->disabled = 1;
4537 printk(KERN_INFO "Disabling %s control group"
4538 " subsystem\n", ss->name);
4539 break;
4540 }
4541 }
4542 }
4543 return 1;
4544}
4545__setup("cgroup_disable=", cgroup_disable);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004546
4547/*
4548 * Functons for CSS ID.
4549 */
4550
4551/*
4552 *To get ID other than 0, this should be called when !cgroup_is_removed().
4553 */
4554unsigned short css_id(struct cgroup_subsys_state *css)
4555{
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004556 struct css_id *cssid;
4557
4558 /*
4559 * This css_id() can return correct value when somone has refcnt
4560 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4561 * it's unchanged until freed.
4562 */
4563 cssid = rcu_dereference_check(css->id,
4564 rcu_read_lock_held() || atomic_read(&css->refcnt));
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004565
4566 if (cssid)
4567 return cssid->id;
4568 return 0;
4569}
Ben Blum67523c42010-03-10 15:22:11 -08004570EXPORT_SYMBOL_GPL(css_id);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004571
4572unsigned short css_depth(struct cgroup_subsys_state *css)
4573{
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004574 struct css_id *cssid;
4575
4576 cssid = rcu_dereference_check(css->id,
4577 rcu_read_lock_held() || atomic_read(&css->refcnt));
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004578
4579 if (cssid)
4580 return cssid->depth;
4581 return 0;
4582}
Ben Blum67523c42010-03-10 15:22:11 -08004583EXPORT_SYMBOL_GPL(css_depth);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004584
KAMEZAWA Hiroyuki747388d2010-05-11 14:06:59 -07004585/**
4586 * css_is_ancestor - test "root" css is an ancestor of "child"
4587 * @child: the css to be tested.
4588 * @root: the css supporsed to be an ancestor of the child.
4589 *
4590 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4591 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4592 * But, considering usual usage, the csses should be valid objects after test.
4593 * Assuming that the caller will do some action to the child if this returns
4594 * returns true, the caller must take "child";s reference count.
4595 * If "child" is valid object and this returns true, "root" is valid, too.
4596 */
4597
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004598bool css_is_ancestor(struct cgroup_subsys_state *child,
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07004599 const struct cgroup_subsys_state *root)
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004600{
KAMEZAWA Hiroyuki747388d2010-05-11 14:06:59 -07004601 struct css_id *child_id;
4602 struct css_id *root_id;
4603 bool ret = true;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004604
KAMEZAWA Hiroyuki747388d2010-05-11 14:06:59 -07004605 rcu_read_lock();
4606 child_id = rcu_dereference(child->id);
4607 root_id = rcu_dereference(root->id);
4608 if (!child_id
4609 || !root_id
4610 || (child_id->depth < root_id->depth)
4611 || (child_id->stack[root_id->depth] != root_id->id))
4612 ret = false;
4613 rcu_read_unlock();
4614 return ret;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004615}
4616
4617static void __free_css_id_cb(struct rcu_head *head)
4618{
4619 struct css_id *id;
4620
4621 id = container_of(head, struct css_id, rcu_head);
4622 kfree(id);
4623}
4624
4625void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
4626{
4627 struct css_id *id = css->id;
4628 /* When this is called before css_id initialization, id can be NULL */
4629 if (!id)
4630 return;
4631
4632 BUG_ON(!ss->use_id);
4633
4634 rcu_assign_pointer(id->css, NULL);
4635 rcu_assign_pointer(css->id, NULL);
4636 spin_lock(&ss->id_lock);
4637 idr_remove(&ss->idr, id->id);
4638 spin_unlock(&ss->id_lock);
4639 call_rcu(&id->rcu_head, __free_css_id_cb);
4640}
Ben Blum67523c42010-03-10 15:22:11 -08004641EXPORT_SYMBOL_GPL(free_css_id);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004642
4643/*
4644 * This is called by init or create(). Then, calls to this function are
4645 * always serialized (By cgroup_mutex() at create()).
4646 */
4647
4648static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
4649{
4650 struct css_id *newid;
4651 int myid, error, size;
4652
4653 BUG_ON(!ss->use_id);
4654
4655 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
4656 newid = kzalloc(size, GFP_KERNEL);
4657 if (!newid)
4658 return ERR_PTR(-ENOMEM);
4659 /* get id */
4660 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
4661 error = -ENOMEM;
4662 goto err_out;
4663 }
4664 spin_lock(&ss->id_lock);
4665 /* Don't use 0. allocates an ID of 1-65535 */
4666 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
4667 spin_unlock(&ss->id_lock);
4668
4669 /* Returns error when there are no free spaces for new ID.*/
4670 if (error) {
4671 error = -ENOSPC;
4672 goto err_out;
4673 }
4674 if (myid > CSS_ID_MAX)
4675 goto remove_idr;
4676
4677 newid->id = myid;
4678 newid->depth = depth;
4679 return newid;
4680remove_idr:
4681 error = -ENOSPC;
4682 spin_lock(&ss->id_lock);
4683 idr_remove(&ss->idr, myid);
4684 spin_unlock(&ss->id_lock);
4685err_out:
4686 kfree(newid);
4687 return ERR_PTR(error);
4688
4689}
4690
Ben Blume6a11052010-03-10 15:22:09 -08004691static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
4692 struct cgroup_subsys_state *rootcss)
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004693{
4694 struct css_id *newid;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004695
4696 spin_lock_init(&ss->id_lock);
4697 idr_init(&ss->idr);
4698
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004699 newid = get_new_cssid(ss, 0);
4700 if (IS_ERR(newid))
4701 return PTR_ERR(newid);
4702
4703 newid->stack[0] = newid->id;
4704 newid->css = rootcss;
4705 rootcss->id = newid;
4706 return 0;
4707}
4708
4709static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
4710 struct cgroup *child)
4711{
4712 int subsys_id, i, depth = 0;
4713 struct cgroup_subsys_state *parent_css, *child_css;
Li Zefanfae9c792010-04-22 17:30:00 +08004714 struct css_id *child_id, *parent_id;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004715
4716 subsys_id = ss->subsys_id;
4717 parent_css = parent->subsys[subsys_id];
4718 child_css = child->subsys[subsys_id];
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004719 parent_id = parent_css->id;
Greg Thelen94b3dd02010-06-04 14:15:03 -07004720 depth = parent_id->depth + 1;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004721
4722 child_id = get_new_cssid(ss, depth);
4723 if (IS_ERR(child_id))
4724 return PTR_ERR(child_id);
4725
4726 for (i = 0; i < depth; i++)
4727 child_id->stack[i] = parent_id->stack[i];
4728 child_id->stack[depth] = child_id->id;
4729 /*
4730 * child_id->css pointer will be set after this cgroup is available
4731 * see cgroup_populate_dir()
4732 */
4733 rcu_assign_pointer(child_css->id, child_id);
4734
4735 return 0;
4736}
4737
4738/**
4739 * css_lookup - lookup css by id
4740 * @ss: cgroup subsys to be looked into.
4741 * @id: the id
4742 *
4743 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4744 * NULL if not. Should be called under rcu_read_lock()
4745 */
4746struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
4747{
4748 struct css_id *cssid = NULL;
4749
4750 BUG_ON(!ss->use_id);
4751 cssid = idr_find(&ss->idr, id);
4752
4753 if (unlikely(!cssid))
4754 return NULL;
4755
4756 return rcu_dereference(cssid->css);
4757}
Ben Blum67523c42010-03-10 15:22:11 -08004758EXPORT_SYMBOL_GPL(css_lookup);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004759
4760/**
4761 * css_get_next - lookup next cgroup under specified hierarchy.
4762 * @ss: pointer to subsystem
4763 * @id: current position of iteration.
4764 * @root: pointer to css. search tree under this.
4765 * @foundid: position of found object.
4766 *
4767 * Search next css under the specified hierarchy of rootid. Calling under
4768 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
4769 */
4770struct cgroup_subsys_state *
4771css_get_next(struct cgroup_subsys *ss, int id,
4772 struct cgroup_subsys_state *root, int *foundid)
4773{
4774 struct cgroup_subsys_state *ret = NULL;
4775 struct css_id *tmp;
4776 int tmpid;
4777 int rootid = css_id(root);
4778 int depth = css_depth(root);
4779
4780 if (!rootid)
4781 return NULL;
4782
4783 BUG_ON(!ss->use_id);
4784 /* fill start point for scan */
4785 tmpid = id;
4786 while (1) {
4787 /*
4788 * scan next entry from bitmap(tree), tmpid is updated after
4789 * idr_get_next().
4790 */
4791 spin_lock(&ss->id_lock);
4792 tmp = idr_get_next(&ss->idr, &tmpid);
4793 spin_unlock(&ss->id_lock);
4794
4795 if (!tmp)
4796 break;
4797 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
4798 ret = rcu_dereference(tmp->css);
4799 if (ret) {
4800 *foundid = tmpid;
4801 break;
4802 }
4803 }
4804 /* continue to scan from next id */
4805 tmpid = tmpid + 1;
4806 }
4807 return ret;
4808}
4809
Paul Menagefe693432009-09-23 15:56:20 -07004810#ifdef CONFIG_CGROUP_DEBUG
4811static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
4812 struct cgroup *cont)
4813{
4814 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
4815
4816 if (!css)
4817 return ERR_PTR(-ENOMEM);
4818
4819 return css;
4820}
4821
4822static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
4823{
4824 kfree(cont->subsys[debug_subsys_id]);
4825}
4826
4827static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
4828{
4829 return atomic_read(&cont->count);
4830}
4831
4832static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
4833{
4834 return cgroup_task_count(cont);
4835}
4836
4837static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
4838{
4839 return (u64)(unsigned long)current->cgroups;
4840}
4841
4842static u64 current_css_set_refcount_read(struct cgroup *cont,
4843 struct cftype *cft)
4844{
4845 u64 count;
4846
4847 rcu_read_lock();
4848 count = atomic_read(&current->cgroups->refcount);
4849 rcu_read_unlock();
4850 return count;
4851}
4852
Paul Menage7717f7b2009-09-23 15:56:22 -07004853static int current_css_set_cg_links_read(struct cgroup *cont,
4854 struct cftype *cft,
4855 struct seq_file *seq)
4856{
4857 struct cg_cgroup_link *link;
4858 struct css_set *cg;
4859
4860 read_lock(&css_set_lock);
4861 rcu_read_lock();
4862 cg = rcu_dereference(current->cgroups);
4863 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
4864 struct cgroup *c = link->cgrp;
4865 const char *name;
4866
4867 if (c->dentry)
4868 name = c->dentry->d_name.name;
4869 else
4870 name = "?";
Paul Menage2c6ab6d2009-09-23 15:56:23 -07004871 seq_printf(seq, "Root %d group %s\n",
4872 c->root->hierarchy_id, name);
Paul Menage7717f7b2009-09-23 15:56:22 -07004873 }
4874 rcu_read_unlock();
4875 read_unlock(&css_set_lock);
4876 return 0;
4877}
4878
4879#define MAX_TASKS_SHOWN_PER_CSS 25
4880static int cgroup_css_links_read(struct cgroup *cont,
4881 struct cftype *cft,
4882 struct seq_file *seq)
4883{
4884 struct cg_cgroup_link *link;
4885
4886 read_lock(&css_set_lock);
4887 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
4888 struct css_set *cg = link->cg;
4889 struct task_struct *task;
4890 int count = 0;
4891 seq_printf(seq, "css_set %p\n", cg);
4892 list_for_each_entry(task, &cg->tasks, cg_list) {
4893 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
4894 seq_puts(seq, " ...\n");
4895 break;
4896 } else {
4897 seq_printf(seq, " task %d\n",
4898 task_pid_vnr(task));
4899 }
4900 }
4901 }
4902 read_unlock(&css_set_lock);
4903 return 0;
4904}
4905
Paul Menagefe693432009-09-23 15:56:20 -07004906static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
4907{
4908 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
4909}
4910
4911static struct cftype debug_files[] = {
4912 {
4913 .name = "cgroup_refcount",
4914 .read_u64 = cgroup_refcount_read,
4915 },
4916 {
4917 .name = "taskcount",
4918 .read_u64 = debug_taskcount_read,
4919 },
4920
4921 {
4922 .name = "current_css_set",
4923 .read_u64 = current_css_set_read,
4924 },
4925
4926 {
4927 .name = "current_css_set_refcount",
4928 .read_u64 = current_css_set_refcount_read,
4929 },
4930
4931 {
Paul Menage7717f7b2009-09-23 15:56:22 -07004932 .name = "current_css_set_cg_links",
4933 .read_seq_string = current_css_set_cg_links_read,
4934 },
4935
4936 {
4937 .name = "cgroup_css_links",
4938 .read_seq_string = cgroup_css_links_read,
4939 },
4940
4941 {
Paul Menagefe693432009-09-23 15:56:20 -07004942 .name = "releasable",
4943 .read_u64 = releasable_read,
4944 },
4945};
4946
4947static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
4948{
4949 return cgroup_add_files(cont, ss, debug_files,
4950 ARRAY_SIZE(debug_files));
4951}
4952
4953struct cgroup_subsys debug_subsys = {
4954 .name = "debug",
4955 .create = debug_create,
4956 .destroy = debug_destroy,
4957 .populate = debug_populate,
4958 .subsys_id = debug_subsys_id,
4959};
4960#endif /* CONFIG_CGROUP_DEBUG */