blob: aeceb88689817482ceadbd51a233d294de5c5a35 [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 *
7 * Copyright notices from the original cpuset code:
8 * --------------------------------------------------
9 * Copyright (C) 2003 BULL SA.
10 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
11 *
12 * Portions derived from Patrick Mochel's sysfs code.
13 * sysfs is Copyright (c) 2001-3 Patrick Mochel
14 *
15 * 2003-10-10 Written by Simon Derr.
16 * 2003-10-22 Updates by Stephen Hemminger.
17 * 2004 May-July Rework by Paul Jackson.
18 * ---------------------------------------------------
19 *
20 * This file is subject to the terms and conditions of the GNU General Public
21 * License. See the file COPYING in the main directory of the Linux
22 * distribution for more details.
23 */
24
25#include <linux/cgroup.h>
26#include <linux/errno.h>
27#include <linux/fs.h>
28#include <linux/kernel.h>
29#include <linux/list.h>
30#include <linux/mm.h>
31#include <linux/mutex.h>
32#include <linux/mount.h>
33#include <linux/pagemap.h>
Paul Menagea4243162007-10-18 23:39:35 -070034#include <linux/proc_fs.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070035#include <linux/rcupdate.h>
36#include <linux/sched.h>
Paul Menage817929e2007-10-18 23:39:36 -070037#include <linux/backing-dev.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070038#include <linux/seq_file.h>
39#include <linux/slab.h>
40#include <linux/magic.h>
41#include <linux/spinlock.h>
42#include <linux/string.h>
Paul Menagebbcb81d2007-10-18 23:39:32 -070043#include <linux/sort.h>
Paul Menage81a6a5c2007-10-18 23:39:38 -070044#include <linux/kmod.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070045#include <linux/delayacct.h>
46#include <linux/cgroupstats.h>
Li Zefan472b1052008-04-29 01:00:11 -070047#include <linux/hash.h>
Balbir Singh846c7bb2007-10-18 23:39:44 -070048
Paul Menageddbcc7e2007-10-18 23:39:30 -070049#include <asm/atomic.h>
50
Paul Menage81a6a5c2007-10-18 23:39:38 -070051static DEFINE_MUTEX(cgroup_mutex);
52
Paul Menageddbcc7e2007-10-18 23:39:30 -070053/* Generate an array of cgroup subsystem pointers */
54#define SUBSYS(_x) &_x ## _subsys,
55
56static struct cgroup_subsys *subsys[] = {
57#include <linux/cgroup_subsys.h>
58};
59
60/*
61 * A cgroupfs_root represents the root of a cgroup hierarchy,
62 * and may be associated with a superblock to form an active
63 * hierarchy
64 */
65struct cgroupfs_root {
66 struct super_block *sb;
67
68 /*
69 * The bitmask of subsystems intended to be attached to this
70 * hierarchy
71 */
72 unsigned long subsys_bits;
73
74 /* The bitmask of subsystems currently attached to this hierarchy */
75 unsigned long actual_subsys_bits;
76
77 /* A list running through the attached subsystems */
78 struct list_head subsys_list;
79
80 /* The root cgroup for this hierarchy */
81 struct cgroup top_cgroup;
82
83 /* Tracks how many cgroups are currently defined in hierarchy.*/
84 int number_of_cgroups;
85
86 /* A list running through the mounted hierarchies */
87 struct list_head root_list;
88
89 /* Hierarchy-specific flags */
90 unsigned long flags;
Paul Menage81a6a5c2007-10-18 23:39:38 -070091
92 /* The path to use for release notifications. No locking
93 * between setting and use - so if userspace updates this
94 * while child cgroups exist, you could miss a
95 * notification. We ensure that it's always a valid
96 * NUL-terminated string */
97 char release_agent_path[PATH_MAX];
Paul Menageddbcc7e2007-10-18 23:39:30 -070098};
99
100
101/*
102 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
103 * subsystems that are otherwise unattached - it never has more than a
104 * single cgroup, and all tasks are part of that cgroup.
105 */
106static struct cgroupfs_root rootnode;
107
108/* The list of hierarchy roots */
109
110static LIST_HEAD(roots);
Paul Menage817929e2007-10-18 23:39:36 -0700111static int root_count;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700112
113/* dummytop is a shorthand for the dummy hierarchy's top cgroup */
114#define dummytop (&rootnode.top_cgroup)
115
116/* This flag indicates whether tasks in the fork and exit paths should
Li Zefana043e3b2008-02-23 15:24:09 -0800117 * check for fork/exit handlers to call. This avoids us having to do
118 * extra work in the fork/exit path if none of the subsystems need to
119 * be called.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700120 */
121static int need_forkexit_callback;
122
Paul Menageddbcc7e2007-10-18 23:39:30 -0700123/* convenient tests for these bits */
Paul Menagebd89aab2007-10-18 23:40:44 -0700124inline int cgroup_is_removed(const struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -0700125{
Paul Menagebd89aab2007-10-18 23:40:44 -0700126 return test_bit(CGRP_REMOVED, &cgrp->flags);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700127}
128
129/* bits in struct cgroupfs_root flags field */
130enum {
131 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
132};
133
Adrian Bunke9685a02008-02-07 00:13:46 -0800134static int cgroup_is_releasable(const struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -0700135{
136 const int bits =
Paul Menagebd89aab2007-10-18 23:40:44 -0700137 (1 << CGRP_RELEASABLE) |
138 (1 << CGRP_NOTIFY_ON_RELEASE);
139 return (cgrp->flags & bits) == bits;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700140}
141
Adrian Bunke9685a02008-02-07 00:13:46 -0800142static int notify_on_release(const struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -0700143{
Paul Menagebd89aab2007-10-18 23:40:44 -0700144 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700145}
146
Paul Menageddbcc7e2007-10-18 23:39:30 -0700147/*
148 * for_each_subsys() allows you to iterate on each subsystem attached to
149 * an active hierarchy
150 */
151#define for_each_subsys(_root, _ss) \
152list_for_each_entry(_ss, &_root->subsys_list, sibling)
153
154/* for_each_root() allows you to iterate across the active hierarchies */
155#define for_each_root(_root) \
156list_for_each_entry(_root, &roots, root_list)
157
Paul Menage81a6a5c2007-10-18 23:39:38 -0700158/* the list of cgroups eligible for automatic release. Protected by
159 * release_list_lock */
160static LIST_HEAD(release_list);
161static DEFINE_SPINLOCK(release_list_lock);
162static void cgroup_release_agent(struct work_struct *work);
163static DECLARE_WORK(release_agent_work, cgroup_release_agent);
Paul Menagebd89aab2007-10-18 23:40:44 -0700164static void check_for_release(struct cgroup *cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700165
Paul Menage817929e2007-10-18 23:39:36 -0700166/* Link structure for associating css_set objects with cgroups */
167struct cg_cgroup_link {
168 /*
169 * List running through cg_cgroup_links associated with a
170 * cgroup, anchored on cgroup->css_sets
171 */
Paul Menagebd89aab2007-10-18 23:40:44 -0700172 struct list_head cgrp_link_list;
Paul Menage817929e2007-10-18 23:39:36 -0700173 /*
174 * List running through cg_cgroup_links pointing at a
175 * single css_set object, anchored on css_set->cg_links
176 */
177 struct list_head cg_link_list;
178 struct css_set *cg;
179};
180
181/* The default css_set - used by init and its children prior to any
182 * hierarchies being mounted. It contains a pointer to the root state
183 * for each subsystem. Also used to anchor the list of css_sets. Not
184 * reference-counted, to improve performance when child cgroups
185 * haven't been created.
186 */
187
188static struct css_set init_css_set;
189static struct cg_cgroup_link init_css_set_link;
190
191/* css_set_lock protects the list of css_set objects, and the
192 * chain of tasks off each css_set. Nests outside task->alloc_lock
193 * due to cgroup_iter_start() */
194static DEFINE_RWLOCK(css_set_lock);
195static int css_set_count;
196
Li Zefan472b1052008-04-29 01:00:11 -0700197/* hash table for cgroup groups. This improves the performance to
198 * find an existing css_set */
199#define CSS_SET_HASH_BITS 7
200#define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
201static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
202
203static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
204{
205 int i;
206 int index;
207 unsigned long tmp = 0UL;
208
209 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
210 tmp += (unsigned long)css[i];
211 tmp = (tmp >> 16) ^ tmp;
212
213 index = hash_long(tmp, CSS_SET_HASH_BITS);
214
215 return &css_set_table[index];
216}
217
Paul Menage817929e2007-10-18 23:39:36 -0700218/* We don't maintain the lists running through each css_set to its
219 * task until after the first call to cgroup_iter_start(). This
220 * reduces the fork()/exit() overhead for people who have cgroups
221 * compiled into their kernel but not actually in use */
222static int use_task_css_set_links;
223
224/* When we create or destroy a css_set, the operation simply
225 * takes/releases a reference count on all the cgroups referenced
226 * by subsystems in this css_set. This can end up multiple-counting
227 * some cgroups, but that's OK - the ref-count is just a
228 * busy/not-busy indicator; ensuring that we only count each cgroup
229 * once would require taking a global lock to ensure that no
Paul Menageb4f48b62007-10-18 23:39:33 -0700230 * subsystems moved between hierarchies while we were doing so.
231 *
232 * Possible TODO: decide at boot time based on the number of
233 * registered subsystems and the number of CPUs or NUMA nodes whether
234 * it's better for performance to ref-count every subsystem, or to
235 * take a global lock and only add one ref count to each hierarchy.
236 */
Paul Menageb4f48b62007-10-18 23:39:33 -0700237
Paul Menage817929e2007-10-18 23:39:36 -0700238/*
239 * unlink a css_set from the list and free it
240 */
Paul Menage81a6a5c2007-10-18 23:39:38 -0700241static void unlink_css_set(struct css_set *cg)
Paul Menageb4f48b62007-10-18 23:39:33 -0700242{
Paul Menage817929e2007-10-18 23:39:36 -0700243 write_lock(&css_set_lock);
Li Zefan472b1052008-04-29 01:00:11 -0700244 hlist_del(&cg->hlist);
Paul Menage817929e2007-10-18 23:39:36 -0700245 css_set_count--;
246 while (!list_empty(&cg->cg_links)) {
247 struct cg_cgroup_link *link;
248 link = list_entry(cg->cg_links.next,
249 struct cg_cgroup_link, cg_link_list);
250 list_del(&link->cg_link_list);
Paul Menagebd89aab2007-10-18 23:40:44 -0700251 list_del(&link->cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -0700252 kfree(link);
253 }
254 write_unlock(&css_set_lock);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700255}
256
257static void __release_css_set(struct kref *k, int taskexit)
258{
259 int i;
260 struct css_set *cg = container_of(k, struct css_set, ref);
261
262 unlink_css_set(cg);
263
264 rcu_read_lock();
265 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
Paul Menagebd89aab2007-10-18 23:40:44 -0700266 struct cgroup *cgrp = cg->subsys[i]->cgroup;
267 if (atomic_dec_and_test(&cgrp->count) &&
268 notify_on_release(cgrp)) {
Paul Menage81a6a5c2007-10-18 23:39:38 -0700269 if (taskexit)
Paul Menagebd89aab2007-10-18 23:40:44 -0700270 set_bit(CGRP_RELEASABLE, &cgrp->flags);
271 check_for_release(cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700272 }
273 }
274 rcu_read_unlock();
Paul Menage817929e2007-10-18 23:39:36 -0700275 kfree(cg);
276}
277
Paul Menage81a6a5c2007-10-18 23:39:38 -0700278static void release_css_set(struct kref *k)
279{
280 __release_css_set(k, 0);
281}
282
283static void release_css_set_taskexit(struct kref *k)
284{
285 __release_css_set(k, 1);
286}
287
Paul Menage817929e2007-10-18 23:39:36 -0700288/*
289 * refcounted get/put for css_set objects
290 */
291static inline void get_css_set(struct css_set *cg)
292{
293 kref_get(&cg->ref);
294}
295
296static inline void put_css_set(struct css_set *cg)
297{
298 kref_put(&cg->ref, release_css_set);
299}
300
Paul Menage81a6a5c2007-10-18 23:39:38 -0700301static inline void put_css_set_taskexit(struct css_set *cg)
302{
303 kref_put(&cg->ref, release_css_set_taskexit);
304}
305
Paul Menage817929e2007-10-18 23:39:36 -0700306/*
307 * find_existing_css_set() is a helper for
308 * find_css_set(), and checks to see whether an existing
Li Zefan472b1052008-04-29 01:00:11 -0700309 * css_set is suitable.
Paul Menage817929e2007-10-18 23:39:36 -0700310 *
311 * oldcg: the cgroup group that we're using before the cgroup
312 * transition
313 *
Paul Menagebd89aab2007-10-18 23:40:44 -0700314 * cgrp: the cgroup that we're moving into
Paul Menage817929e2007-10-18 23:39:36 -0700315 *
316 * template: location in which to build the desired set of subsystem
317 * state objects for the new cgroup group
318 */
Paul Menage817929e2007-10-18 23:39:36 -0700319static struct css_set *find_existing_css_set(
320 struct css_set *oldcg,
Paul Menagebd89aab2007-10-18 23:40:44 -0700321 struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -0700322 struct cgroup_subsys_state *template[])
323{
324 int i;
Paul Menagebd89aab2007-10-18 23:40:44 -0700325 struct cgroupfs_root *root = cgrp->root;
Li Zefan472b1052008-04-29 01:00:11 -0700326 struct hlist_head *hhead;
327 struct hlist_node *node;
328 struct css_set *cg;
Paul Menage817929e2007-10-18 23:39:36 -0700329
330 /* Built the set of subsystem state objects that we want to
331 * see in the new css_set */
332 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
Li Zefan8d53d552008-02-23 15:24:11 -0800333 if (root->subsys_bits & (1UL << i)) {
Paul Menage817929e2007-10-18 23:39:36 -0700334 /* Subsystem is in this hierarchy. So we want
335 * the subsystem state from the new
336 * cgroup */
Paul Menagebd89aab2007-10-18 23:40:44 -0700337 template[i] = cgrp->subsys[i];
Paul Menage817929e2007-10-18 23:39:36 -0700338 } else {
339 /* Subsystem is not in this hierarchy, so we
340 * don't want to change the subsystem state */
341 template[i] = oldcg->subsys[i];
342 }
343 }
344
Li Zefan472b1052008-04-29 01:00:11 -0700345 hhead = css_set_hash(template);
346 hlist_for_each_entry(cg, node, hhead, hlist) {
Paul Menage817929e2007-10-18 23:39:36 -0700347 if (!memcmp(template, cg->subsys, sizeof(cg->subsys))) {
348 /* All subsystems matched */
349 return cg;
350 }
Li Zefan472b1052008-04-29 01:00:11 -0700351 }
Paul Menage817929e2007-10-18 23:39:36 -0700352
353 /* No existing cgroup group matched */
354 return NULL;
355}
356
357/*
358 * allocate_cg_links() allocates "count" cg_cgroup_link structures
Paul Menagebd89aab2007-10-18 23:40:44 -0700359 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
Paul Menage817929e2007-10-18 23:39:36 -0700360 * success or a negative error
361 */
Paul Menage817929e2007-10-18 23:39:36 -0700362static int allocate_cg_links(int count, struct list_head *tmp)
363{
364 struct cg_cgroup_link *link;
365 int i;
366 INIT_LIST_HEAD(tmp);
367 for (i = 0; i < count; i++) {
368 link = kmalloc(sizeof(*link), GFP_KERNEL);
369 if (!link) {
370 while (!list_empty(tmp)) {
371 link = list_entry(tmp->next,
372 struct cg_cgroup_link,
Paul Menagebd89aab2007-10-18 23:40:44 -0700373 cgrp_link_list);
374 list_del(&link->cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -0700375 kfree(link);
376 }
377 return -ENOMEM;
378 }
Paul Menagebd89aab2007-10-18 23:40:44 -0700379 list_add(&link->cgrp_link_list, tmp);
Paul Menage817929e2007-10-18 23:39:36 -0700380 }
381 return 0;
382}
383
384static void free_cg_links(struct list_head *tmp)
385{
386 while (!list_empty(tmp)) {
387 struct cg_cgroup_link *link;
388 link = list_entry(tmp->next,
389 struct cg_cgroup_link,
Paul Menagebd89aab2007-10-18 23:40:44 -0700390 cgrp_link_list);
391 list_del(&link->cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -0700392 kfree(link);
393 }
394}
395
396/*
397 * find_css_set() takes an existing cgroup group and a
398 * cgroup object, and returns a css_set object that's
399 * equivalent to the old group, but with the given cgroup
400 * substituted into the appropriate hierarchy. Must be called with
401 * cgroup_mutex held
402 */
Paul Menage817929e2007-10-18 23:39:36 -0700403static struct css_set *find_css_set(
Paul Menagebd89aab2007-10-18 23:40:44 -0700404 struct css_set *oldcg, struct cgroup *cgrp)
Paul Menage817929e2007-10-18 23:39:36 -0700405{
406 struct css_set *res;
407 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
408 int i;
409
410 struct list_head tmp_cg_links;
411 struct cg_cgroup_link *link;
412
Li Zefan472b1052008-04-29 01:00:11 -0700413 struct hlist_head *hhead;
414
Paul Menage817929e2007-10-18 23:39:36 -0700415 /* First see if we already have a cgroup group that matches
416 * the desired set */
417 write_lock(&css_set_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -0700418 res = find_existing_css_set(oldcg, cgrp, template);
Paul Menage817929e2007-10-18 23:39:36 -0700419 if (res)
420 get_css_set(res);
421 write_unlock(&css_set_lock);
422
423 if (res)
424 return res;
425
426 res = kmalloc(sizeof(*res), GFP_KERNEL);
427 if (!res)
428 return NULL;
429
430 /* Allocate all the cg_cgroup_link objects that we'll need */
431 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
432 kfree(res);
433 return NULL;
434 }
435
436 kref_init(&res->ref);
437 INIT_LIST_HEAD(&res->cg_links);
438 INIT_LIST_HEAD(&res->tasks);
Li Zefan472b1052008-04-29 01:00:11 -0700439 INIT_HLIST_NODE(&res->hlist);
Paul Menage817929e2007-10-18 23:39:36 -0700440
441 /* Copy the set of subsystem state objects generated in
442 * find_existing_css_set() */
443 memcpy(res->subsys, template, sizeof(res->subsys));
444
445 write_lock(&css_set_lock);
446 /* Add reference counts and links from the new css_set. */
447 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
Paul Menagebd89aab2007-10-18 23:40:44 -0700448 struct cgroup *cgrp = res->subsys[i]->cgroup;
Paul Menage817929e2007-10-18 23:39:36 -0700449 struct cgroup_subsys *ss = subsys[i];
Paul Menagebd89aab2007-10-18 23:40:44 -0700450 atomic_inc(&cgrp->count);
Paul Menage817929e2007-10-18 23:39:36 -0700451 /*
452 * We want to add a link once per cgroup, so we
453 * only do it for the first subsystem in each
454 * hierarchy
455 */
456 if (ss->root->subsys_list.next == &ss->sibling) {
457 BUG_ON(list_empty(&tmp_cg_links));
458 link = list_entry(tmp_cg_links.next,
459 struct cg_cgroup_link,
Paul Menagebd89aab2007-10-18 23:40:44 -0700460 cgrp_link_list);
461 list_del(&link->cgrp_link_list);
462 list_add(&link->cgrp_link_list, &cgrp->css_sets);
Paul Menage817929e2007-10-18 23:39:36 -0700463 link->cg = res;
464 list_add(&link->cg_link_list, &res->cg_links);
465 }
466 }
467 if (list_empty(&rootnode.subsys_list)) {
468 link = list_entry(tmp_cg_links.next,
469 struct cg_cgroup_link,
Paul Menagebd89aab2007-10-18 23:40:44 -0700470 cgrp_link_list);
471 list_del(&link->cgrp_link_list);
472 list_add(&link->cgrp_link_list, &dummytop->css_sets);
Paul Menage817929e2007-10-18 23:39:36 -0700473 link->cg = res;
474 list_add(&link->cg_link_list, &res->cg_links);
475 }
476
477 BUG_ON(!list_empty(&tmp_cg_links));
478
Paul Menage817929e2007-10-18 23:39:36 -0700479 css_set_count++;
Li Zefan472b1052008-04-29 01:00:11 -0700480
481 /* Add this cgroup group to the hash table */
482 hhead = css_set_hash(res->subsys);
483 hlist_add_head(&res->hlist, hhead);
484
Paul Menage817929e2007-10-18 23:39:36 -0700485 write_unlock(&css_set_lock);
486
487 return res;
Paul Menageb4f48b62007-10-18 23:39:33 -0700488}
489
Paul Menageddbcc7e2007-10-18 23:39:30 -0700490/*
491 * There is one global cgroup mutex. We also require taking
492 * task_lock() when dereferencing a task's cgroup subsys pointers.
493 * See "The task_lock() exception", at the end of this comment.
494 *
495 * A task must hold cgroup_mutex to modify cgroups.
496 *
497 * Any task can increment and decrement the count field without lock.
498 * So in general, code holding cgroup_mutex can't rely on the count
499 * field not changing. However, if the count goes to zero, then only
Cliff Wickman956db3c2008-02-07 00:14:43 -0800500 * cgroup_attach_task() can increment it again. Because a count of zero
Paul Menageddbcc7e2007-10-18 23:39:30 -0700501 * means that no tasks are currently attached, therefore there is no
502 * way a task attached to that cgroup can fork (the other way to
503 * increment the count). So code holding cgroup_mutex can safely
504 * assume that if the count is zero, it will stay zero. Similarly, if
505 * a task holds cgroup_mutex on a cgroup with zero count, it
506 * knows that the cgroup won't be removed, as cgroup_rmdir()
507 * needs that mutex.
508 *
509 * The cgroup_common_file_write handler for operations that modify
510 * the cgroup hierarchy holds cgroup_mutex across the entire operation,
511 * single threading all such cgroup modifications across the system.
512 *
513 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
514 * (usually) take cgroup_mutex. These are the two most performance
515 * critical pieces of code here. The exception occurs on cgroup_exit(),
516 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
517 * is taken, and if the cgroup count is zero, a usermode call made
Li Zefana043e3b2008-02-23 15:24:09 -0800518 * to the release agent with the name of the cgroup (path relative to
519 * the root of cgroup file system) as the argument.
Paul Menageddbcc7e2007-10-18 23:39:30 -0700520 *
521 * A cgroup can only be deleted if both its 'count' of using tasks
522 * is zero, and its list of 'children' cgroups is empty. Since all
523 * tasks in the system use _some_ cgroup, and since there is always at
524 * least one task in the system (init, pid == 1), therefore, top_cgroup
525 * always has either children cgroups and/or using tasks. So we don't
526 * need a special hack to ensure that top_cgroup cannot be deleted.
527 *
528 * The task_lock() exception
529 *
530 * The need for this exception arises from the action of
Cliff Wickman956db3c2008-02-07 00:14:43 -0800531 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
Li Zefana043e3b2008-02-23 15:24:09 -0800532 * another. It does so using cgroup_mutex, however there are
Paul Menageddbcc7e2007-10-18 23:39:30 -0700533 * several performance critical places that need to reference
534 * task->cgroup without the expense of grabbing a system global
535 * mutex. Therefore except as noted below, when dereferencing or, as
Cliff Wickman956db3c2008-02-07 00:14:43 -0800536 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
Paul Menageddbcc7e2007-10-18 23:39:30 -0700537 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
538 * the task_struct routinely used for such matters.
539 *
540 * P.S. One more locking exception. RCU is used to guard the
Cliff Wickman956db3c2008-02-07 00:14:43 -0800541 * update of a tasks cgroup pointer by cgroup_attach_task()
Paul Menageddbcc7e2007-10-18 23:39:30 -0700542 */
543
Paul Menageddbcc7e2007-10-18 23:39:30 -0700544/**
545 * cgroup_lock - lock out any changes to cgroup structures
546 *
547 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700548void cgroup_lock(void)
549{
550 mutex_lock(&cgroup_mutex);
551}
552
553/**
554 * cgroup_unlock - release lock on cgroup changes
555 *
556 * Undo the lock taken in a previous cgroup_lock() call.
557 */
Paul Menageddbcc7e2007-10-18 23:39:30 -0700558void cgroup_unlock(void)
559{
560 mutex_unlock(&cgroup_mutex);
561}
562
563/*
564 * A couple of forward declarations required, due to cyclic reference loop:
565 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
566 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
567 * -> cgroup_mkdir.
568 */
569
570static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
571static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
Paul Menagebd89aab2007-10-18 23:40:44 -0700572static int cgroup_populate_dir(struct cgroup *cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700573static struct inode_operations cgroup_dir_inode_operations;
Paul Menagea4243162007-10-18 23:39:35 -0700574static struct file_operations proc_cgroupstats_operations;
575
576static struct backing_dev_info cgroup_backing_dev_info = {
577 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
578};
Paul Menageddbcc7e2007-10-18 23:39:30 -0700579
580static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
581{
582 struct inode *inode = new_inode(sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700583
584 if (inode) {
585 inode->i_mode = mode;
586 inode->i_uid = current->fsuid;
587 inode->i_gid = current->fsgid;
588 inode->i_blocks = 0;
589 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
590 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
591 }
592 return inode;
593}
594
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -0800595/*
596 * Call subsys's pre_destroy handler.
597 * This is called before css refcnt check.
598 */
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -0800599static void cgroup_call_pre_destroy(struct cgroup *cgrp)
600{
601 struct cgroup_subsys *ss;
602 for_each_subsys(cgrp->root, ss)
603 if (ss->pre_destroy && cgrp->subsys[ss->subsys_id])
604 ss->pre_destroy(ss, cgrp);
605 return;
606}
607
Paul Menageddbcc7e2007-10-18 23:39:30 -0700608static void cgroup_diput(struct dentry *dentry, struct inode *inode)
609{
610 /* is dentry a directory ? if so, kfree() associated cgroup */
611 if (S_ISDIR(inode->i_mode)) {
Paul Menagebd89aab2007-10-18 23:40:44 -0700612 struct cgroup *cgrp = dentry->d_fsdata;
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800613 struct cgroup_subsys *ss;
Paul Menagebd89aab2007-10-18 23:40:44 -0700614 BUG_ON(!(cgroup_is_removed(cgrp)));
Paul Menage81a6a5c2007-10-18 23:39:38 -0700615 /* It's possible for external users to be holding css
616 * reference counts on a cgroup; css_put() needs to
617 * be able to access the cgroup after decrementing
618 * the reference count in order to know if it needs to
619 * queue the cgroup to be handled by the release
620 * agent */
621 synchronize_rcu();
Paul Menage8dc4f3e2008-02-07 00:13:45 -0800622
623 mutex_lock(&cgroup_mutex);
624 /*
625 * Release the subsystem state objects.
626 */
627 for_each_subsys(cgrp->root, ss) {
628 if (cgrp->subsys[ss->subsys_id])
629 ss->destroy(ss, cgrp);
630 }
631
632 cgrp->root->number_of_cgroups--;
633 mutex_unlock(&cgroup_mutex);
634
635 /* Drop the active superblock reference that we took when we
636 * created the cgroup */
637 deactivate_super(cgrp->root->sb);
638
Paul Menagebd89aab2007-10-18 23:40:44 -0700639 kfree(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700640 }
641 iput(inode);
642}
643
644static void remove_dir(struct dentry *d)
645{
646 struct dentry *parent = dget(d->d_parent);
647
648 d_delete(d);
649 simple_rmdir(parent->d_inode, d);
650 dput(parent);
651}
652
653static void cgroup_clear_directory(struct dentry *dentry)
654{
655 struct list_head *node;
656
657 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
658 spin_lock(&dcache_lock);
659 node = dentry->d_subdirs.next;
660 while (node != &dentry->d_subdirs) {
661 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
662 list_del_init(node);
663 if (d->d_inode) {
664 /* This should never be called on a cgroup
665 * directory with child cgroups */
666 BUG_ON(d->d_inode->i_mode & S_IFDIR);
667 d = dget_locked(d);
668 spin_unlock(&dcache_lock);
669 d_delete(d);
670 simple_unlink(dentry->d_inode, d);
671 dput(d);
672 spin_lock(&dcache_lock);
673 }
674 node = dentry->d_subdirs.next;
675 }
676 spin_unlock(&dcache_lock);
677}
678
679/*
680 * NOTE : the dentry must have been dget()'ed
681 */
682static void cgroup_d_remove_dir(struct dentry *dentry)
683{
684 cgroup_clear_directory(dentry);
685
686 spin_lock(&dcache_lock);
687 list_del_init(&dentry->d_u.d_child);
688 spin_unlock(&dcache_lock);
689 remove_dir(dentry);
690}
691
692static int rebind_subsystems(struct cgroupfs_root *root,
693 unsigned long final_bits)
694{
695 unsigned long added_bits, removed_bits;
Paul Menagebd89aab2007-10-18 23:40:44 -0700696 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700697 int i;
698
699 removed_bits = root->actual_subsys_bits & ~final_bits;
700 added_bits = final_bits & ~root->actual_subsys_bits;
701 /* Check that any added subsystems are currently free */
702 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
Li Zefan8d53d552008-02-23 15:24:11 -0800703 unsigned long bit = 1UL << i;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700704 struct cgroup_subsys *ss = subsys[i];
705 if (!(bit & added_bits))
706 continue;
707 if (ss->root != &rootnode) {
708 /* Subsystem isn't free */
709 return -EBUSY;
710 }
711 }
712
713 /* Currently we don't handle adding/removing subsystems when
714 * any child cgroups exist. This is theoretically supportable
715 * but involves complex error handling, so it's being left until
716 * later */
Paul Menagebd89aab2007-10-18 23:40:44 -0700717 if (!list_empty(&cgrp->children))
Paul Menageddbcc7e2007-10-18 23:39:30 -0700718 return -EBUSY;
719
720 /* Process each subsystem */
721 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
722 struct cgroup_subsys *ss = subsys[i];
723 unsigned long bit = 1UL << i;
724 if (bit & added_bits) {
725 /* We're binding this subsystem to this hierarchy */
Paul Menagebd89aab2007-10-18 23:40:44 -0700726 BUG_ON(cgrp->subsys[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700727 BUG_ON(!dummytop->subsys[i]);
728 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
Paul Menagebd89aab2007-10-18 23:40:44 -0700729 cgrp->subsys[i] = dummytop->subsys[i];
730 cgrp->subsys[i]->cgroup = cgrp;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700731 list_add(&ss->sibling, &root->subsys_list);
732 rcu_assign_pointer(ss->root, root);
733 if (ss->bind)
Paul Menagebd89aab2007-10-18 23:40:44 -0700734 ss->bind(ss, cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700735
736 } else if (bit & removed_bits) {
737 /* We're removing this subsystem */
Paul Menagebd89aab2007-10-18 23:40:44 -0700738 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
739 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700740 if (ss->bind)
741 ss->bind(ss, dummytop);
742 dummytop->subsys[i]->cgroup = dummytop;
Paul Menagebd89aab2007-10-18 23:40:44 -0700743 cgrp->subsys[i] = NULL;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700744 rcu_assign_pointer(subsys[i]->root, &rootnode);
745 list_del(&ss->sibling);
746 } else if (bit & final_bits) {
747 /* Subsystem state should already exist */
Paul Menagebd89aab2007-10-18 23:40:44 -0700748 BUG_ON(!cgrp->subsys[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700749 } else {
750 /* Subsystem state shouldn't exist */
Paul Menagebd89aab2007-10-18 23:40:44 -0700751 BUG_ON(cgrp->subsys[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700752 }
753 }
754 root->subsys_bits = root->actual_subsys_bits = final_bits;
755 synchronize_rcu();
756
757 return 0;
758}
759
760static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
761{
762 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
763 struct cgroup_subsys *ss;
764
765 mutex_lock(&cgroup_mutex);
766 for_each_subsys(root, ss)
767 seq_printf(seq, ",%s", ss->name);
768 if (test_bit(ROOT_NOPREFIX, &root->flags))
769 seq_puts(seq, ",noprefix");
Paul Menage81a6a5c2007-10-18 23:39:38 -0700770 if (strlen(root->release_agent_path))
771 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700772 mutex_unlock(&cgroup_mutex);
773 return 0;
774}
775
776struct cgroup_sb_opts {
777 unsigned long subsys_bits;
778 unsigned long flags;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700779 char *release_agent;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700780};
781
782/* Convert a hierarchy specifier into a bitmask of subsystems and
783 * flags. */
784static int parse_cgroupfs_options(char *data,
785 struct cgroup_sb_opts *opts)
786{
787 char *token, *o = data ?: "all";
788
789 opts->subsys_bits = 0;
790 opts->flags = 0;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700791 opts->release_agent = NULL;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700792
793 while ((token = strsep(&o, ",")) != NULL) {
794 if (!*token)
795 return -EINVAL;
796 if (!strcmp(token, "all")) {
Paul Menage8bab8dd2008-04-04 14:29:57 -0700797 /* Add all non-disabled subsystems */
798 int i;
799 opts->subsys_bits = 0;
800 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
801 struct cgroup_subsys *ss = subsys[i];
802 if (!ss->disabled)
803 opts->subsys_bits |= 1ul << i;
804 }
Paul Menageddbcc7e2007-10-18 23:39:30 -0700805 } else if (!strcmp(token, "noprefix")) {
806 set_bit(ROOT_NOPREFIX, &opts->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700807 } else if (!strncmp(token, "release_agent=", 14)) {
808 /* Specifying two release agents is forbidden */
809 if (opts->release_agent)
810 return -EINVAL;
811 opts->release_agent = kzalloc(PATH_MAX, GFP_KERNEL);
812 if (!opts->release_agent)
813 return -ENOMEM;
814 strncpy(opts->release_agent, token + 14, PATH_MAX - 1);
815 opts->release_agent[PATH_MAX - 1] = 0;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700816 } else {
817 struct cgroup_subsys *ss;
818 int i;
819 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
820 ss = subsys[i];
821 if (!strcmp(token, ss->name)) {
Paul Menage8bab8dd2008-04-04 14:29:57 -0700822 if (!ss->disabled)
823 set_bit(i, &opts->subsys_bits);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700824 break;
825 }
826 }
827 if (i == CGROUP_SUBSYS_COUNT)
828 return -ENOENT;
829 }
830 }
831
832 /* We can't have an empty hierarchy */
833 if (!opts->subsys_bits)
834 return -EINVAL;
835
836 return 0;
837}
838
839static int cgroup_remount(struct super_block *sb, int *flags, char *data)
840{
841 int ret = 0;
842 struct cgroupfs_root *root = sb->s_fs_info;
Paul Menagebd89aab2007-10-18 23:40:44 -0700843 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700844 struct cgroup_sb_opts opts;
845
Paul Menagebd89aab2007-10-18 23:40:44 -0700846 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700847 mutex_lock(&cgroup_mutex);
848
849 /* See what subsystems are wanted */
850 ret = parse_cgroupfs_options(data, &opts);
851 if (ret)
852 goto out_unlock;
853
854 /* Don't allow flags to change at remount */
855 if (opts.flags != root->flags) {
856 ret = -EINVAL;
857 goto out_unlock;
858 }
859
860 ret = rebind_subsystems(root, opts.subsys_bits);
861
862 /* (re)populate subsystem files */
863 if (!ret)
Paul Menagebd89aab2007-10-18 23:40:44 -0700864 cgroup_populate_dir(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700865
Paul Menage81a6a5c2007-10-18 23:39:38 -0700866 if (opts.release_agent)
867 strcpy(root->release_agent_path, opts.release_agent);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700868 out_unlock:
Paul Menage81a6a5c2007-10-18 23:39:38 -0700869 if (opts.release_agent)
870 kfree(opts.release_agent);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700871 mutex_unlock(&cgroup_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -0700872 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700873 return ret;
874}
875
876static struct super_operations cgroup_ops = {
877 .statfs = simple_statfs,
878 .drop_inode = generic_delete_inode,
879 .show_options = cgroup_show_options,
880 .remount_fs = cgroup_remount,
881};
882
883static void init_cgroup_root(struct cgroupfs_root *root)
884{
Paul Menagebd89aab2007-10-18 23:40:44 -0700885 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -0700886 INIT_LIST_HEAD(&root->subsys_list);
887 INIT_LIST_HEAD(&root->root_list);
888 root->number_of_cgroups = 1;
Paul Menagebd89aab2007-10-18 23:40:44 -0700889 cgrp->root = root;
890 cgrp->top_cgroup = cgrp;
891 INIT_LIST_HEAD(&cgrp->sibling);
892 INIT_LIST_HEAD(&cgrp->children);
893 INIT_LIST_HEAD(&cgrp->css_sets);
894 INIT_LIST_HEAD(&cgrp->release_list);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700895}
896
897static int cgroup_test_super(struct super_block *sb, void *data)
898{
899 struct cgroupfs_root *new = data;
900 struct cgroupfs_root *root = sb->s_fs_info;
901
902 /* First check subsystems */
903 if (new->subsys_bits != root->subsys_bits)
904 return 0;
905
906 /* Next check flags */
907 if (new->flags != root->flags)
908 return 0;
909
910 return 1;
911}
912
913static int cgroup_set_super(struct super_block *sb, void *data)
914{
915 int ret;
916 struct cgroupfs_root *root = data;
917
918 ret = set_anon_super(sb, NULL);
919 if (ret)
920 return ret;
921
922 sb->s_fs_info = root;
923 root->sb = sb;
924
925 sb->s_blocksize = PAGE_CACHE_SIZE;
926 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
927 sb->s_magic = CGROUP_SUPER_MAGIC;
928 sb->s_op = &cgroup_ops;
929
930 return 0;
931}
932
933static int cgroup_get_rootdir(struct super_block *sb)
934{
935 struct inode *inode =
936 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
937 struct dentry *dentry;
938
939 if (!inode)
940 return -ENOMEM;
941
Paul Menageddbcc7e2007-10-18 23:39:30 -0700942 inode->i_fop = &simple_dir_operations;
943 inode->i_op = &cgroup_dir_inode_operations;
944 /* directories start off with i_nlink == 2 (for "." entry) */
945 inc_nlink(inode);
946 dentry = d_alloc_root(inode);
947 if (!dentry) {
948 iput(inode);
949 return -ENOMEM;
950 }
951 sb->s_root = dentry;
952 return 0;
953}
954
955static int cgroup_get_sb(struct file_system_type *fs_type,
956 int flags, const char *unused_dev_name,
957 void *data, struct vfsmount *mnt)
958{
959 struct cgroup_sb_opts opts;
960 int ret = 0;
961 struct super_block *sb;
962 struct cgroupfs_root *root;
Li Zefan28fd5df2008-04-29 01:00:13 -0700963 struct list_head tmp_cg_links;
Paul Menage817929e2007-10-18 23:39:36 -0700964 INIT_LIST_HEAD(&tmp_cg_links);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700965
966 /* First find the desired set of subsystems */
967 ret = parse_cgroupfs_options(data, &opts);
Paul Menage81a6a5c2007-10-18 23:39:38 -0700968 if (ret) {
969 if (opts.release_agent)
970 kfree(opts.release_agent);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700971 return ret;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700972 }
Paul Menageddbcc7e2007-10-18 23:39:30 -0700973
974 root = kzalloc(sizeof(*root), GFP_KERNEL);
Li Zefanf7770732008-02-23 15:24:10 -0800975 if (!root) {
976 if (opts.release_agent)
977 kfree(opts.release_agent);
Paul Menageddbcc7e2007-10-18 23:39:30 -0700978 return -ENOMEM;
Li Zefanf7770732008-02-23 15:24:10 -0800979 }
Paul Menageddbcc7e2007-10-18 23:39:30 -0700980
981 init_cgroup_root(root);
982 root->subsys_bits = opts.subsys_bits;
983 root->flags = opts.flags;
Paul Menage81a6a5c2007-10-18 23:39:38 -0700984 if (opts.release_agent) {
985 strcpy(root->release_agent_path, opts.release_agent);
986 kfree(opts.release_agent);
987 }
Paul Menageddbcc7e2007-10-18 23:39:30 -0700988
989 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, root);
990
991 if (IS_ERR(sb)) {
992 kfree(root);
993 return PTR_ERR(sb);
994 }
995
996 if (sb->s_fs_info != root) {
997 /* Reusing an existing superblock */
998 BUG_ON(sb->s_root == NULL);
999 kfree(root);
1000 root = NULL;
1001 } else {
1002 /* New superblock */
Paul Menagebd89aab2007-10-18 23:40:44 -07001003 struct cgroup *cgrp = &root->top_cgroup;
Paul Menage817929e2007-10-18 23:39:36 -07001004 struct inode *inode;
Li Zefan28fd5df2008-04-29 01:00:13 -07001005 int i;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001006
1007 BUG_ON(sb->s_root != NULL);
1008
1009 ret = cgroup_get_rootdir(sb);
1010 if (ret)
1011 goto drop_new_super;
Paul Menage817929e2007-10-18 23:39:36 -07001012 inode = sb->s_root->d_inode;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001013
Paul Menage817929e2007-10-18 23:39:36 -07001014 mutex_lock(&inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001015 mutex_lock(&cgroup_mutex);
1016
Paul Menage817929e2007-10-18 23:39:36 -07001017 /*
1018 * We're accessing css_set_count without locking
1019 * css_set_lock here, but that's OK - it can only be
1020 * increased by someone holding cgroup_lock, and
1021 * that's us. The worst that can happen is that we
1022 * have some link structures left over
1023 */
1024 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1025 if (ret) {
1026 mutex_unlock(&cgroup_mutex);
1027 mutex_unlock(&inode->i_mutex);
1028 goto drop_new_super;
1029 }
1030
Paul Menageddbcc7e2007-10-18 23:39:30 -07001031 ret = rebind_subsystems(root, root->subsys_bits);
1032 if (ret == -EBUSY) {
1033 mutex_unlock(&cgroup_mutex);
Paul Menage817929e2007-10-18 23:39:36 -07001034 mutex_unlock(&inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001035 goto drop_new_super;
1036 }
1037
1038 /* EBUSY should be the only error here */
1039 BUG_ON(ret);
1040
1041 list_add(&root->root_list, &roots);
Paul Menage817929e2007-10-18 23:39:36 -07001042 root_count++;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001043
1044 sb->s_root->d_fsdata = &root->top_cgroup;
1045 root->top_cgroup.dentry = sb->s_root;
1046
Paul Menage817929e2007-10-18 23:39:36 -07001047 /* Link the top cgroup in this hierarchy into all
1048 * the css_set objects */
1049 write_lock(&css_set_lock);
Li Zefan28fd5df2008-04-29 01:00:13 -07001050 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1051 struct hlist_head *hhead = &css_set_table[i];
1052 struct hlist_node *node;
Paul Menage817929e2007-10-18 23:39:36 -07001053 struct css_set *cg;
Li Zefan28fd5df2008-04-29 01:00:13 -07001054
1055 hlist_for_each_entry(cg, node, hhead, hlist) {
1056 struct cg_cgroup_link *link;
1057
1058 BUG_ON(list_empty(&tmp_cg_links));
1059 link = list_entry(tmp_cg_links.next,
1060 struct cg_cgroup_link,
1061 cgrp_link_list);
1062 list_del(&link->cgrp_link_list);
1063 link->cg = cg;
1064 list_add(&link->cgrp_link_list,
1065 &root->top_cgroup.css_sets);
1066 list_add(&link->cg_link_list, &cg->cg_links);
1067 }
1068 }
Paul Menage817929e2007-10-18 23:39:36 -07001069 write_unlock(&css_set_lock);
1070
1071 free_cg_links(&tmp_cg_links);
1072
Paul Menagebd89aab2007-10-18 23:40:44 -07001073 BUG_ON(!list_empty(&cgrp->sibling));
1074 BUG_ON(!list_empty(&cgrp->children));
Paul Menageddbcc7e2007-10-18 23:39:30 -07001075 BUG_ON(root->number_of_cgroups != 1);
1076
Paul Menagebd89aab2007-10-18 23:40:44 -07001077 cgroup_populate_dir(cgrp);
Paul Menage817929e2007-10-18 23:39:36 -07001078 mutex_unlock(&inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001079 mutex_unlock(&cgroup_mutex);
1080 }
1081
1082 return simple_set_mnt(mnt, sb);
1083
1084 drop_new_super:
1085 up_write(&sb->s_umount);
1086 deactivate_super(sb);
Paul Menage817929e2007-10-18 23:39:36 -07001087 free_cg_links(&tmp_cg_links);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001088 return ret;
1089}
1090
1091static void cgroup_kill_sb(struct super_block *sb) {
1092 struct cgroupfs_root *root = sb->s_fs_info;
Paul Menagebd89aab2007-10-18 23:40:44 -07001093 struct cgroup *cgrp = &root->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001094 int ret;
1095
1096 BUG_ON(!root);
1097
1098 BUG_ON(root->number_of_cgroups != 1);
Paul Menagebd89aab2007-10-18 23:40:44 -07001099 BUG_ON(!list_empty(&cgrp->children));
1100 BUG_ON(!list_empty(&cgrp->sibling));
Paul Menageddbcc7e2007-10-18 23:39:30 -07001101
1102 mutex_lock(&cgroup_mutex);
1103
1104 /* Rebind all subsystems back to the default hierarchy */
1105 ret = rebind_subsystems(root, 0);
1106 /* Shouldn't be able to fail ... */
1107 BUG_ON(ret);
1108
Paul Menage817929e2007-10-18 23:39:36 -07001109 /*
1110 * Release all the links from css_sets to this hierarchy's
1111 * root cgroup
1112 */
1113 write_lock(&css_set_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07001114 while (!list_empty(&cgrp->css_sets)) {
Paul Menage817929e2007-10-18 23:39:36 -07001115 struct cg_cgroup_link *link;
Paul Menagebd89aab2007-10-18 23:40:44 -07001116 link = list_entry(cgrp->css_sets.next,
1117 struct cg_cgroup_link, cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -07001118 list_del(&link->cg_link_list);
Paul Menagebd89aab2007-10-18 23:40:44 -07001119 list_del(&link->cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -07001120 kfree(link);
1121 }
1122 write_unlock(&css_set_lock);
1123
1124 if (!list_empty(&root->root_list)) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07001125 list_del(&root->root_list);
Paul Menage817929e2007-10-18 23:39:36 -07001126 root_count--;
1127 }
Paul Menageddbcc7e2007-10-18 23:39:30 -07001128 mutex_unlock(&cgroup_mutex);
1129
1130 kfree(root);
1131 kill_litter_super(sb);
1132}
1133
1134static struct file_system_type cgroup_fs_type = {
1135 .name = "cgroup",
1136 .get_sb = cgroup_get_sb,
1137 .kill_sb = cgroup_kill_sb,
1138};
1139
Paul Menagebd89aab2007-10-18 23:40:44 -07001140static inline struct cgroup *__d_cgrp(struct dentry *dentry)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001141{
1142 return dentry->d_fsdata;
1143}
1144
1145static inline struct cftype *__d_cft(struct dentry *dentry)
1146{
1147 return dentry->d_fsdata;
1148}
1149
Li Zefana043e3b2008-02-23 15:24:09 -08001150/**
1151 * cgroup_path - generate the path of a cgroup
1152 * @cgrp: the cgroup in question
1153 * @buf: the buffer to write the path into
1154 * @buflen: the length of the buffer
1155 *
1156 * Called with cgroup_mutex held. Writes path of cgroup into buf.
Paul Menageddbcc7e2007-10-18 23:39:30 -07001157 * Returns 0 on success, -errno on error.
1158 */
Paul Menagebd89aab2007-10-18 23:40:44 -07001159int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001160{
1161 char *start;
1162
Paul Menagebd89aab2007-10-18 23:40:44 -07001163 if (cgrp == dummytop) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07001164 /*
1165 * Inactive subsystems have no dentry for their root
1166 * cgroup
1167 */
1168 strcpy(buf, "/");
1169 return 0;
1170 }
1171
1172 start = buf + buflen;
1173
1174 *--start = '\0';
1175 for (;;) {
Paul Menagebd89aab2007-10-18 23:40:44 -07001176 int len = cgrp->dentry->d_name.len;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001177 if ((start -= len) < buf)
1178 return -ENAMETOOLONG;
Paul Menagebd89aab2007-10-18 23:40:44 -07001179 memcpy(start, cgrp->dentry->d_name.name, len);
1180 cgrp = cgrp->parent;
1181 if (!cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001182 break;
Paul Menagebd89aab2007-10-18 23:40:44 -07001183 if (!cgrp->parent)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001184 continue;
1185 if (--start < buf)
1186 return -ENAMETOOLONG;
1187 *start = '/';
1188 }
1189 memmove(buf, start, buf + buflen - start);
1190 return 0;
1191}
1192
Paul Menagebbcb81d2007-10-18 23:39:32 -07001193/*
1194 * Return the first subsystem attached to a cgroup's hierarchy, and
1195 * its subsystem id.
1196 */
1197
Paul Menagebd89aab2007-10-18 23:40:44 -07001198static void get_first_subsys(const struct cgroup *cgrp,
Paul Menagebbcb81d2007-10-18 23:39:32 -07001199 struct cgroup_subsys_state **css, int *subsys_id)
1200{
Paul Menagebd89aab2007-10-18 23:40:44 -07001201 const struct cgroupfs_root *root = cgrp->root;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001202 const struct cgroup_subsys *test_ss;
1203 BUG_ON(list_empty(&root->subsys_list));
1204 test_ss = list_entry(root->subsys_list.next,
1205 struct cgroup_subsys, sibling);
1206 if (css) {
Paul Menagebd89aab2007-10-18 23:40:44 -07001207 *css = cgrp->subsys[test_ss->subsys_id];
Paul Menagebbcb81d2007-10-18 23:39:32 -07001208 BUG_ON(!*css);
1209 }
1210 if (subsys_id)
1211 *subsys_id = test_ss->subsys_id;
1212}
1213
Li Zefana043e3b2008-02-23 15:24:09 -08001214/**
1215 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1216 * @cgrp: the cgroup the task is attaching to
1217 * @tsk: the task to be attached
Paul Menagebbcb81d2007-10-18 23:39:32 -07001218 *
Li Zefana043e3b2008-02-23 15:24:09 -08001219 * Call holding cgroup_mutex. May take task_lock of
1220 * the task 'tsk' during call.
Paul Menagebbcb81d2007-10-18 23:39:32 -07001221 */
Cliff Wickman956db3c2008-02-07 00:14:43 -08001222int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
Paul Menagebbcb81d2007-10-18 23:39:32 -07001223{
1224 int retval = 0;
1225 struct cgroup_subsys *ss;
Paul Menagebd89aab2007-10-18 23:40:44 -07001226 struct cgroup *oldcgrp;
Paul Menage817929e2007-10-18 23:39:36 -07001227 struct css_set *cg = tsk->cgroups;
1228 struct css_set *newcg;
Paul Menagebd89aab2007-10-18 23:40:44 -07001229 struct cgroupfs_root *root = cgrp->root;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001230 int subsys_id;
1231
Paul Menagebd89aab2007-10-18 23:40:44 -07001232 get_first_subsys(cgrp, NULL, &subsys_id);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001233
1234 /* Nothing to do if the task is already in that cgroup */
Paul Menagebd89aab2007-10-18 23:40:44 -07001235 oldcgrp = task_cgroup(tsk, subsys_id);
1236 if (cgrp == oldcgrp)
Paul Menagebbcb81d2007-10-18 23:39:32 -07001237 return 0;
1238
1239 for_each_subsys(root, ss) {
1240 if (ss->can_attach) {
Paul Menagebd89aab2007-10-18 23:40:44 -07001241 retval = ss->can_attach(ss, cgrp, tsk);
Paul Jacksone18f6312008-02-07 00:13:44 -08001242 if (retval)
Paul Menagebbcb81d2007-10-18 23:39:32 -07001243 return retval;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001244 }
1245 }
1246
Paul Menage817929e2007-10-18 23:39:36 -07001247 /*
1248 * Locate or allocate a new css_set for this task,
1249 * based on its final set of cgroups
1250 */
Paul Menagebd89aab2007-10-18 23:40:44 -07001251 newcg = find_css_set(cg, cgrp);
Paul Jacksone18f6312008-02-07 00:13:44 -08001252 if (!newcg)
Paul Menage817929e2007-10-18 23:39:36 -07001253 return -ENOMEM;
Paul Menage817929e2007-10-18 23:39:36 -07001254
Paul Menagebbcb81d2007-10-18 23:39:32 -07001255 task_lock(tsk);
1256 if (tsk->flags & PF_EXITING) {
1257 task_unlock(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07001258 put_css_set(newcg);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001259 return -ESRCH;
1260 }
Paul Menage817929e2007-10-18 23:39:36 -07001261 rcu_assign_pointer(tsk->cgroups, newcg);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001262 task_unlock(tsk);
1263
Paul Menage817929e2007-10-18 23:39:36 -07001264 /* Update the css_set linked lists if we're using them */
1265 write_lock(&css_set_lock);
1266 if (!list_empty(&tsk->cg_list)) {
1267 list_del(&tsk->cg_list);
1268 list_add(&tsk->cg_list, &newcg->tasks);
1269 }
1270 write_unlock(&css_set_lock);
1271
Paul Menagebbcb81d2007-10-18 23:39:32 -07001272 for_each_subsys(root, ss) {
Paul Jacksone18f6312008-02-07 00:13:44 -08001273 if (ss->attach)
Paul Menagebd89aab2007-10-18 23:40:44 -07001274 ss->attach(ss, cgrp, oldcgrp, tsk);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001275 }
Paul Menagebd89aab2007-10-18 23:40:44 -07001276 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001277 synchronize_rcu();
Paul Menage817929e2007-10-18 23:39:36 -07001278 put_css_set(cg);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001279 return 0;
1280}
1281
1282/*
Paul Menagebd89aab2007-10-18 23:40:44 -07001283 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with
Paul Menagebbcb81d2007-10-18 23:39:32 -07001284 * cgroup_mutex, may take task_lock of task
1285 */
Paul Menagebd89aab2007-10-18 23:40:44 -07001286static int attach_task_by_pid(struct cgroup *cgrp, char *pidbuf)
Paul Menagebbcb81d2007-10-18 23:39:32 -07001287{
1288 pid_t pid;
1289 struct task_struct *tsk;
1290 int ret;
1291
1292 if (sscanf(pidbuf, "%d", &pid) != 1)
1293 return -EIO;
1294
1295 if (pid) {
1296 rcu_read_lock();
Pavel Emelyanov73507f32008-02-07 00:14:47 -08001297 tsk = find_task_by_vpid(pid);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001298 if (!tsk || tsk->flags & PF_EXITING) {
1299 rcu_read_unlock();
1300 return -ESRCH;
1301 }
1302 get_task_struct(tsk);
1303 rcu_read_unlock();
1304
1305 if ((current->euid) && (current->euid != tsk->uid)
1306 && (current->euid != tsk->suid)) {
1307 put_task_struct(tsk);
1308 return -EACCES;
1309 }
1310 } else {
1311 tsk = current;
1312 get_task_struct(tsk);
1313 }
1314
Cliff Wickman956db3c2008-02-07 00:14:43 -08001315 ret = cgroup_attach_task(cgrp, tsk);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001316 put_task_struct(tsk);
1317 return ret;
1318}
1319
Paul Menageddbcc7e2007-10-18 23:39:30 -07001320/* The various types of files and directories in a cgroup file system */
Paul Menageddbcc7e2007-10-18 23:39:30 -07001321enum cgroup_filetype {
1322 FILE_ROOT,
1323 FILE_DIR,
1324 FILE_TASKLIST,
Paul Menage81a6a5c2007-10-18 23:39:38 -07001325 FILE_NOTIFY_ON_RELEASE,
Paul Menage81a6a5c2007-10-18 23:39:38 -07001326 FILE_RELEASE_AGENT,
Paul Menageddbcc7e2007-10-18 23:39:30 -07001327};
1328
Paul Menagee73d2c62008-04-29 01:00:06 -07001329static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
Paul Menagef4c753b2008-04-29 00:59:56 -07001330 struct file *file,
1331 const char __user *userbuf,
1332 size_t nbytes, loff_t *unused_ppos)
Paul Menage355e0c42007-10-18 23:39:33 -07001333{
1334 char buffer[64];
1335 int retval = 0;
Paul Menage355e0c42007-10-18 23:39:33 -07001336 char *end;
1337
1338 if (!nbytes)
1339 return -EINVAL;
1340 if (nbytes >= sizeof(buffer))
1341 return -E2BIG;
1342 if (copy_from_user(buffer, userbuf, nbytes))
1343 return -EFAULT;
1344
1345 buffer[nbytes] = 0; /* nul-terminate */
Paul Menageb7269df2008-04-29 00:59:59 -07001346 strstrip(buffer);
Paul Menagee73d2c62008-04-29 01:00:06 -07001347 if (cft->write_u64) {
1348 u64 val = simple_strtoull(buffer, &end, 0);
1349 if (*end)
1350 return -EINVAL;
1351 retval = cft->write_u64(cgrp, cft, val);
1352 } else {
1353 s64 val = simple_strtoll(buffer, &end, 0);
1354 if (*end)
1355 return -EINVAL;
1356 retval = cft->write_s64(cgrp, cft, val);
1357 }
Paul Menage355e0c42007-10-18 23:39:33 -07001358 if (!retval)
1359 retval = nbytes;
1360 return retval;
1361}
1362
Paul Menagebd89aab2007-10-18 23:40:44 -07001363static ssize_t cgroup_common_file_write(struct cgroup *cgrp,
Paul Menagebbcb81d2007-10-18 23:39:32 -07001364 struct cftype *cft,
1365 struct file *file,
1366 const char __user *userbuf,
1367 size_t nbytes, loff_t *unused_ppos)
1368{
1369 enum cgroup_filetype type = cft->private;
1370 char *buffer;
1371 int retval = 0;
1372
1373 if (nbytes >= PATH_MAX)
1374 return -E2BIG;
1375
1376 /* +1 for nul-terminator */
1377 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1378 if (buffer == NULL)
1379 return -ENOMEM;
1380
1381 if (copy_from_user(buffer, userbuf, nbytes)) {
1382 retval = -EFAULT;
1383 goto out1;
1384 }
1385 buffer[nbytes] = 0; /* nul-terminate */
Paul Jackson622d42c2008-02-07 00:13:44 -08001386 strstrip(buffer); /* strip -just- trailing whitespace */
Paul Menagebbcb81d2007-10-18 23:39:32 -07001387
1388 mutex_lock(&cgroup_mutex);
1389
Paul Menage8dc4f3e2008-02-07 00:13:45 -08001390 /*
1391 * This was already checked for in cgroup_file_write(), but
1392 * check again now we're holding cgroup_mutex.
1393 */
Paul Menagebd89aab2007-10-18 23:40:44 -07001394 if (cgroup_is_removed(cgrp)) {
Paul Menagebbcb81d2007-10-18 23:39:32 -07001395 retval = -ENODEV;
1396 goto out2;
1397 }
1398
1399 switch (type) {
1400 case FILE_TASKLIST:
Paul Menagebd89aab2007-10-18 23:40:44 -07001401 retval = attach_task_by_pid(cgrp, buffer);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001402 break;
Paul Menage81a6a5c2007-10-18 23:39:38 -07001403 case FILE_NOTIFY_ON_RELEASE:
Paul Menagebd89aab2007-10-18 23:40:44 -07001404 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001405 if (simple_strtoul(buffer, NULL, 10) != 0)
Paul Menagebd89aab2007-10-18 23:40:44 -07001406 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001407 else
Paul Menagebd89aab2007-10-18 23:40:44 -07001408 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001409 break;
1410 case FILE_RELEASE_AGENT:
Paul Jackson622d42c2008-02-07 00:13:44 -08001411 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1412 strcpy(cgrp->root->release_agent_path, buffer);
Paul Menage81a6a5c2007-10-18 23:39:38 -07001413 break;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001414 default:
1415 retval = -EINVAL;
1416 goto out2;
1417 }
1418
1419 if (retval == 0)
1420 retval = nbytes;
1421out2:
1422 mutex_unlock(&cgroup_mutex);
1423out1:
1424 kfree(buffer);
1425 return retval;
1426}
1427
Paul Menageddbcc7e2007-10-18 23:39:30 -07001428static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
1429 size_t nbytes, loff_t *ppos)
1430{
1431 struct cftype *cft = __d_cft(file->f_dentry);
Paul Menagebd89aab2007-10-18 23:40:44 -07001432 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001433
Paul Menage8dc4f3e2008-02-07 00:13:45 -08001434 if (!cft || cgroup_is_removed(cgrp))
Paul Menageddbcc7e2007-10-18 23:39:30 -07001435 return -ENODEV;
Paul Menage355e0c42007-10-18 23:39:33 -07001436 if (cft->write)
Paul Menagebd89aab2007-10-18 23:40:44 -07001437 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
Paul Menagee73d2c62008-04-29 01:00:06 -07001438 if (cft->write_u64 || cft->write_s64)
1439 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
Pavel Emelyanovd447ea22008-04-29 01:00:08 -07001440 if (cft->trigger) {
1441 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
1442 return ret ? ret : nbytes;
1443 }
Paul Menage355e0c42007-10-18 23:39:33 -07001444 return -EINVAL;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001445}
1446
Paul Menagef4c753b2008-04-29 00:59:56 -07001447static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
1448 struct file *file,
1449 char __user *buf, size_t nbytes,
1450 loff_t *ppos)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001451{
1452 char tmp[64];
Paul Menagef4c753b2008-04-29 00:59:56 -07001453 u64 val = cft->read_u64(cgrp, cft);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001454 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
1455
1456 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1457}
1458
Paul Menagee73d2c62008-04-29 01:00:06 -07001459static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
1460 struct file *file,
1461 char __user *buf, size_t nbytes,
1462 loff_t *ppos)
1463{
1464 char tmp[64];
1465 s64 val = cft->read_s64(cgrp, cft);
1466 int len = sprintf(tmp, "%lld\n", (long long) val);
1467
1468 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1469}
1470
Paul Menagebd89aab2007-10-18 23:40:44 -07001471static ssize_t cgroup_common_file_read(struct cgroup *cgrp,
Paul Menage81a6a5c2007-10-18 23:39:38 -07001472 struct cftype *cft,
1473 struct file *file,
1474 char __user *buf,
1475 size_t nbytes, loff_t *ppos)
1476{
1477 enum cgroup_filetype type = cft->private;
1478 char *page;
1479 ssize_t retval = 0;
1480 char *s;
1481
1482 if (!(page = (char *)__get_free_page(GFP_KERNEL)))
1483 return -ENOMEM;
1484
1485 s = page;
1486
1487 switch (type) {
1488 case FILE_RELEASE_AGENT:
1489 {
1490 struct cgroupfs_root *root;
1491 size_t n;
1492 mutex_lock(&cgroup_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -07001493 root = cgrp->root;
Paul Menage81a6a5c2007-10-18 23:39:38 -07001494 n = strnlen(root->release_agent_path,
1495 sizeof(root->release_agent_path));
1496 n = min(n, (size_t) PAGE_SIZE);
1497 strncpy(s, root->release_agent_path, n);
1498 mutex_unlock(&cgroup_mutex);
1499 s += n;
1500 break;
1501 }
1502 default:
1503 retval = -EINVAL;
1504 goto out;
1505 }
1506 *s++ = '\n';
1507
1508 retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1509out:
1510 free_page((unsigned long)page);
1511 return retval;
1512}
1513
Paul Menageddbcc7e2007-10-18 23:39:30 -07001514static ssize_t cgroup_file_read(struct file *file, char __user *buf,
1515 size_t nbytes, loff_t *ppos)
1516{
1517 struct cftype *cft = __d_cft(file->f_dentry);
Paul Menagebd89aab2007-10-18 23:40:44 -07001518 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001519
Paul Menage8dc4f3e2008-02-07 00:13:45 -08001520 if (!cft || cgroup_is_removed(cgrp))
Paul Menageddbcc7e2007-10-18 23:39:30 -07001521 return -ENODEV;
1522
1523 if (cft->read)
Paul Menagebd89aab2007-10-18 23:40:44 -07001524 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
Paul Menagef4c753b2008-04-29 00:59:56 -07001525 if (cft->read_u64)
1526 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
Paul Menagee73d2c62008-04-29 01:00:06 -07001527 if (cft->read_s64)
1528 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001529 return -EINVAL;
1530}
1531
Paul Menage91796562008-04-29 01:00:01 -07001532/*
1533 * seqfile ops/methods for returning structured data. Currently just
1534 * supports string->u64 maps, but can be extended in future.
1535 */
1536
1537struct cgroup_seqfile_state {
1538 struct cftype *cft;
1539 struct cgroup *cgroup;
1540};
1541
1542static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
1543{
1544 struct seq_file *sf = cb->state;
1545 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
1546}
1547
1548static int cgroup_seqfile_show(struct seq_file *m, void *arg)
1549{
1550 struct cgroup_seqfile_state *state = m->private;
1551 struct cftype *cft = state->cft;
1552 struct cgroup_map_cb cb = {
1553 .fill = cgroup_map_add,
1554 .state = m,
1555 };
1556 return cft->read_map(state->cgroup, cft, &cb);
1557}
1558
1559int cgroup_seqfile_release(struct inode *inode, struct file *file)
1560{
1561 struct seq_file *seq = file->private_data;
1562 kfree(seq->private);
1563 return single_release(inode, file);
1564}
1565
1566static struct file_operations cgroup_seqfile_operations = {
1567 .read = seq_read,
1568 .llseek = seq_lseek,
1569 .release = cgroup_seqfile_release,
1570};
1571
Paul Menageddbcc7e2007-10-18 23:39:30 -07001572static int cgroup_file_open(struct inode *inode, struct file *file)
1573{
1574 int err;
1575 struct cftype *cft;
1576
1577 err = generic_file_open(inode, file);
1578 if (err)
1579 return err;
1580
1581 cft = __d_cft(file->f_dentry);
1582 if (!cft)
1583 return -ENODEV;
Paul Menage91796562008-04-29 01:00:01 -07001584 if (cft->read_map) {
1585 struct cgroup_seqfile_state *state =
1586 kzalloc(sizeof(*state), GFP_USER);
1587 if (!state)
1588 return -ENOMEM;
1589 state->cft = cft;
1590 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
1591 file->f_op = &cgroup_seqfile_operations;
1592 err = single_open(file, cgroup_seqfile_show, state);
1593 if (err < 0)
1594 kfree(state);
1595 } else if (cft->open)
Paul Menageddbcc7e2007-10-18 23:39:30 -07001596 err = cft->open(inode, file);
1597 else
1598 err = 0;
1599
1600 return err;
1601}
1602
1603static int cgroup_file_release(struct inode *inode, struct file *file)
1604{
1605 struct cftype *cft = __d_cft(file->f_dentry);
1606 if (cft->release)
1607 return cft->release(inode, file);
1608 return 0;
1609}
1610
1611/*
1612 * cgroup_rename - Only allow simple rename of directories in place.
1613 */
1614static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
1615 struct inode *new_dir, struct dentry *new_dentry)
1616{
1617 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1618 return -ENOTDIR;
1619 if (new_dentry->d_inode)
1620 return -EEXIST;
1621 if (old_dir != new_dir)
1622 return -EIO;
1623 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
1624}
1625
1626static struct file_operations cgroup_file_operations = {
1627 .read = cgroup_file_read,
1628 .write = cgroup_file_write,
1629 .llseek = generic_file_llseek,
1630 .open = cgroup_file_open,
1631 .release = cgroup_file_release,
1632};
1633
1634static struct inode_operations cgroup_dir_inode_operations = {
1635 .lookup = simple_lookup,
1636 .mkdir = cgroup_mkdir,
1637 .rmdir = cgroup_rmdir,
1638 .rename = cgroup_rename,
1639};
1640
1641static int cgroup_create_file(struct dentry *dentry, int mode,
1642 struct super_block *sb)
1643{
1644 static struct dentry_operations cgroup_dops = {
1645 .d_iput = cgroup_diput,
1646 };
1647
1648 struct inode *inode;
1649
1650 if (!dentry)
1651 return -ENOENT;
1652 if (dentry->d_inode)
1653 return -EEXIST;
1654
1655 inode = cgroup_new_inode(mode, sb);
1656 if (!inode)
1657 return -ENOMEM;
1658
1659 if (S_ISDIR(mode)) {
1660 inode->i_op = &cgroup_dir_inode_operations;
1661 inode->i_fop = &simple_dir_operations;
1662
1663 /* start off with i_nlink == 2 (for "." entry) */
1664 inc_nlink(inode);
1665
1666 /* start with the directory inode held, so that we can
1667 * populate it without racing with another mkdir */
Paul Menage817929e2007-10-18 23:39:36 -07001668 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001669 } else if (S_ISREG(mode)) {
1670 inode->i_size = 0;
1671 inode->i_fop = &cgroup_file_operations;
1672 }
1673 dentry->d_op = &cgroup_dops;
1674 d_instantiate(dentry, inode);
1675 dget(dentry); /* Extra count - pin the dentry in core */
1676 return 0;
1677}
1678
1679/*
Li Zefana043e3b2008-02-23 15:24:09 -08001680 * cgroup_create_dir - create a directory for an object.
1681 * @cgrp: the cgroup we create the directory for. It must have a valid
1682 * ->parent field. And we are going to fill its ->dentry field.
1683 * @dentry: dentry of the new cgroup
1684 * @mode: mode to set on new directory.
Paul Menageddbcc7e2007-10-18 23:39:30 -07001685 */
Paul Menagebd89aab2007-10-18 23:40:44 -07001686static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
Paul Menageddbcc7e2007-10-18 23:39:30 -07001687 int mode)
1688{
1689 struct dentry *parent;
1690 int error = 0;
1691
Paul Menagebd89aab2007-10-18 23:40:44 -07001692 parent = cgrp->parent->dentry;
1693 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001694 if (!error) {
Paul Menagebd89aab2007-10-18 23:40:44 -07001695 dentry->d_fsdata = cgrp;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001696 inc_nlink(parent->d_inode);
Paul Menagebd89aab2007-10-18 23:40:44 -07001697 cgrp->dentry = dentry;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001698 dget(dentry);
1699 }
1700 dput(dentry);
1701
1702 return error;
1703}
1704
Paul Menagebd89aab2007-10-18 23:40:44 -07001705int cgroup_add_file(struct cgroup *cgrp,
Paul Menageddbcc7e2007-10-18 23:39:30 -07001706 struct cgroup_subsys *subsys,
1707 const struct cftype *cft)
1708{
Paul Menagebd89aab2007-10-18 23:40:44 -07001709 struct dentry *dir = cgrp->dentry;
Paul Menageddbcc7e2007-10-18 23:39:30 -07001710 struct dentry *dentry;
1711 int error;
1712
1713 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
Paul Menagebd89aab2007-10-18 23:40:44 -07001714 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07001715 strcpy(name, subsys->name);
1716 strcat(name, ".");
1717 }
1718 strcat(name, cft->name);
1719 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
1720 dentry = lookup_one_len(name, dir, strlen(name));
1721 if (!IS_ERR(dentry)) {
1722 error = cgroup_create_file(dentry, 0644 | S_IFREG,
Paul Menagebd89aab2007-10-18 23:40:44 -07001723 cgrp->root->sb);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001724 if (!error)
1725 dentry->d_fsdata = (void *)cft;
1726 dput(dentry);
1727 } else
1728 error = PTR_ERR(dentry);
1729 return error;
1730}
1731
Paul Menagebd89aab2007-10-18 23:40:44 -07001732int cgroup_add_files(struct cgroup *cgrp,
Paul Menageddbcc7e2007-10-18 23:39:30 -07001733 struct cgroup_subsys *subsys,
1734 const struct cftype cft[],
1735 int count)
1736{
1737 int i, err;
1738 for (i = 0; i < count; i++) {
Paul Menagebd89aab2007-10-18 23:40:44 -07001739 err = cgroup_add_file(cgrp, subsys, &cft[i]);
Paul Menageddbcc7e2007-10-18 23:39:30 -07001740 if (err)
1741 return err;
1742 }
1743 return 0;
1744}
1745
Li Zefana043e3b2008-02-23 15:24:09 -08001746/**
1747 * cgroup_task_count - count the number of tasks in a cgroup.
1748 * @cgrp: the cgroup in question
1749 *
1750 * Return the number of tasks in the cgroup.
1751 */
Paul Menagebd89aab2007-10-18 23:40:44 -07001752int cgroup_task_count(const struct cgroup *cgrp)
Paul Menagebbcb81d2007-10-18 23:39:32 -07001753{
1754 int count = 0;
Paul Menage817929e2007-10-18 23:39:36 -07001755 struct list_head *l;
Paul Menagebbcb81d2007-10-18 23:39:32 -07001756
Paul Menage817929e2007-10-18 23:39:36 -07001757 read_lock(&css_set_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07001758 l = cgrp->css_sets.next;
1759 while (l != &cgrp->css_sets) {
Paul Menage817929e2007-10-18 23:39:36 -07001760 struct cg_cgroup_link *link =
Paul Menagebd89aab2007-10-18 23:40:44 -07001761 list_entry(l, struct cg_cgroup_link, cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -07001762 count += atomic_read(&link->cg->ref.refcount);
1763 l = l->next;
1764 }
1765 read_unlock(&css_set_lock);
Paul Menagebbcb81d2007-10-18 23:39:32 -07001766 return count;
1767}
1768
1769/*
Paul Menage817929e2007-10-18 23:39:36 -07001770 * Advance a list_head iterator. The iterator should be positioned at
1771 * the start of a css_set
1772 */
Paul Menagebd89aab2007-10-18 23:40:44 -07001773static void cgroup_advance_iter(struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -07001774 struct cgroup_iter *it)
1775{
1776 struct list_head *l = it->cg_link;
1777 struct cg_cgroup_link *link;
1778 struct css_set *cg;
1779
1780 /* Advance to the next non-empty css_set */
1781 do {
1782 l = l->next;
Paul Menagebd89aab2007-10-18 23:40:44 -07001783 if (l == &cgrp->css_sets) {
Paul Menage817929e2007-10-18 23:39:36 -07001784 it->cg_link = NULL;
1785 return;
1786 }
Paul Menagebd89aab2007-10-18 23:40:44 -07001787 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
Paul Menage817929e2007-10-18 23:39:36 -07001788 cg = link->cg;
1789 } while (list_empty(&cg->tasks));
1790 it->cg_link = l;
1791 it->task = cg->tasks.next;
1792}
1793
Cliff Wickman31a7df02008-02-07 00:14:42 -08001794/*
1795 * To reduce the fork() overhead for systems that are not actually
1796 * using their cgroups capability, we don't maintain the lists running
1797 * through each css_set to its tasks until we see the list actually
1798 * used - in other words after the first call to cgroup_iter_start().
1799 *
1800 * The tasklist_lock is not held here, as do_each_thread() and
1801 * while_each_thread() are protected by RCU.
1802 */
Adrian Bunk3df91fe2008-04-29 00:59:54 -07001803static void cgroup_enable_task_cg_lists(void)
Cliff Wickman31a7df02008-02-07 00:14:42 -08001804{
1805 struct task_struct *p, *g;
1806 write_lock(&css_set_lock);
1807 use_task_css_set_links = 1;
1808 do_each_thread(g, p) {
1809 task_lock(p);
Li Zefan0e043882008-04-17 11:37:15 +08001810 /*
1811 * We should check if the process is exiting, otherwise
1812 * it will race with cgroup_exit() in that the list
1813 * entry won't be deleted though the process has exited.
1814 */
1815 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
Cliff Wickman31a7df02008-02-07 00:14:42 -08001816 list_add(&p->cg_list, &p->cgroups->tasks);
1817 task_unlock(p);
1818 } while_each_thread(g, p);
1819 write_unlock(&css_set_lock);
1820}
1821
Paul Menagebd89aab2007-10-18 23:40:44 -07001822void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07001823{
1824 /*
1825 * The first time anyone tries to iterate across a cgroup,
1826 * we need to enable the list linking each css_set to its
1827 * tasks, and fix up all existing tasks.
1828 */
Cliff Wickman31a7df02008-02-07 00:14:42 -08001829 if (!use_task_css_set_links)
1830 cgroup_enable_task_cg_lists();
1831
Paul Menage817929e2007-10-18 23:39:36 -07001832 read_lock(&css_set_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07001833 it->cg_link = &cgrp->css_sets;
1834 cgroup_advance_iter(cgrp, it);
Paul Menage817929e2007-10-18 23:39:36 -07001835}
1836
Paul Menagebd89aab2007-10-18 23:40:44 -07001837struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
Paul Menage817929e2007-10-18 23:39:36 -07001838 struct cgroup_iter *it)
1839{
1840 struct task_struct *res;
1841 struct list_head *l = it->task;
1842
1843 /* If the iterator cg is NULL, we have no tasks */
1844 if (!it->cg_link)
1845 return NULL;
1846 res = list_entry(l, struct task_struct, cg_list);
1847 /* Advance iterator to find next entry */
1848 l = l->next;
1849 if (l == &res->cgroups->tasks) {
1850 /* We reached the end of this task list - move on to
1851 * the next cg_cgroup_link */
Paul Menagebd89aab2007-10-18 23:40:44 -07001852 cgroup_advance_iter(cgrp, it);
Paul Menage817929e2007-10-18 23:39:36 -07001853 } else {
1854 it->task = l;
1855 }
1856 return res;
1857}
1858
Paul Menagebd89aab2007-10-18 23:40:44 -07001859void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
Paul Menage817929e2007-10-18 23:39:36 -07001860{
1861 read_unlock(&css_set_lock);
1862}
1863
Cliff Wickman31a7df02008-02-07 00:14:42 -08001864static inline int started_after_time(struct task_struct *t1,
1865 struct timespec *time,
1866 struct task_struct *t2)
1867{
1868 int start_diff = timespec_compare(&t1->start_time, time);
1869 if (start_diff > 0) {
1870 return 1;
1871 } else if (start_diff < 0) {
1872 return 0;
1873 } else {
1874 /*
1875 * Arbitrarily, if two processes started at the same
1876 * time, we'll say that the lower pointer value
1877 * started first. Note that t2 may have exited by now
1878 * so this may not be a valid pointer any longer, but
1879 * that's fine - it still serves to distinguish
1880 * between two tasks started (effectively) simultaneously.
1881 */
1882 return t1 > t2;
1883 }
1884}
1885
1886/*
1887 * This function is a callback from heap_insert() and is used to order
1888 * the heap.
1889 * In this case we order the heap in descending task start time.
1890 */
1891static inline int started_after(void *p1, void *p2)
1892{
1893 struct task_struct *t1 = p1;
1894 struct task_struct *t2 = p2;
1895 return started_after_time(t1, &t2->start_time, t2);
1896}
1897
1898/**
1899 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
1900 * @scan: struct cgroup_scanner containing arguments for the scan
1901 *
1902 * Arguments include pointers to callback functions test_task() and
1903 * process_task().
1904 * Iterate through all the tasks in a cgroup, calling test_task() for each,
1905 * and if it returns true, call process_task() for it also.
1906 * The test_task pointer may be NULL, meaning always true (select all tasks).
1907 * Effectively duplicates cgroup_iter_{start,next,end}()
1908 * but does not lock css_set_lock for the call to process_task().
1909 * The struct cgroup_scanner may be embedded in any structure of the caller's
1910 * creation.
1911 * It is guaranteed that process_task() will act on every task that
1912 * is a member of the cgroup for the duration of this call. This
1913 * function may or may not call process_task() for tasks that exit
1914 * or move to a different cgroup during the call, or are forked or
1915 * move into the cgroup during the call.
1916 *
1917 * Note that test_task() may be called with locks held, and may in some
1918 * situations be called multiple times for the same task, so it should
1919 * be cheap.
1920 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
1921 * pre-allocated and will be used for heap operations (and its "gt" member will
1922 * be overwritten), else a temporary heap will be used (allocation of which
1923 * may cause this function to fail).
1924 */
1925int cgroup_scan_tasks(struct cgroup_scanner *scan)
1926{
1927 int retval, i;
1928 struct cgroup_iter it;
1929 struct task_struct *p, *dropped;
1930 /* Never dereference latest_task, since it's not refcounted */
1931 struct task_struct *latest_task = NULL;
1932 struct ptr_heap tmp_heap;
1933 struct ptr_heap *heap;
1934 struct timespec latest_time = { 0, 0 };
1935
1936 if (scan->heap) {
1937 /* The caller supplied our heap and pre-allocated its memory */
1938 heap = scan->heap;
1939 heap->gt = &started_after;
1940 } else {
1941 /* We need to allocate our own heap memory */
1942 heap = &tmp_heap;
1943 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
1944 if (retval)
1945 /* cannot allocate the heap */
1946 return retval;
1947 }
1948
1949 again:
1950 /*
1951 * Scan tasks in the cgroup, using the scanner's "test_task" callback
1952 * to determine which are of interest, and using the scanner's
1953 * "process_task" callback to process any of them that need an update.
1954 * Since we don't want to hold any locks during the task updates,
1955 * gather tasks to be processed in a heap structure.
1956 * The heap is sorted by descending task start time.
1957 * If the statically-sized heap fills up, we overflow tasks that
1958 * started later, and in future iterations only consider tasks that
1959 * started after the latest task in the previous pass. This
1960 * guarantees forward progress and that we don't miss any tasks.
1961 */
1962 heap->size = 0;
1963 cgroup_iter_start(scan->cg, &it);
1964 while ((p = cgroup_iter_next(scan->cg, &it))) {
1965 /*
1966 * Only affect tasks that qualify per the caller's callback,
1967 * if he provided one
1968 */
1969 if (scan->test_task && !scan->test_task(p, scan))
1970 continue;
1971 /*
1972 * Only process tasks that started after the last task
1973 * we processed
1974 */
1975 if (!started_after_time(p, &latest_time, latest_task))
1976 continue;
1977 dropped = heap_insert(heap, p);
1978 if (dropped == NULL) {
1979 /*
1980 * The new task was inserted; the heap wasn't
1981 * previously full
1982 */
1983 get_task_struct(p);
1984 } else if (dropped != p) {
1985 /*
1986 * The new task was inserted, and pushed out a
1987 * different task
1988 */
1989 get_task_struct(p);
1990 put_task_struct(dropped);
1991 }
1992 /*
1993 * Else the new task was newer than anything already in
1994 * the heap and wasn't inserted
1995 */
1996 }
1997 cgroup_iter_end(scan->cg, &it);
1998
1999 if (heap->size) {
2000 for (i = 0; i < heap->size; i++) {
Paul Jackson4fe91d52008-04-29 00:59:55 -07002001 struct task_struct *q = heap->ptrs[i];
Cliff Wickman31a7df02008-02-07 00:14:42 -08002002 if (i == 0) {
Paul Jackson4fe91d52008-04-29 00:59:55 -07002003 latest_time = q->start_time;
2004 latest_task = q;
Cliff Wickman31a7df02008-02-07 00:14:42 -08002005 }
2006 /* Process the task per the caller's callback */
Paul Jackson4fe91d52008-04-29 00:59:55 -07002007 scan->process_task(q, scan);
2008 put_task_struct(q);
Cliff Wickman31a7df02008-02-07 00:14:42 -08002009 }
2010 /*
2011 * If we had to process any tasks at all, scan again
2012 * in case some of them were in the middle of forking
2013 * children that didn't get processed.
2014 * Not the most efficient way to do it, but it avoids
2015 * having to take callback_mutex in the fork path
2016 */
2017 goto again;
2018 }
2019 if (heap == &tmp_heap)
2020 heap_free(&tmp_heap);
2021 return 0;
2022}
2023
Paul Menage817929e2007-10-18 23:39:36 -07002024/*
Paul Menagebbcb81d2007-10-18 23:39:32 -07002025 * Stuff for reading the 'tasks' file.
2026 *
2027 * Reading this file can return large amounts of data if a cgroup has
2028 * *lots* of attached tasks. So it may need several calls to read(),
2029 * but we cannot guarantee that the information we produce is correct
2030 * unless we produce it entirely atomically.
2031 *
2032 * Upon tasks file open(), a struct ctr_struct is allocated, that
2033 * will have a pointer to an array (also allocated here). The struct
2034 * ctr_struct * is stored in file->private_data. Its resources will
2035 * be freed by release() when the file is closed. The array is used
2036 * to sprintf the PIDs and then used by read().
2037 */
2038struct ctr_struct {
2039 char *buf;
2040 int bufsz;
2041};
2042
2043/*
2044 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
Paul Menagebd89aab2007-10-18 23:40:44 -07002045 * 'cgrp'. Return actual number of pids loaded. No need to
Paul Menagebbcb81d2007-10-18 23:39:32 -07002046 * task_lock(p) when reading out p->cgroup, since we're in an RCU
2047 * read section, so the css_set can't go away, and is
2048 * immutable after creation.
2049 */
Paul Menagebd89aab2007-10-18 23:40:44 -07002050static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cgrp)
Paul Menagebbcb81d2007-10-18 23:39:32 -07002051{
2052 int n = 0;
Paul Menage817929e2007-10-18 23:39:36 -07002053 struct cgroup_iter it;
2054 struct task_struct *tsk;
Paul Menagebd89aab2007-10-18 23:40:44 -07002055 cgroup_iter_start(cgrp, &it);
2056 while ((tsk = cgroup_iter_next(cgrp, &it))) {
Paul Menage817929e2007-10-18 23:39:36 -07002057 if (unlikely(n == npids))
2058 break;
Pavel Emelyanov73507f32008-02-07 00:14:47 -08002059 pidarray[n++] = task_pid_vnr(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07002060 }
Paul Menagebd89aab2007-10-18 23:40:44 -07002061 cgroup_iter_end(cgrp, &it);
Paul Menagebbcb81d2007-10-18 23:39:32 -07002062 return n;
2063}
2064
Balbir Singh846c7bb2007-10-18 23:39:44 -07002065/**
Li Zefana043e3b2008-02-23 15:24:09 -08002066 * cgroupstats_build - build and fill cgroupstats
Balbir Singh846c7bb2007-10-18 23:39:44 -07002067 * @stats: cgroupstats to fill information into
2068 * @dentry: A dentry entry belonging to the cgroup for which stats have
2069 * been requested.
Li Zefana043e3b2008-02-23 15:24:09 -08002070 *
2071 * Build and fill cgroupstats so that taskstats can export it to user
2072 * space.
Balbir Singh846c7bb2007-10-18 23:39:44 -07002073 */
2074int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2075{
2076 int ret = -EINVAL;
Paul Menagebd89aab2007-10-18 23:40:44 -07002077 struct cgroup *cgrp;
Balbir Singh846c7bb2007-10-18 23:39:44 -07002078 struct cgroup_iter it;
2079 struct task_struct *tsk;
2080 /*
2081 * Validate dentry by checking the superblock operations
2082 */
2083 if (dentry->d_sb->s_op != &cgroup_ops)
2084 goto err;
2085
2086 ret = 0;
Paul Menagebd89aab2007-10-18 23:40:44 -07002087 cgrp = dentry->d_fsdata;
Balbir Singh846c7bb2007-10-18 23:39:44 -07002088 rcu_read_lock();
2089
Paul Menagebd89aab2007-10-18 23:40:44 -07002090 cgroup_iter_start(cgrp, &it);
2091 while ((tsk = cgroup_iter_next(cgrp, &it))) {
Balbir Singh846c7bb2007-10-18 23:39:44 -07002092 switch (tsk->state) {
2093 case TASK_RUNNING:
2094 stats->nr_running++;
2095 break;
2096 case TASK_INTERRUPTIBLE:
2097 stats->nr_sleeping++;
2098 break;
2099 case TASK_UNINTERRUPTIBLE:
2100 stats->nr_uninterruptible++;
2101 break;
2102 case TASK_STOPPED:
2103 stats->nr_stopped++;
2104 break;
2105 default:
2106 if (delayacct_is_task_waiting_on_io(tsk))
2107 stats->nr_io_wait++;
2108 break;
2109 }
2110 }
Paul Menagebd89aab2007-10-18 23:40:44 -07002111 cgroup_iter_end(cgrp, &it);
Balbir Singh846c7bb2007-10-18 23:39:44 -07002112
2113 rcu_read_unlock();
2114err:
2115 return ret;
2116}
2117
Paul Menagebbcb81d2007-10-18 23:39:32 -07002118static int cmppid(const void *a, const void *b)
2119{
2120 return *(pid_t *)a - *(pid_t *)b;
2121}
2122
2123/*
2124 * Convert array 'a' of 'npids' pid_t's to a string of newline separated
2125 * decimal pids in 'buf'. Don't write more than 'sz' chars, but return
2126 * count 'cnt' of how many chars would be written if buf were large enough.
2127 */
2128static int pid_array_to_buf(char *buf, int sz, pid_t *a, int npids)
2129{
2130 int cnt = 0;
2131 int i;
2132
2133 for (i = 0; i < npids; i++)
2134 cnt += snprintf(buf + cnt, max(sz - cnt, 0), "%d\n", a[i]);
2135 return cnt;
2136}
2137
2138/*
2139 * Handle an open on 'tasks' file. Prepare a buffer listing the
2140 * process id's of tasks currently attached to the cgroup being opened.
2141 *
2142 * Does not require any specific cgroup mutexes, and does not take any.
2143 */
2144static int cgroup_tasks_open(struct inode *unused, struct file *file)
2145{
Paul Menagebd89aab2007-10-18 23:40:44 -07002146 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
Paul Menagebbcb81d2007-10-18 23:39:32 -07002147 struct ctr_struct *ctr;
2148 pid_t *pidarray;
2149 int npids;
2150 char c;
2151
2152 if (!(file->f_mode & FMODE_READ))
2153 return 0;
2154
2155 ctr = kmalloc(sizeof(*ctr), GFP_KERNEL);
2156 if (!ctr)
2157 goto err0;
2158
2159 /*
2160 * If cgroup gets more users after we read count, we won't have
2161 * enough space - tough. This race is indistinguishable to the
2162 * caller from the case that the additional cgroup users didn't
2163 * show up until sometime later on.
2164 */
Paul Menagebd89aab2007-10-18 23:40:44 -07002165 npids = cgroup_task_count(cgrp);
Paul Menagebbcb81d2007-10-18 23:39:32 -07002166 if (npids) {
2167 pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL);
2168 if (!pidarray)
2169 goto err1;
2170
Paul Menagebd89aab2007-10-18 23:40:44 -07002171 npids = pid_array_load(pidarray, npids, cgrp);
Paul Menagebbcb81d2007-10-18 23:39:32 -07002172 sort(pidarray, npids, sizeof(pid_t), cmppid, NULL);
2173
2174 /* Call pid_array_to_buf() twice, first just to get bufsz */
2175 ctr->bufsz = pid_array_to_buf(&c, sizeof(c), pidarray, npids) + 1;
2176 ctr->buf = kmalloc(ctr->bufsz, GFP_KERNEL);
2177 if (!ctr->buf)
2178 goto err2;
2179 ctr->bufsz = pid_array_to_buf(ctr->buf, ctr->bufsz, pidarray, npids);
2180
2181 kfree(pidarray);
2182 } else {
Al Viro9dce07f2008-03-29 03:07:28 +00002183 ctr->buf = NULL;
Paul Menagebbcb81d2007-10-18 23:39:32 -07002184 ctr->bufsz = 0;
2185 }
2186 file->private_data = ctr;
2187 return 0;
2188
2189err2:
2190 kfree(pidarray);
2191err1:
2192 kfree(ctr);
2193err0:
2194 return -ENOMEM;
2195}
2196
Paul Menagebd89aab2007-10-18 23:40:44 -07002197static ssize_t cgroup_tasks_read(struct cgroup *cgrp,
Paul Menagebbcb81d2007-10-18 23:39:32 -07002198 struct cftype *cft,
2199 struct file *file, char __user *buf,
2200 size_t nbytes, loff_t *ppos)
2201{
2202 struct ctr_struct *ctr = file->private_data;
2203
2204 return simple_read_from_buffer(buf, nbytes, ppos, ctr->buf, ctr->bufsz);
2205}
2206
2207static int cgroup_tasks_release(struct inode *unused_inode,
2208 struct file *file)
2209{
2210 struct ctr_struct *ctr;
2211
2212 if (file->f_mode & FMODE_READ) {
2213 ctr = file->private_data;
2214 kfree(ctr->buf);
2215 kfree(ctr);
2216 }
2217 return 0;
2218}
2219
Paul Menagebd89aab2007-10-18 23:40:44 -07002220static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
Paul Menage81a6a5c2007-10-18 23:39:38 -07002221 struct cftype *cft)
2222{
Paul Menagebd89aab2007-10-18 23:40:44 -07002223 return notify_on_release(cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -07002224}
2225
Paul Menagebbcb81d2007-10-18 23:39:32 -07002226/*
2227 * for the common functions, 'private' gives the type of file
2228 */
Paul Menage81a6a5c2007-10-18 23:39:38 -07002229static struct cftype files[] = {
2230 {
2231 .name = "tasks",
2232 .open = cgroup_tasks_open,
2233 .read = cgroup_tasks_read,
2234 .write = cgroup_common_file_write,
2235 .release = cgroup_tasks_release,
2236 .private = FILE_TASKLIST,
2237 },
2238
2239 {
2240 .name = "notify_on_release",
Paul Menagef4c753b2008-04-29 00:59:56 -07002241 .read_u64 = cgroup_read_notify_on_release,
Paul Menage81a6a5c2007-10-18 23:39:38 -07002242 .write = cgroup_common_file_write,
2243 .private = FILE_NOTIFY_ON_RELEASE,
2244 },
Paul Menage81a6a5c2007-10-18 23:39:38 -07002245};
2246
2247static struct cftype cft_release_agent = {
2248 .name = "release_agent",
2249 .read = cgroup_common_file_read,
Paul Menagebbcb81d2007-10-18 23:39:32 -07002250 .write = cgroup_common_file_write,
Paul Menage81a6a5c2007-10-18 23:39:38 -07002251 .private = FILE_RELEASE_AGENT,
Paul Menagebbcb81d2007-10-18 23:39:32 -07002252};
2253
Paul Menagebd89aab2007-10-18 23:40:44 -07002254static int cgroup_populate_dir(struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002255{
2256 int err;
2257 struct cgroup_subsys *ss;
2258
2259 /* First clear out any existing files */
Paul Menagebd89aab2007-10-18 23:40:44 -07002260 cgroup_clear_directory(cgrp->dentry);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002261
Paul Menagebd89aab2007-10-18 23:40:44 -07002262 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
Paul Menagebbcb81d2007-10-18 23:39:32 -07002263 if (err < 0)
2264 return err;
2265
Paul Menagebd89aab2007-10-18 23:40:44 -07002266 if (cgrp == cgrp->top_cgroup) {
2267 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
Paul Menage81a6a5c2007-10-18 23:39:38 -07002268 return err;
2269 }
2270
Paul Menagebd89aab2007-10-18 23:40:44 -07002271 for_each_subsys(cgrp->root, ss) {
2272 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002273 return err;
2274 }
2275
2276 return 0;
2277}
2278
2279static void init_cgroup_css(struct cgroup_subsys_state *css,
2280 struct cgroup_subsys *ss,
Paul Menagebd89aab2007-10-18 23:40:44 -07002281 struct cgroup *cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002282{
Paul Menagebd89aab2007-10-18 23:40:44 -07002283 css->cgroup = cgrp;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002284 atomic_set(&css->refcnt, 0);
2285 css->flags = 0;
Paul Menagebd89aab2007-10-18 23:40:44 -07002286 if (cgrp == dummytop)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002287 set_bit(CSS_ROOT, &css->flags);
Paul Menagebd89aab2007-10-18 23:40:44 -07002288 BUG_ON(cgrp->subsys[ss->subsys_id]);
2289 cgrp->subsys[ss->subsys_id] = css;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002290}
2291
2292/*
Li Zefana043e3b2008-02-23 15:24:09 -08002293 * cgroup_create - create a cgroup
2294 * @parent: cgroup that will be parent of the new cgroup
2295 * @dentry: dentry of the new cgroup
2296 * @mode: mode to set on new inode
Paul Menageddbcc7e2007-10-18 23:39:30 -07002297 *
Li Zefana043e3b2008-02-23 15:24:09 -08002298 * Must be called with the mutex on the parent inode held
Paul Menageddbcc7e2007-10-18 23:39:30 -07002299 */
Paul Menageddbcc7e2007-10-18 23:39:30 -07002300static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
2301 int mode)
2302{
Paul Menagebd89aab2007-10-18 23:40:44 -07002303 struct cgroup *cgrp;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002304 struct cgroupfs_root *root = parent->root;
2305 int err = 0;
2306 struct cgroup_subsys *ss;
2307 struct super_block *sb = root->sb;
2308
Paul Menagebd89aab2007-10-18 23:40:44 -07002309 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
2310 if (!cgrp)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002311 return -ENOMEM;
2312
2313 /* Grab a reference on the superblock so the hierarchy doesn't
2314 * get deleted on unmount if there are child cgroups. This
2315 * can be done outside cgroup_mutex, since the sb can't
2316 * disappear while someone has an open control file on the
2317 * fs */
2318 atomic_inc(&sb->s_active);
2319
2320 mutex_lock(&cgroup_mutex);
2321
Paul Menagebd89aab2007-10-18 23:40:44 -07002322 INIT_LIST_HEAD(&cgrp->sibling);
2323 INIT_LIST_HEAD(&cgrp->children);
2324 INIT_LIST_HEAD(&cgrp->css_sets);
2325 INIT_LIST_HEAD(&cgrp->release_list);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002326
Paul Menagebd89aab2007-10-18 23:40:44 -07002327 cgrp->parent = parent;
2328 cgrp->root = parent->root;
2329 cgrp->top_cgroup = parent->top_cgroup;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002330
Li Zefanb6abdb02008-03-04 14:28:19 -08002331 if (notify_on_release(parent))
2332 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2333
Paul Menageddbcc7e2007-10-18 23:39:30 -07002334 for_each_subsys(root, ss) {
Paul Menagebd89aab2007-10-18 23:40:44 -07002335 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002336 if (IS_ERR(css)) {
2337 err = PTR_ERR(css);
2338 goto err_destroy;
2339 }
Paul Menagebd89aab2007-10-18 23:40:44 -07002340 init_cgroup_css(css, ss, cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002341 }
2342
Paul Menagebd89aab2007-10-18 23:40:44 -07002343 list_add(&cgrp->sibling, &cgrp->parent->children);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002344 root->number_of_cgroups++;
2345
Paul Menagebd89aab2007-10-18 23:40:44 -07002346 err = cgroup_create_dir(cgrp, dentry, mode);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002347 if (err < 0)
2348 goto err_remove;
2349
2350 /* The cgroup directory was pre-locked for us */
Paul Menagebd89aab2007-10-18 23:40:44 -07002351 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
Paul Menageddbcc7e2007-10-18 23:39:30 -07002352
Paul Menagebd89aab2007-10-18 23:40:44 -07002353 err = cgroup_populate_dir(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002354 /* If err < 0, we have a half-filled directory - oh well ;) */
2355
2356 mutex_unlock(&cgroup_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -07002357 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002358
2359 return 0;
2360
2361 err_remove:
2362
Paul Menagebd89aab2007-10-18 23:40:44 -07002363 list_del(&cgrp->sibling);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002364 root->number_of_cgroups--;
2365
2366 err_destroy:
2367
2368 for_each_subsys(root, ss) {
Paul Menagebd89aab2007-10-18 23:40:44 -07002369 if (cgrp->subsys[ss->subsys_id])
2370 ss->destroy(ss, cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002371 }
2372
2373 mutex_unlock(&cgroup_mutex);
2374
2375 /* Release the reference count that we took on the superblock */
2376 deactivate_super(sb);
2377
Paul Menagebd89aab2007-10-18 23:40:44 -07002378 kfree(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002379 return err;
2380}
2381
2382static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2383{
2384 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
2385
2386 /* the vfs holds inode->i_mutex already */
2387 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
2388}
2389
Paul Menagebd89aab2007-10-18 23:40:44 -07002390static inline int cgroup_has_css_refs(struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -07002391{
2392 /* Check the reference count on each subsystem. Since we
2393 * already established that there are no tasks in the
2394 * cgroup, if the css refcount is also 0, then there should
2395 * be no outstanding references, so the subsystem is safe to
2396 * destroy. We scan across all subsystems rather than using
2397 * the per-hierarchy linked list of mounted subsystems since
2398 * we can be called via check_for_release() with no
2399 * synchronization other than RCU, and the subsystem linked
2400 * list isn't RCU-safe */
2401 int i;
2402 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2403 struct cgroup_subsys *ss = subsys[i];
2404 struct cgroup_subsys_state *css;
2405 /* Skip subsystems not in this hierarchy */
Paul Menagebd89aab2007-10-18 23:40:44 -07002406 if (ss->root != cgrp->root)
Paul Menage81a6a5c2007-10-18 23:39:38 -07002407 continue;
Paul Menagebd89aab2007-10-18 23:40:44 -07002408 css = cgrp->subsys[ss->subsys_id];
Paul Menage81a6a5c2007-10-18 23:39:38 -07002409 /* When called from check_for_release() it's possible
2410 * that by this point the cgroup has been removed
2411 * and the css deleted. But a false-positive doesn't
2412 * matter, since it can only happen if the cgroup
2413 * has been deleted and hence no longer needs the
2414 * release agent to be called anyway. */
Paul Jacksone18f6312008-02-07 00:13:44 -08002415 if (css && atomic_read(&css->refcnt))
Paul Menage81a6a5c2007-10-18 23:39:38 -07002416 return 1;
Paul Menage81a6a5c2007-10-18 23:39:38 -07002417 }
2418 return 0;
2419}
2420
Paul Menageddbcc7e2007-10-18 23:39:30 -07002421static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
2422{
Paul Menagebd89aab2007-10-18 23:40:44 -07002423 struct cgroup *cgrp = dentry->d_fsdata;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002424 struct dentry *d;
2425 struct cgroup *parent;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002426 struct super_block *sb;
2427 struct cgroupfs_root *root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002428
2429 /* the vfs holds both inode->i_mutex already */
2430
2431 mutex_lock(&cgroup_mutex);
Paul Menagebd89aab2007-10-18 23:40:44 -07002432 if (atomic_read(&cgrp->count) != 0) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07002433 mutex_unlock(&cgroup_mutex);
2434 return -EBUSY;
2435 }
Paul Menagebd89aab2007-10-18 23:40:44 -07002436 if (!list_empty(&cgrp->children)) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07002437 mutex_unlock(&cgroup_mutex);
2438 return -EBUSY;
2439 }
2440
Paul Menagebd89aab2007-10-18 23:40:44 -07002441 parent = cgrp->parent;
2442 root = cgrp->root;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002443 sb = root->sb;
Li Zefana043e3b2008-02-23 15:24:09 -08002444
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -08002445 /*
Li Zefana043e3b2008-02-23 15:24:09 -08002446 * Call pre_destroy handlers of subsys. Notify subsystems
2447 * that rmdir() request comes.
KAMEZAWA Hiroyuki4fca88c2008-02-07 00:14:27 -08002448 */
2449 cgroup_call_pre_destroy(cgrp);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002450
Paul Menagebd89aab2007-10-18 23:40:44 -07002451 if (cgroup_has_css_refs(cgrp)) {
Paul Menageddbcc7e2007-10-18 23:39:30 -07002452 mutex_unlock(&cgroup_mutex);
2453 return -EBUSY;
2454 }
2455
Paul Menage81a6a5c2007-10-18 23:39:38 -07002456 spin_lock(&release_list_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07002457 set_bit(CGRP_REMOVED, &cgrp->flags);
2458 if (!list_empty(&cgrp->release_list))
2459 list_del(&cgrp->release_list);
Paul Menage81a6a5c2007-10-18 23:39:38 -07002460 spin_unlock(&release_list_lock);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002461 /* delete my sibling from parent->children */
Paul Menagebd89aab2007-10-18 23:40:44 -07002462 list_del(&cgrp->sibling);
2463 spin_lock(&cgrp->dentry->d_lock);
2464 d = dget(cgrp->dentry);
2465 cgrp->dentry = NULL;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002466 spin_unlock(&d->d_lock);
2467
2468 cgroup_d_remove_dir(d);
2469 dput(d);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002470
Paul Menagebd89aab2007-10-18 23:40:44 -07002471 set_bit(CGRP_RELEASABLE, &parent->flags);
Paul Menage81a6a5c2007-10-18 23:39:38 -07002472 check_for_release(parent);
2473
Paul Menageddbcc7e2007-10-18 23:39:30 -07002474 mutex_unlock(&cgroup_mutex);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002475 return 0;
2476}
2477
Li Zefan06a11922008-04-29 01:00:07 -07002478static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
Paul Menageddbcc7e2007-10-18 23:39:30 -07002479{
Paul Menageddbcc7e2007-10-18 23:39:30 -07002480 struct cgroup_subsys_state *css;
Diego Callejacfe36bd2007-11-14 16:58:54 -08002481
2482 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002483
2484 /* Create the top cgroup state for this subsystem */
2485 ss->root = &rootnode;
2486 css = ss->create(ss, dummytop);
2487 /* We don't handle early failures gracefully */
2488 BUG_ON(IS_ERR(css));
2489 init_cgroup_css(css, ss, dummytop);
2490
Li Zefane8d55fd2008-04-29 01:00:13 -07002491 /* Update the init_css_set to contain a subsys
Paul Menage817929e2007-10-18 23:39:36 -07002492 * pointer to this state - since the subsystem is
Li Zefane8d55fd2008-04-29 01:00:13 -07002493 * newly registered, all tasks and hence the
2494 * init_css_set is in the subsystem's top cgroup. */
2495 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
Paul Menageddbcc7e2007-10-18 23:39:30 -07002496
2497 need_forkexit_callback |= ss->fork || ss->exit;
2498
Li Zefane8d55fd2008-04-29 01:00:13 -07002499 /* At system boot, before all subsystems have been
2500 * registered, no tasks have been forked, so we don't
2501 * need to invoke fork callbacks here. */
2502 BUG_ON(!list_empty(&init_task.tasks));
2503
Paul Menageddbcc7e2007-10-18 23:39:30 -07002504 ss->active = 1;
2505}
2506
2507/**
Li Zefana043e3b2008-02-23 15:24:09 -08002508 * cgroup_init_early - cgroup initialization at system boot
2509 *
2510 * Initialize cgroups at system boot, and initialize any
2511 * subsystems that request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07002512 */
2513int __init cgroup_init_early(void)
2514{
2515 int i;
Paul Menage817929e2007-10-18 23:39:36 -07002516 kref_init(&init_css_set.ref);
2517 kref_get(&init_css_set.ref);
Paul Menage817929e2007-10-18 23:39:36 -07002518 INIT_LIST_HEAD(&init_css_set.cg_links);
2519 INIT_LIST_HEAD(&init_css_set.tasks);
Li Zefan472b1052008-04-29 01:00:11 -07002520 INIT_HLIST_NODE(&init_css_set.hlist);
Paul Menage817929e2007-10-18 23:39:36 -07002521 css_set_count = 1;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002522 init_cgroup_root(&rootnode);
2523 list_add(&rootnode.root_list, &roots);
Paul Menage817929e2007-10-18 23:39:36 -07002524 root_count = 1;
2525 init_task.cgroups = &init_css_set;
2526
2527 init_css_set_link.cg = &init_css_set;
Paul Menagebd89aab2007-10-18 23:40:44 -07002528 list_add(&init_css_set_link.cgrp_link_list,
Paul Menage817929e2007-10-18 23:39:36 -07002529 &rootnode.top_cgroup.css_sets);
2530 list_add(&init_css_set_link.cg_link_list,
2531 &init_css_set.cg_links);
Paul Menageddbcc7e2007-10-18 23:39:30 -07002532
Li Zefan472b1052008-04-29 01:00:11 -07002533 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
2534 INIT_HLIST_HEAD(&css_set_table[i]);
2535
Paul Menageddbcc7e2007-10-18 23:39:30 -07002536 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2537 struct cgroup_subsys *ss = subsys[i];
2538
2539 BUG_ON(!ss->name);
2540 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
2541 BUG_ON(!ss->create);
2542 BUG_ON(!ss->destroy);
2543 if (ss->subsys_id != i) {
Diego Callejacfe36bd2007-11-14 16:58:54 -08002544 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
Paul Menageddbcc7e2007-10-18 23:39:30 -07002545 ss->name, ss->subsys_id);
2546 BUG();
2547 }
2548
2549 if (ss->early_init)
2550 cgroup_init_subsys(ss);
2551 }
2552 return 0;
2553}
2554
2555/**
Li Zefana043e3b2008-02-23 15:24:09 -08002556 * cgroup_init - cgroup initialization
2557 *
2558 * Register cgroup filesystem and /proc file, and initialize
2559 * any subsystems that didn't request early init.
Paul Menageddbcc7e2007-10-18 23:39:30 -07002560 */
2561int __init cgroup_init(void)
2562{
2563 int err;
2564 int i;
Li Zefan472b1052008-04-29 01:00:11 -07002565 struct hlist_head *hhead;
Paul Menagea4243162007-10-18 23:39:35 -07002566
2567 err = bdi_init(&cgroup_backing_dev_info);
2568 if (err)
2569 return err;
Paul Menageddbcc7e2007-10-18 23:39:30 -07002570
2571 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2572 struct cgroup_subsys *ss = subsys[i];
2573 if (!ss->early_init)
2574 cgroup_init_subsys(ss);
2575 }
2576
Li Zefan472b1052008-04-29 01:00:11 -07002577 /* Add init_css_set to the hash table */
2578 hhead = css_set_hash(init_css_set.subsys);
2579 hlist_add_head(&init_css_set.hlist, hhead);
2580
Paul Menageddbcc7e2007-10-18 23:39:30 -07002581 err = register_filesystem(&cgroup_fs_type);
2582 if (err < 0)
2583 goto out;
2584
Li Zefan46ae2202008-04-29 01:00:08 -07002585 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
Paul Menagea4243162007-10-18 23:39:35 -07002586
Paul Menageddbcc7e2007-10-18 23:39:30 -07002587out:
Paul Menagea4243162007-10-18 23:39:35 -07002588 if (err)
2589 bdi_destroy(&cgroup_backing_dev_info);
2590
Paul Menageddbcc7e2007-10-18 23:39:30 -07002591 return err;
2592}
Paul Menageb4f48b62007-10-18 23:39:33 -07002593
Paul Menagea4243162007-10-18 23:39:35 -07002594/*
2595 * proc_cgroup_show()
2596 * - Print task's cgroup paths into seq_file, one line for each hierarchy
2597 * - Used for /proc/<pid>/cgroup.
2598 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
2599 * doesn't really matter if tsk->cgroup changes after we read it,
Cliff Wickman956db3c2008-02-07 00:14:43 -08002600 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
Paul Menagea4243162007-10-18 23:39:35 -07002601 * anyway. No need to check that tsk->cgroup != NULL, thanks to
2602 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
2603 * cgroup to top_cgroup.
2604 */
2605
2606/* TODO: Use a proper seq_file iterator */
2607static int proc_cgroup_show(struct seq_file *m, void *v)
2608{
2609 struct pid *pid;
2610 struct task_struct *tsk;
2611 char *buf;
2612 int retval;
2613 struct cgroupfs_root *root;
2614
2615 retval = -ENOMEM;
2616 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2617 if (!buf)
2618 goto out;
2619
2620 retval = -ESRCH;
2621 pid = m->private;
2622 tsk = get_pid_task(pid, PIDTYPE_PID);
2623 if (!tsk)
2624 goto out_free;
2625
2626 retval = 0;
2627
2628 mutex_lock(&cgroup_mutex);
2629
2630 for_each_root(root) {
2631 struct cgroup_subsys *ss;
Paul Menagebd89aab2007-10-18 23:40:44 -07002632 struct cgroup *cgrp;
Paul Menagea4243162007-10-18 23:39:35 -07002633 int subsys_id;
2634 int count = 0;
2635
2636 /* Skip this hierarchy if it has no active subsystems */
2637 if (!root->actual_subsys_bits)
2638 continue;
Paul Menageb6c30062008-04-10 21:29:16 -07002639 seq_printf(m, "%lu:", root->subsys_bits);
Paul Menagea4243162007-10-18 23:39:35 -07002640 for_each_subsys(root, ss)
2641 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
2642 seq_putc(m, ':');
2643 get_first_subsys(&root->top_cgroup, NULL, &subsys_id);
Paul Menagebd89aab2007-10-18 23:40:44 -07002644 cgrp = task_cgroup(tsk, subsys_id);
2645 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
Paul Menagea4243162007-10-18 23:39:35 -07002646 if (retval < 0)
2647 goto out_unlock;
2648 seq_puts(m, buf);
2649 seq_putc(m, '\n');
2650 }
2651
2652out_unlock:
2653 mutex_unlock(&cgroup_mutex);
2654 put_task_struct(tsk);
2655out_free:
2656 kfree(buf);
2657out:
2658 return retval;
2659}
2660
2661static int cgroup_open(struct inode *inode, struct file *file)
2662{
2663 struct pid *pid = PROC_I(inode)->pid;
2664 return single_open(file, proc_cgroup_show, pid);
2665}
2666
2667struct file_operations proc_cgroup_operations = {
2668 .open = cgroup_open,
2669 .read = seq_read,
2670 .llseek = seq_lseek,
2671 .release = single_release,
2672};
2673
2674/* Display information about each subsystem and each hierarchy */
2675static int proc_cgroupstats_show(struct seq_file *m, void *v)
2676{
2677 int i;
Paul Menagea4243162007-10-18 23:39:35 -07002678
Paul Menage8bab8dd2008-04-04 14:29:57 -07002679 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
Paul Menagea4243162007-10-18 23:39:35 -07002680 mutex_lock(&cgroup_mutex);
Paul Menagea4243162007-10-18 23:39:35 -07002681 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2682 struct cgroup_subsys *ss = subsys[i];
Paul Menage8bab8dd2008-04-04 14:29:57 -07002683 seq_printf(m, "%s\t%lu\t%d\t%d\n",
Paul Menage817929e2007-10-18 23:39:36 -07002684 ss->name, ss->root->subsys_bits,
Paul Menage8bab8dd2008-04-04 14:29:57 -07002685 ss->root->number_of_cgroups, !ss->disabled);
Paul Menagea4243162007-10-18 23:39:35 -07002686 }
2687 mutex_unlock(&cgroup_mutex);
2688 return 0;
2689}
2690
2691static int cgroupstats_open(struct inode *inode, struct file *file)
2692{
Al Viro9dce07f2008-03-29 03:07:28 +00002693 return single_open(file, proc_cgroupstats_show, NULL);
Paul Menagea4243162007-10-18 23:39:35 -07002694}
2695
2696static struct file_operations proc_cgroupstats_operations = {
2697 .open = cgroupstats_open,
2698 .read = seq_read,
2699 .llseek = seq_lseek,
2700 .release = single_release,
2701};
2702
Paul Menageb4f48b62007-10-18 23:39:33 -07002703/**
2704 * cgroup_fork - attach newly forked task to its parents cgroup.
Li Zefana043e3b2008-02-23 15:24:09 -08002705 * @child: pointer to task_struct of forking parent process.
Paul Menageb4f48b62007-10-18 23:39:33 -07002706 *
2707 * Description: A task inherits its parent's cgroup at fork().
2708 *
2709 * A pointer to the shared css_set was automatically copied in
2710 * fork.c by dup_task_struct(). However, we ignore that copy, since
2711 * it was not made under the protection of RCU or cgroup_mutex, so
Cliff Wickman956db3c2008-02-07 00:14:43 -08002712 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
Paul Menage817929e2007-10-18 23:39:36 -07002713 * have already changed current->cgroups, allowing the previously
2714 * referenced cgroup group to be removed and freed.
Paul Menageb4f48b62007-10-18 23:39:33 -07002715 *
2716 * At the point that cgroup_fork() is called, 'current' is the parent
2717 * task, and the passed argument 'child' points to the child task.
2718 */
2719void cgroup_fork(struct task_struct *child)
2720{
Paul Menage817929e2007-10-18 23:39:36 -07002721 task_lock(current);
2722 child->cgroups = current->cgroups;
2723 get_css_set(child->cgroups);
2724 task_unlock(current);
2725 INIT_LIST_HEAD(&child->cg_list);
Paul Menageb4f48b62007-10-18 23:39:33 -07002726}
2727
2728/**
Li Zefana043e3b2008-02-23 15:24:09 -08002729 * cgroup_fork_callbacks - run fork callbacks
2730 * @child: the new task
2731 *
2732 * Called on a new task very soon before adding it to the
2733 * tasklist. No need to take any locks since no-one can
2734 * be operating on this task.
Paul Menageb4f48b62007-10-18 23:39:33 -07002735 */
2736void cgroup_fork_callbacks(struct task_struct *child)
2737{
2738 if (need_forkexit_callback) {
2739 int i;
2740 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2741 struct cgroup_subsys *ss = subsys[i];
2742 if (ss->fork)
2743 ss->fork(ss, child);
2744 }
2745 }
2746}
2747
2748/**
Li Zefana043e3b2008-02-23 15:24:09 -08002749 * cgroup_post_fork - called on a new task after adding it to the task list
2750 * @child: the task in question
2751 *
2752 * Adds the task to the list running through its css_set if necessary.
2753 * Has to be after the task is visible on the task list in case we race
2754 * with the first call to cgroup_iter_start() - to guarantee that the
2755 * new task ends up on its list.
2756 */
Paul Menage817929e2007-10-18 23:39:36 -07002757void cgroup_post_fork(struct task_struct *child)
2758{
2759 if (use_task_css_set_links) {
2760 write_lock(&css_set_lock);
2761 if (list_empty(&child->cg_list))
2762 list_add(&child->cg_list, &child->cgroups->tasks);
2763 write_unlock(&css_set_lock);
2764 }
2765}
2766/**
Paul Menageb4f48b62007-10-18 23:39:33 -07002767 * cgroup_exit - detach cgroup from exiting task
2768 * @tsk: pointer to task_struct of exiting process
Li Zefana043e3b2008-02-23 15:24:09 -08002769 * @run_callback: run exit callbacks?
Paul Menageb4f48b62007-10-18 23:39:33 -07002770 *
2771 * Description: Detach cgroup from @tsk and release it.
2772 *
2773 * Note that cgroups marked notify_on_release force every task in
2774 * them to take the global cgroup_mutex mutex when exiting.
2775 * This could impact scaling on very large systems. Be reluctant to
2776 * use notify_on_release cgroups where very high task exit scaling
2777 * is required on large systems.
2778 *
2779 * the_top_cgroup_hack:
2780 *
2781 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
2782 *
2783 * We call cgroup_exit() while the task is still competent to
2784 * handle notify_on_release(), then leave the task attached to the
2785 * root cgroup in each hierarchy for the remainder of its exit.
2786 *
2787 * To do this properly, we would increment the reference count on
2788 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
2789 * code we would add a second cgroup function call, to drop that
2790 * reference. This would just create an unnecessary hot spot on
2791 * the top_cgroup reference count, to no avail.
2792 *
2793 * Normally, holding a reference to a cgroup without bumping its
2794 * count is unsafe. The cgroup could go away, or someone could
2795 * attach us to a different cgroup, decrementing the count on
2796 * the first cgroup that we never incremented. But in this case,
2797 * top_cgroup isn't going away, and either task has PF_EXITING set,
Cliff Wickman956db3c2008-02-07 00:14:43 -08002798 * which wards off any cgroup_attach_task() attempts, or task is a failed
2799 * fork, never visible to cgroup_attach_task.
Paul Menageb4f48b62007-10-18 23:39:33 -07002800 */
2801void cgroup_exit(struct task_struct *tsk, int run_callbacks)
2802{
2803 int i;
Paul Menage817929e2007-10-18 23:39:36 -07002804 struct css_set *cg;
Paul Menageb4f48b62007-10-18 23:39:33 -07002805
2806 if (run_callbacks && need_forkexit_callback) {
2807 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2808 struct cgroup_subsys *ss = subsys[i];
2809 if (ss->exit)
2810 ss->exit(ss, tsk);
2811 }
2812 }
Paul Menage817929e2007-10-18 23:39:36 -07002813
2814 /*
2815 * Unlink from the css_set task list if necessary.
2816 * Optimistically check cg_list before taking
2817 * css_set_lock
2818 */
2819 if (!list_empty(&tsk->cg_list)) {
2820 write_lock(&css_set_lock);
2821 if (!list_empty(&tsk->cg_list))
2822 list_del(&tsk->cg_list);
2823 write_unlock(&css_set_lock);
2824 }
2825
Paul Menageb4f48b62007-10-18 23:39:33 -07002826 /* Reassign the task to the init_css_set. */
2827 task_lock(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07002828 cg = tsk->cgroups;
2829 tsk->cgroups = &init_css_set;
Paul Menageb4f48b62007-10-18 23:39:33 -07002830 task_unlock(tsk);
Paul Menage817929e2007-10-18 23:39:36 -07002831 if (cg)
Paul Menage81a6a5c2007-10-18 23:39:38 -07002832 put_css_set_taskexit(cg);
Paul Menageb4f48b62007-10-18 23:39:33 -07002833}
Paul Menage697f4162007-10-18 23:39:34 -07002834
2835/**
Li Zefana043e3b2008-02-23 15:24:09 -08002836 * cgroup_clone - clone the cgroup the given subsystem is attached to
2837 * @tsk: the task to be moved
2838 * @subsys: the given subsystem
2839 *
2840 * Duplicate the current cgroup in the hierarchy that the given
2841 * subsystem is attached to, and move this task into the new
2842 * child.
Paul Menage697f4162007-10-18 23:39:34 -07002843 */
2844int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys)
2845{
2846 struct dentry *dentry;
2847 int ret = 0;
2848 char nodename[MAX_CGROUP_TYPE_NAMELEN];
2849 struct cgroup *parent, *child;
2850 struct inode *inode;
2851 struct css_set *cg;
2852 struct cgroupfs_root *root;
2853 struct cgroup_subsys *ss;
2854
2855 /* We shouldn't be called by an unregistered subsystem */
2856 BUG_ON(!subsys->active);
2857
2858 /* First figure out what hierarchy and cgroup we're dealing
2859 * with, and pin them so we can drop cgroup_mutex */
2860 mutex_lock(&cgroup_mutex);
2861 again:
2862 root = subsys->root;
2863 if (root == &rootnode) {
2864 printk(KERN_INFO
2865 "Not cloning cgroup for unused subsystem %s\n",
2866 subsys->name);
2867 mutex_unlock(&cgroup_mutex);
2868 return 0;
2869 }
Paul Menage817929e2007-10-18 23:39:36 -07002870 cg = tsk->cgroups;
Paul Menage697f4162007-10-18 23:39:34 -07002871 parent = task_cgroup(tsk, subsys->subsys_id);
2872
2873 snprintf(nodename, MAX_CGROUP_TYPE_NAMELEN, "node_%d", tsk->pid);
2874
2875 /* Pin the hierarchy */
2876 atomic_inc(&parent->root->sb->s_active);
2877
Paul Menage817929e2007-10-18 23:39:36 -07002878 /* Keep the cgroup alive */
2879 get_css_set(cg);
Paul Menage697f4162007-10-18 23:39:34 -07002880 mutex_unlock(&cgroup_mutex);
2881
2882 /* Now do the VFS work to create a cgroup */
2883 inode = parent->dentry->d_inode;
2884
2885 /* Hold the parent directory mutex across this operation to
2886 * stop anyone else deleting the new cgroup */
2887 mutex_lock(&inode->i_mutex);
2888 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
2889 if (IS_ERR(dentry)) {
2890 printk(KERN_INFO
Diego Callejacfe36bd2007-11-14 16:58:54 -08002891 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
Paul Menage697f4162007-10-18 23:39:34 -07002892 PTR_ERR(dentry));
2893 ret = PTR_ERR(dentry);
2894 goto out_release;
2895 }
2896
2897 /* Create the cgroup directory, which also creates the cgroup */
2898 ret = vfs_mkdir(inode, dentry, S_IFDIR | 0755);
Paul Menagebd89aab2007-10-18 23:40:44 -07002899 child = __d_cgrp(dentry);
Paul Menage697f4162007-10-18 23:39:34 -07002900 dput(dentry);
2901 if (ret) {
2902 printk(KERN_INFO
2903 "Failed to create cgroup %s: %d\n", nodename,
2904 ret);
2905 goto out_release;
2906 }
2907
2908 if (!child) {
2909 printk(KERN_INFO
2910 "Couldn't find new cgroup %s\n", nodename);
2911 ret = -ENOMEM;
2912 goto out_release;
2913 }
2914
2915 /* The cgroup now exists. Retake cgroup_mutex and check
2916 * that we're still in the same state that we thought we
2917 * were. */
2918 mutex_lock(&cgroup_mutex);
2919 if ((root != subsys->root) ||
2920 (parent != task_cgroup(tsk, subsys->subsys_id))) {
2921 /* Aargh, we raced ... */
2922 mutex_unlock(&inode->i_mutex);
Paul Menage817929e2007-10-18 23:39:36 -07002923 put_css_set(cg);
Paul Menage697f4162007-10-18 23:39:34 -07002924
2925 deactivate_super(parent->root->sb);
2926 /* The cgroup is still accessible in the VFS, but
2927 * we're not going to try to rmdir() it at this
2928 * point. */
2929 printk(KERN_INFO
2930 "Race in cgroup_clone() - leaking cgroup %s\n",
2931 nodename);
2932 goto again;
2933 }
2934
2935 /* do any required auto-setup */
2936 for_each_subsys(root, ss) {
2937 if (ss->post_clone)
2938 ss->post_clone(ss, child);
2939 }
2940
2941 /* All seems fine. Finish by moving the task into the new cgroup */
Cliff Wickman956db3c2008-02-07 00:14:43 -08002942 ret = cgroup_attach_task(child, tsk);
Paul Menage697f4162007-10-18 23:39:34 -07002943 mutex_unlock(&cgroup_mutex);
2944
2945 out_release:
2946 mutex_unlock(&inode->i_mutex);
Paul Menage81a6a5c2007-10-18 23:39:38 -07002947
2948 mutex_lock(&cgroup_mutex);
Paul Menage817929e2007-10-18 23:39:36 -07002949 put_css_set(cg);
Paul Menage81a6a5c2007-10-18 23:39:38 -07002950 mutex_unlock(&cgroup_mutex);
Paul Menage697f4162007-10-18 23:39:34 -07002951 deactivate_super(parent->root->sb);
2952 return ret;
2953}
2954
Li Zefana043e3b2008-02-23 15:24:09 -08002955/**
2956 * cgroup_is_descendant - see if @cgrp is a descendant of current task's cgrp
2957 * @cgrp: the cgroup in question
2958 *
2959 * See if @cgrp is a descendant of the current task's cgroup in
2960 * the appropriate hierarchy.
Paul Menage697f4162007-10-18 23:39:34 -07002961 *
2962 * If we are sending in dummytop, then presumably we are creating
2963 * the top cgroup in the subsystem.
2964 *
2965 * Called only by the ns (nsproxy) cgroup.
2966 */
Paul Menagebd89aab2007-10-18 23:40:44 -07002967int cgroup_is_descendant(const struct cgroup *cgrp)
Paul Menage697f4162007-10-18 23:39:34 -07002968{
2969 int ret;
2970 struct cgroup *target;
2971 int subsys_id;
2972
Paul Menagebd89aab2007-10-18 23:40:44 -07002973 if (cgrp == dummytop)
Paul Menage697f4162007-10-18 23:39:34 -07002974 return 1;
2975
Paul Menagebd89aab2007-10-18 23:40:44 -07002976 get_first_subsys(cgrp, NULL, &subsys_id);
Paul Menage697f4162007-10-18 23:39:34 -07002977 target = task_cgroup(current, subsys_id);
Paul Menagebd89aab2007-10-18 23:40:44 -07002978 while (cgrp != target && cgrp!= cgrp->top_cgroup)
2979 cgrp = cgrp->parent;
2980 ret = (cgrp == target);
Paul Menage697f4162007-10-18 23:39:34 -07002981 return ret;
2982}
Paul Menage81a6a5c2007-10-18 23:39:38 -07002983
Paul Menagebd89aab2007-10-18 23:40:44 -07002984static void check_for_release(struct cgroup *cgrp)
Paul Menage81a6a5c2007-10-18 23:39:38 -07002985{
2986 /* All of these checks rely on RCU to keep the cgroup
2987 * structure alive */
Paul Menagebd89aab2007-10-18 23:40:44 -07002988 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
2989 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
Paul Menage81a6a5c2007-10-18 23:39:38 -07002990 /* Control Group is currently removeable. If it's not
2991 * already queued for a userspace notification, queue
2992 * it now */
2993 int need_schedule_work = 0;
2994 spin_lock(&release_list_lock);
Paul Menagebd89aab2007-10-18 23:40:44 -07002995 if (!cgroup_is_removed(cgrp) &&
2996 list_empty(&cgrp->release_list)) {
2997 list_add(&cgrp->release_list, &release_list);
Paul Menage81a6a5c2007-10-18 23:39:38 -07002998 need_schedule_work = 1;
2999 }
3000 spin_unlock(&release_list_lock);
3001 if (need_schedule_work)
3002 schedule_work(&release_agent_work);
3003 }
3004}
3005
3006void __css_put(struct cgroup_subsys_state *css)
3007{
Paul Menagebd89aab2007-10-18 23:40:44 -07003008 struct cgroup *cgrp = css->cgroup;
Paul Menage81a6a5c2007-10-18 23:39:38 -07003009 rcu_read_lock();
Paul Menagebd89aab2007-10-18 23:40:44 -07003010 if (atomic_dec_and_test(&css->refcnt) && notify_on_release(cgrp)) {
3011 set_bit(CGRP_RELEASABLE, &cgrp->flags);
3012 check_for_release(cgrp);
Paul Menage81a6a5c2007-10-18 23:39:38 -07003013 }
3014 rcu_read_unlock();
3015}
3016
3017/*
3018 * Notify userspace when a cgroup is released, by running the
3019 * configured release agent with the name of the cgroup (path
3020 * relative to the root of cgroup file system) as the argument.
3021 *
3022 * Most likely, this user command will try to rmdir this cgroup.
3023 *
3024 * This races with the possibility that some other task will be
3025 * attached to this cgroup before it is removed, or that some other
3026 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
3027 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
3028 * unused, and this cgroup will be reprieved from its death sentence,
3029 * to continue to serve a useful existence. Next time it's released,
3030 * we will get notified again, if it still has 'notify_on_release' set.
3031 *
3032 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
3033 * means only wait until the task is successfully execve()'d. The
3034 * separate release agent task is forked by call_usermodehelper(),
3035 * then control in this thread returns here, without waiting for the
3036 * release agent task. We don't bother to wait because the caller of
3037 * this routine has no use for the exit status of the release agent
3038 * task, so no sense holding our caller up for that.
Paul Menage81a6a5c2007-10-18 23:39:38 -07003039 */
Paul Menage81a6a5c2007-10-18 23:39:38 -07003040static void cgroup_release_agent(struct work_struct *work)
3041{
3042 BUG_ON(work != &release_agent_work);
3043 mutex_lock(&cgroup_mutex);
3044 spin_lock(&release_list_lock);
3045 while (!list_empty(&release_list)) {
3046 char *argv[3], *envp[3];
3047 int i;
3048 char *pathbuf;
Paul Menagebd89aab2007-10-18 23:40:44 -07003049 struct cgroup *cgrp = list_entry(release_list.next,
Paul Menage81a6a5c2007-10-18 23:39:38 -07003050 struct cgroup,
3051 release_list);
Paul Menagebd89aab2007-10-18 23:40:44 -07003052 list_del_init(&cgrp->release_list);
Paul Menage81a6a5c2007-10-18 23:39:38 -07003053 spin_unlock(&release_list_lock);
3054 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3055 if (!pathbuf) {
3056 spin_lock(&release_list_lock);
3057 continue;
3058 }
3059
Paul Menagebd89aab2007-10-18 23:40:44 -07003060 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) {
Paul Menage81a6a5c2007-10-18 23:39:38 -07003061 kfree(pathbuf);
3062 spin_lock(&release_list_lock);
3063 continue;
3064 }
3065
3066 i = 0;
Paul Menagebd89aab2007-10-18 23:40:44 -07003067 argv[i++] = cgrp->root->release_agent_path;
Paul Menage81a6a5c2007-10-18 23:39:38 -07003068 argv[i++] = (char *)pathbuf;
3069 argv[i] = NULL;
3070
3071 i = 0;
3072 /* minimal command environment */
3073 envp[i++] = "HOME=/";
3074 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
3075 envp[i] = NULL;
3076
3077 /* Drop the lock while we invoke the usermode helper,
3078 * since the exec could involve hitting disk and hence
3079 * be a slow process */
3080 mutex_unlock(&cgroup_mutex);
3081 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
3082 kfree(pathbuf);
3083 mutex_lock(&cgroup_mutex);
3084 spin_lock(&release_list_lock);
3085 }
3086 spin_unlock(&release_list_lock);
3087 mutex_unlock(&cgroup_mutex);
3088}
Paul Menage8bab8dd2008-04-04 14:29:57 -07003089
3090static int __init cgroup_disable(char *str)
3091{
3092 int i;
3093 char *token;
3094
3095 while ((token = strsep(&str, ",")) != NULL) {
3096 if (!*token)
3097 continue;
3098
3099 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3100 struct cgroup_subsys *ss = subsys[i];
3101
3102 if (!strcmp(token, ss->name)) {
3103 ss->disabled = 1;
3104 printk(KERN_INFO "Disabling %s control group"
3105 " subsystem\n", ss->name);
3106 break;
3107 }
3108 }
3109 }
3110 return 1;
3111}
3112__setup("cgroup_disable=", cgroup_disable);