blob: a8ce099544049e787464c3e4261ac70625cc0653 [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>
Alessio Igor Bogani337eb002009-05-12 15:10:54 +020055#include <linux/smp_lock.h>
Li Zefan096b7fe2009-07-29 15:04:04 -070056#include <linux/pid_namespace.h>
Paul Menage2c6ab6d2009-09-23 15:56:23 -070057#include <linux/idr.h>
Ben Blumd1d9fd32009-09-23 15:56:28 -070058#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -080059#include <linux/eventfd.h>
60#include <linux/poll.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070061
Paul Menageddbcc7e2007-10-18 23:39:30 -070062#include <asm/atomic.h>
63
Paul Menage81a6a5c2007-10-18 23:39:38 -070064static DEFINE_MUTEX(cgroup_mutex);
65
Ben Blumaae8aab2010-03-10 15:22:07 -080066/*
67 * Generate an array of cgroup subsystem pointers. At boot time, this is
68 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
69 * registered after that. The mutable section of this array is protected by
70 * cgroup_mutex.
71 */
Paul Menageddbcc7e2007-10-18 23:39:30 -070072#define SUBSYS(_x) &_x ## _subsys,
Ben Blumaae8aab2010-03-10 15:22:07 -080073static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
Paul Menageddbcc7e2007-10-18 23:39:30 -070074#include <linux/cgroup_subsys.h>
75};
76
Paul Menagec6d57f32009-09-23 15:56:19 -070077#define MAX_CGROUP_ROOT_NAMELEN 64
78
Paul Menageddbcc7e2007-10-18 23:39:30 -070079/*
80 * A cgroupfs_root represents the root of a cgroup hierarchy,
81 * and may be associated with a superblock to form an active
82 * hierarchy
83 */
84struct cgroupfs_root {
85 struct super_block *sb;
86
87 /*
88 * The bitmask of subsystems intended to be attached to this
89 * hierarchy
90 */
91 unsigned long subsys_bits;
92
Paul Menage2c6ab6d2009-09-23 15:56:23 -070093 /* Unique id for this hierarchy. */
94 int hierarchy_id;
95
Paul Menageddbcc7e2007-10-18 23:39:30 -070096 /* The bitmask of subsystems currently attached to this hierarchy */
97 unsigned long actual_subsys_bits;
98
99 /* A list running through the attached subsystems */
100 struct list_head subsys_list;
101
102 /* The root cgroup for this hierarchy */
103 struct cgroup top_cgroup;
104
105 /* Tracks how many cgroups are currently defined in hierarchy.*/
106 int number_of_cgroups;
107
Li Zefane5f6a862009-01-07 18:07:41 -0800108 /* A list running through the active hierarchies */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700109 struct list_head root_list;
110
111 /* Hierarchy-specific flags */
112 unsigned long flags;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700113
Paul Menagee788e062008-07-25 01:46:59 -0700114 /* The path to use for release notifications. */
Paul Menage81a6a5c2007-10-18 23:39:38 -0700115 char release_agent_path[PATH_MAX];
Paul Menagec6d57f32009-09-23 15:56:19 -0700116
117 /* The name for this hierarchy - may be empty */
118 char name[MAX_CGROUP_ROOT_NAMELEN];
Paul Menageddbcc7e2007-10-18 23:39:30 -0700119};
120
Paul Menageddbcc7e2007-10-18 23:39:30 -0700121/*
122 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
123 * subsystems that are otherwise unattached - it never has more than a
124 * single cgroup, and all tasks are part of that cgroup.
125 */
126static struct cgroupfs_root rootnode;
127
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700128/*
129 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
130 * cgroup_subsys->use_id != 0.
131 */
132#define CSS_ID_MAX (65535)
133struct css_id {
134 /*
135 * The css to which this ID points. This pointer is set to valid value
136 * after cgroup is populated. If cgroup is removed, this will be NULL.
137 * This pointer is expected to be RCU-safe because destroy()
138 * is called after synchronize_rcu(). But for safe use, css_is_removed()
139 * css_tryget() should be used for avoiding race.
140 */
141 struct cgroup_subsys_state *css;
142 /*
143 * ID of this css.
144 */
145 unsigned short id;
146 /*
147 * Depth in hierarchy which this ID belongs to.
148 */
149 unsigned short depth;
150 /*
151 * ID is freed by RCU. (and lookup routine is RCU safe.)
152 */
153 struct rcu_head rcu_head;
154 /*
155 * Hierarchy of CSS ID belongs to.
156 */
157 unsigned short stack[0]; /* Array of Length (depth+1) */
158};
159
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800160/*
161 * cgroup_event represents events which userspace want to recieve.
162 */
163struct cgroup_event {
164 /*
165 * Cgroup which the event belongs to.
166 */
167 struct cgroup *cgrp;
168 /*
169 * Control file which the event associated.
170 */
171 struct cftype *cft;
172 /*
173 * eventfd to signal userspace about the event.
174 */
175 struct eventfd_ctx *eventfd;
176 /*
177 * Each of these stored in a list by the cgroup.
178 */
179 struct list_head list;
180 /*
181 * All fields below needed to unregister event when
182 * userspace closes eventfd.
183 */
184 poll_table pt;
185 wait_queue_head_t *wqh;
186 wait_queue_t wait;
187 struct work_struct remove;
188};
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700189
Paul Menageddbcc7e2007-10-18 23:39:30 -0700190/* The list of hierarchy roots */
191
192static LIST_HEAD(roots);
Paul Menage817929e2007-10-18 23:39:36 -0700193static int root_count;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700194
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700195static DEFINE_IDA(hierarchy_ida);
196static int next_hierarchy_id;
197static DEFINE_SPINLOCK(hierarchy_id_lock);
198
Paul Menageddbcc7e2007-10-18 23:39:30 -0700199/* dummytop is a shorthand for the dummy hierarchy's top cgroup */
200#define dummytop (&rootnode.top_cgroup)
201
202/* This flag indicates whether tasks in the fork and exit paths should
Li Zefana043e3b2008-02-23 15:24:09 -0800203 * check for fork/exit handlers to call. This avoids us having to do
204 * extra work in the fork/exit path if none of the subsystems need to
205 * be called.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700206 */
Li Zefan8947f9d2008-07-25 01:46:56 -0700207static int need_forkexit_callback __read_mostly;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700208
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800209#ifdef CONFIG_PROVE_LOCKING
210int cgroup_lock_is_held(void)
211{
212 return lockdep_is_held(&cgroup_mutex);
213}
214#else /* #ifdef CONFIG_PROVE_LOCKING */
215int cgroup_lock_is_held(void)
216{
217 return mutex_is_locked(&cgroup_mutex);
218}
219#endif /* #else #ifdef CONFIG_PROVE_LOCKING */
220
221EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
222
Paul Menageddbcc7e2007-10-18 23:39:30 -0700223/* convenient tests for these bits */
Paul Menagebd89aab2007-10-18 23:40:44 -0700224inline int cgroup_is_removed(const struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700225{
Paul Menagebd89aab2007-10-18 23:40:44 -0700226 return test_bit(CGRP_REMOVED, &cgrp->flags);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700227}
228
229/* bits in struct cgroupfs_root flags field */
230enum {
231 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
232};
233
Adrian Bunke9685a02008-02-07 00:13:46 -0800234static int cgroup_is_releasable(const struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -0700235{
236 const int bits =
Paul Menagebd89aab2007-10-18 23:40:44 -0700237 (1 << CGRP_RELEASABLE) |
238 (1 << CGRP_NOTIFY_ON_RELEASE);
239 return (cgrp->flags & bits) == bits;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700240}
241
Adrian Bunke9685a02008-02-07 00:13:46 -0800242static int notify_on_release(const struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -0700243{
Paul Menagebd89aab2007-10-18 23:40:44 -0700244 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700245}
246
Paul Menageddbcc7e2007-10-18 23:39:30 -0700247/*
248 * for_each_subsys() allows you to iterate on each subsystem attached to
249 * an active hierarchy
250 */
251#define for_each_subsys(_root, _ss) \
252list_for_each_entry(_ss, &_root->subsys_list, sibling)
253
Li Zefane5f6a862009-01-07 18:07:41 -0800254/* for_each_active_root() allows you to iterate across the active hierarchies */
255#define for_each_active_root(_root) \
Paul Menageddbcc7e2007-10-18 23:39:30 -0700256list_for_each_entry(_root, &roots, root_list)
257
Paul Menage81a6a5c2007-10-18 23:39:38 -0700258/* the list of cgroups eligible for automatic release. Protected by
259 * release_list_lock */
260static LIST_HEAD(release_list);
261static DEFINE_SPINLOCK(release_list_lock);
262static void cgroup_release_agent(struct work_struct *work);
263static DECLARE_WORK(release_agent_work, cgroup_release_agent);
Paul Menagebd89aab2007-10-18 23:40:44 -0700264static void check_for_release(struct cgroup *cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700265
Paul Menage817929e2007-10-18 23:39:36 -0700266/* Link structure for associating css_set objects with cgroups */
267struct cg_cgroup_link {
268 /*
269 * List running through cg_cgroup_links associated with a
270 * cgroup, anchored on cgroup->css_sets
271 */
Paul Menagebd89aab2007-10-18 23:40:44 -0700272 struct list_head cgrp_link_list;
Paul Menage7717f7b2009-09-23 15:56:22 -0700273 struct cgroup *cgrp;
Paul Menage817929e2007-10-18 23:39:36 -0700274 /*
275 * List running through cg_cgroup_links pointing at a
276 * single css_set object, anchored on css_set->cg_links
277 */
278 struct list_head cg_link_list;
279 struct css_set *cg;
280};
281
282/* The default css_set - used by init and its children prior to any
283 * hierarchies being mounted. It contains a pointer to the root state
284 * for each subsystem. Also used to anchor the list of css_sets. Not
285 * reference-counted, to improve performance when child cgroups
286 * haven't been created.
287 */
288
289static struct css_set init_css_set;
290static struct cg_cgroup_link init_css_set_link;
291
Ben Blume6a11052010-03-10 15:22:09 -0800292static int cgroup_init_idr(struct cgroup_subsys *ss,
293 struct cgroup_subsys_state *css);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700294
Paul Menage817929e2007-10-18 23:39:36 -0700295/* css_set_lock protects the list of css_set objects, and the
296 * chain of tasks off each css_set. Nests outside task->alloc_lock
297 * due to cgroup_iter_start() */
298static DEFINE_RWLOCK(css_set_lock);
299static int css_set_count;
300
Paul Menage7717f7b2009-09-23 15:56:22 -0700301/*
302 * hash table for cgroup groups. This improves the performance to find
303 * an existing css_set. This hash doesn't (currently) take into
304 * account cgroups in empty hierarchies.
305 */
Li Zefan472b1052008-04-29 01:00:11 -0700306#define CSS_SET_HASH_BITS 7
307#define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
308static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
309
310static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
311{
312 int i;
313 int index;
314 unsigned long tmp = 0UL;
315
316 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
317 tmp += (unsigned long)css[i];
318 tmp = (tmp >> 16) ^ tmp;
319
320 index = hash_long(tmp, CSS_SET_HASH_BITS);
321
322 return &css_set_table[index];
323}
324
Ben Blumc3783692009-09-23 15:56:29 -0700325static void free_css_set_rcu(struct rcu_head *obj)
326{
327 struct css_set *cg = container_of(obj, struct css_set, rcu_head);
328 kfree(cg);
329}
330
Paul Menage817929e2007-10-18 23:39:36 -0700331/* We don't maintain the lists running through each css_set to its
332 * task until after the first call to cgroup_iter_start(). This
333 * reduces the fork()/exit() overhead for people who have cgroups
334 * compiled into their kernel but not actually in use */
Li Zefan8947f9d2008-07-25 01:46:56 -0700335static int use_task_css_set_links __read_mostly;
Paul Menage817929e2007-10-18 23:39:36 -0700336
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700337static void __put_css_set(struct css_set *cg, int taskexit)
Paul Menageb4f48b62007-10-18 23:39:33 -0700338{
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -0700339 struct cg_cgroup_link *link;
340 struct cg_cgroup_link *saved_link;
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700341 /*
342 * Ensure that the refcount doesn't hit zero while any readers
343 * can see it. Similar to atomic_dec_and_lock(), but for an
344 * rwlock
345 */
346 if (atomic_add_unless(&cg->refcount, -1, 1))
347 return;
348 write_lock(&css_set_lock);
349 if (!atomic_dec_and_test(&cg->refcount)) {
350 write_unlock(&css_set_lock);
351 return;
352 }
Paul Menage81a6a5c2007-10-18 23:39:38 -0700353
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700354 /* This css_set is dead. unlink it and release cgroup refcounts */
355 hlist_del(&cg->hlist);
356 css_set_count--;
357
358 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
359 cg_link_list) {
360 struct cgroup *cgrp = link->cgrp;
361 list_del(&link->cg_link_list);
362 list_del(&link->cgrp_link_list);
Paul Menagebd89aab2007-10-18 23:40:44 -0700363 if (atomic_dec_and_test(&cgrp->count) &&
364 notify_on_release(cgrp)) {
Paul Menage81a6a5c2007-10-18 23:39:38 -0700365 if (taskexit)
Paul Menagebd89aab2007-10-18 23:40:44 -0700366 set_bit(CGRP_RELEASABLE, &cgrp->flags);
367 check_for_release(cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700368 }
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700369
370 kfree(link);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700371 }
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700372
373 write_unlock(&css_set_lock);
Ben Blumc3783692009-09-23 15:56:29 -0700374 call_rcu(&cg->rcu_head, free_css_set_rcu);
Paul Menage817929e2007-10-18 23:39:36 -0700375}
376
377/*
378 * refcounted get/put for css_set objects
379 */
380static inline void get_css_set(struct css_set *cg)
381{
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700382 atomic_inc(&cg->refcount);
Paul Menage817929e2007-10-18 23:39:36 -0700383}
384
385static inline void put_css_set(struct css_set *cg)
386{
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700387 __put_css_set(cg, 0);
Paul Menage817929e2007-10-18 23:39:36 -0700388}
389
Paul Menage81a6a5c2007-10-18 23:39:38 -0700390static inline void put_css_set_taskexit(struct css_set *cg)
391{
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700392 __put_css_set(cg, 1);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700393}
394
Paul Menage817929e2007-10-18 23:39:36 -0700395/*
Paul Menage7717f7b2009-09-23 15:56:22 -0700396 * compare_css_sets - helper function for find_existing_css_set().
397 * @cg: candidate css_set being tested
398 * @old_cg: existing css_set for a task
399 * @new_cgrp: cgroup that's being entered by the task
400 * @template: desired set of css pointers in css_set (pre-calculated)
401 *
402 * Returns true if "cg" matches "old_cg" except for the hierarchy
403 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
404 */
405static bool compare_css_sets(struct css_set *cg,
406 struct css_set *old_cg,
407 struct cgroup *new_cgrp,
408 struct cgroup_subsys_state *template[])
409{
410 struct list_head *l1, *l2;
411
412 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
413 /* Not all subsystems matched */
414 return false;
415 }
416
417 /*
418 * Compare cgroup pointers in order to distinguish between
419 * different cgroups in heirarchies with no subsystems. We
420 * could get by with just this check alone (and skip the
421 * memcmp above) but on most setups the memcmp check will
422 * avoid the need for this more expensive check on almost all
423 * candidates.
424 */
425
426 l1 = &cg->cg_links;
427 l2 = &old_cg->cg_links;
428 while (1) {
429 struct cg_cgroup_link *cgl1, *cgl2;
430 struct cgroup *cg1, *cg2;
431
432 l1 = l1->next;
433 l2 = l2->next;
434 /* See if we reached the end - both lists are equal length. */
435 if (l1 == &cg->cg_links) {
436 BUG_ON(l2 != &old_cg->cg_links);
437 break;
438 } else {
439 BUG_ON(l2 == &old_cg->cg_links);
440 }
441 /* Locate the cgroups associated with these links. */
442 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
443 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
444 cg1 = cgl1->cgrp;
445 cg2 = cgl2->cgrp;
446 /* Hierarchies should be linked in the same order. */
447 BUG_ON(cg1->root != cg2->root);
448
449 /*
450 * If this hierarchy is the hierarchy of the cgroup
451 * that's changing, then we need to check that this
452 * css_set points to the new cgroup; if it's any other
453 * hierarchy, then this css_set should point to the
454 * same cgroup as the old css_set.
455 */
456 if (cg1->root == new_cgrp->root) {
457 if (cg1 != new_cgrp)
458 return false;
459 } else {
460 if (cg1 != cg2)
461 return false;
462 }
463 }
464 return true;
465}
466
467/*
Paul Menage817929e2007-10-18 23:39:36 -0700468 * find_existing_css_set() is a helper for
469 * find_css_set(), and checks to see whether an existing
Li Zefan472b1052008-04-29 01:00:11 -0700470 * css_set is suitable.
Paul Menage817929e2007-10-18 23:39:36 -0700471 *
472 * oldcg: the cgroup group that we're using before the cgroup
473 * transition
474 *
Paul Menagebd89aab2007-10-18 23:40:44 -0700475 * cgrp: the cgroup that we're moving into
Paul Menage817929e2007-10-18 23:39:36 -0700476 *
477 * template: location in which to build the desired set of subsystem
478 * state objects for the new cgroup group
479 */
Paul Menage817929e2007-10-18 23:39:36 -0700480static struct css_set *find_existing_css_set(
481 struct css_set *oldcg,
Paul Menagebd89aab2007-10-18 23:40:44 -0700482 struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -0700483 struct cgroup_subsys_state *template[])
484{
485 int i;
Paul Menagebd89aab2007-10-18 23:40:44 -0700486 struct cgroupfs_root *root = cgrp->root;
Li Zefan472b1052008-04-29 01:00:11 -0700487 struct hlist_head *hhead;
488 struct hlist_node *node;
489 struct css_set *cg;
Paul Menage817929e2007-10-18 23:39:36 -0700490
Ben Blumaae8aab2010-03-10 15:22:07 -0800491 /*
492 * Build the set of subsystem state objects that we want to see in the
493 * new css_set. while subsystems can change globally, the entries here
494 * won't change, so no need for locking.
495 */
Paul Menage817929e2007-10-18 23:39:36 -0700496 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
Li Zefan8d53d552008-02-23 15:24:11 -0800497 if (root->subsys_bits & (1UL << i)) {
Paul Menage817929e2007-10-18 23:39:36 -0700498 /* Subsystem is in this hierarchy. So we want
499 * the subsystem state from the new
500 * cgroup */
Paul Menagebd89aab2007-10-18 23:40:44 -0700501 template[i] = cgrp->subsys[i];
Paul Menage817929e2007-10-18 23:39:36 -0700502 } else {
503 /* Subsystem is not in this hierarchy, so we
504 * don't want to change the subsystem state */
505 template[i] = oldcg->subsys[i];
506 }
507 }
508
Li Zefan472b1052008-04-29 01:00:11 -0700509 hhead = css_set_hash(template);
510 hlist_for_each_entry(cg, node, hhead, hlist) {
Paul Menage7717f7b2009-09-23 15:56:22 -0700511 if (!compare_css_sets(cg, oldcg, cgrp, template))
512 continue;
513
514 /* This css_set matches what we need */
515 return cg;
Li Zefan472b1052008-04-29 01:00:11 -0700516 }
Paul Menage817929e2007-10-18 23:39:36 -0700517
518 /* No existing cgroup group matched */
519 return NULL;
520}
521
Paul Menage817929e2007-10-18 23:39:36 -0700522static void free_cg_links(struct list_head *tmp)
523{
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -0700524 struct cg_cgroup_link *link;
525 struct cg_cgroup_link *saved_link;
526
527 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
Paul Menagebd89aab2007-10-18 23:40:44 -0700528 list_del(&link->cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -0700529 kfree(link);
530 }
531}
532
533/*
Li Zefan36553432008-07-29 22:33:19 -0700534 * allocate_cg_links() allocates "count" cg_cgroup_link structures
535 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
536 * success or a negative error
537 */
538static int allocate_cg_links(int count, struct list_head *tmp)
539{
540 struct cg_cgroup_link *link;
541 int i;
542 INIT_LIST_HEAD(tmp);
543 for (i = 0; i < count; i++) {
544 link = kmalloc(sizeof(*link), GFP_KERNEL);
545 if (!link) {
546 free_cg_links(tmp);
547 return -ENOMEM;
548 }
549 list_add(&link->cgrp_link_list, tmp);
550 }
551 return 0;
552}
553
Li Zefanc12f65d2009-01-07 18:07:42 -0800554/**
555 * link_css_set - a helper function to link a css_set to a cgroup
556 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
557 * @cg: the css_set to be linked
558 * @cgrp: the destination cgroup
559 */
560static void link_css_set(struct list_head *tmp_cg_links,
561 struct css_set *cg, struct cgroup *cgrp)
562{
563 struct cg_cgroup_link *link;
564
565 BUG_ON(list_empty(tmp_cg_links));
566 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
567 cgrp_link_list);
568 link->cg = cg;
Paul Menage7717f7b2009-09-23 15:56:22 -0700569 link->cgrp = cgrp;
Paul Menage2c6ab6d2009-09-23 15:56:23 -0700570 atomic_inc(&cgrp->count);
Li Zefanc12f65d2009-01-07 18:07:42 -0800571 list_move(&link->cgrp_link_list, &cgrp->css_sets);
Paul Menage7717f7b2009-09-23 15:56:22 -0700572 /*
573 * Always add links to the tail of the list so that the list
574 * is sorted by order of hierarchy creation
575 */
576 list_add_tail(&link->cg_link_list, &cg->cg_links);
Li Zefanc12f65d2009-01-07 18:07:42 -0800577}
578
Li Zefan36553432008-07-29 22:33:19 -0700579/*
Paul Menage817929e2007-10-18 23:39:36 -0700580 * find_css_set() takes an existing cgroup group and a
581 * cgroup object, and returns a css_set object that's
582 * equivalent to the old group, but with the given cgroup
583 * substituted into the appropriate hierarchy. Must be called with
584 * cgroup_mutex held
585 */
Paul Menage817929e2007-10-18 23:39:36 -0700586static struct css_set *find_css_set(
Paul Menagebd89aab2007-10-18 23:40:44 -0700587 struct css_set *oldcg, struct cgroup *cgrp)
Paul Menage817929e2007-10-18 23:39:36 -0700588{
589 struct css_set *res;
590 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
Paul Menage817929e2007-10-18 23:39:36 -0700591
592 struct list_head tmp_cg_links;
Paul Menage817929e2007-10-18 23:39:36 -0700593
Li Zefan472b1052008-04-29 01:00:11 -0700594 struct hlist_head *hhead;
Paul Menage7717f7b2009-09-23 15:56:22 -0700595 struct cg_cgroup_link *link;
Li Zefan472b1052008-04-29 01:00:11 -0700596
Paul Menage817929e2007-10-18 23:39:36 -0700597 /* First see if we already have a cgroup group that matches
598 * the desired set */
Li Zefan7e9abd82008-07-25 01:46:54 -0700599 read_lock(&css_set_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -0700600 res = find_existing_css_set(oldcg, cgrp, template);
Paul Menage817929e2007-10-18 23:39:36 -0700601 if (res)
602 get_css_set(res);
Li Zefan7e9abd82008-07-25 01:46:54 -0700603 read_unlock(&css_set_lock);
Paul Menage817929e2007-10-18 23:39:36 -0700604
605 if (res)
606 return res;
607
608 res = kmalloc(sizeof(*res), GFP_KERNEL);
609 if (!res)
610 return NULL;
611
612 /* Allocate all the cg_cgroup_link objects that we'll need */
613 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
614 kfree(res);
615 return NULL;
616 }
617
Lai Jiangshan146aa1b2008-10-18 20:28:03 -0700618 atomic_set(&res->refcount, 1);
Paul Menage817929e2007-10-18 23:39:36 -0700619 INIT_LIST_HEAD(&res->cg_links);
620 INIT_LIST_HEAD(&res->tasks);
Li Zefan472b1052008-04-29 01:00:11 -0700621 INIT_HLIST_NODE(&res->hlist);
Paul Menage817929e2007-10-18 23:39:36 -0700622
623 /* Copy the set of subsystem state objects generated in
624 * find_existing_css_set() */
625 memcpy(res->subsys, template, sizeof(res->subsys));
626
627 write_lock(&css_set_lock);
628 /* Add reference counts and links from the new css_set. */
Paul Menage7717f7b2009-09-23 15:56:22 -0700629 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
630 struct cgroup *c = link->cgrp;
631 if (c->root == cgrp->root)
632 c = cgrp;
633 link_css_set(&tmp_cg_links, res, c);
634 }
Paul Menage817929e2007-10-18 23:39:36 -0700635
636 BUG_ON(!list_empty(&tmp_cg_links));
637
Paul Menage817929e2007-10-18 23:39:36 -0700638 css_set_count++;
Li Zefan472b1052008-04-29 01:00:11 -0700639
640 /* Add this cgroup group to the hash table */
641 hhead = css_set_hash(res->subsys);
642 hlist_add_head(&res->hlist, hhead);
643
Paul Menage817929e2007-10-18 23:39:36 -0700644 write_unlock(&css_set_lock);
645
646 return res;
Paul Menageb4f48b62007-10-18 23:39:33 -0700647}
648
Paul Menageddbcc7e2007-10-18 23:39:30 -0700649/*
Paul Menage7717f7b2009-09-23 15:56:22 -0700650 * Return the cgroup for "task" from the given hierarchy. Must be
651 * called with cgroup_mutex held.
652 */
653static struct cgroup *task_cgroup_from_root(struct task_struct *task,
654 struct cgroupfs_root *root)
655{
656 struct css_set *css;
657 struct cgroup *res = NULL;
658
659 BUG_ON(!mutex_is_locked(&cgroup_mutex));
660 read_lock(&css_set_lock);
661 /*
662 * No need to lock the task - since we hold cgroup_mutex the
663 * task can't change groups, so the only thing that can happen
664 * is that it exits and its css is set back to init_css_set.
665 */
666 css = task->cgroups;
667 if (css == &init_css_set) {
668 res = &root->top_cgroup;
669 } else {
670 struct cg_cgroup_link *link;
671 list_for_each_entry(link, &css->cg_links, cg_link_list) {
672 struct cgroup *c = link->cgrp;
673 if (c->root == root) {
674 res = c;
675 break;
676 }
677 }
678 }
679 read_unlock(&css_set_lock);
680 BUG_ON(!res);
681 return res;
682}
683
684/*
Paul Menageddbcc7e2007-10-18 23:39:30 -0700685 * There is one global cgroup mutex. We also require taking
686 * task_lock() when dereferencing a task's cgroup subsys pointers.
687 * See "The task_lock() exception", at the end of this comment.
688 *
689 * A task must hold cgroup_mutex to modify cgroups.
690 *
691 * Any task can increment and decrement the count field without lock.
692 * So in general, code holding cgroup_mutex can't rely on the count
693 * field not changing. However, if the count goes to zero, then only
Cliff Wickman956db3c2008-02-07 00:14:43 -0800694 * cgroup_attach_task() can increment it again. Because a count of zero
Paul Menageddbcc7e2007-10-18 23:39:30 -0700695 * means that no tasks are currently attached, therefore there is no
696 * way a task attached to that cgroup can fork (the other way to
697 * increment the count). So code holding cgroup_mutex can safely
698 * assume that if the count is zero, it will stay zero. Similarly, if
699 * a task holds cgroup_mutex on a cgroup with zero count, it
700 * knows that the cgroup won't be removed, as cgroup_rmdir()
701 * needs that mutex.
702 *
Paul Menageddbcc7e2007-10-18 23:39:30 -0700703 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
704 * (usually) take cgroup_mutex. These are the two most performance
705 * critical pieces of code here. The exception occurs on cgroup_exit(),
706 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
707 * is taken, and if the cgroup count is zero, a usermode call made
Li Zefana043e3b2008-02-23 15:24:09 -0800708 * to the release agent with the name of the cgroup (path relative to
709 * the root of cgroup file system) as the argument.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700710 *
711 * A cgroup can only be deleted if both its 'count' of using tasks
712 * is zero, and its list of 'children' cgroups is empty. Since all
713 * tasks in the system use _some_ cgroup, and since there is always at
714 * least one task in the system (init, pid == 1), therefore, top_cgroup
715 * always has either children cgroups and/or using tasks. So we don't
716 * need a special hack to ensure that top_cgroup cannot be deleted.
717 *
718 * The task_lock() exception
719 *
720 * The need for this exception arises from the action of
Cliff Wickman956db3c2008-02-07 00:14:43 -0800721 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
Li Zefana043e3b2008-02-23 15:24:09 -0800722 * another. It does so using cgroup_mutex, however there are
Paul Menageddbcc7e2007-10-18 23:39:30 -0700723 * several performance critical places that need to reference
724 * task->cgroup without the expense of grabbing a system global
725 * mutex. Therefore except as noted below, when dereferencing or, as
Cliff Wickman956db3c2008-02-07 00:14:43 -0800726 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
Paul Menageddbcc7e2007-10-18 23:39:30 -0700727 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
728 * the task_struct routinely used for such matters.
729 *
730 * P.S. One more locking exception. RCU is used to guard the
Cliff Wickman956db3c2008-02-07 00:14:43 -0800731 * update of a tasks cgroup pointer by cgroup_attach_task()
Paul Menageddbcc7e2007-10-18 23:39:30 -0700732 */
733
Paul Menageddbcc7e2007-10-18 23:39:30 -0700734/**
735 * cgroup_lock - lock out any changes to cgroup structures
736 *
737 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700738void cgroup_lock(void)
739{
740 mutex_lock(&cgroup_mutex);
741}
Ben Blum67523c42010-03-10 15:22:11 -0800742EXPORT_SYMBOL_GPL(cgroup_lock);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700743
744/**
745 * cgroup_unlock - release lock on cgroup changes
746 *
747 * Undo the lock taken in a previous cgroup_lock() call.
748 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700749void cgroup_unlock(void)
750{
751 mutex_unlock(&cgroup_mutex);
752}
Ben Blum67523c42010-03-10 15:22:11 -0800753EXPORT_SYMBOL_GPL(cgroup_unlock);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700754
755/*
756 * A couple of forward declarations required, due to cyclic reference loop:
757 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
758 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
759 * -> cgroup_mkdir.
760 */
761
762static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
763static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
Paul Menagebd89aab2007-10-18 23:40:44 -0700764static int cgroup_populate_dir(struct cgroup *cgrp);
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -0700765static const struct inode_operations cgroup_dir_inode_operations;
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700766static const struct file_operations proc_cgroupstats_operations;
Paul Menagea4243162007-10-18 23:39:35 -0700767
768static struct backing_dev_info cgroup_backing_dev_info = {
Jens Axboed9938312009-06-12 14:45:52 +0200769 .name = "cgroup",
Miklos Szeredie4ad08f2008-04-30 00:54:37 -0700770 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Paul Menagea4243162007-10-18 23:39:35 -0700771};
Paul Menageddbcc7e2007-10-18 23:39:30 -0700772
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -0700773static int alloc_css_id(struct cgroup_subsys *ss,
774 struct cgroup *parent, struct cgroup *child);
775
Paul Menageddbcc7e2007-10-18 23:39:30 -0700776static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
777{
778 struct inode *inode = new_inode(sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700779
780 if (inode) {
781 inode->i_mode = mode;
David Howells76aac0e2008-11-14 10:39:12 +1100782 inode->i_uid = current_fsuid();
783 inode->i_gid = current_fsgid();
Paul Menageddbcc7e2007-10-18 23:39:30 -0700784 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
785 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
786 }
787 return inode;
788}
789
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -0800790/*
791 * Call subsys's pre_destroy handler.
792 * This is called before css refcnt check.
793 */
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700794static int cgroup_call_pre_destroy(struct cgroup *cgrp)
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -0800795{
796 struct cgroup_subsys *ss;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700797 int ret = 0;
798
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -0800799 for_each_subsys(cgrp->root, ss)
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700800 if (ss->pre_destroy) {
801 ret = ss->pre_destroy(ss, cgrp);
802 if (ret)
Kirill A. Shutemov4ab78682010-03-10 15:22:34 -0800803 break;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700804 }
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -0800805
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700806 return ret;
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -0800807}
808
Paul Menagea47295e2009-01-07 18:07:44 -0800809static void free_cgroup_rcu(struct rcu_head *obj)
810{
811 struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head);
812
813 kfree(cgrp);
814}
815
Paul Menageddbcc7e2007-10-18 23:39:30 -0700816static void cgroup_diput(struct dentry *dentry, struct inode *inode)
817{
818 /* is dentry a directory ? if so, kfree() associated cgroup */
819 if (S_ISDIR(inode->i_mode)) {
Paul Menagebd89aab2007-10-18 23:40:44 -0700820 struct cgroup *cgrp = dentry->d_fsdata;
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800821 struct cgroup_subsys *ss;
Paul Menagebd89aab2007-10-18 23:40:44 -0700822 BUG_ON(!(cgroup_is_removed(cgrp)));
Paul Menage81a6a5c2007-10-18 23:39:38 -0700823 /* It's possible for external users to be holding css
824 * reference counts on a cgroup; css_put() needs to
825 * be able to access the cgroup after decrementing
826 * the reference count in order to know if it needs to
827 * queue the cgroup to be handled by the release
828 * agent */
829 synchronize_rcu();
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800830
831 mutex_lock(&cgroup_mutex);
832 /*
833 * Release the subsystem state objects.
834 */
Li Zefan75139b82009-01-07 18:07:33 -0800835 for_each_subsys(cgrp->root, ss)
836 ss->destroy(ss, cgrp);
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800837
838 cgrp->root->number_of_cgroups--;
839 mutex_unlock(&cgroup_mutex);
840
Paul Menagea47295e2009-01-07 18:07:44 -0800841 /*
842 * Drop the active superblock reference that we took when we
843 * created the cgroup
844 */
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800845 deactivate_super(cgrp->root->sb);
846
Ben Blum72a8cb32009-09-23 15:56:27 -0700847 /*
848 * if we're getting rid of the cgroup, refcount should ensure
849 * that there are no pidlists left.
850 */
851 BUG_ON(!list_empty(&cgrp->pidlists));
852
Paul Menagea47295e2009-01-07 18:07:44 -0800853 call_rcu(&cgrp->rcu_head, free_cgroup_rcu);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700854 }
855 iput(inode);
856}
857
858static void remove_dir(struct dentry *d)
859{
860 struct dentry *parent = dget(d->d_parent);
861
862 d_delete(d);
863 simple_rmdir(parent->d_inode, d);
864 dput(parent);
865}
866
867static void cgroup_clear_directory(struct dentry *dentry)
868{
869 struct list_head *node;
870
871 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
872 spin_lock(&dcache_lock);
873 node = dentry->d_subdirs.next;
874 while (node != &dentry->d_subdirs) {
875 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
876 list_del_init(node);
877 if (d->d_inode) {
878 /* This should never be called on a cgroup
879 * directory with child cgroups */
880 BUG_ON(d->d_inode->i_mode & S_IFDIR);
881 d = dget_locked(d);
882 spin_unlock(&dcache_lock);
883 d_delete(d);
884 simple_unlink(dentry->d_inode, d);
885 dput(d);
886 spin_lock(&dcache_lock);
887 }
888 node = dentry->d_subdirs.next;
889 }
890 spin_unlock(&dcache_lock);
891}
892
893/*
894 * NOTE : the dentry must have been dget()'ed
895 */
896static void cgroup_d_remove_dir(struct dentry *dentry)
897{
898 cgroup_clear_directory(dentry);
899
900 spin_lock(&dcache_lock);
901 list_del_init(&dentry->d_u.d_child);
902 spin_unlock(&dcache_lock);
903 remove_dir(dentry);
904}
905
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700906/*
907 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
908 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
909 * reference to css->refcnt. In general, this refcnt is expected to goes down
910 * to zero, soon.
911 *
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -0700912 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700913 */
914DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
915
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -0700916static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700917{
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -0700918 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -0700919 wake_up_all(&cgroup_rmdir_waitq);
920}
921
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -0700922void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
923{
924 css_get(css);
925}
926
927void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
928{
929 cgroup_wakeup_rmdir_waiter(css->cgroup);
930 css_put(css);
931}
932
Ben Blumaae8aab2010-03-10 15:22:07 -0800933/*
Ben Blumcf5d5942010-03-10 15:22:09 -0800934 * Call with cgroup_mutex held. Drops reference counts on modules, including
935 * any duplicate ones that parse_cgroupfs_options took. If this function
936 * returns an error, no reference counts are touched.
Ben Blumaae8aab2010-03-10 15:22:07 -0800937 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700938static int rebind_subsystems(struct cgroupfs_root *root,
939 unsigned long final_bits)
940{
941 unsigned long added_bits, removed_bits;
Paul Menagebd89aab2007-10-18 23:40:44 -0700942 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700943 int i;
944
Ben Blumaae8aab2010-03-10 15:22:07 -0800945 BUG_ON(!mutex_is_locked(&cgroup_mutex));
946
Paul Menageddbcc7e2007-10-18 23:39:30 -0700947 removed_bits = root->actual_subsys_bits & ~final_bits;
948 added_bits = final_bits & ~root->actual_subsys_bits;
949 /* Check that any added subsystems are currently free */
950 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
Li Zefan8d53d552008-02-23 15:24:11 -0800951 unsigned long bit = 1UL << i;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700952 struct cgroup_subsys *ss = subsys[i];
953 if (!(bit & added_bits))
954 continue;
Ben Blumaae8aab2010-03-10 15:22:07 -0800955 /*
956 * Nobody should tell us to do a subsys that doesn't exist:
957 * parse_cgroupfs_options should catch that case and refcounts
958 * ensure that subsystems won't disappear once selected.
959 */
960 BUG_ON(ss == NULL);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700961 if (ss->root != &rootnode) {
962 /* Subsystem isn't free */
963 return -EBUSY;
964 }
965 }
966
967 /* Currently we don't handle adding/removing subsystems when
968 * any child cgroups exist. This is theoretically supportable
969 * but involves complex error handling, so it's being left until
970 * later */
Paul Menage307257c2008-12-15 13:54:22 -0800971 if (root->number_of_cgroups > 1)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700972 return -EBUSY;
973
974 /* Process each subsystem */
975 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
976 struct cgroup_subsys *ss = subsys[i];
977 unsigned long bit = 1UL << i;
978 if (bit & added_bits) {
979 /* We're binding this subsystem to this hierarchy */
Ben Blumaae8aab2010-03-10 15:22:07 -0800980 BUG_ON(ss == NULL);
Paul Menagebd89aab2007-10-18 23:40:44 -0700981 BUG_ON(cgrp->subsys[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700982 BUG_ON(!dummytop->subsys[i]);
983 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
Paul Menage999cd8a2009-01-07 18:08:36 -0800984 mutex_lock(&ss->hierarchy_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -0700985 cgrp->subsys[i] = dummytop->subsys[i];
986 cgrp->subsys[i]->cgroup = cgrp;
Li Zefan33a68ac2009-01-07 18:07:42 -0800987 list_move(&ss->sibling, &root->subsys_list);
Lai Jiangshanb2aa30f2009-01-07 18:07:37 -0800988 ss->root = root;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700989 if (ss->bind)
Paul Menagebd89aab2007-10-18 23:40:44 -0700990 ss->bind(ss, cgrp);
Paul Menage999cd8a2009-01-07 18:08:36 -0800991 mutex_unlock(&ss->hierarchy_mutex);
Ben Blumcf5d5942010-03-10 15:22:09 -0800992 /* refcount was already taken, and we're keeping it */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700993 } else if (bit & removed_bits) {
994 /* We're removing this subsystem */
Ben Blumaae8aab2010-03-10 15:22:07 -0800995 BUG_ON(ss == NULL);
Paul Menagebd89aab2007-10-18 23:40:44 -0700996 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
997 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
Paul Menage999cd8a2009-01-07 18:08:36 -0800998 mutex_lock(&ss->hierarchy_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700999 if (ss->bind)
1000 ss->bind(ss, dummytop);
1001 dummytop->subsys[i]->cgroup = dummytop;
Paul Menagebd89aab2007-10-18 23:40:44 -07001002 cgrp->subsys[i] = NULL;
Lai Jiangshanb2aa30f2009-01-07 18:07:37 -08001003 subsys[i]->root = &rootnode;
Li Zefan33a68ac2009-01-07 18:07:42 -08001004 list_move(&ss->sibling, &rootnode.subsys_list);
Paul Menage999cd8a2009-01-07 18:08:36 -08001005 mutex_unlock(&ss->hierarchy_mutex);
Ben Blumcf5d5942010-03-10 15:22:09 -08001006 /* subsystem is now free - drop reference on module */
1007 module_put(ss->module);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001008 } else if (bit & final_bits) {
1009 /* Subsystem state should already exist */
Ben Blumaae8aab2010-03-10 15:22:07 -08001010 BUG_ON(ss == NULL);
Paul Menagebd89aab2007-10-18 23:40:44 -07001011 BUG_ON(!cgrp->subsys[i]);
Ben Blumcf5d5942010-03-10 15:22:09 -08001012 /*
1013 * a refcount was taken, but we already had one, so
1014 * drop the extra reference.
1015 */
1016 module_put(ss->module);
1017#ifdef CONFIG_MODULE_UNLOAD
1018 BUG_ON(ss->module && !module_refcount(ss->module));
1019#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -07001020 } else {
1021 /* Subsystem state shouldn't exist */
Paul Menagebd89aab2007-10-18 23:40:44 -07001022 BUG_ON(cgrp->subsys[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001023 }
1024 }
1025 root->subsys_bits = root->actual_subsys_bits = final_bits;
1026 synchronize_rcu();
1027
1028 return 0;
1029}
1030
1031static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
1032{
1033 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
1034 struct cgroup_subsys *ss;
1035
1036 mutex_lock(&cgroup_mutex);
1037 for_each_subsys(root, ss)
1038 seq_printf(seq, ",%s", ss->name);
1039 if (test_bit(ROOT_NOPREFIX, &root->flags))
1040 seq_puts(seq, ",noprefix");
Paul Menage81a6a5c2007-10-18 23:39:38 -07001041 if (strlen(root->release_agent_path))
1042 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
Paul Menagec6d57f32009-09-23 15:56:19 -07001043 if (strlen(root->name))
1044 seq_printf(seq, ",name=%s", root->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001045 mutex_unlock(&cgroup_mutex);
1046 return 0;
1047}
1048
1049struct cgroup_sb_opts {
1050 unsigned long subsys_bits;
1051 unsigned long flags;
Paul Menage81a6a5c2007-10-18 23:39:38 -07001052 char *release_agent;
Paul Menagec6d57f32009-09-23 15:56:19 -07001053 char *name;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001054 /* User explicitly requested empty subsystem */
1055 bool none;
Paul Menagec6d57f32009-09-23 15:56:19 -07001056
1057 struct cgroupfs_root *new_root;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001058
Paul Menageddbcc7e2007-10-18 23:39:30 -07001059};
1060
Ben Blumaae8aab2010-03-10 15:22:07 -08001061/*
1062 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
Ben Blumcf5d5942010-03-10 15:22:09 -08001063 * with cgroup_mutex held to protect the subsys[] array. This function takes
1064 * refcounts on subsystems to be used, unless it returns error, in which case
1065 * no refcounts are taken.
Ben Blumaae8aab2010-03-10 15:22:07 -08001066 */
Ben Blumcf5d5942010-03-10 15:22:09 -08001067static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001068{
1069 char *token, *o = data ?: "all";
Li Zefanf9ab5b52009-06-17 16:26:33 -07001070 unsigned long mask = (unsigned long)-1;
Ben Blumcf5d5942010-03-10 15:22:09 -08001071 int i;
1072 bool module_pin_failed = false;
Li Zefanf9ab5b52009-06-17 16:26:33 -07001073
Ben Blumaae8aab2010-03-10 15:22:07 -08001074 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1075
Li Zefanf9ab5b52009-06-17 16:26:33 -07001076#ifdef CONFIG_CPUSETS
1077 mask = ~(1UL << cpuset_subsys_id);
1078#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -07001079
Paul Menagec6d57f32009-09-23 15:56:19 -07001080 memset(opts, 0, sizeof(*opts));
Paul Menageddbcc7e2007-10-18 23:39:30 -07001081
1082 while ((token = strsep(&o, ",")) != NULL) {
1083 if (!*token)
1084 return -EINVAL;
1085 if (!strcmp(token, "all")) {
Paul Menage8bab8dd2008-04-04 14:29:57 -07001086 /* Add all non-disabled subsystems */
Paul Menage8bab8dd2008-04-04 14:29:57 -07001087 opts->subsys_bits = 0;
1088 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1089 struct cgroup_subsys *ss = subsys[i];
Ben Blumaae8aab2010-03-10 15:22:07 -08001090 if (ss == NULL)
1091 continue;
Paul Menage8bab8dd2008-04-04 14:29:57 -07001092 if (!ss->disabled)
1093 opts->subsys_bits |= 1ul << i;
1094 }
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001095 } else if (!strcmp(token, "none")) {
1096 /* Explicitly have no subsystems */
1097 opts->none = true;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001098 } else if (!strcmp(token, "noprefix")) {
1099 set_bit(ROOT_NOPREFIX, &opts->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001100 } else if (!strncmp(token, "release_agent=", 14)) {
1101 /* Specifying two release agents is forbidden */
1102 if (opts->release_agent)
1103 return -EINVAL;
Paul Menagec6d57f32009-09-23 15:56:19 -07001104 opts->release_agent =
1105 kstrndup(token + 14, PATH_MAX, GFP_KERNEL);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001106 if (!opts->release_agent)
1107 return -ENOMEM;
Paul Menagec6d57f32009-09-23 15:56:19 -07001108 } else if (!strncmp(token, "name=", 5)) {
Paul Menagec6d57f32009-09-23 15:56:19 -07001109 const char *name = token + 5;
1110 /* Can't specify an empty name */
1111 if (!strlen(name))
1112 return -EINVAL;
1113 /* Must match [\w.-]+ */
1114 for (i = 0; i < strlen(name); i++) {
1115 char c = name[i];
1116 if (isalnum(c))
1117 continue;
1118 if ((c == '.') || (c == '-') || (c == '_'))
1119 continue;
1120 return -EINVAL;
1121 }
1122 /* Specifying two names is forbidden */
1123 if (opts->name)
1124 return -EINVAL;
1125 opts->name = kstrndup(name,
1126 MAX_CGROUP_ROOT_NAMELEN,
1127 GFP_KERNEL);
1128 if (!opts->name)
1129 return -ENOMEM;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001130 } else {
1131 struct cgroup_subsys *ss;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001132 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1133 ss = subsys[i];
Ben Blumaae8aab2010-03-10 15:22:07 -08001134 if (ss == NULL)
1135 continue;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001136 if (!strcmp(token, ss->name)) {
Paul Menage8bab8dd2008-04-04 14:29:57 -07001137 if (!ss->disabled)
1138 set_bit(i, &opts->subsys_bits);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001139 break;
1140 }
1141 }
1142 if (i == CGROUP_SUBSYS_COUNT)
1143 return -ENOENT;
1144 }
1145 }
1146
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001147 /* Consistency checks */
1148
Li Zefanf9ab5b52009-06-17 16:26:33 -07001149 /*
1150 * Option noprefix was introduced just for backward compatibility
1151 * with the old cpuset, so we allow noprefix only if mounting just
1152 * the cpuset subsystem.
1153 */
1154 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1155 (opts->subsys_bits & mask))
1156 return -EINVAL;
1157
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001158
1159 /* Can't specify "none" and some subsystems */
1160 if (opts->subsys_bits && opts->none)
1161 return -EINVAL;
1162
1163 /*
1164 * We either have to specify by name or by subsystems. (So all
1165 * empty hierarchies must have a name).
1166 */
Paul Menagec6d57f32009-09-23 15:56:19 -07001167 if (!opts->subsys_bits && !opts->name)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001168 return -EINVAL;
1169
Ben Blumcf5d5942010-03-10 15:22:09 -08001170 /*
1171 * Grab references on all the modules we'll need, so the subsystems
1172 * don't dance around before rebind_subsystems attaches them. This may
1173 * take duplicate reference counts on a subsystem that's already used,
1174 * but rebind_subsystems handles this case.
1175 */
1176 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1177 unsigned long bit = 1UL << i;
1178
1179 if (!(bit & opts->subsys_bits))
1180 continue;
1181 if (!try_module_get(subsys[i]->module)) {
1182 module_pin_failed = true;
1183 break;
1184 }
1185 }
1186 if (module_pin_failed) {
1187 /*
1188 * oops, one of the modules was going away. this means that we
1189 * raced with a module_delete call, and to the user this is
1190 * essentially a "subsystem doesn't exist" case.
1191 */
1192 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1193 /* drop refcounts only on the ones we took */
1194 unsigned long bit = 1UL << i;
1195
1196 if (!(bit & opts->subsys_bits))
1197 continue;
1198 module_put(subsys[i]->module);
1199 }
1200 return -ENOENT;
1201 }
1202
Paul Menageddbcc7e2007-10-18 23:39:30 -07001203 return 0;
1204}
1205
Ben Blumcf5d5942010-03-10 15:22:09 -08001206static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1207{
1208 int i;
1209 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1210 unsigned long bit = 1UL << i;
1211
1212 if (!(bit & subsys_bits))
1213 continue;
1214 module_put(subsys[i]->module);
1215 }
1216}
1217
Paul Menageddbcc7e2007-10-18 23:39:30 -07001218static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1219{
1220 int ret = 0;
1221 struct cgroupfs_root *root = sb->s_fs_info;
Paul Menagebd89aab2007-10-18 23:40:44 -07001222 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001223 struct cgroup_sb_opts opts;
1224
Alessio Igor Bogani337eb002009-05-12 15:10:54 +02001225 lock_kernel();
Paul Menagebd89aab2007-10-18 23:40:44 -07001226 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001227 mutex_lock(&cgroup_mutex);
1228
1229 /* See what subsystems are wanted */
1230 ret = parse_cgroupfs_options(data, &opts);
1231 if (ret)
1232 goto out_unlock;
1233
Ben Blumcf5d5942010-03-10 15:22:09 -08001234 /* Don't allow flags or name to change at remount */
1235 if (opts.flags != root->flags ||
1236 (opts.name && strcmp(opts.name, root->name))) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07001237 ret = -EINVAL;
Ben Blumcf5d5942010-03-10 15:22:09 -08001238 drop_parsed_module_refcounts(opts.subsys_bits);
Paul Menagec6d57f32009-09-23 15:56:19 -07001239 goto out_unlock;
1240 }
1241
Paul Menageddbcc7e2007-10-18 23:39:30 -07001242 ret = rebind_subsystems(root, opts.subsys_bits);
Ben Blumcf5d5942010-03-10 15:22:09 -08001243 if (ret) {
1244 drop_parsed_module_refcounts(opts.subsys_bits);
Li Zefan0670e082009-04-02 16:57:30 -07001245 goto out_unlock;
Ben Blumcf5d5942010-03-10 15:22:09 -08001246 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001247
1248 /* (re)populate subsystem files */
Li Zefan0670e082009-04-02 16:57:30 -07001249 cgroup_populate_dir(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001250
Paul Menage81a6a5c2007-10-18 23:39:38 -07001251 if (opts.release_agent)
1252 strcpy(root->release_agent_path, opts.release_agent);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001253 out_unlock:
Jesper Juhl66bdc9c2009-04-02 16:57:27 -07001254 kfree(opts.release_agent);
Paul Menagec6d57f32009-09-23 15:56:19 -07001255 kfree(opts.name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001256 mutex_unlock(&cgroup_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -07001257 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
Alessio Igor Bogani337eb002009-05-12 15:10:54 +02001258 unlock_kernel();
Paul Menageddbcc7e2007-10-18 23:39:30 -07001259 return ret;
1260}
1261
Alexey Dobriyanb87221d2009-09-21 17:01:09 -07001262static const struct super_operations cgroup_ops = {
Paul Menageddbcc7e2007-10-18 23:39:30 -07001263 .statfs = simple_statfs,
1264 .drop_inode = generic_delete_inode,
1265 .show_options = cgroup_show_options,
1266 .remount_fs = cgroup_remount,
1267};
1268
Paul Menagecc31edc2008-10-18 20:28:04 -07001269static void init_cgroup_housekeeping(struct cgroup *cgrp)
1270{
1271 INIT_LIST_HEAD(&cgrp->sibling);
1272 INIT_LIST_HEAD(&cgrp->children);
1273 INIT_LIST_HEAD(&cgrp->css_sets);
1274 INIT_LIST_HEAD(&cgrp->release_list);
Ben Blum72a8cb32009-09-23 15:56:27 -07001275 INIT_LIST_HEAD(&cgrp->pidlists);
1276 mutex_init(&cgrp->pidlist_mutex);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08001277 INIT_LIST_HEAD(&cgrp->event_list);
1278 spin_lock_init(&cgrp->event_list_lock);
Paul Menagecc31edc2008-10-18 20:28:04 -07001279}
Paul Menagec6d57f32009-09-23 15:56:19 -07001280
Paul Menageddbcc7e2007-10-18 23:39:30 -07001281static void init_cgroup_root(struct cgroupfs_root *root)
1282{
Paul Menagebd89aab2007-10-18 23:40:44 -07001283 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001284 INIT_LIST_HEAD(&root->subsys_list);
1285 INIT_LIST_HEAD(&root->root_list);
1286 root->number_of_cgroups = 1;
Paul Menagebd89aab2007-10-18 23:40:44 -07001287 cgrp->root = root;
1288 cgrp->top_cgroup = cgrp;
Paul Menagecc31edc2008-10-18 20:28:04 -07001289 init_cgroup_housekeeping(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001290}
1291
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001292static bool init_root_id(struct cgroupfs_root *root)
1293{
1294 int ret = 0;
1295
1296 do {
1297 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1298 return false;
1299 spin_lock(&hierarchy_id_lock);
1300 /* Try to allocate the next unused ID */
1301 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1302 &root->hierarchy_id);
1303 if (ret == -ENOSPC)
1304 /* Try again starting from 0 */
1305 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1306 if (!ret) {
1307 next_hierarchy_id = root->hierarchy_id + 1;
1308 } else if (ret != -EAGAIN) {
1309 /* Can only get here if the 31-bit IDR is full ... */
1310 BUG_ON(ret);
1311 }
1312 spin_unlock(&hierarchy_id_lock);
1313 } while (ret);
1314 return true;
1315}
1316
Paul Menageddbcc7e2007-10-18 23:39:30 -07001317static int cgroup_test_super(struct super_block *sb, void *data)
1318{
Paul Menagec6d57f32009-09-23 15:56:19 -07001319 struct cgroup_sb_opts *opts = data;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001320 struct cgroupfs_root *root = sb->s_fs_info;
1321
Paul Menagec6d57f32009-09-23 15:56:19 -07001322 /* If we asked for a name then it must match */
1323 if (opts->name && strcmp(opts->name, root->name))
1324 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001325
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001326 /*
1327 * If we asked for subsystems (or explicitly for no
1328 * subsystems) then they must match
1329 */
1330 if ((opts->subsys_bits || opts->none)
1331 && (opts->subsys_bits != root->subsys_bits))
Paul Menageddbcc7e2007-10-18 23:39:30 -07001332 return 0;
1333
1334 return 1;
1335}
1336
Paul Menagec6d57f32009-09-23 15:56:19 -07001337static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1338{
1339 struct cgroupfs_root *root;
1340
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001341 if (!opts->subsys_bits && !opts->none)
Paul Menagec6d57f32009-09-23 15:56:19 -07001342 return NULL;
1343
1344 root = kzalloc(sizeof(*root), GFP_KERNEL);
1345 if (!root)
1346 return ERR_PTR(-ENOMEM);
1347
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001348 if (!init_root_id(root)) {
1349 kfree(root);
1350 return ERR_PTR(-ENOMEM);
1351 }
Paul Menagec6d57f32009-09-23 15:56:19 -07001352 init_cgroup_root(root);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001353
Paul Menagec6d57f32009-09-23 15:56:19 -07001354 root->subsys_bits = opts->subsys_bits;
1355 root->flags = opts->flags;
1356 if (opts->release_agent)
1357 strcpy(root->release_agent_path, opts->release_agent);
1358 if (opts->name)
1359 strcpy(root->name, opts->name);
1360 return root;
1361}
1362
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001363static void cgroup_drop_root(struct cgroupfs_root *root)
1364{
1365 if (!root)
1366 return;
1367
1368 BUG_ON(!root->hierarchy_id);
1369 spin_lock(&hierarchy_id_lock);
1370 ida_remove(&hierarchy_ida, root->hierarchy_id);
1371 spin_unlock(&hierarchy_id_lock);
1372 kfree(root);
1373}
1374
Paul Menageddbcc7e2007-10-18 23:39:30 -07001375static int cgroup_set_super(struct super_block *sb, void *data)
1376{
1377 int ret;
Paul Menagec6d57f32009-09-23 15:56:19 -07001378 struct cgroup_sb_opts *opts = data;
1379
1380 /* If we don't have a new root, we can't set up a new sb */
1381 if (!opts->new_root)
1382 return -EINVAL;
1383
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001384 BUG_ON(!opts->subsys_bits && !opts->none);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001385
1386 ret = set_anon_super(sb, NULL);
1387 if (ret)
1388 return ret;
1389
Paul Menagec6d57f32009-09-23 15:56:19 -07001390 sb->s_fs_info = opts->new_root;
1391 opts->new_root->sb = sb;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001392
1393 sb->s_blocksize = PAGE_CACHE_SIZE;
1394 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1395 sb->s_magic = CGROUP_SUPER_MAGIC;
1396 sb->s_op = &cgroup_ops;
1397
1398 return 0;
1399}
1400
1401static int cgroup_get_rootdir(struct super_block *sb)
1402{
1403 struct inode *inode =
1404 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1405 struct dentry *dentry;
1406
1407 if (!inode)
1408 return -ENOMEM;
1409
Paul Menageddbcc7e2007-10-18 23:39:30 -07001410 inode->i_fop = &simple_dir_operations;
1411 inode->i_op = &cgroup_dir_inode_operations;
1412 /* directories start off with i_nlink == 2 (for "." entry) */
1413 inc_nlink(inode);
1414 dentry = d_alloc_root(inode);
1415 if (!dentry) {
1416 iput(inode);
1417 return -ENOMEM;
1418 }
1419 sb->s_root = dentry;
1420 return 0;
1421}
1422
1423static int cgroup_get_sb(struct file_system_type *fs_type,
1424 int flags, const char *unused_dev_name,
1425 void *data, struct vfsmount *mnt)
1426{
1427 struct cgroup_sb_opts opts;
Paul Menagec6d57f32009-09-23 15:56:19 -07001428 struct cgroupfs_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001429 int ret = 0;
1430 struct super_block *sb;
Paul Menagec6d57f32009-09-23 15:56:19 -07001431 struct cgroupfs_root *new_root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001432
1433 /* First find the desired set of subsystems */
Ben Blumaae8aab2010-03-10 15:22:07 -08001434 mutex_lock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001435 ret = parse_cgroupfs_options(data, &opts);
Ben Blumaae8aab2010-03-10 15:22:07 -08001436 mutex_unlock(&cgroup_mutex);
Paul Menagec6d57f32009-09-23 15:56:19 -07001437 if (ret)
1438 goto out_err;
1439
1440 /*
1441 * Allocate a new cgroup root. We may not need it if we're
1442 * reusing an existing hierarchy.
1443 */
1444 new_root = cgroup_root_from_opts(&opts);
1445 if (IS_ERR(new_root)) {
1446 ret = PTR_ERR(new_root);
Ben Blumcf5d5942010-03-10 15:22:09 -08001447 goto drop_modules;
Paul Menage81a6a5c2007-10-18 23:39:38 -07001448 }
Paul Menagec6d57f32009-09-23 15:56:19 -07001449 opts.new_root = new_root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001450
Paul Menagec6d57f32009-09-23 15:56:19 -07001451 /* Locate an existing or new sb for this hierarchy */
1452 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001453 if (IS_ERR(sb)) {
Paul Menagec6d57f32009-09-23 15:56:19 -07001454 ret = PTR_ERR(sb);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001455 cgroup_drop_root(opts.new_root);
Ben Blumcf5d5942010-03-10 15:22:09 -08001456 goto drop_modules;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001457 }
1458
Paul Menagec6d57f32009-09-23 15:56:19 -07001459 root = sb->s_fs_info;
1460 BUG_ON(!root);
1461 if (root == opts.new_root) {
1462 /* We used the new root structure, so this is a new hierarchy */
1463 struct list_head tmp_cg_links;
Li Zefanc12f65d2009-01-07 18:07:42 -08001464 struct cgroup *root_cgrp = &root->top_cgroup;
Paul Menage817929e2007-10-18 23:39:36 -07001465 struct inode *inode;
Paul Menagec6d57f32009-09-23 15:56:19 -07001466 struct cgroupfs_root *existing_root;
Li Zefan28fd5df2008-04-29 01:00:13 -07001467 int i;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001468
1469 BUG_ON(sb->s_root != NULL);
1470
1471 ret = cgroup_get_rootdir(sb);
1472 if (ret)
1473 goto drop_new_super;
Paul Menage817929e2007-10-18 23:39:36 -07001474 inode = sb->s_root->d_inode;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001475
Paul Menage817929e2007-10-18 23:39:36 -07001476 mutex_lock(&inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001477 mutex_lock(&cgroup_mutex);
1478
Paul Menagec6d57f32009-09-23 15:56:19 -07001479 if (strlen(root->name)) {
1480 /* Check for name clashes with existing mounts */
1481 for_each_active_root(existing_root) {
1482 if (!strcmp(existing_root->name, root->name)) {
1483 ret = -EBUSY;
1484 mutex_unlock(&cgroup_mutex);
1485 mutex_unlock(&inode->i_mutex);
1486 goto drop_new_super;
1487 }
1488 }
1489 }
1490
Paul Menage817929e2007-10-18 23:39:36 -07001491 /*
1492 * We're accessing css_set_count without locking
1493 * css_set_lock here, but that's OK - it can only be
1494 * increased by someone holding cgroup_lock, and
1495 * that's us. The worst that can happen is that we
1496 * have some link structures left over
1497 */
1498 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1499 if (ret) {
1500 mutex_unlock(&cgroup_mutex);
1501 mutex_unlock(&inode->i_mutex);
1502 goto drop_new_super;
1503 }
1504
Paul Menageddbcc7e2007-10-18 23:39:30 -07001505 ret = rebind_subsystems(root, root->subsys_bits);
1506 if (ret == -EBUSY) {
1507 mutex_unlock(&cgroup_mutex);
Paul Menage817929e2007-10-18 23:39:36 -07001508 mutex_unlock(&inode->i_mutex);
Paul Menagec6d57f32009-09-23 15:56:19 -07001509 free_cg_links(&tmp_cg_links);
1510 goto drop_new_super;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001511 }
Ben Blumcf5d5942010-03-10 15:22:09 -08001512 /*
1513 * There must be no failure case after here, since rebinding
1514 * takes care of subsystems' refcounts, which are explicitly
1515 * dropped in the failure exit path.
1516 */
Paul Menageddbcc7e2007-10-18 23:39:30 -07001517
1518 /* EBUSY should be the only error here */
1519 BUG_ON(ret);
1520
1521 list_add(&root->root_list, &roots);
Paul Menage817929e2007-10-18 23:39:36 -07001522 root_count++;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001523
Li Zefanc12f65d2009-01-07 18:07:42 -08001524 sb->s_root->d_fsdata = root_cgrp;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001525 root->top_cgroup.dentry = sb->s_root;
1526
Paul Menage817929e2007-10-18 23:39:36 -07001527 /* Link the top cgroup in this hierarchy into all
1528 * the css_set objects */
1529 write_lock(&css_set_lock);
Li Zefan28fd5df2008-04-29 01:00:13 -07001530 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1531 struct hlist_head *hhead = &css_set_table[i];
1532 struct hlist_node *node;
Paul Menage817929e2007-10-18 23:39:36 -07001533 struct css_set *cg;
Li Zefan28fd5df2008-04-29 01:00:13 -07001534
Li Zefanc12f65d2009-01-07 18:07:42 -08001535 hlist_for_each_entry(cg, node, hhead, hlist)
1536 link_css_set(&tmp_cg_links, cg, root_cgrp);
Li Zefan28fd5df2008-04-29 01:00:13 -07001537 }
Paul Menage817929e2007-10-18 23:39:36 -07001538 write_unlock(&css_set_lock);
1539
1540 free_cg_links(&tmp_cg_links);
1541
Li Zefanc12f65d2009-01-07 18:07:42 -08001542 BUG_ON(!list_empty(&root_cgrp->sibling));
1543 BUG_ON(!list_empty(&root_cgrp->children));
Paul Menageddbcc7e2007-10-18 23:39:30 -07001544 BUG_ON(root->number_of_cgroups != 1);
1545
Li Zefanc12f65d2009-01-07 18:07:42 -08001546 cgroup_populate_dir(root_cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001547 mutex_unlock(&cgroup_mutex);
Xiaotian Feng34f77a92009-09-23 15:56:18 -07001548 mutex_unlock(&inode->i_mutex);
Paul Menagec6d57f32009-09-23 15:56:19 -07001549 } else {
1550 /*
1551 * We re-used an existing hierarchy - the new root (if
1552 * any) is not needed
1553 */
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001554 cgroup_drop_root(opts.new_root);
Ben Blumcf5d5942010-03-10 15:22:09 -08001555 /* no subsys rebinding, so refcounts don't change */
1556 drop_parsed_module_refcounts(opts.subsys_bits);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001557 }
1558
Sukadev Bhattiprolua3ec9472009-03-04 12:06:34 -08001559 simple_set_mnt(mnt, sb);
Paul Menagec6d57f32009-09-23 15:56:19 -07001560 kfree(opts.release_agent);
1561 kfree(opts.name);
Sukadev Bhattiprolua3ec9472009-03-04 12:06:34 -08001562 return 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001563
1564 drop_new_super:
Al Viro6f5bbff2009-05-06 01:34:22 -04001565 deactivate_locked_super(sb);
Ben Blumcf5d5942010-03-10 15:22:09 -08001566 drop_modules:
1567 drop_parsed_module_refcounts(opts.subsys_bits);
Paul Menagec6d57f32009-09-23 15:56:19 -07001568 out_err:
1569 kfree(opts.release_agent);
1570 kfree(opts.name);
1571
Paul Menageddbcc7e2007-10-18 23:39:30 -07001572 return ret;
1573}
1574
1575static void cgroup_kill_sb(struct super_block *sb) {
1576 struct cgroupfs_root *root = sb->s_fs_info;
Paul Menagebd89aab2007-10-18 23:40:44 -07001577 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001578 int ret;
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -07001579 struct cg_cgroup_link *link;
1580 struct cg_cgroup_link *saved_link;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001581
1582 BUG_ON(!root);
1583
1584 BUG_ON(root->number_of_cgroups != 1);
Paul Menagebd89aab2007-10-18 23:40:44 -07001585 BUG_ON(!list_empty(&cgrp->children));
1586 BUG_ON(!list_empty(&cgrp->sibling));
Paul Menageddbcc7e2007-10-18 23:39:30 -07001587
1588 mutex_lock(&cgroup_mutex);
1589
1590 /* Rebind all subsystems back to the default hierarchy */
1591 ret = rebind_subsystems(root, 0);
1592 /* Shouldn't be able to fail ... */
1593 BUG_ON(ret);
1594
Paul Menage817929e2007-10-18 23:39:36 -07001595 /*
1596 * Release all the links from css_sets to this hierarchy's
1597 * root cgroup
1598 */
1599 write_lock(&css_set_lock);
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -07001600
1601 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1602 cgrp_link_list) {
Paul Menage817929e2007-10-18 23:39:36 -07001603 list_del(&link->cg_link_list);
Paul Menagebd89aab2007-10-18 23:40:44 -07001604 list_del(&link->cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -07001605 kfree(link);
1606 }
1607 write_unlock(&css_set_lock);
1608
Paul Menage839ec542009-01-29 14:25:22 -08001609 if (!list_empty(&root->root_list)) {
1610 list_del(&root->root_list);
1611 root_count--;
1612 }
Li Zefane5f6a862009-01-07 18:07:41 -08001613
Paul Menageddbcc7e2007-10-18 23:39:30 -07001614 mutex_unlock(&cgroup_mutex);
1615
Paul Menageddbcc7e2007-10-18 23:39:30 -07001616 kill_litter_super(sb);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07001617 cgroup_drop_root(root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001618}
1619
1620static struct file_system_type cgroup_fs_type = {
1621 .name = "cgroup",
1622 .get_sb = cgroup_get_sb,
1623 .kill_sb = cgroup_kill_sb,
1624};
1625
Paul Menagebd89aab2007-10-18 23:40:44 -07001626static inline struct cgroup *__d_cgrp(struct dentry *dentry)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001627{
1628 return dentry->d_fsdata;
1629}
1630
1631static inline struct cftype *__d_cft(struct dentry *dentry)
1632{
1633 return dentry->d_fsdata;
1634}
1635
Li Zefana043e3b2008-02-23 15:24:09 -08001636/**
1637 * cgroup_path - generate the path of a cgroup
1638 * @cgrp: the cgroup in question
1639 * @buf: the buffer to write the path into
1640 * @buflen: the length of the buffer
1641 *
Paul Menagea47295e2009-01-07 18:07:44 -08001642 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1643 * reference. Writes path of cgroup into buf. Returns 0 on success,
1644 * -errno on error.
Paul Menageddbcc7e2007-10-18 23:39:30 -07001645 */
Paul Menagebd89aab2007-10-18 23:40:44 -07001646int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001647{
1648 char *start;
Li Zefan9a9686b2010-04-22 17:29:24 +08001649 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1650 rcu_read_lock_held() ||
1651 cgroup_lock_is_held());
Paul Menageddbcc7e2007-10-18 23:39:30 -07001652
Paul Menagea47295e2009-01-07 18:07:44 -08001653 if (!dentry || cgrp == dummytop) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07001654 /*
1655 * Inactive subsystems have no dentry for their root
1656 * cgroup
1657 */
1658 strcpy(buf, "/");
1659 return 0;
1660 }
1661
1662 start = buf + buflen;
1663
1664 *--start = '\0';
1665 for (;;) {
Paul Menagea47295e2009-01-07 18:07:44 -08001666 int len = dentry->d_name.len;
Li Zefan9a9686b2010-04-22 17:29:24 +08001667
Paul Menageddbcc7e2007-10-18 23:39:30 -07001668 if ((start -= len) < buf)
1669 return -ENAMETOOLONG;
Li Zefan9a9686b2010-04-22 17:29:24 +08001670 memcpy(start, dentry->d_name.name, len);
Paul Menagebd89aab2007-10-18 23:40:44 -07001671 cgrp = cgrp->parent;
1672 if (!cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001673 break;
Li Zefan9a9686b2010-04-22 17:29:24 +08001674
1675 dentry = rcu_dereference_check(cgrp->dentry,
1676 rcu_read_lock_held() ||
1677 cgroup_lock_is_held());
Paul Menagebd89aab2007-10-18 23:40:44 -07001678 if (!cgrp->parent)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001679 continue;
1680 if (--start < buf)
1681 return -ENAMETOOLONG;
1682 *start = '/';
1683 }
1684 memmove(buf, start, buf + buflen - start);
1685 return 0;
1686}
Ben Blum67523c42010-03-10 15:22:11 -08001687EXPORT_SYMBOL_GPL(cgroup_path);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001688
Li Zefana043e3b2008-02-23 15:24:09 -08001689/**
1690 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1691 * @cgrp: the cgroup the task is attaching to
1692 * @tsk: the task to be attached
Paul Menagebbcb81d2007-10-18 23:39:32 -07001693 *
Li Zefana043e3b2008-02-23 15:24:09 -08001694 * Call holding cgroup_mutex. May take task_lock of
1695 * the task 'tsk' during call.
Paul Menagebbcb81d2007-10-18 23:39:32 -07001696 */
Cliff Wickman956db3c2008-02-07 00:14:43 -08001697int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
Paul Menagebbcb81d2007-10-18 23:39:32 -07001698{
1699 int retval = 0;
Daisuke Nishimura2468c722010-03-10 15:22:03 -08001700 struct cgroup_subsys *ss, *failed_ss = NULL;
Paul Menagebd89aab2007-10-18 23:40:44 -07001701 struct cgroup *oldcgrp;
Lai Jiangshan77efecd2009-01-07 18:07:39 -08001702 struct css_set *cg;
Paul Menage817929e2007-10-18 23:39:36 -07001703 struct css_set *newcg;
Paul Menagebd89aab2007-10-18 23:40:44 -07001704 struct cgroupfs_root *root = cgrp->root;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001705
1706 /* Nothing to do if the task is already in that cgroup */
Paul Menage7717f7b2009-09-23 15:56:22 -07001707 oldcgrp = task_cgroup_from_root(tsk, root);
Paul Menagebd89aab2007-10-18 23:40:44 -07001708 if (cgrp == oldcgrp)
Paul Menagebbcb81d2007-10-18 23:39:32 -07001709 return 0;
1710
1711 for_each_subsys(root, ss) {
1712 if (ss->can_attach) {
Ben Blumbe367d02009-09-23 15:56:31 -07001713 retval = ss->can_attach(ss, cgrp, tsk, false);
Daisuke Nishimura2468c722010-03-10 15:22:03 -08001714 if (retval) {
1715 /*
1716 * Remember on which subsystem the can_attach()
1717 * failed, so that we only call cancel_attach()
1718 * against the subsystems whose can_attach()
1719 * succeeded. (See below)
1720 */
1721 failed_ss = ss;
1722 goto out;
1723 }
Paul Menagebbcb81d2007-10-18 23:39:32 -07001724 }
1725 }
1726
Lai Jiangshan77efecd2009-01-07 18:07:39 -08001727 task_lock(tsk);
1728 cg = tsk->cgroups;
1729 get_css_set(cg);
1730 task_unlock(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07001731 /*
1732 * Locate or allocate a new css_set for this task,
1733 * based on its final set of cgroups
1734 */
Paul Menagebd89aab2007-10-18 23:40:44 -07001735 newcg = find_css_set(cg, cgrp);
Lai Jiangshan77efecd2009-01-07 18:07:39 -08001736 put_css_set(cg);
Daisuke Nishimura2468c722010-03-10 15:22:03 -08001737 if (!newcg) {
1738 retval = -ENOMEM;
1739 goto out;
1740 }
Paul Menage817929e2007-10-18 23:39:36 -07001741
Paul Menagebbcb81d2007-10-18 23:39:32 -07001742 task_lock(tsk);
1743 if (tsk->flags & PF_EXITING) {
1744 task_unlock(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07001745 put_css_set(newcg);
Daisuke Nishimura2468c722010-03-10 15:22:03 -08001746 retval = -ESRCH;
1747 goto out;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001748 }
Paul Menage817929e2007-10-18 23:39:36 -07001749 rcu_assign_pointer(tsk->cgroups, newcg);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001750 task_unlock(tsk);
1751
Paul Menage817929e2007-10-18 23:39:36 -07001752 /* Update the css_set linked lists if we're using them */
1753 write_lock(&css_set_lock);
1754 if (!list_empty(&tsk->cg_list)) {
1755 list_del(&tsk->cg_list);
1756 list_add(&tsk->cg_list, &newcg->tasks);
1757 }
1758 write_unlock(&css_set_lock);
1759
Paul Menagebbcb81d2007-10-18 23:39:32 -07001760 for_each_subsys(root, ss) {
Paul Jacksone18f6312008-02-07 00:13:44 -08001761 if (ss->attach)
Ben Blumbe367d02009-09-23 15:56:31 -07001762 ss->attach(ss, cgrp, oldcgrp, tsk, false);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001763 }
Paul Menagebd89aab2007-10-18 23:40:44 -07001764 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001765 synchronize_rcu();
Paul Menage817929e2007-10-18 23:39:36 -07001766 put_css_set(cg);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07001767
1768 /*
1769 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1770 * is no longer empty.
1771 */
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07001772 cgroup_wakeup_rmdir_waiter(cgrp);
Daisuke Nishimura2468c722010-03-10 15:22:03 -08001773out:
1774 if (retval) {
1775 for_each_subsys(root, ss) {
1776 if (ss == failed_ss)
1777 /*
1778 * This subsystem was the one that failed the
1779 * can_attach() check earlier, so we don't need
1780 * to call cancel_attach() against it or any
1781 * remaining subsystems.
1782 */
1783 break;
1784 if (ss->cancel_attach)
1785 ss->cancel_attach(ss, cgrp, tsk, false);
1786 }
1787 }
1788 return retval;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001789}
1790
Sridhar Samudralad7926ee2010-05-30 22:24:39 +02001791/**
1792 * cgroup_attach_task_current_cg - attach task 'tsk' to current task's cgroup
1793 * @tsk: the task to be attached
1794 */
1795int cgroup_attach_task_current_cg(struct task_struct *tsk)
1796{
1797 struct cgroupfs_root *root;
1798 struct cgroup *cur_cg;
1799 int retval = 0;
1800
1801 cgroup_lock();
1802 for_each_active_root(root) {
1803 cur_cg = task_cgroup_from_root(current, root);
1804 retval = cgroup_attach_task(cur_cg, tsk);
1805 if (retval)
1806 break;
1807 }
1808 cgroup_unlock();
1809
1810 return retval;
1811}
1812EXPORT_SYMBOL_GPL(cgroup_attach_task_current_cg);
1813
Paul Menagebbcb81d2007-10-18 23:39:32 -07001814/*
Paul Menageaf351022008-07-25 01:47:01 -07001815 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1816 * held. May take task_lock of task
Paul Menagebbcb81d2007-10-18 23:39:32 -07001817 */
Paul Menageaf351022008-07-25 01:47:01 -07001818static int attach_task_by_pid(struct cgroup *cgrp, u64 pid)
Paul Menagebbcb81d2007-10-18 23:39:32 -07001819{
Paul Menagebbcb81d2007-10-18 23:39:32 -07001820 struct task_struct *tsk;
David Howellsc69e8d92008-11-14 10:39:19 +11001821 const struct cred *cred = current_cred(), *tcred;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001822 int ret;
1823
Paul Menagebbcb81d2007-10-18 23:39:32 -07001824 if (pid) {
1825 rcu_read_lock();
Pavel Emelyanov73507f32008-02-07 00:14:47 -08001826 tsk = find_task_by_vpid(pid);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001827 if (!tsk || tsk->flags & PF_EXITING) {
1828 rcu_read_unlock();
1829 return -ESRCH;
1830 }
Paul Menagebbcb81d2007-10-18 23:39:32 -07001831
David Howellsc69e8d92008-11-14 10:39:19 +11001832 tcred = __task_cred(tsk);
1833 if (cred->euid &&
1834 cred->euid != tcred->uid &&
1835 cred->euid != tcred->suid) {
1836 rcu_read_unlock();
Paul Menagebbcb81d2007-10-18 23:39:32 -07001837 return -EACCES;
1838 }
David Howellsc69e8d92008-11-14 10:39:19 +11001839 get_task_struct(tsk);
1840 rcu_read_unlock();
Paul Menagebbcb81d2007-10-18 23:39:32 -07001841 } else {
1842 tsk = current;
1843 get_task_struct(tsk);
1844 }
1845
Cliff Wickman956db3c2008-02-07 00:14:43 -08001846 ret = cgroup_attach_task(cgrp, tsk);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001847 put_task_struct(tsk);
1848 return ret;
1849}
1850
Paul Menageaf351022008-07-25 01:47:01 -07001851static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
1852{
1853 int ret;
1854 if (!cgroup_lock_live_group(cgrp))
1855 return -ENODEV;
1856 ret = attach_task_by_pid(cgrp, pid);
1857 cgroup_unlock();
1858 return ret;
1859}
1860
Paul Menagee788e062008-07-25 01:46:59 -07001861/**
1862 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1863 * @cgrp: the cgroup to be checked for liveness
1864 *
Paul Menage84eea842008-07-25 01:47:00 -07001865 * On success, returns true; the lock should be later released with
1866 * cgroup_unlock(). On failure returns false with no lock held.
Paul Menagee788e062008-07-25 01:46:59 -07001867 */
Paul Menage84eea842008-07-25 01:47:00 -07001868bool cgroup_lock_live_group(struct cgroup *cgrp)
Paul Menagee788e062008-07-25 01:46:59 -07001869{
1870 mutex_lock(&cgroup_mutex);
1871 if (cgroup_is_removed(cgrp)) {
1872 mutex_unlock(&cgroup_mutex);
1873 return false;
1874 }
1875 return true;
1876}
Ben Blum67523c42010-03-10 15:22:11 -08001877EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
Paul Menagee788e062008-07-25 01:46:59 -07001878
1879static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
1880 const char *buffer)
1881{
1882 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1883 if (!cgroup_lock_live_group(cgrp))
1884 return -ENODEV;
1885 strcpy(cgrp->root->release_agent_path, buffer);
Paul Menage84eea842008-07-25 01:47:00 -07001886 cgroup_unlock();
Paul Menagee788e062008-07-25 01:46:59 -07001887 return 0;
1888}
1889
1890static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
1891 struct seq_file *seq)
1892{
1893 if (!cgroup_lock_live_group(cgrp))
1894 return -ENODEV;
1895 seq_puts(seq, cgrp->root->release_agent_path);
1896 seq_putc(seq, '\n');
Paul Menage84eea842008-07-25 01:47:00 -07001897 cgroup_unlock();
Paul Menagee788e062008-07-25 01:46:59 -07001898 return 0;
1899}
1900
Paul Menage84eea842008-07-25 01:47:00 -07001901/* A buffer size big enough for numbers or short strings */
1902#define CGROUP_LOCAL_BUFFER_SIZE 64
1903
Paul Menagee73d2c62008-04-29 01:00:06 -07001904static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
Paul Menagef4c753b2008-04-29 00:59:56 -07001905 struct file *file,
1906 const char __user *userbuf,
1907 size_t nbytes, loff_t *unused_ppos)
Paul Menage355e0c42007-10-18 23:39:33 -07001908{
Paul Menage84eea842008-07-25 01:47:00 -07001909 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
Paul Menage355e0c42007-10-18 23:39:33 -07001910 int retval = 0;
Paul Menage355e0c42007-10-18 23:39:33 -07001911 char *end;
1912
1913 if (!nbytes)
1914 return -EINVAL;
1915 if (nbytes >= sizeof(buffer))
1916 return -E2BIG;
1917 if (copy_from_user(buffer, userbuf, nbytes))
1918 return -EFAULT;
1919
1920 buffer[nbytes] = 0; /* nul-terminate */
Paul Menagee73d2c62008-04-29 01:00:06 -07001921 if (cft->write_u64) {
KOSAKI Motohiro478988d2009-10-26 16:49:36 -07001922 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
Paul Menagee73d2c62008-04-29 01:00:06 -07001923 if (*end)
1924 return -EINVAL;
1925 retval = cft->write_u64(cgrp, cft, val);
1926 } else {
KOSAKI Motohiro478988d2009-10-26 16:49:36 -07001927 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
Paul Menagee73d2c62008-04-29 01:00:06 -07001928 if (*end)
1929 return -EINVAL;
1930 retval = cft->write_s64(cgrp, cft, val);
1931 }
Paul Menage355e0c42007-10-18 23:39:33 -07001932 if (!retval)
1933 retval = nbytes;
1934 return retval;
1935}
1936
Paul Menagedb3b1492008-07-25 01:46:58 -07001937static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
1938 struct file *file,
1939 const char __user *userbuf,
1940 size_t nbytes, loff_t *unused_ppos)
1941{
Paul Menage84eea842008-07-25 01:47:00 -07001942 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
Paul Menagedb3b1492008-07-25 01:46:58 -07001943 int retval = 0;
1944 size_t max_bytes = cft->max_write_len;
1945 char *buffer = local_buffer;
1946
1947 if (!max_bytes)
1948 max_bytes = sizeof(local_buffer) - 1;
1949 if (nbytes >= max_bytes)
1950 return -E2BIG;
1951 /* Allocate a dynamic buffer if we need one */
1952 if (nbytes >= sizeof(local_buffer)) {
1953 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1954 if (buffer == NULL)
1955 return -ENOMEM;
1956 }
Li Zefan5a3eb9f2008-07-29 22:33:18 -07001957 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
1958 retval = -EFAULT;
1959 goto out;
1960 }
Paul Menagedb3b1492008-07-25 01:46:58 -07001961
1962 buffer[nbytes] = 0; /* nul-terminate */
KOSAKI Motohiro478988d2009-10-26 16:49:36 -07001963 retval = cft->write_string(cgrp, cft, strstrip(buffer));
Paul Menagedb3b1492008-07-25 01:46:58 -07001964 if (!retval)
1965 retval = nbytes;
Li Zefan5a3eb9f2008-07-29 22:33:18 -07001966out:
Paul Menagedb3b1492008-07-25 01:46:58 -07001967 if (buffer != local_buffer)
1968 kfree(buffer);
1969 return retval;
1970}
1971
Paul Menageddbcc7e2007-10-18 23:39:30 -07001972static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
1973 size_t nbytes, loff_t *ppos)
1974{
1975 struct cftype *cft = __d_cft(file->f_dentry);
Paul Menagebd89aab2007-10-18 23:40:44 -07001976 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001977
Li Zefan75139b82009-01-07 18:07:33 -08001978 if (cgroup_is_removed(cgrp))
Paul Menageddbcc7e2007-10-18 23:39:30 -07001979 return -ENODEV;
Paul Menage355e0c42007-10-18 23:39:33 -07001980 if (cft->write)
Paul Menagebd89aab2007-10-18 23:40:44 -07001981 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
Paul Menagee73d2c62008-04-29 01:00:06 -07001982 if (cft->write_u64 || cft->write_s64)
1983 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
Paul Menagedb3b1492008-07-25 01:46:58 -07001984 if (cft->write_string)
1985 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
Pavel Emelyanovd447ea22008-04-29 01:00:08 -07001986 if (cft->trigger) {
1987 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
1988 return ret ? ret : nbytes;
1989 }
Paul Menage355e0c42007-10-18 23:39:33 -07001990 return -EINVAL;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001991}
1992
Paul Menagef4c753b2008-04-29 00:59:56 -07001993static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
1994 struct file *file,
1995 char __user *buf, size_t nbytes,
1996 loff_t *ppos)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001997{
Paul Menage84eea842008-07-25 01:47:00 -07001998 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
Paul Menagef4c753b2008-04-29 00:59:56 -07001999 u64 val = cft->read_u64(cgrp, cft);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002000 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2001
2002 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2003}
2004
Paul Menagee73d2c62008-04-29 01:00:06 -07002005static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2006 struct file *file,
2007 char __user *buf, size_t nbytes,
2008 loff_t *ppos)
2009{
Paul Menage84eea842008-07-25 01:47:00 -07002010 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
Paul Menagee73d2c62008-04-29 01:00:06 -07002011 s64 val = cft->read_s64(cgrp, cft);
2012 int len = sprintf(tmp, "%lld\n", (long long) val);
2013
2014 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2015}
2016
Paul Menageddbcc7e2007-10-18 23:39:30 -07002017static ssize_t cgroup_file_read(struct file *file, 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;
2025
2026 if (cft->read)
Paul Menagebd89aab2007-10-18 23:40:44 -07002027 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
Paul Menagef4c753b2008-04-29 00:59:56 -07002028 if (cft->read_u64)
2029 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
Paul Menagee73d2c62008-04-29 01:00:06 -07002030 if (cft->read_s64)
2031 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002032 return -EINVAL;
2033}
2034
Paul Menage91796562008-04-29 01:00:01 -07002035/*
2036 * seqfile ops/methods for returning structured data. Currently just
2037 * supports string->u64 maps, but can be extended in future.
2038 */
2039
2040struct cgroup_seqfile_state {
2041 struct cftype *cft;
2042 struct cgroup *cgroup;
2043};
2044
2045static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2046{
2047 struct seq_file *sf = cb->state;
2048 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2049}
2050
2051static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2052{
2053 struct cgroup_seqfile_state *state = m->private;
2054 struct cftype *cft = state->cft;
Serge E. Hallyn29486df2008-04-29 01:00:14 -07002055 if (cft->read_map) {
2056 struct cgroup_map_cb cb = {
2057 .fill = cgroup_map_add,
2058 .state = m,
2059 };
2060 return cft->read_map(state->cgroup, cft, &cb);
2061 }
2062 return cft->read_seq_string(state->cgroup, cft, m);
Paul Menage91796562008-04-29 01:00:01 -07002063}
2064
Adrian Bunk96930a62008-07-25 19:46:21 -07002065static int cgroup_seqfile_release(struct inode *inode, struct file *file)
Paul Menage91796562008-04-29 01:00:01 -07002066{
2067 struct seq_file *seq = file->private_data;
2068 kfree(seq->private);
2069 return single_release(inode, file);
2070}
2071
Alexey Dobriyan828c0952009-10-01 15:43:56 -07002072static const struct file_operations cgroup_seqfile_operations = {
Paul Menage91796562008-04-29 01:00:01 -07002073 .read = seq_read,
Paul Menagee788e062008-07-25 01:46:59 -07002074 .write = cgroup_file_write,
Paul Menage91796562008-04-29 01:00:01 -07002075 .llseek = seq_lseek,
2076 .release = cgroup_seqfile_release,
2077};
2078
Paul Menageddbcc7e2007-10-18 23:39:30 -07002079static int cgroup_file_open(struct inode *inode, struct file *file)
2080{
2081 int err;
2082 struct cftype *cft;
2083
2084 err = generic_file_open(inode, file);
2085 if (err)
2086 return err;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002087 cft = __d_cft(file->f_dentry);
Li Zefan75139b82009-01-07 18:07:33 -08002088
Serge E. Hallyn29486df2008-04-29 01:00:14 -07002089 if (cft->read_map || cft->read_seq_string) {
Paul Menage91796562008-04-29 01:00:01 -07002090 struct cgroup_seqfile_state *state =
2091 kzalloc(sizeof(*state), GFP_USER);
2092 if (!state)
2093 return -ENOMEM;
2094 state->cft = cft;
2095 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2096 file->f_op = &cgroup_seqfile_operations;
2097 err = single_open(file, cgroup_seqfile_show, state);
2098 if (err < 0)
2099 kfree(state);
2100 } else if (cft->open)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002101 err = cft->open(inode, file);
2102 else
2103 err = 0;
2104
2105 return err;
2106}
2107
2108static int cgroup_file_release(struct inode *inode, struct file *file)
2109{
2110 struct cftype *cft = __d_cft(file->f_dentry);
2111 if (cft->release)
2112 return cft->release(inode, file);
2113 return 0;
2114}
2115
2116/*
2117 * cgroup_rename - Only allow simple rename of directories in place.
2118 */
2119static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2120 struct inode *new_dir, struct dentry *new_dentry)
2121{
2122 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2123 return -ENOTDIR;
2124 if (new_dentry->d_inode)
2125 return -EEXIST;
2126 if (old_dir != new_dir)
2127 return -EIO;
2128 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2129}
2130
Alexey Dobriyan828c0952009-10-01 15:43:56 -07002131static const struct file_operations cgroup_file_operations = {
Paul Menageddbcc7e2007-10-18 23:39:30 -07002132 .read = cgroup_file_read,
2133 .write = cgroup_file_write,
2134 .llseek = generic_file_llseek,
2135 .open = cgroup_file_open,
2136 .release = cgroup_file_release,
2137};
2138
Alexey Dobriyan6e1d5dc2009-09-21 17:01:11 -07002139static const struct inode_operations cgroup_dir_inode_operations = {
Paul Menageddbcc7e2007-10-18 23:39:30 -07002140 .lookup = simple_lookup,
2141 .mkdir = cgroup_mkdir,
2142 .rmdir = cgroup_rmdir,
2143 .rename = cgroup_rename,
2144};
2145
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08002146/*
2147 * Check if a file is a control file
2148 */
2149static inline struct cftype *__file_cft(struct file *file)
2150{
2151 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2152 return ERR_PTR(-EINVAL);
2153 return __d_cft(file->f_dentry);
2154}
2155
Li Zefan099fca32009-04-02 16:57:29 -07002156static int cgroup_create_file(struct dentry *dentry, mode_t mode,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002157 struct super_block *sb)
2158{
Al Viro3ba13d12009-02-20 06:02:22 +00002159 static const struct dentry_operations cgroup_dops = {
Paul Menageddbcc7e2007-10-18 23:39:30 -07002160 .d_iput = cgroup_diput,
2161 };
2162
2163 struct inode *inode;
2164
2165 if (!dentry)
2166 return -ENOENT;
2167 if (dentry->d_inode)
2168 return -EEXIST;
2169
2170 inode = cgroup_new_inode(mode, sb);
2171 if (!inode)
2172 return -ENOMEM;
2173
2174 if (S_ISDIR(mode)) {
2175 inode->i_op = &cgroup_dir_inode_operations;
2176 inode->i_fop = &simple_dir_operations;
2177
2178 /* start off with i_nlink == 2 (for "." entry) */
2179 inc_nlink(inode);
2180
2181 /* start with the directory inode held, so that we can
2182 * populate it without racing with another mkdir */
Paul Menage817929e2007-10-18 23:39:36 -07002183 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002184 } else if (S_ISREG(mode)) {
2185 inode->i_size = 0;
2186 inode->i_fop = &cgroup_file_operations;
2187 }
2188 dentry->d_op = &cgroup_dops;
2189 d_instantiate(dentry, inode);
2190 dget(dentry); /* Extra count - pin the dentry in core */
2191 return 0;
2192}
2193
2194/*
Li Zefana043e3b2008-02-23 15:24:09 -08002195 * cgroup_create_dir - create a directory for an object.
2196 * @cgrp: the cgroup we create the directory for. It must have a valid
2197 * ->parent field. And we are going to fill its ->dentry field.
2198 * @dentry: dentry of the new cgroup
2199 * @mode: mode to set on new directory.
Paul Menageddbcc7e2007-10-18 23:39:30 -07002200 */
Paul Menagebd89aab2007-10-18 23:40:44 -07002201static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
Li Zefan099fca32009-04-02 16:57:29 -07002202 mode_t mode)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002203{
2204 struct dentry *parent;
2205 int error = 0;
2206
Paul Menagebd89aab2007-10-18 23:40:44 -07002207 parent = cgrp->parent->dentry;
2208 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002209 if (!error) {
Paul Menagebd89aab2007-10-18 23:40:44 -07002210 dentry->d_fsdata = cgrp;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002211 inc_nlink(parent->d_inode);
Paul Menagea47295e2009-01-07 18:07:44 -08002212 rcu_assign_pointer(cgrp->dentry, dentry);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002213 dget(dentry);
2214 }
2215 dput(dentry);
2216
2217 return error;
2218}
2219
Li Zefan099fca32009-04-02 16:57:29 -07002220/**
2221 * cgroup_file_mode - deduce file mode of a control file
2222 * @cft: the control file in question
2223 *
2224 * returns cft->mode if ->mode is not 0
2225 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2226 * returns S_IRUGO if it has only a read handler
2227 * returns S_IWUSR if it has only a write hander
2228 */
2229static mode_t cgroup_file_mode(const struct cftype *cft)
2230{
2231 mode_t mode = 0;
2232
2233 if (cft->mode)
2234 return cft->mode;
2235
2236 if (cft->read || cft->read_u64 || cft->read_s64 ||
2237 cft->read_map || cft->read_seq_string)
2238 mode |= S_IRUGO;
2239
2240 if (cft->write || cft->write_u64 || cft->write_s64 ||
2241 cft->write_string || cft->trigger)
2242 mode |= S_IWUSR;
2243
2244 return mode;
2245}
2246
Paul Menagebd89aab2007-10-18 23:40:44 -07002247int cgroup_add_file(struct cgroup *cgrp,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002248 struct cgroup_subsys *subsys,
2249 const struct cftype *cft)
2250{
Paul Menagebd89aab2007-10-18 23:40:44 -07002251 struct dentry *dir = cgrp->dentry;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002252 struct dentry *dentry;
2253 int error;
Li Zefan099fca32009-04-02 16:57:29 -07002254 mode_t mode;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002255
2256 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
Paul Menagebd89aab2007-10-18 23:40:44 -07002257 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07002258 strcpy(name, subsys->name);
2259 strcat(name, ".");
2260 }
2261 strcat(name, cft->name);
2262 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2263 dentry = lookup_one_len(name, dir, strlen(name));
2264 if (!IS_ERR(dentry)) {
Li Zefan099fca32009-04-02 16:57:29 -07002265 mode = cgroup_file_mode(cft);
2266 error = cgroup_create_file(dentry, mode | S_IFREG,
Paul Menagebd89aab2007-10-18 23:40:44 -07002267 cgrp->root->sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002268 if (!error)
2269 dentry->d_fsdata = (void *)cft;
2270 dput(dentry);
2271 } else
2272 error = PTR_ERR(dentry);
2273 return error;
2274}
Ben Blume6a11052010-03-10 15:22:09 -08002275EXPORT_SYMBOL_GPL(cgroup_add_file);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002276
Paul Menagebd89aab2007-10-18 23:40:44 -07002277int cgroup_add_files(struct cgroup *cgrp,
Paul Menageddbcc7e2007-10-18 23:39:30 -07002278 struct cgroup_subsys *subsys,
2279 const struct cftype cft[],
2280 int count)
2281{
2282 int i, err;
2283 for (i = 0; i < count; i++) {
Paul Menagebd89aab2007-10-18 23:40:44 -07002284 err = cgroup_add_file(cgrp, subsys, &cft[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002285 if (err)
2286 return err;
2287 }
2288 return 0;
2289}
Ben Blume6a11052010-03-10 15:22:09 -08002290EXPORT_SYMBOL_GPL(cgroup_add_files);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002291
Li Zefana043e3b2008-02-23 15:24:09 -08002292/**
2293 * cgroup_task_count - count the number of tasks in a cgroup.
2294 * @cgrp: the cgroup in question
2295 *
2296 * Return the number of tasks in the cgroup.
2297 */
Paul Menagebd89aab2007-10-18 23:40:44 -07002298int cgroup_task_count(const struct cgroup *cgrp)
Paul Menagebbcb81d2007-10-18 23:39:32 -07002299{
2300 int count = 0;
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -07002301 struct cg_cgroup_link *link;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002302
Paul Menage817929e2007-10-18 23:39:36 -07002303 read_lock(&css_set_lock);
KOSAKI Motohiro71cbb942008-07-25 01:46:55 -07002304 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
Lai Jiangshan146aa1b2008-10-18 20:28:03 -07002305 count += atomic_read(&link->cg->refcount);
Paul Menage817929e2007-10-18 23:39:36 -07002306 }
2307 read_unlock(&css_set_lock);
Paul Menagebbcb81d2007-10-18 23:39:32 -07002308 return count;
2309}
2310
2311/*
Paul Menage817929e2007-10-18 23:39:36 -07002312 * Advance a list_head iterator. The iterator should be positioned at
2313 * the start of a css_set
2314 */
Paul Menagebd89aab2007-10-18 23:40:44 -07002315static void cgroup_advance_iter(struct cgroup *cgrp,
Paul Menage7717f7b2009-09-23 15:56:22 -07002316 struct cgroup_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07002317{
2318 struct list_head *l = it->cg_link;
2319 struct cg_cgroup_link *link;
2320 struct css_set *cg;
2321
2322 /* Advance to the next non-empty css_set */
2323 do {
2324 l = l->next;
Paul Menagebd89aab2007-10-18 23:40:44 -07002325 if (l == &cgrp->css_sets) {
Paul Menage817929e2007-10-18 23:39:36 -07002326 it->cg_link = NULL;
2327 return;
2328 }
Paul Menagebd89aab2007-10-18 23:40:44 -07002329 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -07002330 cg = link->cg;
2331 } while (list_empty(&cg->tasks));
2332 it->cg_link = l;
2333 it->task = cg->tasks.next;
2334}
2335
Cliff Wickman31a7df02008-02-07 00:14:42 -08002336/*
2337 * To reduce the fork() overhead for systems that are not actually
2338 * using their cgroups capability, we don't maintain the lists running
2339 * through each css_set to its tasks until we see the list actually
2340 * used - in other words after the first call to cgroup_iter_start().
2341 *
2342 * The tasklist_lock is not held here, as do_each_thread() and
2343 * while_each_thread() are protected by RCU.
2344 */
Adrian Bunk3df91fe2008-04-29 00:59:54 -07002345static void cgroup_enable_task_cg_lists(void)
Cliff Wickman31a7df02008-02-07 00:14:42 -08002346{
2347 struct task_struct *p, *g;
2348 write_lock(&css_set_lock);
2349 use_task_css_set_links = 1;
2350 do_each_thread(g, p) {
2351 task_lock(p);
Li Zefan0e043882008-04-17 11:37:15 +08002352 /*
2353 * We should check if the process is exiting, otherwise
2354 * it will race with cgroup_exit() in that the list
2355 * entry won't be deleted though the process has exited.
2356 */
2357 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
Cliff Wickman31a7df02008-02-07 00:14:42 -08002358 list_add(&p->cg_list, &p->cgroups->tasks);
2359 task_unlock(p);
2360 } while_each_thread(g, p);
2361 write_unlock(&css_set_lock);
2362}
2363
Paul Menagebd89aab2007-10-18 23:40:44 -07002364void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07002365{
2366 /*
2367 * The first time anyone tries to iterate across a cgroup,
2368 * we need to enable the list linking each css_set to its
2369 * tasks, and fix up all existing tasks.
2370 */
Cliff Wickman31a7df02008-02-07 00:14:42 -08002371 if (!use_task_css_set_links)
2372 cgroup_enable_task_cg_lists();
2373
Paul Menage817929e2007-10-18 23:39:36 -07002374 read_lock(&css_set_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07002375 it->cg_link = &cgrp->css_sets;
2376 cgroup_advance_iter(cgrp, it);
Paul Menage817929e2007-10-18 23:39:36 -07002377}
2378
Paul Menagebd89aab2007-10-18 23:40:44 -07002379struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -07002380 struct cgroup_iter *it)
2381{
2382 struct task_struct *res;
2383 struct list_head *l = it->task;
Lai Jiangshan2019f632009-01-07 18:07:36 -08002384 struct cg_cgroup_link *link;
Paul Menage817929e2007-10-18 23:39:36 -07002385
2386 /* If the iterator cg is NULL, we have no tasks */
2387 if (!it->cg_link)
2388 return NULL;
2389 res = list_entry(l, struct task_struct, cg_list);
2390 /* Advance iterator to find next entry */
2391 l = l->next;
Lai Jiangshan2019f632009-01-07 18:07:36 -08002392 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2393 if (l == &link->cg->tasks) {
Paul Menage817929e2007-10-18 23:39:36 -07002394 /* We reached the end of this task list - move on to
2395 * the next cg_cgroup_link */
Paul Menagebd89aab2007-10-18 23:40:44 -07002396 cgroup_advance_iter(cgrp, it);
Paul Menage817929e2007-10-18 23:39:36 -07002397 } else {
2398 it->task = l;
2399 }
2400 return res;
2401}
2402
Paul Menagebd89aab2007-10-18 23:40:44 -07002403void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07002404{
2405 read_unlock(&css_set_lock);
2406}
2407
Cliff Wickman31a7df02008-02-07 00:14:42 -08002408static inline int started_after_time(struct task_struct *t1,
2409 struct timespec *time,
2410 struct task_struct *t2)
2411{
2412 int start_diff = timespec_compare(&t1->start_time, time);
2413 if (start_diff > 0) {
2414 return 1;
2415 } else if (start_diff < 0) {
2416 return 0;
2417 } else {
2418 /*
2419 * Arbitrarily, if two processes started at the same
2420 * time, we'll say that the lower pointer value
2421 * started first. Note that t2 may have exited by now
2422 * so this may not be a valid pointer any longer, but
2423 * that's fine - it still serves to distinguish
2424 * between two tasks started (effectively) simultaneously.
2425 */
2426 return t1 > t2;
2427 }
2428}
2429
2430/*
2431 * This function is a callback from heap_insert() and is used to order
2432 * the heap.
2433 * In this case we order the heap in descending task start time.
2434 */
2435static inline int started_after(void *p1, void *p2)
2436{
2437 struct task_struct *t1 = p1;
2438 struct task_struct *t2 = p2;
2439 return started_after_time(t1, &t2->start_time, t2);
2440}
2441
2442/**
2443 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2444 * @scan: struct cgroup_scanner containing arguments for the scan
2445 *
2446 * Arguments include pointers to callback functions test_task() and
2447 * process_task().
2448 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2449 * and if it returns true, call process_task() for it also.
2450 * The test_task pointer may be NULL, meaning always true (select all tasks).
2451 * Effectively duplicates cgroup_iter_{start,next,end}()
2452 * but does not lock css_set_lock for the call to process_task().
2453 * The struct cgroup_scanner may be embedded in any structure of the caller's
2454 * creation.
2455 * It is guaranteed that process_task() will act on every task that
2456 * is a member of the cgroup for the duration of this call. This
2457 * function may or may not call process_task() for tasks that exit
2458 * or move to a different cgroup during the call, or are forked or
2459 * move into the cgroup during the call.
2460 *
2461 * Note that test_task() may be called with locks held, and may in some
2462 * situations be called multiple times for the same task, so it should
2463 * be cheap.
2464 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2465 * pre-allocated and will be used for heap operations (and its "gt" member will
2466 * be overwritten), else a temporary heap will be used (allocation of which
2467 * may cause this function to fail).
2468 */
2469int cgroup_scan_tasks(struct cgroup_scanner *scan)
2470{
2471 int retval, i;
2472 struct cgroup_iter it;
2473 struct task_struct *p, *dropped;
2474 /* Never dereference latest_task, since it's not refcounted */
2475 struct task_struct *latest_task = NULL;
2476 struct ptr_heap tmp_heap;
2477 struct ptr_heap *heap;
2478 struct timespec latest_time = { 0, 0 };
2479
2480 if (scan->heap) {
2481 /* The caller supplied our heap and pre-allocated its memory */
2482 heap = scan->heap;
2483 heap->gt = &started_after;
2484 } else {
2485 /* We need to allocate our own heap memory */
2486 heap = &tmp_heap;
2487 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2488 if (retval)
2489 /* cannot allocate the heap */
2490 return retval;
2491 }
2492
2493 again:
2494 /*
2495 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2496 * to determine which are of interest, and using the scanner's
2497 * "process_task" callback to process any of them that need an update.
2498 * Since we don't want to hold any locks during the task updates,
2499 * gather tasks to be processed in a heap structure.
2500 * The heap is sorted by descending task start time.
2501 * If the statically-sized heap fills up, we overflow tasks that
2502 * started later, and in future iterations only consider tasks that
2503 * started after the latest task in the previous pass. This
2504 * guarantees forward progress and that we don't miss any tasks.
2505 */
2506 heap->size = 0;
2507 cgroup_iter_start(scan->cg, &it);
2508 while ((p = cgroup_iter_next(scan->cg, &it))) {
2509 /*
2510 * Only affect tasks that qualify per the caller's callback,
2511 * if he provided one
2512 */
2513 if (scan->test_task && !scan->test_task(p, scan))
2514 continue;
2515 /*
2516 * Only process tasks that started after the last task
2517 * we processed
2518 */
2519 if (!started_after_time(p, &latest_time, latest_task))
2520 continue;
2521 dropped = heap_insert(heap, p);
2522 if (dropped == NULL) {
2523 /*
2524 * The new task was inserted; the heap wasn't
2525 * previously full
2526 */
2527 get_task_struct(p);
2528 } else if (dropped != p) {
2529 /*
2530 * The new task was inserted, and pushed out a
2531 * different task
2532 */
2533 get_task_struct(p);
2534 put_task_struct(dropped);
2535 }
2536 /*
2537 * Else the new task was newer than anything already in
2538 * the heap and wasn't inserted
2539 */
2540 }
2541 cgroup_iter_end(scan->cg, &it);
2542
2543 if (heap->size) {
2544 for (i = 0; i < heap->size; i++) {
Paul Jackson4fe91d52008-04-29 00:59:55 -07002545 struct task_struct *q = heap->ptrs[i];
Cliff Wickman31a7df02008-02-07 00:14:42 -08002546 if (i == 0) {
Paul Jackson4fe91d52008-04-29 00:59:55 -07002547 latest_time = q->start_time;
2548 latest_task = q;
Cliff Wickman31a7df02008-02-07 00:14:42 -08002549 }
2550 /* Process the task per the caller's callback */
Paul Jackson4fe91d52008-04-29 00:59:55 -07002551 scan->process_task(q, scan);
2552 put_task_struct(q);
Cliff Wickman31a7df02008-02-07 00:14:42 -08002553 }
2554 /*
2555 * If we had to process any tasks at all, scan again
2556 * in case some of them were in the middle of forking
2557 * children that didn't get processed.
2558 * Not the most efficient way to do it, but it avoids
2559 * having to take callback_mutex in the fork path
2560 */
2561 goto again;
2562 }
2563 if (heap == &tmp_heap)
2564 heap_free(&tmp_heap);
2565 return 0;
2566}
2567
Paul Menage817929e2007-10-18 23:39:36 -07002568/*
Ben Blum102a7752009-09-23 15:56:26 -07002569 * Stuff for reading the 'tasks'/'procs' files.
Paul Menagebbcb81d2007-10-18 23:39:32 -07002570 *
2571 * Reading this file can return large amounts of data if a cgroup has
2572 * *lots* of attached tasks. So it may need several calls to read(),
2573 * but we cannot guarantee that the information we produce is correct
2574 * unless we produce it entirely atomically.
2575 *
Paul Menagebbcb81d2007-10-18 23:39:32 -07002576 */
Paul Menagebbcb81d2007-10-18 23:39:32 -07002577
2578/*
Ben Blumd1d9fd32009-09-23 15:56:28 -07002579 * The following two functions "fix" the issue where there are more pids
2580 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2581 * TODO: replace with a kernel-wide solution to this problem
2582 */
2583#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2584static void *pidlist_allocate(int count)
2585{
2586 if (PIDLIST_TOO_LARGE(count))
2587 return vmalloc(count * sizeof(pid_t));
2588 else
2589 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
2590}
2591static void pidlist_free(void *p)
2592{
2593 if (is_vmalloc_addr(p))
2594 vfree(p);
2595 else
2596 kfree(p);
2597}
2598static void *pidlist_resize(void *p, int newcount)
2599{
2600 void *newlist;
2601 /* note: if new alloc fails, old p will still be valid either way */
2602 if (is_vmalloc_addr(p)) {
2603 newlist = vmalloc(newcount * sizeof(pid_t));
2604 if (!newlist)
2605 return NULL;
2606 memcpy(newlist, p, newcount * sizeof(pid_t));
2607 vfree(p);
2608 } else {
2609 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
2610 }
2611 return newlist;
2612}
2613
2614/*
Ben Blum102a7752009-09-23 15:56:26 -07002615 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2616 * If the new stripped list is sufficiently smaller and there's enough memory
2617 * to allocate a new buffer, will let go of the unneeded memory. Returns the
2618 * number of unique elements.
Paul Menagebbcb81d2007-10-18 23:39:32 -07002619 */
Ben Blum102a7752009-09-23 15:56:26 -07002620/* is the size difference enough that we should re-allocate the array? */
2621#define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
2622static int pidlist_uniq(pid_t **p, int length)
Paul Menagebbcb81d2007-10-18 23:39:32 -07002623{
Ben Blum102a7752009-09-23 15:56:26 -07002624 int src, dest = 1;
2625 pid_t *list = *p;
2626 pid_t *newlist;
2627
2628 /*
2629 * we presume the 0th element is unique, so i starts at 1. trivial
2630 * edge cases first; no work needs to be done for either
2631 */
2632 if (length == 0 || length == 1)
2633 return length;
2634 /* src and dest walk down the list; dest counts unique elements */
2635 for (src = 1; src < length; src++) {
2636 /* find next unique element */
2637 while (list[src] == list[src-1]) {
2638 src++;
2639 if (src == length)
2640 goto after;
2641 }
2642 /* dest always points to where the next unique element goes */
2643 list[dest] = list[src];
2644 dest++;
2645 }
2646after:
2647 /*
2648 * if the length difference is large enough, we want to allocate a
2649 * smaller buffer to save memory. if this fails due to out of memory,
2650 * we'll just stay with what we've got.
2651 */
2652 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
Ben Blumd1d9fd32009-09-23 15:56:28 -07002653 newlist = pidlist_resize(list, dest);
Ben Blum102a7752009-09-23 15:56:26 -07002654 if (newlist)
2655 *p = newlist;
2656 }
2657 return dest;
2658}
2659
2660static int cmppid(const void *a, const void *b)
2661{
2662 return *(pid_t *)a - *(pid_t *)b;
2663}
2664
2665/*
Ben Blum72a8cb32009-09-23 15:56:27 -07002666 * find the appropriate pidlist for our purpose (given procs vs tasks)
2667 * returns with the lock on that pidlist already held, and takes care
2668 * of the use count, or returns NULL with no locks held if we're out of
2669 * memory.
2670 */
2671static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
2672 enum cgroup_filetype type)
2673{
2674 struct cgroup_pidlist *l;
2675 /* don't need task_nsproxy() if we're looking at ourself */
Li Zefanb70cc5f2010-03-10 15:22:12 -08002676 struct pid_namespace *ns = current->nsproxy->pid_ns;
2677
Ben Blum72a8cb32009-09-23 15:56:27 -07002678 /*
2679 * We can't drop the pidlist_mutex before taking the l->mutex in case
2680 * the last ref-holder is trying to remove l from the list at the same
2681 * time. Holding the pidlist_mutex precludes somebody taking whichever
2682 * list we find out from under us - compare release_pid_array().
2683 */
2684 mutex_lock(&cgrp->pidlist_mutex);
2685 list_for_each_entry(l, &cgrp->pidlists, links) {
2686 if (l->key.type == type && l->key.ns == ns) {
Ben Blum72a8cb32009-09-23 15:56:27 -07002687 /* make sure l doesn't vanish out from under us */
2688 down_write(&l->mutex);
2689 mutex_unlock(&cgrp->pidlist_mutex);
Ben Blum72a8cb32009-09-23 15:56:27 -07002690 return l;
2691 }
2692 }
2693 /* entry not found; create a new one */
2694 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
2695 if (!l) {
2696 mutex_unlock(&cgrp->pidlist_mutex);
Ben Blum72a8cb32009-09-23 15:56:27 -07002697 return l;
2698 }
2699 init_rwsem(&l->mutex);
2700 down_write(&l->mutex);
2701 l->key.type = type;
Li Zefanb70cc5f2010-03-10 15:22:12 -08002702 l->key.ns = get_pid_ns(ns);
Ben Blum72a8cb32009-09-23 15:56:27 -07002703 l->use_count = 0; /* don't increment here */
2704 l->list = NULL;
2705 l->owner = cgrp;
2706 list_add(&l->links, &cgrp->pidlists);
2707 mutex_unlock(&cgrp->pidlist_mutex);
2708 return l;
2709}
2710
2711/*
Ben Blum102a7752009-09-23 15:56:26 -07002712 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
2713 */
Ben Blum72a8cb32009-09-23 15:56:27 -07002714static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
2715 struct cgroup_pidlist **lp)
Ben Blum102a7752009-09-23 15:56:26 -07002716{
2717 pid_t *array;
2718 int length;
2719 int pid, n = 0; /* used for populating the array */
Paul Menage817929e2007-10-18 23:39:36 -07002720 struct cgroup_iter it;
2721 struct task_struct *tsk;
Ben Blum102a7752009-09-23 15:56:26 -07002722 struct cgroup_pidlist *l;
2723
2724 /*
2725 * If cgroup gets more users after we read count, we won't have
2726 * enough space - tough. This race is indistinguishable to the
2727 * caller from the case that the additional cgroup users didn't
2728 * show up until sometime later on.
2729 */
2730 length = cgroup_task_count(cgrp);
Ben Blumd1d9fd32009-09-23 15:56:28 -07002731 array = pidlist_allocate(length);
Ben Blum102a7752009-09-23 15:56:26 -07002732 if (!array)
2733 return -ENOMEM;
2734 /* now, populate the array */
Paul Menagebd89aab2007-10-18 23:40:44 -07002735 cgroup_iter_start(cgrp, &it);
2736 while ((tsk = cgroup_iter_next(cgrp, &it))) {
Ben Blum102a7752009-09-23 15:56:26 -07002737 if (unlikely(n == length))
Paul Menage817929e2007-10-18 23:39:36 -07002738 break;
Ben Blum102a7752009-09-23 15:56:26 -07002739 /* get tgid or pid for procs or tasks file respectively */
Ben Blum72a8cb32009-09-23 15:56:27 -07002740 if (type == CGROUP_FILE_PROCS)
2741 pid = task_tgid_vnr(tsk);
2742 else
2743 pid = task_pid_vnr(tsk);
Ben Blum102a7752009-09-23 15:56:26 -07002744 if (pid > 0) /* make sure to only use valid results */
2745 array[n++] = pid;
Paul Menage817929e2007-10-18 23:39:36 -07002746 }
Paul Menagebd89aab2007-10-18 23:40:44 -07002747 cgroup_iter_end(cgrp, &it);
Ben Blum102a7752009-09-23 15:56:26 -07002748 length = n;
2749 /* now sort & (if procs) strip out duplicates */
2750 sort(array, length, sizeof(pid_t), cmppid, NULL);
Ben Blum72a8cb32009-09-23 15:56:27 -07002751 if (type == CGROUP_FILE_PROCS)
Ben Blum102a7752009-09-23 15:56:26 -07002752 length = pidlist_uniq(&array, length);
Ben Blum72a8cb32009-09-23 15:56:27 -07002753 l = cgroup_pidlist_find(cgrp, type);
2754 if (!l) {
Ben Blumd1d9fd32009-09-23 15:56:28 -07002755 pidlist_free(array);
Ben Blum72a8cb32009-09-23 15:56:27 -07002756 return -ENOMEM;
Ben Blum102a7752009-09-23 15:56:26 -07002757 }
Ben Blum72a8cb32009-09-23 15:56:27 -07002758 /* store array, freeing old if necessary - lock already held */
Ben Blumd1d9fd32009-09-23 15:56:28 -07002759 pidlist_free(l->list);
Ben Blum102a7752009-09-23 15:56:26 -07002760 l->list = array;
2761 l->length = length;
2762 l->use_count++;
2763 up_write(&l->mutex);
Ben Blum72a8cb32009-09-23 15:56:27 -07002764 *lp = l;
Ben Blum102a7752009-09-23 15:56:26 -07002765 return 0;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002766}
2767
Balbir Singh846c7bb2007-10-18 23:39:44 -07002768/**
Li Zefana043e3b2008-02-23 15:24:09 -08002769 * cgroupstats_build - build and fill cgroupstats
Balbir Singh846c7bb2007-10-18 23:39:44 -07002770 * @stats: cgroupstats to fill information into
2771 * @dentry: A dentry entry belonging to the cgroup for which stats have
2772 * been requested.
Li Zefana043e3b2008-02-23 15:24:09 -08002773 *
2774 * Build and fill cgroupstats so that taskstats can export it to user
2775 * space.
Balbir Singh846c7bb2007-10-18 23:39:44 -07002776 */
2777int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2778{
2779 int ret = -EINVAL;
Paul Menagebd89aab2007-10-18 23:40:44 -07002780 struct cgroup *cgrp;
Balbir Singh846c7bb2007-10-18 23:39:44 -07002781 struct cgroup_iter it;
2782 struct task_struct *tsk;
Li Zefan33d283b2008-11-19 15:36:48 -08002783
Balbir Singh846c7bb2007-10-18 23:39:44 -07002784 /*
Li Zefan33d283b2008-11-19 15:36:48 -08002785 * Validate dentry by checking the superblock operations,
2786 * and make sure it's a directory.
Balbir Singh846c7bb2007-10-18 23:39:44 -07002787 */
Li Zefan33d283b2008-11-19 15:36:48 -08002788 if (dentry->d_sb->s_op != &cgroup_ops ||
2789 !S_ISDIR(dentry->d_inode->i_mode))
Balbir Singh846c7bb2007-10-18 23:39:44 -07002790 goto err;
2791
2792 ret = 0;
Paul Menagebd89aab2007-10-18 23:40:44 -07002793 cgrp = dentry->d_fsdata;
Balbir Singh846c7bb2007-10-18 23:39:44 -07002794
Paul Menagebd89aab2007-10-18 23:40:44 -07002795 cgroup_iter_start(cgrp, &it);
2796 while ((tsk = cgroup_iter_next(cgrp, &it))) {
Balbir Singh846c7bb2007-10-18 23:39:44 -07002797 switch (tsk->state) {
2798 case TASK_RUNNING:
2799 stats->nr_running++;
2800 break;
2801 case TASK_INTERRUPTIBLE:
2802 stats->nr_sleeping++;
2803 break;
2804 case TASK_UNINTERRUPTIBLE:
2805 stats->nr_uninterruptible++;
2806 break;
2807 case TASK_STOPPED:
2808 stats->nr_stopped++;
2809 break;
2810 default:
2811 if (delayacct_is_task_waiting_on_io(tsk))
2812 stats->nr_io_wait++;
2813 break;
2814 }
2815 }
Paul Menagebd89aab2007-10-18 23:40:44 -07002816 cgroup_iter_end(cgrp, &it);
Balbir Singh846c7bb2007-10-18 23:39:44 -07002817
Balbir Singh846c7bb2007-10-18 23:39:44 -07002818err:
2819 return ret;
2820}
2821
Paul Menage8f3ff202009-09-23 15:56:25 -07002822
Paul Menagecc31edc2008-10-18 20:28:04 -07002823/*
Ben Blum102a7752009-09-23 15:56:26 -07002824 * seq_file methods for the tasks/procs files. The seq_file position is the
Paul Menagecc31edc2008-10-18 20:28:04 -07002825 * next pid to display; the seq_file iterator is a pointer to the pid
Ben Blum102a7752009-09-23 15:56:26 -07002826 * in the cgroup->l->list array.
Paul Menagecc31edc2008-10-18 20:28:04 -07002827 */
2828
Ben Blum102a7752009-09-23 15:56:26 -07002829static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
Paul Menagecc31edc2008-10-18 20:28:04 -07002830{
2831 /*
2832 * Initially we receive a position value that corresponds to
2833 * one more than the last pid shown (or 0 on the first call or
2834 * after a seek to the start). Use a binary-search to find the
2835 * next pid to display, if any
2836 */
Ben Blum102a7752009-09-23 15:56:26 -07002837 struct cgroup_pidlist *l = s->private;
Paul Menagecc31edc2008-10-18 20:28:04 -07002838 int index = 0, pid = *pos;
2839 int *iter;
2840
Ben Blum102a7752009-09-23 15:56:26 -07002841 down_read(&l->mutex);
Paul Menagecc31edc2008-10-18 20:28:04 -07002842 if (pid) {
Ben Blum102a7752009-09-23 15:56:26 -07002843 int end = l->length;
Stephen Rothwell20777762008-10-21 16:11:20 +11002844
Paul Menagecc31edc2008-10-18 20:28:04 -07002845 while (index < end) {
2846 int mid = (index + end) / 2;
Ben Blum102a7752009-09-23 15:56:26 -07002847 if (l->list[mid] == pid) {
Paul Menagecc31edc2008-10-18 20:28:04 -07002848 index = mid;
2849 break;
Ben Blum102a7752009-09-23 15:56:26 -07002850 } else if (l->list[mid] <= pid)
Paul Menagecc31edc2008-10-18 20:28:04 -07002851 index = mid + 1;
2852 else
2853 end = mid;
2854 }
2855 }
2856 /* If we're off the end of the array, we're done */
Ben Blum102a7752009-09-23 15:56:26 -07002857 if (index >= l->length)
Paul Menagecc31edc2008-10-18 20:28:04 -07002858 return NULL;
2859 /* Update the abstract position to be the actual pid that we found */
Ben Blum102a7752009-09-23 15:56:26 -07002860 iter = l->list + index;
Paul Menagecc31edc2008-10-18 20:28:04 -07002861 *pos = *iter;
2862 return iter;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002863}
2864
Ben Blum102a7752009-09-23 15:56:26 -07002865static void cgroup_pidlist_stop(struct seq_file *s, void *v)
Paul Menagecc31edc2008-10-18 20:28:04 -07002866{
Ben Blum102a7752009-09-23 15:56:26 -07002867 struct cgroup_pidlist *l = s->private;
2868 up_read(&l->mutex);
Paul Menagecc31edc2008-10-18 20:28:04 -07002869}
2870
Ben Blum102a7752009-09-23 15:56:26 -07002871static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
Paul Menagecc31edc2008-10-18 20:28:04 -07002872{
Ben Blum102a7752009-09-23 15:56:26 -07002873 struct cgroup_pidlist *l = s->private;
2874 pid_t *p = v;
2875 pid_t *end = l->list + l->length;
Paul Menagecc31edc2008-10-18 20:28:04 -07002876 /*
2877 * Advance to the next pid in the array. If this goes off the
2878 * end, we're done
2879 */
2880 p++;
2881 if (p >= end) {
2882 return NULL;
2883 } else {
2884 *pos = *p;
2885 return p;
2886 }
2887}
2888
Ben Blum102a7752009-09-23 15:56:26 -07002889static int cgroup_pidlist_show(struct seq_file *s, void *v)
Paul Menagecc31edc2008-10-18 20:28:04 -07002890{
2891 return seq_printf(s, "%d\n", *(int *)v);
2892}
2893
Ben Blum102a7752009-09-23 15:56:26 -07002894/*
2895 * seq_operations functions for iterating on pidlists through seq_file -
2896 * independent of whether it's tasks or procs
2897 */
2898static const struct seq_operations cgroup_pidlist_seq_operations = {
2899 .start = cgroup_pidlist_start,
2900 .stop = cgroup_pidlist_stop,
2901 .next = cgroup_pidlist_next,
2902 .show = cgroup_pidlist_show,
Paul Menagecc31edc2008-10-18 20:28:04 -07002903};
2904
Ben Blum102a7752009-09-23 15:56:26 -07002905static void cgroup_release_pid_array(struct cgroup_pidlist *l)
Paul Menagecc31edc2008-10-18 20:28:04 -07002906{
Ben Blum72a8cb32009-09-23 15:56:27 -07002907 /*
2908 * the case where we're the last user of this particular pidlist will
2909 * have us remove it from the cgroup's list, which entails taking the
2910 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
2911 * pidlist_mutex, we have to take pidlist_mutex first.
2912 */
2913 mutex_lock(&l->owner->pidlist_mutex);
Ben Blum102a7752009-09-23 15:56:26 -07002914 down_write(&l->mutex);
2915 BUG_ON(!l->use_count);
2916 if (!--l->use_count) {
Ben Blum72a8cb32009-09-23 15:56:27 -07002917 /* we're the last user if refcount is 0; remove and free */
2918 list_del(&l->links);
2919 mutex_unlock(&l->owner->pidlist_mutex);
Ben Blumd1d9fd32009-09-23 15:56:28 -07002920 pidlist_free(l->list);
Ben Blum72a8cb32009-09-23 15:56:27 -07002921 put_pid_ns(l->key.ns);
2922 up_write(&l->mutex);
2923 kfree(l);
2924 return;
Paul Menagecc31edc2008-10-18 20:28:04 -07002925 }
Ben Blum72a8cb32009-09-23 15:56:27 -07002926 mutex_unlock(&l->owner->pidlist_mutex);
Ben Blum102a7752009-09-23 15:56:26 -07002927 up_write(&l->mutex);
Paul Menagecc31edc2008-10-18 20:28:04 -07002928}
2929
Ben Blum102a7752009-09-23 15:56:26 -07002930static int cgroup_pidlist_release(struct inode *inode, struct file *file)
Paul Menagebbcb81d2007-10-18 23:39:32 -07002931{
Ben Blum102a7752009-09-23 15:56:26 -07002932 struct cgroup_pidlist *l;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002933 if (!(file->f_mode & FMODE_READ))
2934 return 0;
Ben Blum102a7752009-09-23 15:56:26 -07002935 /*
2936 * the seq_file will only be initialized if the file was opened for
2937 * reading; hence we check if it's not null only in that case.
2938 */
2939 l = ((struct seq_file *)file->private_data)->private;
2940 cgroup_release_pid_array(l);
Paul Menagecc31edc2008-10-18 20:28:04 -07002941 return seq_release(inode, file);
2942}
2943
Ben Blum102a7752009-09-23 15:56:26 -07002944static const struct file_operations cgroup_pidlist_operations = {
Paul Menagecc31edc2008-10-18 20:28:04 -07002945 .read = seq_read,
2946 .llseek = seq_lseek,
2947 .write = cgroup_file_write,
Ben Blum102a7752009-09-23 15:56:26 -07002948 .release = cgroup_pidlist_release,
Paul Menagecc31edc2008-10-18 20:28:04 -07002949};
2950
2951/*
Ben Blum102a7752009-09-23 15:56:26 -07002952 * The following functions handle opens on a file that displays a pidlist
2953 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
2954 * in the cgroup.
Paul Menagecc31edc2008-10-18 20:28:04 -07002955 */
Ben Blum102a7752009-09-23 15:56:26 -07002956/* helper function for the two below it */
Ben Blum72a8cb32009-09-23 15:56:27 -07002957static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
Paul Menagecc31edc2008-10-18 20:28:04 -07002958{
2959 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
Ben Blum72a8cb32009-09-23 15:56:27 -07002960 struct cgroup_pidlist *l;
Paul Menagecc31edc2008-10-18 20:28:04 -07002961 int retval;
2962
2963 /* Nothing to do for write-only files */
2964 if (!(file->f_mode & FMODE_READ))
2965 return 0;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002966
Ben Blum102a7752009-09-23 15:56:26 -07002967 /* have the array populated */
Ben Blum72a8cb32009-09-23 15:56:27 -07002968 retval = pidlist_array_load(cgrp, type, &l);
Ben Blum102a7752009-09-23 15:56:26 -07002969 if (retval)
2970 return retval;
2971 /* configure file information */
2972 file->f_op = &cgroup_pidlist_operations;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002973
Ben Blum102a7752009-09-23 15:56:26 -07002974 retval = seq_open(file, &cgroup_pidlist_seq_operations);
Paul Menagecc31edc2008-10-18 20:28:04 -07002975 if (retval) {
Ben Blum102a7752009-09-23 15:56:26 -07002976 cgroup_release_pid_array(l);
Paul Menagecc31edc2008-10-18 20:28:04 -07002977 return retval;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002978 }
Ben Blum102a7752009-09-23 15:56:26 -07002979 ((struct seq_file *)file->private_data)->private = l;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002980 return 0;
2981}
Ben Blum102a7752009-09-23 15:56:26 -07002982static int cgroup_tasks_open(struct inode *unused, struct file *file)
2983{
Ben Blum72a8cb32009-09-23 15:56:27 -07002984 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
Ben Blum102a7752009-09-23 15:56:26 -07002985}
2986static int cgroup_procs_open(struct inode *unused, struct file *file)
2987{
Ben Blum72a8cb32009-09-23 15:56:27 -07002988 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
Ben Blum102a7752009-09-23 15:56:26 -07002989}
Paul Menagebbcb81d2007-10-18 23:39:32 -07002990
Paul Menagebd89aab2007-10-18 23:40:44 -07002991static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
Paul Menage81a6a5c2007-10-18 23:39:38 -07002992 struct cftype *cft)
2993{
Paul Menagebd89aab2007-10-18 23:40:44 -07002994 return notify_on_release(cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -07002995}
2996
Paul Menage6379c102008-07-25 01:47:01 -07002997static int cgroup_write_notify_on_release(struct cgroup *cgrp,
2998 struct cftype *cft,
2999 u64 val)
3000{
3001 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3002 if (val)
3003 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3004 else
3005 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3006 return 0;
3007}
3008
Paul Menagebbcb81d2007-10-18 23:39:32 -07003009/*
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003010 * Unregister event and free resources.
3011 *
3012 * Gets called from workqueue.
3013 */
3014static void cgroup_event_remove(struct work_struct *work)
3015{
3016 struct cgroup_event *event = container_of(work, struct cgroup_event,
3017 remove);
3018 struct cgroup *cgrp = event->cgrp;
3019
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003020 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3021
3022 eventfd_ctx_put(event->eventfd);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003023 kfree(event);
Kirill A. Shutemova0a4db52010-03-10 15:22:34 -08003024 dput(cgrp->dentry);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003025}
3026
3027/*
3028 * Gets called on POLLHUP on eventfd when user closes it.
3029 *
3030 * Called with wqh->lock held and interrupts disabled.
3031 */
3032static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3033 int sync, void *key)
3034{
3035 struct cgroup_event *event = container_of(wait,
3036 struct cgroup_event, wait);
3037 struct cgroup *cgrp = event->cgrp;
3038 unsigned long flags = (unsigned long)key;
3039
3040 if (flags & POLLHUP) {
Changli Gaoa93d2f12010-05-07 14:33:26 +08003041 __remove_wait_queue(event->wqh, &event->wait);
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003042 spin_lock(&cgrp->event_list_lock);
3043 list_del(&event->list);
3044 spin_unlock(&cgrp->event_list_lock);
3045 /*
3046 * We are in atomic context, but cgroup_event_remove() may
3047 * sleep, so we have to call it in workqueue.
3048 */
3049 schedule_work(&event->remove);
3050 }
3051
3052 return 0;
3053}
3054
3055static void cgroup_event_ptable_queue_proc(struct file *file,
3056 wait_queue_head_t *wqh, poll_table *pt)
3057{
3058 struct cgroup_event *event = container_of(pt,
3059 struct cgroup_event, pt);
3060
3061 event->wqh = wqh;
3062 add_wait_queue(wqh, &event->wait);
3063}
3064
3065/*
3066 * Parse input and register new cgroup event handler.
3067 *
3068 * Input must be in format '<event_fd> <control_fd> <args>'.
3069 * Interpretation of args is defined by control file implementation.
3070 */
3071static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3072 const char *buffer)
3073{
3074 struct cgroup_event *event = NULL;
3075 unsigned int efd, cfd;
3076 struct file *efile = NULL;
3077 struct file *cfile = NULL;
3078 char *endp;
3079 int ret;
3080
3081 efd = simple_strtoul(buffer, &endp, 10);
3082 if (*endp != ' ')
3083 return -EINVAL;
3084 buffer = endp + 1;
3085
3086 cfd = simple_strtoul(buffer, &endp, 10);
3087 if ((*endp != ' ') && (*endp != '\0'))
3088 return -EINVAL;
3089 buffer = endp + 1;
3090
3091 event = kzalloc(sizeof(*event), GFP_KERNEL);
3092 if (!event)
3093 return -ENOMEM;
3094 event->cgrp = cgrp;
3095 INIT_LIST_HEAD(&event->list);
3096 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3097 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3098 INIT_WORK(&event->remove, cgroup_event_remove);
3099
3100 efile = eventfd_fget(efd);
3101 if (IS_ERR(efile)) {
3102 ret = PTR_ERR(efile);
3103 goto fail;
3104 }
3105
3106 event->eventfd = eventfd_ctx_fileget(efile);
3107 if (IS_ERR(event->eventfd)) {
3108 ret = PTR_ERR(event->eventfd);
3109 goto fail;
3110 }
3111
3112 cfile = fget(cfd);
3113 if (!cfile) {
3114 ret = -EBADF;
3115 goto fail;
3116 }
3117
3118 /* the process need read permission on control file */
3119 ret = file_permission(cfile, MAY_READ);
3120 if (ret < 0)
3121 goto fail;
3122
3123 event->cft = __file_cft(cfile);
3124 if (IS_ERR(event->cft)) {
3125 ret = PTR_ERR(event->cft);
3126 goto fail;
3127 }
3128
3129 if (!event->cft->register_event || !event->cft->unregister_event) {
3130 ret = -EINVAL;
3131 goto fail;
3132 }
3133
3134 ret = event->cft->register_event(cgrp, event->cft,
3135 event->eventfd, buffer);
3136 if (ret)
3137 goto fail;
3138
3139 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3140 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3141 ret = 0;
3142 goto fail;
3143 }
3144
Kirill A. Shutemova0a4db52010-03-10 15:22:34 -08003145 /*
3146 * Events should be removed after rmdir of cgroup directory, but before
3147 * destroying subsystem state objects. Let's take reference to cgroup
3148 * directory dentry to do that.
3149 */
3150 dget(cgrp->dentry);
3151
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003152 spin_lock(&cgrp->event_list_lock);
3153 list_add(&event->list, &cgrp->event_list);
3154 spin_unlock(&cgrp->event_list_lock);
3155
3156 fput(cfile);
3157 fput(efile);
3158
3159 return 0;
3160
3161fail:
3162 if (cfile)
3163 fput(cfile);
3164
3165 if (event && event->eventfd && !IS_ERR(event->eventfd))
3166 eventfd_ctx_put(event->eventfd);
3167
3168 if (!IS_ERR_OR_NULL(efile))
3169 fput(efile);
3170
3171 kfree(event);
3172
3173 return ret;
3174}
3175
3176/*
Paul Menagebbcb81d2007-10-18 23:39:32 -07003177 * for the common functions, 'private' gives the type of file
3178 */
Ben Blum102a7752009-09-23 15:56:26 -07003179/* for hysterical raisins, we can't put this on the older files */
3180#define CGROUP_FILE_GENERIC_PREFIX "cgroup."
Paul Menage81a6a5c2007-10-18 23:39:38 -07003181static struct cftype files[] = {
3182 {
3183 .name = "tasks",
3184 .open = cgroup_tasks_open,
Paul Menageaf351022008-07-25 01:47:01 -07003185 .write_u64 = cgroup_tasks_write,
Ben Blum102a7752009-09-23 15:56:26 -07003186 .release = cgroup_pidlist_release,
Li Zefan099fca32009-04-02 16:57:29 -07003187 .mode = S_IRUGO | S_IWUSR,
Paul Menage81a6a5c2007-10-18 23:39:38 -07003188 },
Ben Blum102a7752009-09-23 15:56:26 -07003189 {
3190 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3191 .open = cgroup_procs_open,
3192 /* .write_u64 = cgroup_procs_write, TODO */
3193 .release = cgroup_pidlist_release,
3194 .mode = S_IRUGO,
3195 },
Paul Menage81a6a5c2007-10-18 23:39:38 -07003196 {
3197 .name = "notify_on_release",
Paul Menagef4c753b2008-04-29 00:59:56 -07003198 .read_u64 = cgroup_read_notify_on_release,
Paul Menage6379c102008-07-25 01:47:01 -07003199 .write_u64 = cgroup_write_notify_on_release,
Paul Menage81a6a5c2007-10-18 23:39:38 -07003200 },
Kirill A. Shutemov0dea1162010-03-10 15:22:20 -08003201 {
3202 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3203 .write_string = cgroup_write_event_control,
3204 .mode = S_IWUGO,
3205 },
Paul Menage81a6a5c2007-10-18 23:39:38 -07003206};
3207
3208static struct cftype cft_release_agent = {
3209 .name = "release_agent",
Paul Menagee788e062008-07-25 01:46:59 -07003210 .read_seq_string = cgroup_release_agent_show,
3211 .write_string = cgroup_release_agent_write,
3212 .max_write_len = PATH_MAX,
Paul Menagebbcb81d2007-10-18 23:39:32 -07003213};
3214
Paul Menagebd89aab2007-10-18 23:40:44 -07003215static int cgroup_populate_dir(struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003216{
3217 int err;
3218 struct cgroup_subsys *ss;
3219
3220 /* First clear out any existing files */
Paul Menagebd89aab2007-10-18 23:40:44 -07003221 cgroup_clear_directory(cgrp->dentry);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003222
Paul Menagebd89aab2007-10-18 23:40:44 -07003223 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
Paul Menagebbcb81d2007-10-18 23:39:32 -07003224 if (err < 0)
3225 return err;
3226
Paul Menagebd89aab2007-10-18 23:40:44 -07003227 if (cgrp == cgrp->top_cgroup) {
3228 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
Paul Menage81a6a5c2007-10-18 23:39:38 -07003229 return err;
3230 }
3231
Paul Menagebd89aab2007-10-18 23:40:44 -07003232 for_each_subsys(cgrp->root, ss) {
3233 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003234 return err;
3235 }
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07003236 /* This cgroup is ready now */
3237 for_each_subsys(cgrp->root, ss) {
3238 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3239 /*
3240 * Update id->css pointer and make this css visible from
3241 * CSS ID functions. This pointer will be dereferened
3242 * from RCU-read-side without locks.
3243 */
3244 if (css->id)
3245 rcu_assign_pointer(css->id->css, css);
3246 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07003247
3248 return 0;
3249}
3250
3251static void init_cgroup_css(struct cgroup_subsys_state *css,
3252 struct cgroup_subsys *ss,
Paul Menagebd89aab2007-10-18 23:40:44 -07003253 struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003254{
Paul Menagebd89aab2007-10-18 23:40:44 -07003255 css->cgroup = cgrp;
Paul Menagee7c5ec92009-01-07 18:08:38 -08003256 atomic_set(&css->refcnt, 1);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003257 css->flags = 0;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07003258 css->id = NULL;
Paul Menagebd89aab2007-10-18 23:40:44 -07003259 if (cgrp == dummytop)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003260 set_bit(CSS_ROOT, &css->flags);
Paul Menagebd89aab2007-10-18 23:40:44 -07003261 BUG_ON(cgrp->subsys[ss->subsys_id]);
3262 cgrp->subsys[ss->subsys_id] = css;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003263}
3264
Paul Menage999cd8a2009-01-07 18:08:36 -08003265static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3266{
3267 /* We need to take each hierarchy_mutex in a consistent order */
3268 int i;
3269
Ben Blumaae8aab2010-03-10 15:22:07 -08003270 /*
3271 * No worry about a race with rebind_subsystems that might mess up the
3272 * locking order, since both parties are under cgroup_mutex.
3273 */
Paul Menage999cd8a2009-01-07 18:08:36 -08003274 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3275 struct cgroup_subsys *ss = subsys[i];
Ben Blumaae8aab2010-03-10 15:22:07 -08003276 if (ss == NULL)
3277 continue;
Paul Menage999cd8a2009-01-07 18:08:36 -08003278 if (ss->root == root)
Li Zefancfebe562009-02-11 13:04:36 -08003279 mutex_lock(&ss->hierarchy_mutex);
Paul Menage999cd8a2009-01-07 18:08:36 -08003280 }
3281}
3282
3283static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3284{
3285 int i;
3286
3287 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3288 struct cgroup_subsys *ss = subsys[i];
Ben Blumaae8aab2010-03-10 15:22:07 -08003289 if (ss == NULL)
3290 continue;
Paul Menage999cd8a2009-01-07 18:08:36 -08003291 if (ss->root == root)
3292 mutex_unlock(&ss->hierarchy_mutex);
3293 }
3294}
3295
Paul Menageddbcc7e2007-10-18 23:39:30 -07003296/*
Li Zefana043e3b2008-02-23 15:24:09 -08003297 * cgroup_create - create a cgroup
3298 * @parent: cgroup that will be parent of the new cgroup
3299 * @dentry: dentry of the new cgroup
3300 * @mode: mode to set on new inode
Paul Menageddbcc7e2007-10-18 23:39:30 -07003301 *
Li Zefana043e3b2008-02-23 15:24:09 -08003302 * Must be called with the mutex on the parent inode held
Paul Menageddbcc7e2007-10-18 23:39:30 -07003303 */
Paul Menageddbcc7e2007-10-18 23:39:30 -07003304static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
Li Zefan099fca32009-04-02 16:57:29 -07003305 mode_t mode)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003306{
Paul Menagebd89aab2007-10-18 23:40:44 -07003307 struct cgroup *cgrp;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003308 struct cgroupfs_root *root = parent->root;
3309 int err = 0;
3310 struct cgroup_subsys *ss;
3311 struct super_block *sb = root->sb;
3312
Paul Menagebd89aab2007-10-18 23:40:44 -07003313 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3314 if (!cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003315 return -ENOMEM;
3316
3317 /* Grab a reference on the superblock so the hierarchy doesn't
3318 * get deleted on unmount if there are child cgroups. This
3319 * can be done outside cgroup_mutex, since the sb can't
3320 * disappear while someone has an open control file on the
3321 * fs */
3322 atomic_inc(&sb->s_active);
3323
3324 mutex_lock(&cgroup_mutex);
3325
Paul Menagecc31edc2008-10-18 20:28:04 -07003326 init_cgroup_housekeeping(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003327
Paul Menagebd89aab2007-10-18 23:40:44 -07003328 cgrp->parent = parent;
3329 cgrp->root = parent->root;
3330 cgrp->top_cgroup = parent->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003331
Li Zefanb6abdb02008-03-04 14:28:19 -08003332 if (notify_on_release(parent))
3333 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3334
Paul Menageddbcc7e2007-10-18 23:39:30 -07003335 for_each_subsys(root, ss) {
Paul Menagebd89aab2007-10-18 23:40:44 -07003336 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
Li Zefan4528fd02010-02-02 13:44:10 -08003337
Paul Menageddbcc7e2007-10-18 23:39:30 -07003338 if (IS_ERR(css)) {
3339 err = PTR_ERR(css);
3340 goto err_destroy;
3341 }
Paul Menagebd89aab2007-10-18 23:40:44 -07003342 init_cgroup_css(css, ss, cgrp);
Li Zefan4528fd02010-02-02 13:44:10 -08003343 if (ss->use_id) {
3344 err = alloc_css_id(ss, parent, cgrp);
3345 if (err)
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07003346 goto err_destroy;
Li Zefan4528fd02010-02-02 13:44:10 -08003347 }
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07003348 /* At error, ->destroy() callback has to free assigned ID. */
Paul Menageddbcc7e2007-10-18 23:39:30 -07003349 }
3350
Paul Menage999cd8a2009-01-07 18:08:36 -08003351 cgroup_lock_hierarchy(root);
Paul Menagebd89aab2007-10-18 23:40:44 -07003352 list_add(&cgrp->sibling, &cgrp->parent->children);
Paul Menage999cd8a2009-01-07 18:08:36 -08003353 cgroup_unlock_hierarchy(root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003354 root->number_of_cgroups++;
3355
Paul Menagebd89aab2007-10-18 23:40:44 -07003356 err = cgroup_create_dir(cgrp, dentry, mode);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003357 if (err < 0)
3358 goto err_remove;
3359
3360 /* The cgroup directory was pre-locked for us */
Paul Menagebd89aab2007-10-18 23:40:44 -07003361 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
Paul Menageddbcc7e2007-10-18 23:39:30 -07003362
Paul Menagebd89aab2007-10-18 23:40:44 -07003363 err = cgroup_populate_dir(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003364 /* If err < 0, we have a half-filled directory - oh well ;) */
3365
3366 mutex_unlock(&cgroup_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -07003367 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003368
3369 return 0;
3370
3371 err_remove:
3372
KAMEZAWA Hiroyukibaef99a2009-01-29 14:25:10 -08003373 cgroup_lock_hierarchy(root);
Paul Menagebd89aab2007-10-18 23:40:44 -07003374 list_del(&cgrp->sibling);
KAMEZAWA Hiroyukibaef99a2009-01-29 14:25:10 -08003375 cgroup_unlock_hierarchy(root);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003376 root->number_of_cgroups--;
3377
3378 err_destroy:
3379
3380 for_each_subsys(root, ss) {
Paul Menagebd89aab2007-10-18 23:40:44 -07003381 if (cgrp->subsys[ss->subsys_id])
3382 ss->destroy(ss, cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003383 }
3384
3385 mutex_unlock(&cgroup_mutex);
3386
3387 /* Release the reference count that we took on the superblock */
3388 deactivate_super(sb);
3389
Paul Menagebd89aab2007-10-18 23:40:44 -07003390 kfree(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003391 return err;
3392}
3393
3394static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3395{
3396 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3397
3398 /* the vfs holds inode->i_mutex already */
3399 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3400}
3401
Li Zefan55b6fd02008-07-29 22:33:20 -07003402static int cgroup_has_css_refs(struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -07003403{
3404 /* Check the reference count on each subsystem. Since we
3405 * already established that there are no tasks in the
Paul Menagee7c5ec92009-01-07 18:08:38 -08003406 * cgroup, if the css refcount is also 1, then there should
Paul Menage81a6a5c2007-10-18 23:39:38 -07003407 * be no outstanding references, so the subsystem is safe to
3408 * destroy. We scan across all subsystems rather than using
3409 * the per-hierarchy linked list of mounted subsystems since
3410 * we can be called via check_for_release() with no
3411 * synchronization other than RCU, and the subsystem linked
3412 * list isn't RCU-safe */
3413 int i;
Ben Blumaae8aab2010-03-10 15:22:07 -08003414 /*
3415 * We won't need to lock the subsys array, because the subsystems
3416 * we're concerned about aren't going anywhere since our cgroup root
3417 * has a reference on them.
3418 */
Paul Menage81a6a5c2007-10-18 23:39:38 -07003419 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3420 struct cgroup_subsys *ss = subsys[i];
3421 struct cgroup_subsys_state *css;
Ben Blumaae8aab2010-03-10 15:22:07 -08003422 /* Skip subsystems not present or not in this hierarchy */
3423 if (ss == NULL || ss->root != cgrp->root)
Paul Menage81a6a5c2007-10-18 23:39:38 -07003424 continue;
Paul Menagebd89aab2007-10-18 23:40:44 -07003425 css = cgrp->subsys[ss->subsys_id];
Paul Menage81a6a5c2007-10-18 23:39:38 -07003426 /* When called from check_for_release() it's possible
3427 * that by this point the cgroup has been removed
3428 * and the css deleted. But a false-positive doesn't
3429 * matter, since it can only happen if the cgroup
3430 * has been deleted and hence no longer needs the
3431 * release agent to be called anyway. */
Paul Menagee7c5ec92009-01-07 18:08:38 -08003432 if (css && (atomic_read(&css->refcnt) > 1))
Paul Menage81a6a5c2007-10-18 23:39:38 -07003433 return 1;
Paul Menage81a6a5c2007-10-18 23:39:38 -07003434 }
3435 return 0;
3436}
3437
Paul Menagee7c5ec92009-01-07 18:08:38 -08003438/*
3439 * Atomically mark all (or else none) of the cgroup's CSS objects as
3440 * CSS_REMOVED. Return true on success, or false if the cgroup has
3441 * busy subsystems. Call with cgroup_mutex held
3442 */
3443
3444static int cgroup_clear_css_refs(struct cgroup *cgrp)
3445{
3446 struct cgroup_subsys *ss;
3447 unsigned long flags;
3448 bool failed = false;
3449 local_irq_save(flags);
3450 for_each_subsys(cgrp->root, ss) {
3451 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3452 int refcnt;
Paul Menage804b3c22009-01-29 14:25:21 -08003453 while (1) {
Paul Menagee7c5ec92009-01-07 18:08:38 -08003454 /* We can only remove a CSS with a refcnt==1 */
3455 refcnt = atomic_read(&css->refcnt);
3456 if (refcnt > 1) {
3457 failed = true;
3458 goto done;
3459 }
3460 BUG_ON(!refcnt);
3461 /*
3462 * Drop the refcnt to 0 while we check other
3463 * subsystems. This will cause any racing
3464 * css_tryget() to spin until we set the
3465 * CSS_REMOVED bits or abort
3466 */
Paul Menage804b3c22009-01-29 14:25:21 -08003467 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3468 break;
3469 cpu_relax();
3470 }
Paul Menagee7c5ec92009-01-07 18:08:38 -08003471 }
3472 done:
3473 for_each_subsys(cgrp->root, ss) {
3474 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3475 if (failed) {
3476 /*
3477 * Restore old refcnt if we previously managed
3478 * to clear it from 1 to 0
3479 */
3480 if (!atomic_read(&css->refcnt))
3481 atomic_set(&css->refcnt, 1);
3482 } else {
3483 /* Commit the fact that the CSS is removed */
3484 set_bit(CSS_REMOVED, &css->flags);
3485 }
3486 }
3487 local_irq_restore(flags);
3488 return !failed;
3489}
3490
Paul Menageddbcc7e2007-10-18 23:39:30 -07003491static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3492{
Paul Menagebd89aab2007-10-18 23:40:44 -07003493 struct cgroup *cgrp = dentry->d_fsdata;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003494 struct dentry *d;
3495 struct cgroup *parent;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003496 DEFINE_WAIT(wait);
Kirill A. Shutemov4ab78682010-03-10 15:22:34 -08003497 struct cgroup_event *event, *tmp;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003498 int ret;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003499
3500 /* the vfs holds both inode->i_mutex already */
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003501again:
Paul Menageddbcc7e2007-10-18 23:39:30 -07003502 mutex_lock(&cgroup_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -07003503 if (atomic_read(&cgrp->count) != 0) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07003504 mutex_unlock(&cgroup_mutex);
3505 return -EBUSY;
3506 }
Paul Menagebd89aab2007-10-18 23:40:44 -07003507 if (!list_empty(&cgrp->children)) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07003508 mutex_unlock(&cgroup_mutex);
3509 return -EBUSY;
3510 }
KAMEZAWA Hiroyuki3fa59df2008-11-19 15:36:34 -08003511 mutex_unlock(&cgroup_mutex);
Li Zefana043e3b2008-02-23 15:24:09 -08003512
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -08003513 /*
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003514 * In general, subsystem has no css->refcnt after pre_destroy(). But
3515 * in racy cases, subsystem may have to get css->refcnt after
3516 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3517 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3518 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3519 * and subsystem's reference count handling. Please see css_get/put
3520 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3521 */
3522 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3523
3524 /*
Li Zefana043e3b2008-02-23 15:24:09 -08003525 * Call pre_destroy handlers of subsys. Notify subsystems
3526 * that rmdir() request comes.
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -08003527 */
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003528 ret = cgroup_call_pre_destroy(cgrp);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003529 if (ret) {
3530 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003531 return ret;
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003532 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07003533
KAMEZAWA Hiroyuki3fa59df2008-11-19 15:36:34 -08003534 mutex_lock(&cgroup_mutex);
3535 parent = cgrp->parent;
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003536 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003537 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003538 mutex_unlock(&cgroup_mutex);
3539 return -EBUSY;
3540 }
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003541 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003542 if (!cgroup_clear_css_refs(cgrp)) {
3543 mutex_unlock(&cgroup_mutex);
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07003544 /*
3545 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3546 * prepare_to_wait(), we need to check this flag.
3547 */
3548 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
3549 schedule();
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07003550 finish_wait(&cgroup_rmdir_waitq, &wait);
3551 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3552 if (signal_pending(current))
3553 return -EINTR;
3554 goto again;
3555 }
3556 /* NO css_tryget() can success after here. */
3557 finish_wait(&cgroup_rmdir_waitq, &wait);
3558 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003559
Paul Menage81a6a5c2007-10-18 23:39:38 -07003560 spin_lock(&release_list_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07003561 set_bit(CGRP_REMOVED, &cgrp->flags);
3562 if (!list_empty(&cgrp->release_list))
3563 list_del(&cgrp->release_list);
Paul Menage81a6a5c2007-10-18 23:39:38 -07003564 spin_unlock(&release_list_lock);
Paul Menage999cd8a2009-01-07 18:08:36 -08003565
3566 cgroup_lock_hierarchy(cgrp->root);
3567 /* delete this cgroup from parent->children */
Paul Menagebd89aab2007-10-18 23:40:44 -07003568 list_del(&cgrp->sibling);
Paul Menage999cd8a2009-01-07 18:08:36 -08003569 cgroup_unlock_hierarchy(cgrp->root);
3570
Paul Menagebd89aab2007-10-18 23:40:44 -07003571 spin_lock(&cgrp->dentry->d_lock);
3572 d = dget(cgrp->dentry);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003573 spin_unlock(&d->d_lock);
3574
3575 cgroup_d_remove_dir(d);
3576 dput(d);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003577
Paul Menagebd89aab2007-10-18 23:40:44 -07003578 set_bit(CGRP_RELEASABLE, &parent->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -07003579 check_for_release(parent);
3580
Kirill A. Shutemov4ab78682010-03-10 15:22:34 -08003581 /*
3582 * Unregister events and notify userspace.
3583 * Notify userspace about cgroup removing only after rmdir of cgroup
3584 * directory to avoid race between userspace and kernelspace
3585 */
3586 spin_lock(&cgrp->event_list_lock);
3587 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
3588 list_del(&event->list);
3589 remove_wait_queue(event->wqh, &event->wait);
3590 eventfd_signal(event->eventfd, 1);
3591 schedule_work(&event->remove);
3592 }
3593 spin_unlock(&cgrp->event_list_lock);
3594
Paul Menageddbcc7e2007-10-18 23:39:30 -07003595 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003596 return 0;
3597}
3598
Li Zefan06a11922008-04-29 01:00:07 -07003599static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
Paul Menageddbcc7e2007-10-18 23:39:30 -07003600{
Paul Menageddbcc7e2007-10-18 23:39:30 -07003601 struct cgroup_subsys_state *css;
Diego Callejacfe36bd2007-11-14 16:58:54 -08003602
3603 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003604
3605 /* Create the top cgroup state for this subsystem */
Li Zefan33a68ac2009-01-07 18:07:42 -08003606 list_add(&ss->sibling, &rootnode.subsys_list);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003607 ss->root = &rootnode;
3608 css = ss->create(ss, dummytop);
3609 /* We don't handle early failures gracefully */
3610 BUG_ON(IS_ERR(css));
3611 init_cgroup_css(css, ss, dummytop);
3612
Li Zefane8d55fd2008-04-29 01:00:13 -07003613 /* Update the init_css_set to contain a subsys
Paul Menage817929e2007-10-18 23:39:36 -07003614 * pointer to this state - since the subsystem is
Li Zefane8d55fd2008-04-29 01:00:13 -07003615 * newly registered, all tasks and hence the
3616 * init_css_set is in the subsystem's top cgroup. */
3617 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
Paul Menageddbcc7e2007-10-18 23:39:30 -07003618
3619 need_forkexit_callback |= ss->fork || ss->exit;
3620
Li Zefane8d55fd2008-04-29 01:00:13 -07003621 /* At system boot, before all subsystems have been
3622 * registered, no tasks have been forked, so we don't
3623 * need to invoke fork callbacks here. */
3624 BUG_ON(!list_empty(&init_task.tasks));
3625
Paul Menage999cd8a2009-01-07 18:08:36 -08003626 mutex_init(&ss->hierarchy_mutex);
Li Zefancfebe562009-02-11 13:04:36 -08003627 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003628 ss->active = 1;
Ben Blume6a11052010-03-10 15:22:09 -08003629
3630 /* this function shouldn't be used with modular subsystems, since they
3631 * need to register a subsys_id, among other things */
3632 BUG_ON(ss->module);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003633}
3634
3635/**
Ben Blume6a11052010-03-10 15:22:09 -08003636 * cgroup_load_subsys: load and register a modular subsystem at runtime
3637 * @ss: the subsystem to load
3638 *
3639 * This function should be called in a modular subsystem's initcall. If the
Thomas Weber88393162010-03-16 11:47:56 +01003640 * subsystem is built as a module, it will be assigned a new subsys_id and set
Ben Blume6a11052010-03-10 15:22:09 -08003641 * up for use. If the subsystem is built-in anyway, work is delegated to the
3642 * simpler cgroup_init_subsys.
3643 */
3644int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
3645{
3646 int i;
3647 struct cgroup_subsys_state *css;
3648
3649 /* check name and function validity */
3650 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
3651 ss->create == NULL || ss->destroy == NULL)
3652 return -EINVAL;
3653
3654 /*
3655 * we don't support callbacks in modular subsystems. this check is
3656 * before the ss->module check for consistency; a subsystem that could
3657 * be a module should still have no callbacks even if the user isn't
3658 * compiling it as one.
3659 */
3660 if (ss->fork || ss->exit)
3661 return -EINVAL;
3662
3663 /*
3664 * an optionally modular subsystem is built-in: we want to do nothing,
3665 * since cgroup_init_subsys will have already taken care of it.
3666 */
3667 if (ss->module == NULL) {
3668 /* a few sanity checks */
3669 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
3670 BUG_ON(subsys[ss->subsys_id] != ss);
3671 return 0;
3672 }
3673
3674 /*
3675 * need to register a subsys id before anything else - for example,
3676 * init_cgroup_css needs it.
3677 */
3678 mutex_lock(&cgroup_mutex);
3679 /* find the first empty slot in the array */
3680 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
3681 if (subsys[i] == NULL)
3682 break;
3683 }
3684 if (i == CGROUP_SUBSYS_COUNT) {
3685 /* maximum number of subsystems already registered! */
3686 mutex_unlock(&cgroup_mutex);
3687 return -EBUSY;
3688 }
3689 /* assign ourselves the subsys_id */
3690 ss->subsys_id = i;
3691 subsys[i] = ss;
3692
3693 /*
3694 * no ss->create seems to need anything important in the ss struct, so
3695 * this can happen first (i.e. before the rootnode attachment).
3696 */
3697 css = ss->create(ss, dummytop);
3698 if (IS_ERR(css)) {
3699 /* failure case - need to deassign the subsys[] slot. */
3700 subsys[i] = NULL;
3701 mutex_unlock(&cgroup_mutex);
3702 return PTR_ERR(css);
3703 }
3704
3705 list_add(&ss->sibling, &rootnode.subsys_list);
3706 ss->root = &rootnode;
3707
3708 /* our new subsystem will be attached to the dummy hierarchy. */
3709 init_cgroup_css(css, ss, dummytop);
3710 /* init_idr must be after init_cgroup_css because it sets css->id. */
3711 if (ss->use_id) {
3712 int ret = cgroup_init_idr(ss, css);
3713 if (ret) {
3714 dummytop->subsys[ss->subsys_id] = NULL;
3715 ss->destroy(ss, dummytop);
3716 subsys[i] = NULL;
3717 mutex_unlock(&cgroup_mutex);
3718 return ret;
3719 }
3720 }
3721
3722 /*
3723 * Now we need to entangle the css into the existing css_sets. unlike
3724 * in cgroup_init_subsys, there are now multiple css_sets, so each one
3725 * will need a new pointer to it; done by iterating the css_set_table.
3726 * furthermore, modifying the existing css_sets will corrupt the hash
3727 * table state, so each changed css_set will need its hash recomputed.
3728 * this is all done under the css_set_lock.
3729 */
3730 write_lock(&css_set_lock);
3731 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
3732 struct css_set *cg;
3733 struct hlist_node *node, *tmp;
3734 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
3735
3736 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
3737 /* skip entries that we already rehashed */
3738 if (cg->subsys[ss->subsys_id])
3739 continue;
3740 /* remove existing entry */
3741 hlist_del(&cg->hlist);
3742 /* set new value */
3743 cg->subsys[ss->subsys_id] = css;
3744 /* recompute hash and restore entry */
3745 new_bucket = css_set_hash(cg->subsys);
3746 hlist_add_head(&cg->hlist, new_bucket);
3747 }
3748 }
3749 write_unlock(&css_set_lock);
3750
3751 mutex_init(&ss->hierarchy_mutex);
3752 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3753 ss->active = 1;
3754
Ben Blume6a11052010-03-10 15:22:09 -08003755 /* success! */
3756 mutex_unlock(&cgroup_mutex);
3757 return 0;
3758}
3759EXPORT_SYMBOL_GPL(cgroup_load_subsys);
3760
3761/**
Ben Blumcf5d5942010-03-10 15:22:09 -08003762 * cgroup_unload_subsys: unload a modular subsystem
3763 * @ss: the subsystem to unload
3764 *
3765 * This function should be called in a modular subsystem's exitcall. When this
3766 * function is invoked, the refcount on the subsystem's module will be 0, so
3767 * the subsystem will not be attached to any hierarchy.
3768 */
3769void cgroup_unload_subsys(struct cgroup_subsys *ss)
3770{
3771 struct cg_cgroup_link *link;
3772 struct hlist_head *hhead;
3773
3774 BUG_ON(ss->module == NULL);
3775
3776 /*
3777 * we shouldn't be called if the subsystem is in use, and the use of
3778 * try_module_get in parse_cgroupfs_options should ensure that it
3779 * doesn't start being used while we're killing it off.
3780 */
3781 BUG_ON(ss->root != &rootnode);
3782
3783 mutex_lock(&cgroup_mutex);
3784 /* deassign the subsys_id */
3785 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
3786 subsys[ss->subsys_id] = NULL;
3787
3788 /* remove subsystem from rootnode's list of subsystems */
3789 list_del(&ss->sibling);
3790
3791 /*
3792 * disentangle the css from all css_sets attached to the dummytop. as
3793 * in loading, we need to pay our respects to the hashtable gods.
3794 */
3795 write_lock(&css_set_lock);
3796 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
3797 struct css_set *cg = link->cg;
3798
3799 hlist_del(&cg->hlist);
3800 BUG_ON(!cg->subsys[ss->subsys_id]);
3801 cg->subsys[ss->subsys_id] = NULL;
3802 hhead = css_set_hash(cg->subsys);
3803 hlist_add_head(&cg->hlist, hhead);
3804 }
3805 write_unlock(&css_set_lock);
3806
3807 /*
3808 * remove subsystem's css from the dummytop and free it - need to free
3809 * before marking as null because ss->destroy needs the cgrp->subsys
3810 * pointer to find their state. note that this also takes care of
3811 * freeing the css_id.
3812 */
3813 ss->destroy(ss, dummytop);
3814 dummytop->subsys[ss->subsys_id] = NULL;
3815
3816 mutex_unlock(&cgroup_mutex);
3817}
3818EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
3819
3820/**
Li Zefana043e3b2008-02-23 15:24:09 -08003821 * cgroup_init_early - cgroup initialization at system boot
3822 *
3823 * Initialize cgroups at system boot, and initialize any
3824 * subsystems that request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07003825 */
3826int __init cgroup_init_early(void)
3827{
3828 int i;
Lai Jiangshan146aa1b2008-10-18 20:28:03 -07003829 atomic_set(&init_css_set.refcount, 1);
Paul Menage817929e2007-10-18 23:39:36 -07003830 INIT_LIST_HEAD(&init_css_set.cg_links);
3831 INIT_LIST_HEAD(&init_css_set.tasks);
Li Zefan472b1052008-04-29 01:00:11 -07003832 INIT_HLIST_NODE(&init_css_set.hlist);
Paul Menage817929e2007-10-18 23:39:36 -07003833 css_set_count = 1;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003834 init_cgroup_root(&rootnode);
Paul Menage817929e2007-10-18 23:39:36 -07003835 root_count = 1;
3836 init_task.cgroups = &init_css_set;
3837
3838 init_css_set_link.cg = &init_css_set;
Paul Menage7717f7b2009-09-23 15:56:22 -07003839 init_css_set_link.cgrp = dummytop;
Paul Menagebd89aab2007-10-18 23:40:44 -07003840 list_add(&init_css_set_link.cgrp_link_list,
Paul Menage817929e2007-10-18 23:39:36 -07003841 &rootnode.top_cgroup.css_sets);
3842 list_add(&init_css_set_link.cg_link_list,
3843 &init_css_set.cg_links);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003844
Li Zefan472b1052008-04-29 01:00:11 -07003845 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
3846 INIT_HLIST_HEAD(&css_set_table[i]);
3847
Ben Blumaae8aab2010-03-10 15:22:07 -08003848 /* at bootup time, we don't worry about modular subsystems */
3849 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07003850 struct cgroup_subsys *ss = subsys[i];
3851
3852 BUG_ON(!ss->name);
3853 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
3854 BUG_ON(!ss->create);
3855 BUG_ON(!ss->destroy);
3856 if (ss->subsys_id != i) {
Diego Callejacfe36bd2007-11-14 16:58:54 -08003857 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
Paul Menageddbcc7e2007-10-18 23:39:30 -07003858 ss->name, ss->subsys_id);
3859 BUG();
3860 }
3861
3862 if (ss->early_init)
3863 cgroup_init_subsys(ss);
3864 }
3865 return 0;
3866}
3867
3868/**
Li Zefana043e3b2008-02-23 15:24:09 -08003869 * cgroup_init - cgroup initialization
3870 *
3871 * Register cgroup filesystem and /proc file, and initialize
3872 * any subsystems that didn't request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07003873 */
3874int __init cgroup_init(void)
3875{
3876 int err;
3877 int i;
Li Zefan472b1052008-04-29 01:00:11 -07003878 struct hlist_head *hhead;
Paul Menagea4243162007-10-18 23:39:35 -07003879
3880 err = bdi_init(&cgroup_backing_dev_info);
3881 if (err)
3882 return err;
Paul Menageddbcc7e2007-10-18 23:39:30 -07003883
Ben Blumaae8aab2010-03-10 15:22:07 -08003884 /* at bootup time, we don't worry about modular subsystems */
3885 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07003886 struct cgroup_subsys *ss = subsys[i];
3887 if (!ss->early_init)
3888 cgroup_init_subsys(ss);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07003889 if (ss->use_id)
Ben Blume6a11052010-03-10 15:22:09 -08003890 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
Paul Menageddbcc7e2007-10-18 23:39:30 -07003891 }
3892
Li Zefan472b1052008-04-29 01:00:11 -07003893 /* Add init_css_set to the hash table */
3894 hhead = css_set_hash(init_css_set.subsys);
3895 hlist_add_head(&init_css_set.hlist, hhead);
Paul Menage2c6ab6d2009-09-23 15:56:23 -07003896 BUG_ON(!init_root_id(&rootnode));
Paul Menageddbcc7e2007-10-18 23:39:30 -07003897 err = register_filesystem(&cgroup_fs_type);
3898 if (err < 0)
3899 goto out;
3900
Li Zefan46ae2202008-04-29 01:00:08 -07003901 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
Paul Menagea4243162007-10-18 23:39:35 -07003902
Paul Menageddbcc7e2007-10-18 23:39:30 -07003903out:
Paul Menagea4243162007-10-18 23:39:35 -07003904 if (err)
3905 bdi_destroy(&cgroup_backing_dev_info);
3906
Paul Menageddbcc7e2007-10-18 23:39:30 -07003907 return err;
3908}
Paul Menageb4f48b62007-10-18 23:39:33 -07003909
Paul Menagea4243162007-10-18 23:39:35 -07003910/*
3911 * proc_cgroup_show()
3912 * - Print task's cgroup paths into seq_file, one line for each hierarchy
3913 * - Used for /proc/<pid>/cgroup.
3914 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
3915 * doesn't really matter if tsk->cgroup changes after we read it,
Cliff Wickman956db3c2008-02-07 00:14:43 -08003916 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
Paul Menagea4243162007-10-18 23:39:35 -07003917 * anyway. No need to check that tsk->cgroup != NULL, thanks to
3918 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
3919 * cgroup to top_cgroup.
3920 */
3921
3922/* TODO: Use a proper seq_file iterator */
3923static int proc_cgroup_show(struct seq_file *m, void *v)
3924{
3925 struct pid *pid;
3926 struct task_struct *tsk;
3927 char *buf;
3928 int retval;
3929 struct cgroupfs_root *root;
3930
3931 retval = -ENOMEM;
3932 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3933 if (!buf)
3934 goto out;
3935
3936 retval = -ESRCH;
3937 pid = m->private;
3938 tsk = get_pid_task(pid, PIDTYPE_PID);
3939 if (!tsk)
3940 goto out_free;
3941
3942 retval = 0;
3943
3944 mutex_lock(&cgroup_mutex);
3945
Li Zefane5f6a862009-01-07 18:07:41 -08003946 for_each_active_root(root) {
Paul Menagea4243162007-10-18 23:39:35 -07003947 struct cgroup_subsys *ss;
Paul Menagebd89aab2007-10-18 23:40:44 -07003948 struct cgroup *cgrp;
Paul Menagea4243162007-10-18 23:39:35 -07003949 int count = 0;
3950
Paul Menage2c6ab6d2009-09-23 15:56:23 -07003951 seq_printf(m, "%d:", root->hierarchy_id);
Paul Menagea4243162007-10-18 23:39:35 -07003952 for_each_subsys(root, ss)
3953 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
Paul Menagec6d57f32009-09-23 15:56:19 -07003954 if (strlen(root->name))
3955 seq_printf(m, "%sname=%s", count ? "," : "",
3956 root->name);
Paul Menagea4243162007-10-18 23:39:35 -07003957 seq_putc(m, ':');
Paul Menage7717f7b2009-09-23 15:56:22 -07003958 cgrp = task_cgroup_from_root(tsk, root);
Paul Menagebd89aab2007-10-18 23:40:44 -07003959 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
Paul Menagea4243162007-10-18 23:39:35 -07003960 if (retval < 0)
3961 goto out_unlock;
3962 seq_puts(m, buf);
3963 seq_putc(m, '\n');
3964 }
3965
3966out_unlock:
3967 mutex_unlock(&cgroup_mutex);
3968 put_task_struct(tsk);
3969out_free:
3970 kfree(buf);
3971out:
3972 return retval;
3973}
3974
3975static int cgroup_open(struct inode *inode, struct file *file)
3976{
3977 struct pid *pid = PROC_I(inode)->pid;
3978 return single_open(file, proc_cgroup_show, pid);
3979}
3980
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003981const struct file_operations proc_cgroup_operations = {
Paul Menagea4243162007-10-18 23:39:35 -07003982 .open = cgroup_open,
3983 .read = seq_read,
3984 .llseek = seq_lseek,
3985 .release = single_release,
3986};
3987
3988/* Display information about each subsystem and each hierarchy */
3989static int proc_cgroupstats_show(struct seq_file *m, void *v)
3990{
3991 int i;
Paul Menagea4243162007-10-18 23:39:35 -07003992
Paul Menage8bab8dd2008-04-04 14:29:57 -07003993 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
Ben Blumaae8aab2010-03-10 15:22:07 -08003994 /*
3995 * ideally we don't want subsystems moving around while we do this.
3996 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
3997 * subsys/hierarchy state.
3998 */
Paul Menagea4243162007-10-18 23:39:35 -07003999 mutex_lock(&cgroup_mutex);
Paul Menagea4243162007-10-18 23:39:35 -07004000 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4001 struct cgroup_subsys *ss = subsys[i];
Ben Blumaae8aab2010-03-10 15:22:07 -08004002 if (ss == NULL)
4003 continue;
Paul Menage2c6ab6d2009-09-23 15:56:23 -07004004 seq_printf(m, "%s\t%d\t%d\t%d\n",
4005 ss->name, ss->root->hierarchy_id,
Paul Menage8bab8dd2008-04-04 14:29:57 -07004006 ss->root->number_of_cgroups, !ss->disabled);
Paul Menagea4243162007-10-18 23:39:35 -07004007 }
4008 mutex_unlock(&cgroup_mutex);
4009 return 0;
4010}
4011
4012static int cgroupstats_open(struct inode *inode, struct file *file)
4013{
Al Viro9dce07f2008-03-29 03:07:28 +00004014 return single_open(file, proc_cgroupstats_show, NULL);
Paul Menagea4243162007-10-18 23:39:35 -07004015}
4016
Alexey Dobriyan828c0952009-10-01 15:43:56 -07004017static const struct file_operations proc_cgroupstats_operations = {
Paul Menagea4243162007-10-18 23:39:35 -07004018 .open = cgroupstats_open,
4019 .read = seq_read,
4020 .llseek = seq_lseek,
4021 .release = single_release,
4022};
4023
Paul Menageb4f48b62007-10-18 23:39:33 -07004024/**
4025 * cgroup_fork - attach newly forked task to its parents cgroup.
Li Zefana043e3b2008-02-23 15:24:09 -08004026 * @child: pointer to task_struct of forking parent process.
Paul Menageb4f48b62007-10-18 23:39:33 -07004027 *
4028 * Description: A task inherits its parent's cgroup at fork().
4029 *
4030 * A pointer to the shared css_set was automatically copied in
4031 * fork.c by dup_task_struct(). However, we ignore that copy, since
4032 * it was not made under the protection of RCU or cgroup_mutex, so
Cliff Wickman956db3c2008-02-07 00:14:43 -08004033 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
Paul Menage817929e2007-10-18 23:39:36 -07004034 * have already changed current->cgroups, allowing the previously
4035 * referenced cgroup group to be removed and freed.
Paul Menageb4f48b62007-10-18 23:39:33 -07004036 *
4037 * At the point that cgroup_fork() is called, 'current' is the parent
4038 * task, and the passed argument 'child' points to the child task.
4039 */
4040void cgroup_fork(struct task_struct *child)
4041{
Paul Menage817929e2007-10-18 23:39:36 -07004042 task_lock(current);
4043 child->cgroups = current->cgroups;
4044 get_css_set(child->cgroups);
4045 task_unlock(current);
4046 INIT_LIST_HEAD(&child->cg_list);
Paul Menageb4f48b62007-10-18 23:39:33 -07004047}
4048
4049/**
Li Zefana043e3b2008-02-23 15:24:09 -08004050 * cgroup_fork_callbacks - run fork callbacks
4051 * @child: the new task
4052 *
4053 * Called on a new task very soon before adding it to the
4054 * tasklist. No need to take any locks since no-one can
4055 * be operating on this task.
Paul Menageb4f48b62007-10-18 23:39:33 -07004056 */
4057void cgroup_fork_callbacks(struct task_struct *child)
4058{
4059 if (need_forkexit_callback) {
4060 int i;
Ben Blumaae8aab2010-03-10 15:22:07 -08004061 /*
4062 * forkexit callbacks are only supported for builtin
4063 * subsystems, and the builtin section of the subsys array is
4064 * immutable, so we don't need to lock the subsys array here.
4065 */
4066 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
Paul Menageb4f48b62007-10-18 23:39:33 -07004067 struct cgroup_subsys *ss = subsys[i];
4068 if (ss->fork)
4069 ss->fork(ss, child);
4070 }
4071 }
4072}
4073
4074/**
Li Zefana043e3b2008-02-23 15:24:09 -08004075 * cgroup_post_fork - called on a new task after adding it to the task list
4076 * @child: the task in question
4077 *
4078 * Adds the task to the list running through its css_set if necessary.
4079 * Has to be after the task is visible on the task list in case we race
4080 * with the first call to cgroup_iter_start() - to guarantee that the
4081 * new task ends up on its list.
4082 */
Paul Menage817929e2007-10-18 23:39:36 -07004083void cgroup_post_fork(struct task_struct *child)
4084{
4085 if (use_task_css_set_links) {
4086 write_lock(&css_set_lock);
Lai Jiangshanb12b5332009-01-07 18:07:36 -08004087 task_lock(child);
Paul Menage817929e2007-10-18 23:39:36 -07004088 if (list_empty(&child->cg_list))
4089 list_add(&child->cg_list, &child->cgroups->tasks);
Lai Jiangshanb12b5332009-01-07 18:07:36 -08004090 task_unlock(child);
Paul Menage817929e2007-10-18 23:39:36 -07004091 write_unlock(&css_set_lock);
4092 }
4093}
4094/**
Paul Menageb4f48b62007-10-18 23:39:33 -07004095 * cgroup_exit - detach cgroup from exiting task
4096 * @tsk: pointer to task_struct of exiting process
Li Zefana043e3b2008-02-23 15:24:09 -08004097 * @run_callback: run exit callbacks?
Paul Menageb4f48b62007-10-18 23:39:33 -07004098 *
4099 * Description: Detach cgroup from @tsk and release it.
4100 *
4101 * Note that cgroups marked notify_on_release force every task in
4102 * them to take the global cgroup_mutex mutex when exiting.
4103 * This could impact scaling on very large systems. Be reluctant to
4104 * use notify_on_release cgroups where very high task exit scaling
4105 * is required on large systems.
4106 *
4107 * the_top_cgroup_hack:
4108 *
4109 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4110 *
4111 * We call cgroup_exit() while the task is still competent to
4112 * handle notify_on_release(), then leave the task attached to the
4113 * root cgroup in each hierarchy for the remainder of its exit.
4114 *
4115 * To do this properly, we would increment the reference count on
4116 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4117 * code we would add a second cgroup function call, to drop that
4118 * reference. This would just create an unnecessary hot spot on
4119 * the top_cgroup reference count, to no avail.
4120 *
4121 * Normally, holding a reference to a cgroup without bumping its
4122 * count is unsafe. The cgroup could go away, or someone could
4123 * attach us to a different cgroup, decrementing the count on
4124 * the first cgroup that we never incremented. But in this case,
4125 * top_cgroup isn't going away, and either task has PF_EXITING set,
Cliff Wickman956db3c2008-02-07 00:14:43 -08004126 * which wards off any cgroup_attach_task() attempts, or task is a failed
4127 * fork, never visible to cgroup_attach_task.
Paul Menageb4f48b62007-10-18 23:39:33 -07004128 */
4129void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4130{
4131 int i;
Paul Menage817929e2007-10-18 23:39:36 -07004132 struct css_set *cg;
Paul Menageb4f48b62007-10-18 23:39:33 -07004133
4134 if (run_callbacks && need_forkexit_callback) {
Ben Blumaae8aab2010-03-10 15:22:07 -08004135 /*
4136 * modular subsystems can't use callbacks, so no need to lock
4137 * the subsys array
4138 */
4139 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
Paul Menageb4f48b62007-10-18 23:39:33 -07004140 struct cgroup_subsys *ss = subsys[i];
4141 if (ss->exit)
4142 ss->exit(ss, tsk);
4143 }
4144 }
Paul Menage817929e2007-10-18 23:39:36 -07004145
4146 /*
4147 * Unlink from the css_set task list if necessary.
4148 * Optimistically check cg_list before taking
4149 * css_set_lock
4150 */
4151 if (!list_empty(&tsk->cg_list)) {
4152 write_lock(&css_set_lock);
4153 if (!list_empty(&tsk->cg_list))
4154 list_del(&tsk->cg_list);
4155 write_unlock(&css_set_lock);
4156 }
4157
Paul Menageb4f48b62007-10-18 23:39:33 -07004158 /* Reassign the task to the init_css_set. */
4159 task_lock(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07004160 cg = tsk->cgroups;
4161 tsk->cgroups = &init_css_set;
Paul Menageb4f48b62007-10-18 23:39:33 -07004162 task_unlock(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07004163 if (cg)
Paul Menage81a6a5c2007-10-18 23:39:38 -07004164 put_css_set_taskexit(cg);
Paul Menageb4f48b62007-10-18 23:39:33 -07004165}
Paul Menage697f4162007-10-18 23:39:34 -07004166
4167/**
Li Zefana043e3b2008-02-23 15:24:09 -08004168 * cgroup_clone - clone the cgroup the given subsystem is attached to
4169 * @tsk: the task to be moved
4170 * @subsys: the given subsystem
Serge E. Hallyne885dcd2008-07-25 01:47:06 -07004171 * @nodename: the name for the new cgroup
Li Zefana043e3b2008-02-23 15:24:09 -08004172 *
4173 * Duplicate the current cgroup in the hierarchy that the given
4174 * subsystem is attached to, and move this task into the new
4175 * child.
Paul Menage697f4162007-10-18 23:39:34 -07004176 */
Serge E. Hallyne885dcd2008-07-25 01:47:06 -07004177int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys,
4178 char *nodename)
Paul Menage697f4162007-10-18 23:39:34 -07004179{
4180 struct dentry *dentry;
4181 int ret = 0;
Paul Menage697f4162007-10-18 23:39:34 -07004182 struct cgroup *parent, *child;
4183 struct inode *inode;
4184 struct css_set *cg;
4185 struct cgroupfs_root *root;
4186 struct cgroup_subsys *ss;
4187
4188 /* We shouldn't be called by an unregistered subsystem */
4189 BUG_ON(!subsys->active);
4190
4191 /* First figure out what hierarchy and cgroup we're dealing
4192 * with, and pin them so we can drop cgroup_mutex */
4193 mutex_lock(&cgroup_mutex);
4194 again:
4195 root = subsys->root;
4196 if (root == &rootnode) {
Paul Menage697f4162007-10-18 23:39:34 -07004197 mutex_unlock(&cgroup_mutex);
4198 return 0;
4199 }
Paul Menage697f4162007-10-18 23:39:34 -07004200
Paul Menage697f4162007-10-18 23:39:34 -07004201 /* Pin the hierarchy */
Li Zefan1404f062009-01-29 14:25:21 -08004202 if (!atomic_inc_not_zero(&root->sb->s_active)) {
Li Zefan7b574b72009-01-04 12:00:45 -08004203 /* We race with the final deactivate_super() */
4204 mutex_unlock(&cgroup_mutex);
4205 return 0;
4206 }
Paul Menage697f4162007-10-18 23:39:34 -07004207
Paul Menage817929e2007-10-18 23:39:36 -07004208 /* Keep the cgroup alive */
Li Zefan1404f062009-01-29 14:25:21 -08004209 task_lock(tsk);
4210 parent = task_cgroup(tsk, subsys->subsys_id);
4211 cg = tsk->cgroups;
Paul Menage817929e2007-10-18 23:39:36 -07004212 get_css_set(cg);
Lai Jiangshan104cbd52009-01-07 18:07:38 -08004213 task_unlock(tsk);
Li Zefan1404f062009-01-29 14:25:21 -08004214
Paul Menage697f4162007-10-18 23:39:34 -07004215 mutex_unlock(&cgroup_mutex);
4216
4217 /* Now do the VFS work to create a cgroup */
4218 inode = parent->dentry->d_inode;
4219
4220 /* Hold the parent directory mutex across this operation to
4221 * stop anyone else deleting the new cgroup */
4222 mutex_lock(&inode->i_mutex);
4223 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
4224 if (IS_ERR(dentry)) {
4225 printk(KERN_INFO
Diego Callejacfe36bd2007-11-14 16:58:54 -08004226 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
Paul Menage697f4162007-10-18 23:39:34 -07004227 PTR_ERR(dentry));
4228 ret = PTR_ERR(dentry);
4229 goto out_release;
4230 }
4231
4232 /* Create the cgroup directory, which also creates the cgroup */
Li Zefan75139b82009-01-07 18:07:33 -08004233 ret = vfs_mkdir(inode, dentry, 0755);
Paul Menagebd89aab2007-10-18 23:40:44 -07004234 child = __d_cgrp(dentry);
Paul Menage697f4162007-10-18 23:39:34 -07004235 dput(dentry);
4236 if (ret) {
4237 printk(KERN_INFO
4238 "Failed to create cgroup %s: %d\n", nodename,
4239 ret);
4240 goto out_release;
4241 }
4242
Paul Menage697f4162007-10-18 23:39:34 -07004243 /* The cgroup now exists. Retake cgroup_mutex and check
4244 * that we're still in the same state that we thought we
4245 * were. */
4246 mutex_lock(&cgroup_mutex);
4247 if ((root != subsys->root) ||
4248 (parent != task_cgroup(tsk, subsys->subsys_id))) {
4249 /* Aargh, we raced ... */
4250 mutex_unlock(&inode->i_mutex);
Paul Menage817929e2007-10-18 23:39:36 -07004251 put_css_set(cg);
Paul Menage697f4162007-10-18 23:39:34 -07004252
Li Zefan1404f062009-01-29 14:25:21 -08004253 deactivate_super(root->sb);
Paul Menage697f4162007-10-18 23:39:34 -07004254 /* The cgroup is still accessible in the VFS, but
4255 * we're not going to try to rmdir() it at this
4256 * point. */
4257 printk(KERN_INFO
4258 "Race in cgroup_clone() - leaking cgroup %s\n",
4259 nodename);
4260 goto again;
4261 }
4262
4263 /* do any required auto-setup */
4264 for_each_subsys(root, ss) {
4265 if (ss->post_clone)
4266 ss->post_clone(ss, child);
4267 }
4268
4269 /* All seems fine. Finish by moving the task into the new cgroup */
Cliff Wickman956db3c2008-02-07 00:14:43 -08004270 ret = cgroup_attach_task(child, tsk);
Paul Menage697f4162007-10-18 23:39:34 -07004271 mutex_unlock(&cgroup_mutex);
4272
4273 out_release:
4274 mutex_unlock(&inode->i_mutex);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004275
4276 mutex_lock(&cgroup_mutex);
Paul Menage817929e2007-10-18 23:39:36 -07004277 put_css_set(cg);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004278 mutex_unlock(&cgroup_mutex);
Li Zefan1404f062009-01-29 14:25:21 -08004279 deactivate_super(root->sb);
Paul Menage697f4162007-10-18 23:39:34 -07004280 return ret;
4281}
4282
Li Zefana043e3b2008-02-23 15:24:09 -08004283/**
Grzegorz Nosek313e9242009-04-02 16:57:23 -07004284 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
Li Zefana043e3b2008-02-23 15:24:09 -08004285 * @cgrp: the cgroup in question
Grzegorz Nosek313e9242009-04-02 16:57:23 -07004286 * @task: the task in question
Li Zefana043e3b2008-02-23 15:24:09 -08004287 *
Grzegorz Nosek313e9242009-04-02 16:57:23 -07004288 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4289 * hierarchy.
Paul Menage697f4162007-10-18 23:39:34 -07004290 *
4291 * If we are sending in dummytop, then presumably we are creating
4292 * the top cgroup in the subsystem.
4293 *
4294 * Called only by the ns (nsproxy) cgroup.
4295 */
Grzegorz Nosek313e9242009-04-02 16:57:23 -07004296int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
Paul Menage697f4162007-10-18 23:39:34 -07004297{
4298 int ret;
4299 struct cgroup *target;
Paul Menage697f4162007-10-18 23:39:34 -07004300
Paul Menagebd89aab2007-10-18 23:40:44 -07004301 if (cgrp == dummytop)
Paul Menage697f4162007-10-18 23:39:34 -07004302 return 1;
4303
Paul Menage7717f7b2009-09-23 15:56:22 -07004304 target = task_cgroup_from_root(task, cgrp->root);
Paul Menagebd89aab2007-10-18 23:40:44 -07004305 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4306 cgrp = cgrp->parent;
4307 ret = (cgrp == target);
Paul Menage697f4162007-10-18 23:39:34 -07004308 return ret;
4309}
Paul Menage81a6a5c2007-10-18 23:39:38 -07004310
Paul Menagebd89aab2007-10-18 23:40:44 -07004311static void check_for_release(struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -07004312{
4313 /* All of these checks rely on RCU to keep the cgroup
4314 * structure alive */
Paul Menagebd89aab2007-10-18 23:40:44 -07004315 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4316 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
Paul Menage81a6a5c2007-10-18 23:39:38 -07004317 /* Control Group is currently removeable. If it's not
4318 * already queued for a userspace notification, queue
4319 * it now */
4320 int need_schedule_work = 0;
4321 spin_lock(&release_list_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07004322 if (!cgroup_is_removed(cgrp) &&
4323 list_empty(&cgrp->release_list)) {
4324 list_add(&cgrp->release_list, &release_list);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004325 need_schedule_work = 1;
4326 }
4327 spin_unlock(&release_list_lock);
4328 if (need_schedule_work)
4329 schedule_work(&release_agent_work);
4330 }
4331}
4332
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -08004333/* Caller must verify that the css is not for root cgroup */
4334void __css_put(struct cgroup_subsys_state *css, int count)
Paul Menage81a6a5c2007-10-18 23:39:38 -07004335{
Paul Menagebd89aab2007-10-18 23:40:44 -07004336 struct cgroup *cgrp = css->cgroup;
KAMEZAWA Hiroyuki3dece832009-10-01 15:44:09 -07004337 int val;
Paul Menage81a6a5c2007-10-18 23:39:38 -07004338 rcu_read_lock();
Daisuke Nishimurad7b9fff2010-03-10 15:22:05 -08004339 val = atomic_sub_return(count, &css->refcnt);
KAMEZAWA Hiroyuki3dece832009-10-01 15:44:09 -07004340 if (val == 1) {
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004341 if (notify_on_release(cgrp)) {
4342 set_bit(CGRP_RELEASABLE, &cgrp->flags);
4343 check_for_release(cgrp);
4344 }
KAMEZAWA Hiroyuki88703262009-07-29 15:04:06 -07004345 cgroup_wakeup_rmdir_waiter(cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004346 }
4347 rcu_read_unlock();
KAMEZAWA Hiroyuki3dece832009-10-01 15:44:09 -07004348 WARN_ON_ONCE(val < 1);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004349}
Ben Blum67523c42010-03-10 15:22:11 -08004350EXPORT_SYMBOL_GPL(__css_put);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004351
4352/*
4353 * Notify userspace when a cgroup is released, by running the
4354 * configured release agent with the name of the cgroup (path
4355 * relative to the root of cgroup file system) as the argument.
4356 *
4357 * Most likely, this user command will try to rmdir this cgroup.
4358 *
4359 * This races with the possibility that some other task will be
4360 * attached to this cgroup before it is removed, or that some other
4361 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4362 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4363 * unused, and this cgroup will be reprieved from its death sentence,
4364 * to continue to serve a useful existence. Next time it's released,
4365 * we will get notified again, if it still has 'notify_on_release' set.
4366 *
4367 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4368 * means only wait until the task is successfully execve()'d. The
4369 * separate release agent task is forked by call_usermodehelper(),
4370 * then control in this thread returns here, without waiting for the
4371 * release agent task. We don't bother to wait because the caller of
4372 * this routine has no use for the exit status of the release agent
4373 * task, so no sense holding our caller up for that.
Paul Menage81a6a5c2007-10-18 23:39:38 -07004374 */
Paul Menage81a6a5c2007-10-18 23:39:38 -07004375static void cgroup_release_agent(struct work_struct *work)
4376{
4377 BUG_ON(work != &release_agent_work);
4378 mutex_lock(&cgroup_mutex);
4379 spin_lock(&release_list_lock);
4380 while (!list_empty(&release_list)) {
4381 char *argv[3], *envp[3];
4382 int i;
Paul Menagee788e062008-07-25 01:46:59 -07004383 char *pathbuf = NULL, *agentbuf = NULL;
Paul Menagebd89aab2007-10-18 23:40:44 -07004384 struct cgroup *cgrp = list_entry(release_list.next,
Paul Menage81a6a5c2007-10-18 23:39:38 -07004385 struct cgroup,
4386 release_list);
Paul Menagebd89aab2007-10-18 23:40:44 -07004387 list_del_init(&cgrp->release_list);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004388 spin_unlock(&release_list_lock);
4389 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
Paul Menagee788e062008-07-25 01:46:59 -07004390 if (!pathbuf)
4391 goto continue_free;
4392 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4393 goto continue_free;
4394 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4395 if (!agentbuf)
4396 goto continue_free;
Paul Menage81a6a5c2007-10-18 23:39:38 -07004397
4398 i = 0;
Paul Menagee788e062008-07-25 01:46:59 -07004399 argv[i++] = agentbuf;
4400 argv[i++] = pathbuf;
Paul Menage81a6a5c2007-10-18 23:39:38 -07004401 argv[i] = NULL;
4402
4403 i = 0;
4404 /* minimal command environment */
4405 envp[i++] = "HOME=/";
4406 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4407 envp[i] = NULL;
4408
4409 /* Drop the lock while we invoke the usermode helper,
4410 * since the exec could involve hitting disk and hence
4411 * be a slow process */
4412 mutex_unlock(&cgroup_mutex);
4413 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004414 mutex_lock(&cgroup_mutex);
Paul Menagee788e062008-07-25 01:46:59 -07004415 continue_free:
4416 kfree(pathbuf);
4417 kfree(agentbuf);
Paul Menage81a6a5c2007-10-18 23:39:38 -07004418 spin_lock(&release_list_lock);
4419 }
4420 spin_unlock(&release_list_lock);
4421 mutex_unlock(&cgroup_mutex);
4422}
Paul Menage8bab8dd2008-04-04 14:29:57 -07004423
4424static int __init cgroup_disable(char *str)
4425{
4426 int i;
4427 char *token;
4428
4429 while ((token = strsep(&str, ",")) != NULL) {
4430 if (!*token)
4431 continue;
Ben Blumaae8aab2010-03-10 15:22:07 -08004432 /*
4433 * cgroup_disable, being at boot time, can't know about module
4434 * subsystems, so we don't worry about them.
4435 */
4436 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
Paul Menage8bab8dd2008-04-04 14:29:57 -07004437 struct cgroup_subsys *ss = subsys[i];
4438
4439 if (!strcmp(token, ss->name)) {
4440 ss->disabled = 1;
4441 printk(KERN_INFO "Disabling %s control group"
4442 " subsystem\n", ss->name);
4443 break;
4444 }
4445 }
4446 }
4447 return 1;
4448}
4449__setup("cgroup_disable=", cgroup_disable);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004450
4451/*
4452 * Functons for CSS ID.
4453 */
4454
4455/*
4456 *To get ID other than 0, this should be called when !cgroup_is_removed().
4457 */
4458unsigned short css_id(struct cgroup_subsys_state *css)
4459{
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004460 struct css_id *cssid;
4461
4462 /*
4463 * This css_id() can return correct value when somone has refcnt
4464 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4465 * it's unchanged until freed.
4466 */
4467 cssid = rcu_dereference_check(css->id,
4468 rcu_read_lock_held() || atomic_read(&css->refcnt));
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004469
4470 if (cssid)
4471 return cssid->id;
4472 return 0;
4473}
Ben Blum67523c42010-03-10 15:22:11 -08004474EXPORT_SYMBOL_GPL(css_id);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004475
4476unsigned short css_depth(struct cgroup_subsys_state *css)
4477{
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004478 struct css_id *cssid;
4479
4480 cssid = rcu_dereference_check(css->id,
4481 rcu_read_lock_held() || atomic_read(&css->refcnt));
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004482
4483 if (cssid)
4484 return cssid->depth;
4485 return 0;
4486}
Ben Blum67523c42010-03-10 15:22:11 -08004487EXPORT_SYMBOL_GPL(css_depth);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004488
KAMEZAWA Hiroyuki747388d2010-05-11 14:06:59 -07004489/**
4490 * css_is_ancestor - test "root" css is an ancestor of "child"
4491 * @child: the css to be tested.
4492 * @root: the css supporsed to be an ancestor of the child.
4493 *
4494 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4495 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4496 * But, considering usual usage, the csses should be valid objects after test.
4497 * Assuming that the caller will do some action to the child if this returns
4498 * returns true, the caller must take "child";s reference count.
4499 * If "child" is valid object and this returns true, "root" is valid, too.
4500 */
4501
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004502bool css_is_ancestor(struct cgroup_subsys_state *child,
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07004503 const struct cgroup_subsys_state *root)
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004504{
KAMEZAWA Hiroyuki747388d2010-05-11 14:06:59 -07004505 struct css_id *child_id;
4506 struct css_id *root_id;
4507 bool ret = true;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004508
KAMEZAWA Hiroyuki747388d2010-05-11 14:06:59 -07004509 rcu_read_lock();
4510 child_id = rcu_dereference(child->id);
4511 root_id = rcu_dereference(root->id);
4512 if (!child_id
4513 || !root_id
4514 || (child_id->depth < root_id->depth)
4515 || (child_id->stack[root_id->depth] != root_id->id))
4516 ret = false;
4517 rcu_read_unlock();
4518 return ret;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004519}
4520
4521static void __free_css_id_cb(struct rcu_head *head)
4522{
4523 struct css_id *id;
4524
4525 id = container_of(head, struct css_id, rcu_head);
4526 kfree(id);
4527}
4528
4529void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
4530{
4531 struct css_id *id = css->id;
4532 /* When this is called before css_id initialization, id can be NULL */
4533 if (!id)
4534 return;
4535
4536 BUG_ON(!ss->use_id);
4537
4538 rcu_assign_pointer(id->css, NULL);
4539 rcu_assign_pointer(css->id, NULL);
4540 spin_lock(&ss->id_lock);
4541 idr_remove(&ss->idr, id->id);
4542 spin_unlock(&ss->id_lock);
4543 call_rcu(&id->rcu_head, __free_css_id_cb);
4544}
Ben Blum67523c42010-03-10 15:22:11 -08004545EXPORT_SYMBOL_GPL(free_css_id);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004546
4547/*
4548 * This is called by init or create(). Then, calls to this function are
4549 * always serialized (By cgroup_mutex() at create()).
4550 */
4551
4552static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
4553{
4554 struct css_id *newid;
4555 int myid, error, size;
4556
4557 BUG_ON(!ss->use_id);
4558
4559 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
4560 newid = kzalloc(size, GFP_KERNEL);
4561 if (!newid)
4562 return ERR_PTR(-ENOMEM);
4563 /* get id */
4564 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
4565 error = -ENOMEM;
4566 goto err_out;
4567 }
4568 spin_lock(&ss->id_lock);
4569 /* Don't use 0. allocates an ID of 1-65535 */
4570 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
4571 spin_unlock(&ss->id_lock);
4572
4573 /* Returns error when there are no free spaces for new ID.*/
4574 if (error) {
4575 error = -ENOSPC;
4576 goto err_out;
4577 }
4578 if (myid > CSS_ID_MAX)
4579 goto remove_idr;
4580
4581 newid->id = myid;
4582 newid->depth = depth;
4583 return newid;
4584remove_idr:
4585 error = -ENOSPC;
4586 spin_lock(&ss->id_lock);
4587 idr_remove(&ss->idr, myid);
4588 spin_unlock(&ss->id_lock);
4589err_out:
4590 kfree(newid);
4591 return ERR_PTR(error);
4592
4593}
4594
Ben Blume6a11052010-03-10 15:22:09 -08004595static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
4596 struct cgroup_subsys_state *rootcss)
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004597{
4598 struct css_id *newid;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004599
4600 spin_lock_init(&ss->id_lock);
4601 idr_init(&ss->idr);
4602
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004603 newid = get_new_cssid(ss, 0);
4604 if (IS_ERR(newid))
4605 return PTR_ERR(newid);
4606
4607 newid->stack[0] = newid->id;
4608 newid->css = rootcss;
4609 rootcss->id = newid;
4610 return 0;
4611}
4612
4613static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
4614 struct cgroup *child)
4615{
4616 int subsys_id, i, depth = 0;
4617 struct cgroup_subsys_state *parent_css, *child_css;
Li Zefanfae9c792010-04-22 17:30:00 +08004618 struct css_id *child_id, *parent_id;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004619
4620 subsys_id = ss->subsys_id;
4621 parent_css = parent->subsys[subsys_id];
4622 child_css = child->subsys[subsys_id];
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004623 parent_id = parent_css->id;
Greg Thelen94b3dd02010-06-04 14:15:03 -07004624 depth = parent_id->depth + 1;
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004625
4626 child_id = get_new_cssid(ss, depth);
4627 if (IS_ERR(child_id))
4628 return PTR_ERR(child_id);
4629
4630 for (i = 0; i < depth; i++)
4631 child_id->stack[i] = parent_id->stack[i];
4632 child_id->stack[depth] = child_id->id;
4633 /*
4634 * child_id->css pointer will be set after this cgroup is available
4635 * see cgroup_populate_dir()
4636 */
4637 rcu_assign_pointer(child_css->id, child_id);
4638
4639 return 0;
4640}
4641
4642/**
4643 * css_lookup - lookup css by id
4644 * @ss: cgroup subsys to be looked into.
4645 * @id: the id
4646 *
4647 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4648 * NULL if not. Should be called under rcu_read_lock()
4649 */
4650struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
4651{
4652 struct css_id *cssid = NULL;
4653
4654 BUG_ON(!ss->use_id);
4655 cssid = idr_find(&ss->idr, id);
4656
4657 if (unlikely(!cssid))
4658 return NULL;
4659
4660 return rcu_dereference(cssid->css);
4661}
Ben Blum67523c42010-03-10 15:22:11 -08004662EXPORT_SYMBOL_GPL(css_lookup);
KAMEZAWA Hiroyuki38460b42009-04-02 16:57:25 -07004663
4664/**
4665 * css_get_next - lookup next cgroup under specified hierarchy.
4666 * @ss: pointer to subsystem
4667 * @id: current position of iteration.
4668 * @root: pointer to css. search tree under this.
4669 * @foundid: position of found object.
4670 *
4671 * Search next css under the specified hierarchy of rootid. Calling under
4672 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
4673 */
4674struct cgroup_subsys_state *
4675css_get_next(struct cgroup_subsys *ss, int id,
4676 struct cgroup_subsys_state *root, int *foundid)
4677{
4678 struct cgroup_subsys_state *ret = NULL;
4679 struct css_id *tmp;
4680 int tmpid;
4681 int rootid = css_id(root);
4682 int depth = css_depth(root);
4683
4684 if (!rootid)
4685 return NULL;
4686
4687 BUG_ON(!ss->use_id);
4688 /* fill start point for scan */
4689 tmpid = id;
4690 while (1) {
4691 /*
4692 * scan next entry from bitmap(tree), tmpid is updated after
4693 * idr_get_next().
4694 */
4695 spin_lock(&ss->id_lock);
4696 tmp = idr_get_next(&ss->idr, &tmpid);
4697 spin_unlock(&ss->id_lock);
4698
4699 if (!tmp)
4700 break;
4701 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
4702 ret = rcu_dereference(tmp->css);
4703 if (ret) {
4704 *foundid = tmpid;
4705 break;
4706 }
4707 }
4708 /* continue to scan from next id */
4709 tmpid = tmpid + 1;
4710 }
4711 return ret;
4712}
4713
Paul Menagefe693432009-09-23 15:56:20 -07004714#ifdef CONFIG_CGROUP_DEBUG
4715static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
4716 struct cgroup *cont)
4717{
4718 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
4719
4720 if (!css)
4721 return ERR_PTR(-ENOMEM);
4722
4723 return css;
4724}
4725
4726static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
4727{
4728 kfree(cont->subsys[debug_subsys_id]);
4729}
4730
4731static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
4732{
4733 return atomic_read(&cont->count);
4734}
4735
4736static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
4737{
4738 return cgroup_task_count(cont);
4739}
4740
4741static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
4742{
4743 return (u64)(unsigned long)current->cgroups;
4744}
4745
4746static u64 current_css_set_refcount_read(struct cgroup *cont,
4747 struct cftype *cft)
4748{
4749 u64 count;
4750
4751 rcu_read_lock();
4752 count = atomic_read(&current->cgroups->refcount);
4753 rcu_read_unlock();
4754 return count;
4755}
4756
Paul Menage7717f7b2009-09-23 15:56:22 -07004757static int current_css_set_cg_links_read(struct cgroup *cont,
4758 struct cftype *cft,
4759 struct seq_file *seq)
4760{
4761 struct cg_cgroup_link *link;
4762 struct css_set *cg;
4763
4764 read_lock(&css_set_lock);
4765 rcu_read_lock();
4766 cg = rcu_dereference(current->cgroups);
4767 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
4768 struct cgroup *c = link->cgrp;
4769 const char *name;
4770
4771 if (c->dentry)
4772 name = c->dentry->d_name.name;
4773 else
4774 name = "?";
Paul Menage2c6ab6d2009-09-23 15:56:23 -07004775 seq_printf(seq, "Root %d group %s\n",
4776 c->root->hierarchy_id, name);
Paul Menage7717f7b2009-09-23 15:56:22 -07004777 }
4778 rcu_read_unlock();
4779 read_unlock(&css_set_lock);
4780 return 0;
4781}
4782
4783#define MAX_TASKS_SHOWN_PER_CSS 25
4784static int cgroup_css_links_read(struct cgroup *cont,
4785 struct cftype *cft,
4786 struct seq_file *seq)
4787{
4788 struct cg_cgroup_link *link;
4789
4790 read_lock(&css_set_lock);
4791 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
4792 struct css_set *cg = link->cg;
4793 struct task_struct *task;
4794 int count = 0;
4795 seq_printf(seq, "css_set %p\n", cg);
4796 list_for_each_entry(task, &cg->tasks, cg_list) {
4797 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
4798 seq_puts(seq, " ...\n");
4799 break;
4800 } else {
4801 seq_printf(seq, " task %d\n",
4802 task_pid_vnr(task));
4803 }
4804 }
4805 }
4806 read_unlock(&css_set_lock);
4807 return 0;
4808}
4809
Paul Menagefe693432009-09-23 15:56:20 -07004810static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
4811{
4812 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
4813}
4814
4815static struct cftype debug_files[] = {
4816 {
4817 .name = "cgroup_refcount",
4818 .read_u64 = cgroup_refcount_read,
4819 },
4820 {
4821 .name = "taskcount",
4822 .read_u64 = debug_taskcount_read,
4823 },
4824
4825 {
4826 .name = "current_css_set",
4827 .read_u64 = current_css_set_read,
4828 },
4829
4830 {
4831 .name = "current_css_set_refcount",
4832 .read_u64 = current_css_set_refcount_read,
4833 },
4834
4835 {
Paul Menage7717f7b2009-09-23 15:56:22 -07004836 .name = "current_css_set_cg_links",
4837 .read_seq_string = current_css_set_cg_links_read,
4838 },
4839
4840 {
4841 .name = "cgroup_css_links",
4842 .read_seq_string = cgroup_css_links_read,
4843 },
4844
4845 {
Paul Menagefe693432009-09-23 15:56:20 -07004846 .name = "releasable",
4847 .read_u64 = releasable_read,
4848 },
4849};
4850
4851static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
4852{
4853 return cgroup_add_files(cont, ss, debug_files,
4854 ARRAY_SIZE(debug_files));
4855}
4856
4857struct cgroup_subsys debug_subsys = {
4858 .name = "debug",
4859 .create = debug_create,
4860 .destroy = debug_destroy,
4861 .populate = debug_populate,
4862 .subsys_id = debug_subsys_id,
4863};
4864#endif /* CONFIG_CGROUP_DEBUG */