Commit | Line | Data |
---|---|---|
ddbcc7e8 | 1 | /* |
ddbcc7e8 PM |
2 | * 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> | |
a424316c | 34 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
35 | #include <linux/rcupdate.h> |
36 | #include <linux/sched.h> | |
817929ec | 37 | #include <linux/backing-dev.h> |
ddbcc7e8 PM |
38 | #include <linux/seq_file.h> |
39 | #include <linux/slab.h> | |
40 | #include <linux/magic.h> | |
41 | #include <linux/spinlock.h> | |
42 | #include <linux/string.h> | |
bbcb81d0 | 43 | #include <linux/sort.h> |
81a6a5cd | 44 | #include <linux/kmod.h> |
846c7bb0 BS |
45 | #include <linux/delayacct.h> |
46 | #include <linux/cgroupstats.h> | |
472b1053 | 47 | #include <linux/hash.h> |
3f8206d4 | 48 | #include <linux/namei.h> |
846c7bb0 | 49 | |
ddbcc7e8 PM |
50 | #include <asm/atomic.h> |
51 | ||
81a6a5cd PM |
52 | static DEFINE_MUTEX(cgroup_mutex); |
53 | ||
ddbcc7e8 PM |
54 | /* Generate an array of cgroup subsystem pointers */ |
55 | #define SUBSYS(_x) &_x ## _subsys, | |
56 | ||
57 | static struct cgroup_subsys *subsys[] = { | |
58 | #include <linux/cgroup_subsys.h> | |
59 | }; | |
60 | ||
61 | /* | |
62 | * A cgroupfs_root represents the root of a cgroup hierarchy, | |
63 | * and may be associated with a superblock to form an active | |
64 | * hierarchy | |
65 | */ | |
66 | struct cgroupfs_root { | |
67 | struct super_block *sb; | |
68 | ||
69 | /* | |
70 | * The bitmask of subsystems intended to be attached to this | |
71 | * hierarchy | |
72 | */ | |
73 | unsigned long subsys_bits; | |
74 | ||
75 | /* The bitmask of subsystems currently attached to this hierarchy */ | |
76 | unsigned long actual_subsys_bits; | |
77 | ||
78 | /* A list running through the attached subsystems */ | |
79 | struct list_head subsys_list; | |
80 | ||
81 | /* The root cgroup for this hierarchy */ | |
82 | struct cgroup top_cgroup; | |
83 | ||
84 | /* Tracks how many cgroups are currently defined in hierarchy.*/ | |
85 | int number_of_cgroups; | |
86 | ||
87 | /* A list running through the mounted hierarchies */ | |
88 | struct list_head root_list; | |
89 | ||
90 | /* Hierarchy-specific flags */ | |
91 | unsigned long flags; | |
81a6a5cd | 92 | |
e788e066 | 93 | /* The path to use for release notifications. */ |
81a6a5cd | 94 | char release_agent_path[PATH_MAX]; |
ddbcc7e8 PM |
95 | }; |
96 | ||
97 | ||
98 | /* | |
99 | * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the | |
100 | * subsystems that are otherwise unattached - it never has more than a | |
101 | * single cgroup, and all tasks are part of that cgroup. | |
102 | */ | |
103 | static struct cgroupfs_root rootnode; | |
104 | ||
105 | /* The list of hierarchy roots */ | |
106 | ||
107 | static LIST_HEAD(roots); | |
817929ec | 108 | static int root_count; |
ddbcc7e8 PM |
109 | |
110 | /* dummytop is a shorthand for the dummy hierarchy's top cgroup */ | |
111 | #define dummytop (&rootnode.top_cgroup) | |
112 | ||
113 | /* This flag indicates whether tasks in the fork and exit paths should | |
a043e3b2 LZ |
114 | * check for fork/exit handlers to call. This avoids us having to do |
115 | * extra work in the fork/exit path if none of the subsystems need to | |
116 | * be called. | |
ddbcc7e8 | 117 | */ |
8947f9d5 | 118 | static int need_forkexit_callback __read_mostly; |
cf475ad2 | 119 | static int need_mm_owner_callback __read_mostly; |
ddbcc7e8 | 120 | |
ddbcc7e8 | 121 | /* convenient tests for these bits */ |
bd89aabc | 122 | inline int cgroup_is_removed(const struct cgroup *cgrp) |
ddbcc7e8 | 123 | { |
bd89aabc | 124 | return test_bit(CGRP_REMOVED, &cgrp->flags); |
ddbcc7e8 PM |
125 | } |
126 | ||
127 | /* bits in struct cgroupfs_root flags field */ | |
128 | enum { | |
129 | ROOT_NOPREFIX, /* mounted subsystems have no named prefix */ | |
130 | }; | |
131 | ||
e9685a03 | 132 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
133 | { |
134 | const int bits = | |
bd89aabc PM |
135 | (1 << CGRP_RELEASABLE) | |
136 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
137 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
138 | } |
139 | ||
e9685a03 | 140 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 141 | { |
bd89aabc | 142 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
143 | } |
144 | ||
ddbcc7e8 PM |
145 | /* |
146 | * for_each_subsys() allows you to iterate on each subsystem attached to | |
147 | * an active hierarchy | |
148 | */ | |
149 | #define for_each_subsys(_root, _ss) \ | |
150 | list_for_each_entry(_ss, &_root->subsys_list, sibling) | |
151 | ||
152 | /* for_each_root() allows you to iterate across the active hierarchies */ | |
153 | #define for_each_root(_root) \ | |
154 | list_for_each_entry(_root, &roots, root_list) | |
155 | ||
81a6a5cd PM |
156 | /* the list of cgroups eligible for automatic release. Protected by |
157 | * release_list_lock */ | |
158 | static LIST_HEAD(release_list); | |
159 | static DEFINE_SPINLOCK(release_list_lock); | |
160 | static void cgroup_release_agent(struct work_struct *work); | |
161 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 162 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 163 | |
817929ec PM |
164 | /* Link structure for associating css_set objects with cgroups */ |
165 | struct cg_cgroup_link { | |
166 | /* | |
167 | * List running through cg_cgroup_links associated with a | |
168 | * cgroup, anchored on cgroup->css_sets | |
169 | */ | |
bd89aabc | 170 | struct list_head cgrp_link_list; |
817929ec PM |
171 | /* |
172 | * List running through cg_cgroup_links pointing at a | |
173 | * single css_set object, anchored on css_set->cg_links | |
174 | */ | |
175 | struct list_head cg_link_list; | |
176 | struct css_set *cg; | |
177 | }; | |
178 | ||
179 | /* The default css_set - used by init and its children prior to any | |
180 | * hierarchies being mounted. It contains a pointer to the root state | |
181 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
182 | * reference-counted, to improve performance when child cgroups | |
183 | * haven't been created. | |
184 | */ | |
185 | ||
186 | static struct css_set init_css_set; | |
187 | static struct cg_cgroup_link init_css_set_link; | |
188 | ||
189 | /* css_set_lock protects the list of css_set objects, and the | |
190 | * chain of tasks off each css_set. Nests outside task->alloc_lock | |
191 | * due to cgroup_iter_start() */ | |
192 | static DEFINE_RWLOCK(css_set_lock); | |
193 | static int css_set_count; | |
194 | ||
472b1053 LZ |
195 | /* hash table for cgroup groups. This improves the performance to |
196 | * find an existing css_set */ | |
197 | #define CSS_SET_HASH_BITS 7 | |
198 | #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS) | |
199 | static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE]; | |
200 | ||
201 | static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[]) | |
202 | { | |
203 | int i; | |
204 | int index; | |
205 | unsigned long tmp = 0UL; | |
206 | ||
207 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) | |
208 | tmp += (unsigned long)css[i]; | |
209 | tmp = (tmp >> 16) ^ tmp; | |
210 | ||
211 | index = hash_long(tmp, CSS_SET_HASH_BITS); | |
212 | ||
213 | return &css_set_table[index]; | |
214 | } | |
215 | ||
817929ec PM |
216 | /* We don't maintain the lists running through each css_set to its |
217 | * task until after the first call to cgroup_iter_start(). This | |
218 | * reduces the fork()/exit() overhead for people who have cgroups | |
219 | * compiled into their kernel but not actually in use */ | |
8947f9d5 | 220 | static int use_task_css_set_links __read_mostly; |
817929ec PM |
221 | |
222 | /* When we create or destroy a css_set, the operation simply | |
223 | * takes/releases a reference count on all the cgroups referenced | |
224 | * by subsystems in this css_set. This can end up multiple-counting | |
225 | * some cgroups, but that's OK - the ref-count is just a | |
226 | * busy/not-busy indicator; ensuring that we only count each cgroup | |
227 | * once would require taking a global lock to ensure that no | |
b4f48b63 PM |
228 | * subsystems moved between hierarchies while we were doing so. |
229 | * | |
230 | * Possible TODO: decide at boot time based on the number of | |
231 | * registered subsystems and the number of CPUs or NUMA nodes whether | |
232 | * it's better for performance to ref-count every subsystem, or to | |
233 | * take a global lock and only add one ref count to each hierarchy. | |
234 | */ | |
817929ec PM |
235 | |
236 | /* | |
237 | * unlink a css_set from the list and free it | |
238 | */ | |
81a6a5cd | 239 | static void unlink_css_set(struct css_set *cg) |
b4f48b63 | 240 | { |
71cbb949 KM |
241 | struct cg_cgroup_link *link; |
242 | struct cg_cgroup_link *saved_link; | |
243 | ||
472b1053 | 244 | hlist_del(&cg->hlist); |
817929ec | 245 | css_set_count--; |
71cbb949 KM |
246 | |
247 | list_for_each_entry_safe(link, saved_link, &cg->cg_links, | |
248 | cg_link_list) { | |
817929ec | 249 | list_del(&link->cg_link_list); |
bd89aabc | 250 | list_del(&link->cgrp_link_list); |
817929ec PM |
251 | kfree(link); |
252 | } | |
81a6a5cd PM |
253 | } |
254 | ||
146aa1bd | 255 | static void __put_css_set(struct css_set *cg, int taskexit) |
81a6a5cd PM |
256 | { |
257 | int i; | |
146aa1bd LJ |
258 | /* |
259 | * Ensure that the refcount doesn't hit zero while any readers | |
260 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
261 | * rwlock | |
262 | */ | |
263 | if (atomic_add_unless(&cg->refcount, -1, 1)) | |
264 | return; | |
265 | write_lock(&css_set_lock); | |
266 | if (!atomic_dec_and_test(&cg->refcount)) { | |
267 | write_unlock(&css_set_lock); | |
268 | return; | |
269 | } | |
81a6a5cd | 270 | unlink_css_set(cg); |
146aa1bd | 271 | write_unlock(&css_set_lock); |
81a6a5cd PM |
272 | |
273 | rcu_read_lock(); | |
274 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
bd89aabc PM |
275 | struct cgroup *cgrp = cg->subsys[i]->cgroup; |
276 | if (atomic_dec_and_test(&cgrp->count) && | |
277 | notify_on_release(cgrp)) { | |
81a6a5cd | 278 | if (taskexit) |
bd89aabc PM |
279 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
280 | check_for_release(cgrp); | |
81a6a5cd PM |
281 | } |
282 | } | |
283 | rcu_read_unlock(); | |
817929ec | 284 | kfree(cg); |
b4f48b63 PM |
285 | } |
286 | ||
817929ec PM |
287 | /* |
288 | * refcounted get/put for css_set objects | |
289 | */ | |
290 | static inline void get_css_set(struct css_set *cg) | |
291 | { | |
146aa1bd | 292 | atomic_inc(&cg->refcount); |
817929ec PM |
293 | } |
294 | ||
295 | static inline void put_css_set(struct css_set *cg) | |
296 | { | |
146aa1bd | 297 | __put_css_set(cg, 0); |
817929ec PM |
298 | } |
299 | ||
81a6a5cd PM |
300 | static inline void put_css_set_taskexit(struct css_set *cg) |
301 | { | |
146aa1bd | 302 | __put_css_set(cg, 1); |
81a6a5cd PM |
303 | } |
304 | ||
817929ec PM |
305 | /* |
306 | * find_existing_css_set() is a helper for | |
307 | * find_css_set(), and checks to see whether an existing | |
472b1053 | 308 | * css_set is suitable. |
817929ec PM |
309 | * |
310 | * oldcg: the cgroup group that we're using before the cgroup | |
311 | * transition | |
312 | * | |
bd89aabc | 313 | * cgrp: the cgroup that we're moving into |
817929ec PM |
314 | * |
315 | * template: location in which to build the desired set of subsystem | |
316 | * state objects for the new cgroup group | |
317 | */ | |
817929ec PM |
318 | static struct css_set *find_existing_css_set( |
319 | struct css_set *oldcg, | |
bd89aabc | 320 | struct cgroup *cgrp, |
817929ec | 321 | struct cgroup_subsys_state *template[]) |
b4f48b63 PM |
322 | { |
323 | int i; | |
bd89aabc | 324 | struct cgroupfs_root *root = cgrp->root; |
472b1053 LZ |
325 | struct hlist_head *hhead; |
326 | struct hlist_node *node; | |
327 | struct css_set *cg; | |
817929ec PM |
328 | |
329 | /* Built the set of subsystem state objects that we want to | |
330 | * see in the new css_set */ | |
331 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
8d53d55d | 332 | if (root->subsys_bits & (1UL << i)) { |
817929ec PM |
333 | /* Subsystem is in this hierarchy. So we want |
334 | * the subsystem state from the new | |
335 | * cgroup */ | |
bd89aabc | 336 | template[i] = cgrp->subsys[i]; |
817929ec PM |
337 | } else { |
338 | /* Subsystem is not in this hierarchy, so we | |
339 | * don't want to change the subsystem state */ | |
340 | template[i] = oldcg->subsys[i]; | |
341 | } | |
342 | } | |
343 | ||
472b1053 LZ |
344 | hhead = css_set_hash(template); |
345 | hlist_for_each_entry(cg, node, hhead, hlist) { | |
817929ec PM |
346 | if (!memcmp(template, cg->subsys, sizeof(cg->subsys))) { |
347 | /* All subsystems matched */ | |
348 | return cg; | |
349 | } | |
472b1053 | 350 | } |
817929ec PM |
351 | |
352 | /* No existing cgroup group matched */ | |
353 | return NULL; | |
354 | } | |
355 | ||
36553434 LZ |
356 | static void free_cg_links(struct list_head *tmp) |
357 | { | |
358 | struct cg_cgroup_link *link; | |
359 | struct cg_cgroup_link *saved_link; | |
360 | ||
361 | list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) { | |
362 | list_del(&link->cgrp_link_list); | |
363 | kfree(link); | |
364 | } | |
365 | } | |
366 | ||
817929ec PM |
367 | /* |
368 | * allocate_cg_links() allocates "count" cg_cgroup_link structures | |
bd89aabc | 369 | * and chains them on tmp through their cgrp_link_list fields. Returns 0 on |
817929ec PM |
370 | * success or a negative error |
371 | */ | |
817929ec PM |
372 | static int allocate_cg_links(int count, struct list_head *tmp) |
373 | { | |
374 | struct cg_cgroup_link *link; | |
375 | int i; | |
376 | INIT_LIST_HEAD(tmp); | |
377 | for (i = 0; i < count; i++) { | |
378 | link = kmalloc(sizeof(*link), GFP_KERNEL); | |
379 | if (!link) { | |
36553434 | 380 | free_cg_links(tmp); |
817929ec PM |
381 | return -ENOMEM; |
382 | } | |
bd89aabc | 383 | list_add(&link->cgrp_link_list, tmp); |
817929ec PM |
384 | } |
385 | return 0; | |
386 | } | |
387 | ||
817929ec PM |
388 | /* |
389 | * find_css_set() takes an existing cgroup group and a | |
390 | * cgroup object, and returns a css_set object that's | |
391 | * equivalent to the old group, but with the given cgroup | |
392 | * substituted into the appropriate hierarchy. Must be called with | |
393 | * cgroup_mutex held | |
394 | */ | |
817929ec | 395 | static struct css_set *find_css_set( |
bd89aabc | 396 | struct css_set *oldcg, struct cgroup *cgrp) |
817929ec PM |
397 | { |
398 | struct css_set *res; | |
399 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT]; | |
400 | int i; | |
401 | ||
402 | struct list_head tmp_cg_links; | |
403 | struct cg_cgroup_link *link; | |
404 | ||
472b1053 LZ |
405 | struct hlist_head *hhead; |
406 | ||
817929ec PM |
407 | /* First see if we already have a cgroup group that matches |
408 | * the desired set */ | |
7e9abd89 | 409 | read_lock(&css_set_lock); |
bd89aabc | 410 | res = find_existing_css_set(oldcg, cgrp, template); |
817929ec PM |
411 | if (res) |
412 | get_css_set(res); | |
7e9abd89 | 413 | read_unlock(&css_set_lock); |
817929ec PM |
414 | |
415 | if (res) | |
416 | return res; | |
417 | ||
418 | res = kmalloc(sizeof(*res), GFP_KERNEL); | |
419 | if (!res) | |
420 | return NULL; | |
421 | ||
422 | /* Allocate all the cg_cgroup_link objects that we'll need */ | |
423 | if (allocate_cg_links(root_count, &tmp_cg_links) < 0) { | |
424 | kfree(res); | |
425 | return NULL; | |
426 | } | |
427 | ||
146aa1bd | 428 | atomic_set(&res->refcount, 1); |
817929ec PM |
429 | INIT_LIST_HEAD(&res->cg_links); |
430 | INIT_LIST_HEAD(&res->tasks); | |
472b1053 | 431 | INIT_HLIST_NODE(&res->hlist); |
817929ec PM |
432 | |
433 | /* Copy the set of subsystem state objects generated in | |
434 | * find_existing_css_set() */ | |
435 | memcpy(res->subsys, template, sizeof(res->subsys)); | |
436 | ||
437 | write_lock(&css_set_lock); | |
438 | /* Add reference counts and links from the new css_set. */ | |
439 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
bd89aabc | 440 | struct cgroup *cgrp = res->subsys[i]->cgroup; |
817929ec | 441 | struct cgroup_subsys *ss = subsys[i]; |
bd89aabc | 442 | atomic_inc(&cgrp->count); |
817929ec PM |
443 | /* |
444 | * We want to add a link once per cgroup, so we | |
445 | * only do it for the first subsystem in each | |
446 | * hierarchy | |
447 | */ | |
448 | if (ss->root->subsys_list.next == &ss->sibling) { | |
449 | BUG_ON(list_empty(&tmp_cg_links)); | |
450 | link = list_entry(tmp_cg_links.next, | |
451 | struct cg_cgroup_link, | |
bd89aabc PM |
452 | cgrp_link_list); |
453 | list_del(&link->cgrp_link_list); | |
454 | list_add(&link->cgrp_link_list, &cgrp->css_sets); | |
817929ec PM |
455 | link->cg = res; |
456 | list_add(&link->cg_link_list, &res->cg_links); | |
457 | } | |
458 | } | |
459 | if (list_empty(&rootnode.subsys_list)) { | |
460 | link = list_entry(tmp_cg_links.next, | |
461 | struct cg_cgroup_link, | |
bd89aabc PM |
462 | cgrp_link_list); |
463 | list_del(&link->cgrp_link_list); | |
464 | list_add(&link->cgrp_link_list, &dummytop->css_sets); | |
817929ec PM |
465 | link->cg = res; |
466 | list_add(&link->cg_link_list, &res->cg_links); | |
467 | } | |
468 | ||
469 | BUG_ON(!list_empty(&tmp_cg_links)); | |
470 | ||
817929ec | 471 | css_set_count++; |
472b1053 LZ |
472 | |
473 | /* Add this cgroup group to the hash table */ | |
474 | hhead = css_set_hash(res->subsys); | |
475 | hlist_add_head(&res->hlist, hhead); | |
476 | ||
817929ec PM |
477 | write_unlock(&css_set_lock); |
478 | ||
479 | return res; | |
b4f48b63 PM |
480 | } |
481 | ||
ddbcc7e8 PM |
482 | /* |
483 | * There is one global cgroup mutex. We also require taking | |
484 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
485 | * See "The task_lock() exception", at the end of this comment. | |
486 | * | |
487 | * A task must hold cgroup_mutex to modify cgroups. | |
488 | * | |
489 | * Any task can increment and decrement the count field without lock. | |
490 | * So in general, code holding cgroup_mutex can't rely on the count | |
491 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 492 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
493 | * means that no tasks are currently attached, therefore there is no |
494 | * way a task attached to that cgroup can fork (the other way to | |
495 | * increment the count). So code holding cgroup_mutex can safely | |
496 | * assume that if the count is zero, it will stay zero. Similarly, if | |
497 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
498 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
499 | * needs that mutex. | |
500 | * | |
ddbcc7e8 PM |
501 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
502 | * (usually) take cgroup_mutex. These are the two most performance | |
503 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
504 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
505 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
506 | * to the release agent with the name of the cgroup (path relative to |
507 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
508 | * |
509 | * A cgroup can only be deleted if both its 'count' of using tasks | |
510 | * is zero, and its list of 'children' cgroups is empty. Since all | |
511 | * tasks in the system use _some_ cgroup, and since there is always at | |
512 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
513 | * always has either children cgroups and/or using tasks. So we don't | |
514 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
515 | * | |
516 | * The task_lock() exception | |
517 | * | |
518 | * The need for this exception arises from the action of | |
956db3ca | 519 | * cgroup_attach_task(), which overwrites one tasks cgroup pointer with |
a043e3b2 | 520 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
521 | * several performance critical places that need to reference |
522 | * task->cgroup without the expense of grabbing a system global | |
523 | * mutex. Therefore except as noted below, when dereferencing or, as | |
956db3ca | 524 | * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use |
ddbcc7e8 PM |
525 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
526 | * the task_struct routinely used for such matters. | |
527 | * | |
528 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 529 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
530 | */ |
531 | ||
ddbcc7e8 PM |
532 | /** |
533 | * cgroup_lock - lock out any changes to cgroup structures | |
534 | * | |
535 | */ | |
ddbcc7e8 PM |
536 | void cgroup_lock(void) |
537 | { | |
538 | mutex_lock(&cgroup_mutex); | |
539 | } | |
540 | ||
541 | /** | |
542 | * cgroup_unlock - release lock on cgroup changes | |
543 | * | |
544 | * Undo the lock taken in a previous cgroup_lock() call. | |
545 | */ | |
ddbcc7e8 PM |
546 | void cgroup_unlock(void) |
547 | { | |
548 | mutex_unlock(&cgroup_mutex); | |
549 | } | |
550 | ||
551 | /* | |
552 | * A couple of forward declarations required, due to cyclic reference loop: | |
553 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> | |
554 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations | |
555 | * -> cgroup_mkdir. | |
556 | */ | |
557 | ||
558 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode); | |
559 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); | |
bd89aabc | 560 | static int cgroup_populate_dir(struct cgroup *cgrp); |
ddbcc7e8 | 561 | static struct inode_operations cgroup_dir_inode_operations; |
a424316c PM |
562 | static struct file_operations proc_cgroupstats_operations; |
563 | ||
564 | static struct backing_dev_info cgroup_backing_dev_info = { | |
e4ad08fe | 565 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
a424316c | 566 | }; |
ddbcc7e8 PM |
567 | |
568 | static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb) | |
569 | { | |
570 | struct inode *inode = new_inode(sb); | |
ddbcc7e8 PM |
571 | |
572 | if (inode) { | |
573 | inode->i_mode = mode; | |
76aac0e9 DH |
574 | inode->i_uid = current_fsuid(); |
575 | inode->i_gid = current_fsgid(); | |
ddbcc7e8 PM |
576 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
577 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; | |
578 | } | |
579 | return inode; | |
580 | } | |
581 | ||
4fca88c8 KH |
582 | /* |
583 | * Call subsys's pre_destroy handler. | |
584 | * This is called before css refcnt check. | |
585 | */ | |
4fca88c8 KH |
586 | static void cgroup_call_pre_destroy(struct cgroup *cgrp) |
587 | { | |
588 | struct cgroup_subsys *ss; | |
589 | for_each_subsys(cgrp->root, ss) | |
590 | if (ss->pre_destroy && cgrp->subsys[ss->subsys_id]) | |
591 | ss->pre_destroy(ss, cgrp); | |
592 | return; | |
593 | } | |
594 | ||
ddbcc7e8 PM |
595 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) |
596 | { | |
597 | /* is dentry a directory ? if so, kfree() associated cgroup */ | |
598 | if (S_ISDIR(inode->i_mode)) { | |
bd89aabc | 599 | struct cgroup *cgrp = dentry->d_fsdata; |
8dc4f3e1 | 600 | struct cgroup_subsys *ss; |
bd89aabc | 601 | BUG_ON(!(cgroup_is_removed(cgrp))); |
81a6a5cd PM |
602 | /* It's possible for external users to be holding css |
603 | * reference counts on a cgroup; css_put() needs to | |
604 | * be able to access the cgroup after decrementing | |
605 | * the reference count in order to know if it needs to | |
606 | * queue the cgroup to be handled by the release | |
607 | * agent */ | |
608 | synchronize_rcu(); | |
8dc4f3e1 PM |
609 | |
610 | mutex_lock(&cgroup_mutex); | |
611 | /* | |
612 | * Release the subsystem state objects. | |
613 | */ | |
614 | for_each_subsys(cgrp->root, ss) { | |
615 | if (cgrp->subsys[ss->subsys_id]) | |
616 | ss->destroy(ss, cgrp); | |
617 | } | |
618 | ||
619 | cgrp->root->number_of_cgroups--; | |
620 | mutex_unlock(&cgroup_mutex); | |
621 | ||
622 | /* Drop the active superblock reference that we took when we | |
623 | * created the cgroup */ | |
624 | deactivate_super(cgrp->root->sb); | |
625 | ||
bd89aabc | 626 | kfree(cgrp); |
ddbcc7e8 PM |
627 | } |
628 | iput(inode); | |
629 | } | |
630 | ||
631 | static void remove_dir(struct dentry *d) | |
632 | { | |
633 | struct dentry *parent = dget(d->d_parent); | |
634 | ||
635 | d_delete(d); | |
636 | simple_rmdir(parent->d_inode, d); | |
637 | dput(parent); | |
638 | } | |
639 | ||
640 | static void cgroup_clear_directory(struct dentry *dentry) | |
641 | { | |
642 | struct list_head *node; | |
643 | ||
644 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
645 | spin_lock(&dcache_lock); | |
646 | node = dentry->d_subdirs.next; | |
647 | while (node != &dentry->d_subdirs) { | |
648 | struct dentry *d = list_entry(node, struct dentry, d_u.d_child); | |
649 | list_del_init(node); | |
650 | if (d->d_inode) { | |
651 | /* This should never be called on a cgroup | |
652 | * directory with child cgroups */ | |
653 | BUG_ON(d->d_inode->i_mode & S_IFDIR); | |
654 | d = dget_locked(d); | |
655 | spin_unlock(&dcache_lock); | |
656 | d_delete(d); | |
657 | simple_unlink(dentry->d_inode, d); | |
658 | dput(d); | |
659 | spin_lock(&dcache_lock); | |
660 | } | |
661 | node = dentry->d_subdirs.next; | |
662 | } | |
663 | spin_unlock(&dcache_lock); | |
664 | } | |
665 | ||
666 | /* | |
667 | * NOTE : the dentry must have been dget()'ed | |
668 | */ | |
669 | static void cgroup_d_remove_dir(struct dentry *dentry) | |
670 | { | |
671 | cgroup_clear_directory(dentry); | |
672 | ||
673 | spin_lock(&dcache_lock); | |
674 | list_del_init(&dentry->d_u.d_child); | |
675 | spin_unlock(&dcache_lock); | |
676 | remove_dir(dentry); | |
677 | } | |
678 | ||
679 | static int rebind_subsystems(struct cgroupfs_root *root, | |
680 | unsigned long final_bits) | |
681 | { | |
682 | unsigned long added_bits, removed_bits; | |
bd89aabc | 683 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
684 | int i; |
685 | ||
686 | removed_bits = root->actual_subsys_bits & ~final_bits; | |
687 | added_bits = final_bits & ~root->actual_subsys_bits; | |
688 | /* Check that any added subsystems are currently free */ | |
689 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
8d53d55d | 690 | unsigned long bit = 1UL << i; |
ddbcc7e8 PM |
691 | struct cgroup_subsys *ss = subsys[i]; |
692 | if (!(bit & added_bits)) | |
693 | continue; | |
694 | if (ss->root != &rootnode) { | |
695 | /* Subsystem isn't free */ | |
696 | return -EBUSY; | |
697 | } | |
698 | } | |
699 | ||
700 | /* Currently we don't handle adding/removing subsystems when | |
701 | * any child cgroups exist. This is theoretically supportable | |
702 | * but involves complex error handling, so it's being left until | |
703 | * later */ | |
307257cf | 704 | if (root->number_of_cgroups > 1) |
ddbcc7e8 PM |
705 | return -EBUSY; |
706 | ||
707 | /* Process each subsystem */ | |
708 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
709 | struct cgroup_subsys *ss = subsys[i]; | |
710 | unsigned long bit = 1UL << i; | |
711 | if (bit & added_bits) { | |
712 | /* We're binding this subsystem to this hierarchy */ | |
bd89aabc | 713 | BUG_ON(cgrp->subsys[i]); |
ddbcc7e8 PM |
714 | BUG_ON(!dummytop->subsys[i]); |
715 | BUG_ON(dummytop->subsys[i]->cgroup != dummytop); | |
bd89aabc PM |
716 | cgrp->subsys[i] = dummytop->subsys[i]; |
717 | cgrp->subsys[i]->cgroup = cgrp; | |
ddbcc7e8 PM |
718 | list_add(&ss->sibling, &root->subsys_list); |
719 | rcu_assign_pointer(ss->root, root); | |
720 | if (ss->bind) | |
bd89aabc | 721 | ss->bind(ss, cgrp); |
ddbcc7e8 PM |
722 | |
723 | } else if (bit & removed_bits) { | |
724 | /* We're removing this subsystem */ | |
bd89aabc PM |
725 | BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]); |
726 | BUG_ON(cgrp->subsys[i]->cgroup != cgrp); | |
ddbcc7e8 PM |
727 | if (ss->bind) |
728 | ss->bind(ss, dummytop); | |
729 | dummytop->subsys[i]->cgroup = dummytop; | |
bd89aabc | 730 | cgrp->subsys[i] = NULL; |
ddbcc7e8 PM |
731 | rcu_assign_pointer(subsys[i]->root, &rootnode); |
732 | list_del(&ss->sibling); | |
733 | } else if (bit & final_bits) { | |
734 | /* Subsystem state should already exist */ | |
bd89aabc | 735 | BUG_ON(!cgrp->subsys[i]); |
ddbcc7e8 PM |
736 | } else { |
737 | /* Subsystem state shouldn't exist */ | |
bd89aabc | 738 | BUG_ON(cgrp->subsys[i]); |
ddbcc7e8 PM |
739 | } |
740 | } | |
741 | root->subsys_bits = root->actual_subsys_bits = final_bits; | |
742 | synchronize_rcu(); | |
743 | ||
744 | return 0; | |
745 | } | |
746 | ||
747 | static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs) | |
748 | { | |
749 | struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info; | |
750 | struct cgroup_subsys *ss; | |
751 | ||
752 | mutex_lock(&cgroup_mutex); | |
753 | for_each_subsys(root, ss) | |
754 | seq_printf(seq, ",%s", ss->name); | |
755 | if (test_bit(ROOT_NOPREFIX, &root->flags)) | |
756 | seq_puts(seq, ",noprefix"); | |
81a6a5cd PM |
757 | if (strlen(root->release_agent_path)) |
758 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
ddbcc7e8 PM |
759 | mutex_unlock(&cgroup_mutex); |
760 | return 0; | |
761 | } | |
762 | ||
763 | struct cgroup_sb_opts { | |
764 | unsigned long subsys_bits; | |
765 | unsigned long flags; | |
81a6a5cd | 766 | char *release_agent; |
ddbcc7e8 PM |
767 | }; |
768 | ||
769 | /* Convert a hierarchy specifier into a bitmask of subsystems and | |
770 | * flags. */ | |
771 | static int parse_cgroupfs_options(char *data, | |
772 | struct cgroup_sb_opts *opts) | |
773 | { | |
774 | char *token, *o = data ?: "all"; | |
775 | ||
776 | opts->subsys_bits = 0; | |
777 | opts->flags = 0; | |
81a6a5cd | 778 | opts->release_agent = NULL; |
ddbcc7e8 PM |
779 | |
780 | while ((token = strsep(&o, ",")) != NULL) { | |
781 | if (!*token) | |
782 | return -EINVAL; | |
783 | if (!strcmp(token, "all")) { | |
8bab8dde PM |
784 | /* Add all non-disabled subsystems */ |
785 | int i; | |
786 | opts->subsys_bits = 0; | |
787 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
788 | struct cgroup_subsys *ss = subsys[i]; | |
789 | if (!ss->disabled) | |
790 | opts->subsys_bits |= 1ul << i; | |
791 | } | |
ddbcc7e8 PM |
792 | } else if (!strcmp(token, "noprefix")) { |
793 | set_bit(ROOT_NOPREFIX, &opts->flags); | |
81a6a5cd PM |
794 | } else if (!strncmp(token, "release_agent=", 14)) { |
795 | /* Specifying two release agents is forbidden */ | |
796 | if (opts->release_agent) | |
797 | return -EINVAL; | |
798 | opts->release_agent = kzalloc(PATH_MAX, GFP_KERNEL); | |
799 | if (!opts->release_agent) | |
800 | return -ENOMEM; | |
801 | strncpy(opts->release_agent, token + 14, PATH_MAX - 1); | |
802 | opts->release_agent[PATH_MAX - 1] = 0; | |
ddbcc7e8 PM |
803 | } else { |
804 | struct cgroup_subsys *ss; | |
805 | int i; | |
806 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
807 | ss = subsys[i]; | |
808 | if (!strcmp(token, ss->name)) { | |
8bab8dde PM |
809 | if (!ss->disabled) |
810 | set_bit(i, &opts->subsys_bits); | |
ddbcc7e8 PM |
811 | break; |
812 | } | |
813 | } | |
814 | if (i == CGROUP_SUBSYS_COUNT) | |
815 | return -ENOENT; | |
816 | } | |
817 | } | |
818 | ||
819 | /* We can't have an empty hierarchy */ | |
820 | if (!opts->subsys_bits) | |
821 | return -EINVAL; | |
822 | ||
823 | return 0; | |
824 | } | |
825 | ||
826 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) | |
827 | { | |
828 | int ret = 0; | |
829 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 830 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
831 | struct cgroup_sb_opts opts; |
832 | ||
bd89aabc | 833 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
834 | mutex_lock(&cgroup_mutex); |
835 | ||
836 | /* See what subsystems are wanted */ | |
837 | ret = parse_cgroupfs_options(data, &opts); | |
838 | if (ret) | |
839 | goto out_unlock; | |
840 | ||
841 | /* Don't allow flags to change at remount */ | |
842 | if (opts.flags != root->flags) { | |
843 | ret = -EINVAL; | |
844 | goto out_unlock; | |
845 | } | |
846 | ||
847 | ret = rebind_subsystems(root, opts.subsys_bits); | |
848 | ||
849 | /* (re)populate subsystem files */ | |
850 | if (!ret) | |
bd89aabc | 851 | cgroup_populate_dir(cgrp); |
ddbcc7e8 | 852 | |
81a6a5cd PM |
853 | if (opts.release_agent) |
854 | strcpy(root->release_agent_path, opts.release_agent); | |
ddbcc7e8 | 855 | out_unlock: |
81a6a5cd PM |
856 | if (opts.release_agent) |
857 | kfree(opts.release_agent); | |
ddbcc7e8 | 858 | mutex_unlock(&cgroup_mutex); |
bd89aabc | 859 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
860 | return ret; |
861 | } | |
862 | ||
863 | static struct super_operations cgroup_ops = { | |
864 | .statfs = simple_statfs, | |
865 | .drop_inode = generic_delete_inode, | |
866 | .show_options = cgroup_show_options, | |
867 | .remount_fs = cgroup_remount, | |
868 | }; | |
869 | ||
cc31edce PM |
870 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
871 | { | |
872 | INIT_LIST_HEAD(&cgrp->sibling); | |
873 | INIT_LIST_HEAD(&cgrp->children); | |
874 | INIT_LIST_HEAD(&cgrp->css_sets); | |
875 | INIT_LIST_HEAD(&cgrp->release_list); | |
876 | init_rwsem(&cgrp->pids_mutex); | |
877 | } | |
ddbcc7e8 PM |
878 | static void init_cgroup_root(struct cgroupfs_root *root) |
879 | { | |
bd89aabc | 880 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
881 | INIT_LIST_HEAD(&root->subsys_list); |
882 | INIT_LIST_HEAD(&root->root_list); | |
883 | root->number_of_cgroups = 1; | |
bd89aabc PM |
884 | cgrp->root = root; |
885 | cgrp->top_cgroup = cgrp; | |
cc31edce | 886 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 PM |
887 | } |
888 | ||
889 | static int cgroup_test_super(struct super_block *sb, void *data) | |
890 | { | |
891 | struct cgroupfs_root *new = data; | |
892 | struct cgroupfs_root *root = sb->s_fs_info; | |
893 | ||
894 | /* First check subsystems */ | |
895 | if (new->subsys_bits != root->subsys_bits) | |
896 | return 0; | |
897 | ||
898 | /* Next check flags */ | |
899 | if (new->flags != root->flags) | |
900 | return 0; | |
901 | ||
902 | return 1; | |
903 | } | |
904 | ||
905 | static int cgroup_set_super(struct super_block *sb, void *data) | |
906 | { | |
907 | int ret; | |
908 | struct cgroupfs_root *root = data; | |
909 | ||
910 | ret = set_anon_super(sb, NULL); | |
911 | if (ret) | |
912 | return ret; | |
913 | ||
914 | sb->s_fs_info = root; | |
915 | root->sb = sb; | |
916 | ||
917 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
918 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
919 | sb->s_magic = CGROUP_SUPER_MAGIC; | |
920 | sb->s_op = &cgroup_ops; | |
921 | ||
922 | return 0; | |
923 | } | |
924 | ||
925 | static int cgroup_get_rootdir(struct super_block *sb) | |
926 | { | |
927 | struct inode *inode = | |
928 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); | |
929 | struct dentry *dentry; | |
930 | ||
931 | if (!inode) | |
932 | return -ENOMEM; | |
933 | ||
ddbcc7e8 PM |
934 | inode->i_fop = &simple_dir_operations; |
935 | inode->i_op = &cgroup_dir_inode_operations; | |
936 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
937 | inc_nlink(inode); | |
938 | dentry = d_alloc_root(inode); | |
939 | if (!dentry) { | |
940 | iput(inode); | |
941 | return -ENOMEM; | |
942 | } | |
943 | sb->s_root = dentry; | |
944 | return 0; | |
945 | } | |
946 | ||
947 | static int cgroup_get_sb(struct file_system_type *fs_type, | |
948 | int flags, const char *unused_dev_name, | |
949 | void *data, struct vfsmount *mnt) | |
950 | { | |
951 | struct cgroup_sb_opts opts; | |
952 | int ret = 0; | |
953 | struct super_block *sb; | |
954 | struct cgroupfs_root *root; | |
28fd5dfc | 955 | struct list_head tmp_cg_links; |
ddbcc7e8 PM |
956 | |
957 | /* First find the desired set of subsystems */ | |
958 | ret = parse_cgroupfs_options(data, &opts); | |
81a6a5cd PM |
959 | if (ret) { |
960 | if (opts.release_agent) | |
961 | kfree(opts.release_agent); | |
ddbcc7e8 | 962 | return ret; |
81a6a5cd | 963 | } |
ddbcc7e8 PM |
964 | |
965 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
f7770738 LZ |
966 | if (!root) { |
967 | if (opts.release_agent) | |
968 | kfree(opts.release_agent); | |
ddbcc7e8 | 969 | return -ENOMEM; |
f7770738 | 970 | } |
ddbcc7e8 PM |
971 | |
972 | init_cgroup_root(root); | |
973 | root->subsys_bits = opts.subsys_bits; | |
974 | root->flags = opts.flags; | |
81a6a5cd PM |
975 | if (opts.release_agent) { |
976 | strcpy(root->release_agent_path, opts.release_agent); | |
977 | kfree(opts.release_agent); | |
978 | } | |
ddbcc7e8 PM |
979 | |
980 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, root); | |
981 | ||
982 | if (IS_ERR(sb)) { | |
983 | kfree(root); | |
984 | return PTR_ERR(sb); | |
985 | } | |
986 | ||
987 | if (sb->s_fs_info != root) { | |
988 | /* Reusing an existing superblock */ | |
989 | BUG_ON(sb->s_root == NULL); | |
990 | kfree(root); | |
991 | root = NULL; | |
992 | } else { | |
993 | /* New superblock */ | |
bd89aabc | 994 | struct cgroup *cgrp = &root->top_cgroup; |
817929ec | 995 | struct inode *inode; |
28fd5dfc | 996 | int i; |
ddbcc7e8 PM |
997 | |
998 | BUG_ON(sb->s_root != NULL); | |
999 | ||
1000 | ret = cgroup_get_rootdir(sb); | |
1001 | if (ret) | |
1002 | goto drop_new_super; | |
817929ec | 1003 | inode = sb->s_root->d_inode; |
ddbcc7e8 | 1004 | |
817929ec | 1005 | mutex_lock(&inode->i_mutex); |
ddbcc7e8 PM |
1006 | mutex_lock(&cgroup_mutex); |
1007 | ||
817929ec PM |
1008 | /* |
1009 | * We're accessing css_set_count without locking | |
1010 | * css_set_lock here, but that's OK - it can only be | |
1011 | * increased by someone holding cgroup_lock, and | |
1012 | * that's us. The worst that can happen is that we | |
1013 | * have some link structures left over | |
1014 | */ | |
1015 | ret = allocate_cg_links(css_set_count, &tmp_cg_links); | |
1016 | if (ret) { | |
1017 | mutex_unlock(&cgroup_mutex); | |
1018 | mutex_unlock(&inode->i_mutex); | |
1019 | goto drop_new_super; | |
1020 | } | |
1021 | ||
ddbcc7e8 PM |
1022 | ret = rebind_subsystems(root, root->subsys_bits); |
1023 | if (ret == -EBUSY) { | |
1024 | mutex_unlock(&cgroup_mutex); | |
817929ec | 1025 | mutex_unlock(&inode->i_mutex); |
20ca9b3f | 1026 | goto free_cg_links; |
ddbcc7e8 PM |
1027 | } |
1028 | ||
1029 | /* EBUSY should be the only error here */ | |
1030 | BUG_ON(ret); | |
1031 | ||
1032 | list_add(&root->root_list, &roots); | |
817929ec | 1033 | root_count++; |
ddbcc7e8 PM |
1034 | |
1035 | sb->s_root->d_fsdata = &root->top_cgroup; | |
1036 | root->top_cgroup.dentry = sb->s_root; | |
1037 | ||
817929ec PM |
1038 | /* Link the top cgroup in this hierarchy into all |
1039 | * the css_set objects */ | |
1040 | write_lock(&css_set_lock); | |
28fd5dfc LZ |
1041 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { |
1042 | struct hlist_head *hhead = &css_set_table[i]; | |
1043 | struct hlist_node *node; | |
817929ec | 1044 | struct css_set *cg; |
28fd5dfc LZ |
1045 | |
1046 | hlist_for_each_entry(cg, node, hhead, hlist) { | |
1047 | struct cg_cgroup_link *link; | |
1048 | ||
1049 | BUG_ON(list_empty(&tmp_cg_links)); | |
1050 | link = list_entry(tmp_cg_links.next, | |
1051 | struct cg_cgroup_link, | |
1052 | cgrp_link_list); | |
1053 | list_del(&link->cgrp_link_list); | |
1054 | link->cg = cg; | |
1055 | list_add(&link->cgrp_link_list, | |
1056 | &root->top_cgroup.css_sets); | |
1057 | list_add(&link->cg_link_list, &cg->cg_links); | |
1058 | } | |
1059 | } | |
817929ec PM |
1060 | write_unlock(&css_set_lock); |
1061 | ||
1062 | free_cg_links(&tmp_cg_links); | |
1063 | ||
bd89aabc PM |
1064 | BUG_ON(!list_empty(&cgrp->sibling)); |
1065 | BUG_ON(!list_empty(&cgrp->children)); | |
ddbcc7e8 PM |
1066 | BUG_ON(root->number_of_cgroups != 1); |
1067 | ||
bd89aabc | 1068 | cgroup_populate_dir(cgrp); |
817929ec | 1069 | mutex_unlock(&inode->i_mutex); |
ddbcc7e8 PM |
1070 | mutex_unlock(&cgroup_mutex); |
1071 | } | |
1072 | ||
1073 | return simple_set_mnt(mnt, sb); | |
1074 | ||
20ca9b3f LZ |
1075 | free_cg_links: |
1076 | free_cg_links(&tmp_cg_links); | |
ddbcc7e8 PM |
1077 | drop_new_super: |
1078 | up_write(&sb->s_umount); | |
1079 | deactivate_super(sb); | |
1080 | return ret; | |
1081 | } | |
1082 | ||
1083 | static void cgroup_kill_sb(struct super_block *sb) { | |
1084 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1085 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 | 1086 | int ret; |
71cbb949 KM |
1087 | struct cg_cgroup_link *link; |
1088 | struct cg_cgroup_link *saved_link; | |
ddbcc7e8 PM |
1089 | |
1090 | BUG_ON(!root); | |
1091 | ||
1092 | BUG_ON(root->number_of_cgroups != 1); | |
bd89aabc PM |
1093 | BUG_ON(!list_empty(&cgrp->children)); |
1094 | BUG_ON(!list_empty(&cgrp->sibling)); | |
ddbcc7e8 PM |
1095 | |
1096 | mutex_lock(&cgroup_mutex); | |
1097 | ||
1098 | /* Rebind all subsystems back to the default hierarchy */ | |
1099 | ret = rebind_subsystems(root, 0); | |
1100 | /* Shouldn't be able to fail ... */ | |
1101 | BUG_ON(ret); | |
1102 | ||
817929ec PM |
1103 | /* |
1104 | * Release all the links from css_sets to this hierarchy's | |
1105 | * root cgroup | |
1106 | */ | |
1107 | write_lock(&css_set_lock); | |
71cbb949 KM |
1108 | |
1109 | list_for_each_entry_safe(link, saved_link, &cgrp->css_sets, | |
1110 | cgrp_link_list) { | |
817929ec | 1111 | list_del(&link->cg_link_list); |
bd89aabc | 1112 | list_del(&link->cgrp_link_list); |
817929ec PM |
1113 | kfree(link); |
1114 | } | |
1115 | write_unlock(&css_set_lock); | |
1116 | ||
1117 | if (!list_empty(&root->root_list)) { | |
ddbcc7e8 | 1118 | list_del(&root->root_list); |
817929ec PM |
1119 | root_count--; |
1120 | } | |
ddbcc7e8 PM |
1121 | mutex_unlock(&cgroup_mutex); |
1122 | ||
1123 | kfree(root); | |
1124 | kill_litter_super(sb); | |
1125 | } | |
1126 | ||
1127 | static struct file_system_type cgroup_fs_type = { | |
1128 | .name = "cgroup", | |
1129 | .get_sb = cgroup_get_sb, | |
1130 | .kill_sb = cgroup_kill_sb, | |
1131 | }; | |
1132 | ||
bd89aabc | 1133 | static inline struct cgroup *__d_cgrp(struct dentry *dentry) |
ddbcc7e8 PM |
1134 | { |
1135 | return dentry->d_fsdata; | |
1136 | } | |
1137 | ||
1138 | static inline struct cftype *__d_cft(struct dentry *dentry) | |
1139 | { | |
1140 | return dentry->d_fsdata; | |
1141 | } | |
1142 | ||
a043e3b2 LZ |
1143 | /** |
1144 | * cgroup_path - generate the path of a cgroup | |
1145 | * @cgrp: the cgroup in question | |
1146 | * @buf: the buffer to write the path into | |
1147 | * @buflen: the length of the buffer | |
1148 | * | |
1149 | * Called with cgroup_mutex held. Writes path of cgroup into buf. | |
ddbcc7e8 PM |
1150 | * Returns 0 on success, -errno on error. |
1151 | */ | |
bd89aabc | 1152 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
ddbcc7e8 PM |
1153 | { |
1154 | char *start; | |
1155 | ||
bd89aabc | 1156 | if (cgrp == dummytop) { |
ddbcc7e8 PM |
1157 | /* |
1158 | * Inactive subsystems have no dentry for their root | |
1159 | * cgroup | |
1160 | */ | |
1161 | strcpy(buf, "/"); | |
1162 | return 0; | |
1163 | } | |
1164 | ||
1165 | start = buf + buflen; | |
1166 | ||
1167 | *--start = '\0'; | |
1168 | for (;;) { | |
bd89aabc | 1169 | int len = cgrp->dentry->d_name.len; |
ddbcc7e8 PM |
1170 | if ((start -= len) < buf) |
1171 | return -ENAMETOOLONG; | |
bd89aabc PM |
1172 | memcpy(start, cgrp->dentry->d_name.name, len); |
1173 | cgrp = cgrp->parent; | |
1174 | if (!cgrp) | |
ddbcc7e8 | 1175 | break; |
bd89aabc | 1176 | if (!cgrp->parent) |
ddbcc7e8 PM |
1177 | continue; |
1178 | if (--start < buf) | |
1179 | return -ENAMETOOLONG; | |
1180 | *start = '/'; | |
1181 | } | |
1182 | memmove(buf, start, buf + buflen - start); | |
1183 | return 0; | |
1184 | } | |
1185 | ||
bbcb81d0 PM |
1186 | /* |
1187 | * Return the first subsystem attached to a cgroup's hierarchy, and | |
1188 | * its subsystem id. | |
1189 | */ | |
1190 | ||
bd89aabc | 1191 | static void get_first_subsys(const struct cgroup *cgrp, |
bbcb81d0 PM |
1192 | struct cgroup_subsys_state **css, int *subsys_id) |
1193 | { | |
bd89aabc | 1194 | const struct cgroupfs_root *root = cgrp->root; |
bbcb81d0 PM |
1195 | const struct cgroup_subsys *test_ss; |
1196 | BUG_ON(list_empty(&root->subsys_list)); | |
1197 | test_ss = list_entry(root->subsys_list.next, | |
1198 | struct cgroup_subsys, sibling); | |
1199 | if (css) { | |
bd89aabc | 1200 | *css = cgrp->subsys[test_ss->subsys_id]; |
bbcb81d0 PM |
1201 | BUG_ON(!*css); |
1202 | } | |
1203 | if (subsys_id) | |
1204 | *subsys_id = test_ss->subsys_id; | |
1205 | } | |
1206 | ||
a043e3b2 LZ |
1207 | /** |
1208 | * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp' | |
1209 | * @cgrp: the cgroup the task is attaching to | |
1210 | * @tsk: the task to be attached | |
bbcb81d0 | 1211 | * |
a043e3b2 LZ |
1212 | * Call holding cgroup_mutex. May take task_lock of |
1213 | * the task 'tsk' during call. | |
bbcb81d0 | 1214 | */ |
956db3ca | 1215 | int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
bbcb81d0 PM |
1216 | { |
1217 | int retval = 0; | |
1218 | struct cgroup_subsys *ss; | |
bd89aabc | 1219 | struct cgroup *oldcgrp; |
817929ec PM |
1220 | struct css_set *cg = tsk->cgroups; |
1221 | struct css_set *newcg; | |
bd89aabc | 1222 | struct cgroupfs_root *root = cgrp->root; |
bbcb81d0 PM |
1223 | int subsys_id; |
1224 | ||
bd89aabc | 1225 | get_first_subsys(cgrp, NULL, &subsys_id); |
bbcb81d0 PM |
1226 | |
1227 | /* Nothing to do if the task is already in that cgroup */ | |
bd89aabc PM |
1228 | oldcgrp = task_cgroup(tsk, subsys_id); |
1229 | if (cgrp == oldcgrp) | |
bbcb81d0 PM |
1230 | return 0; |
1231 | ||
1232 | for_each_subsys(root, ss) { | |
1233 | if (ss->can_attach) { | |
bd89aabc | 1234 | retval = ss->can_attach(ss, cgrp, tsk); |
e18f6318 | 1235 | if (retval) |
bbcb81d0 | 1236 | return retval; |
bbcb81d0 PM |
1237 | } |
1238 | } | |
1239 | ||
817929ec PM |
1240 | /* |
1241 | * Locate or allocate a new css_set for this task, | |
1242 | * based on its final set of cgroups | |
1243 | */ | |
bd89aabc | 1244 | newcg = find_css_set(cg, cgrp); |
e18f6318 | 1245 | if (!newcg) |
817929ec | 1246 | return -ENOMEM; |
817929ec | 1247 | |
bbcb81d0 PM |
1248 | task_lock(tsk); |
1249 | if (tsk->flags & PF_EXITING) { | |
1250 | task_unlock(tsk); | |
817929ec | 1251 | put_css_set(newcg); |
bbcb81d0 PM |
1252 | return -ESRCH; |
1253 | } | |
817929ec | 1254 | rcu_assign_pointer(tsk->cgroups, newcg); |
bbcb81d0 PM |
1255 | task_unlock(tsk); |
1256 | ||
817929ec PM |
1257 | /* Update the css_set linked lists if we're using them */ |
1258 | write_lock(&css_set_lock); | |
1259 | if (!list_empty(&tsk->cg_list)) { | |
1260 | list_del(&tsk->cg_list); | |
1261 | list_add(&tsk->cg_list, &newcg->tasks); | |
1262 | } | |
1263 | write_unlock(&css_set_lock); | |
1264 | ||
bbcb81d0 | 1265 | for_each_subsys(root, ss) { |
e18f6318 | 1266 | if (ss->attach) |
bd89aabc | 1267 | ss->attach(ss, cgrp, oldcgrp, tsk); |
bbcb81d0 | 1268 | } |
bd89aabc | 1269 | set_bit(CGRP_RELEASABLE, &oldcgrp->flags); |
bbcb81d0 | 1270 | synchronize_rcu(); |
817929ec | 1271 | put_css_set(cg); |
bbcb81d0 PM |
1272 | return 0; |
1273 | } | |
1274 | ||
1275 | /* | |
af351026 PM |
1276 | * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex |
1277 | * held. May take task_lock of task | |
bbcb81d0 | 1278 | */ |
af351026 | 1279 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid) |
bbcb81d0 | 1280 | { |
bbcb81d0 | 1281 | struct task_struct *tsk; |
c69e8d9c | 1282 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
1283 | int ret; |
1284 | ||
bbcb81d0 PM |
1285 | if (pid) { |
1286 | rcu_read_lock(); | |
73507f33 | 1287 | tsk = find_task_by_vpid(pid); |
bbcb81d0 PM |
1288 | if (!tsk || tsk->flags & PF_EXITING) { |
1289 | rcu_read_unlock(); | |
1290 | return -ESRCH; | |
1291 | } | |
bbcb81d0 | 1292 | |
c69e8d9c DH |
1293 | tcred = __task_cred(tsk); |
1294 | if (cred->euid && | |
1295 | cred->euid != tcred->uid && | |
1296 | cred->euid != tcred->suid) { | |
1297 | rcu_read_unlock(); | |
bbcb81d0 PM |
1298 | return -EACCES; |
1299 | } | |
c69e8d9c DH |
1300 | get_task_struct(tsk); |
1301 | rcu_read_unlock(); | |
bbcb81d0 PM |
1302 | } else { |
1303 | tsk = current; | |
1304 | get_task_struct(tsk); | |
1305 | } | |
1306 | ||
956db3ca | 1307 | ret = cgroup_attach_task(cgrp, tsk); |
bbcb81d0 PM |
1308 | put_task_struct(tsk); |
1309 | return ret; | |
1310 | } | |
1311 | ||
af351026 PM |
1312 | static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid) |
1313 | { | |
1314 | int ret; | |
1315 | if (!cgroup_lock_live_group(cgrp)) | |
1316 | return -ENODEV; | |
1317 | ret = attach_task_by_pid(cgrp, pid); | |
1318 | cgroup_unlock(); | |
1319 | return ret; | |
1320 | } | |
1321 | ||
ddbcc7e8 | 1322 | /* The various types of files and directories in a cgroup file system */ |
ddbcc7e8 PM |
1323 | enum cgroup_filetype { |
1324 | FILE_ROOT, | |
1325 | FILE_DIR, | |
1326 | FILE_TASKLIST, | |
81a6a5cd | 1327 | FILE_NOTIFY_ON_RELEASE, |
81a6a5cd | 1328 | FILE_RELEASE_AGENT, |
ddbcc7e8 PM |
1329 | }; |
1330 | ||
e788e066 PM |
1331 | /** |
1332 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
1333 | * @cgrp: the cgroup to be checked for liveness | |
1334 | * | |
84eea842 PM |
1335 | * On success, returns true; the lock should be later released with |
1336 | * cgroup_unlock(). On failure returns false with no lock held. | |
e788e066 | 1337 | */ |
84eea842 | 1338 | bool cgroup_lock_live_group(struct cgroup *cgrp) |
e788e066 PM |
1339 | { |
1340 | mutex_lock(&cgroup_mutex); | |
1341 | if (cgroup_is_removed(cgrp)) { | |
1342 | mutex_unlock(&cgroup_mutex); | |
1343 | return false; | |
1344 | } | |
1345 | return true; | |
1346 | } | |
1347 | ||
1348 | static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft, | |
1349 | const char *buffer) | |
1350 | { | |
1351 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
1352 | if (!cgroup_lock_live_group(cgrp)) | |
1353 | return -ENODEV; | |
1354 | strcpy(cgrp->root->release_agent_path, buffer); | |
84eea842 | 1355 | cgroup_unlock(); |
e788e066 PM |
1356 | return 0; |
1357 | } | |
1358 | ||
1359 | static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft, | |
1360 | struct seq_file *seq) | |
1361 | { | |
1362 | if (!cgroup_lock_live_group(cgrp)) | |
1363 | return -ENODEV; | |
1364 | seq_puts(seq, cgrp->root->release_agent_path); | |
1365 | seq_putc(seq, '\n'); | |
84eea842 | 1366 | cgroup_unlock(); |
e788e066 PM |
1367 | return 0; |
1368 | } | |
1369 | ||
84eea842 PM |
1370 | /* A buffer size big enough for numbers or short strings */ |
1371 | #define CGROUP_LOCAL_BUFFER_SIZE 64 | |
1372 | ||
e73d2c61 | 1373 | static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft, |
f4c753b7 PM |
1374 | struct file *file, |
1375 | const char __user *userbuf, | |
1376 | size_t nbytes, loff_t *unused_ppos) | |
355e0c48 | 1377 | { |
84eea842 | 1378 | char buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
355e0c48 | 1379 | int retval = 0; |
355e0c48 PM |
1380 | char *end; |
1381 | ||
1382 | if (!nbytes) | |
1383 | return -EINVAL; | |
1384 | if (nbytes >= sizeof(buffer)) | |
1385 | return -E2BIG; | |
1386 | if (copy_from_user(buffer, userbuf, nbytes)) | |
1387 | return -EFAULT; | |
1388 | ||
1389 | buffer[nbytes] = 0; /* nul-terminate */ | |
b7269dfc | 1390 | strstrip(buffer); |
e73d2c61 PM |
1391 | if (cft->write_u64) { |
1392 | u64 val = simple_strtoull(buffer, &end, 0); | |
1393 | if (*end) | |
1394 | return -EINVAL; | |
1395 | retval = cft->write_u64(cgrp, cft, val); | |
1396 | } else { | |
1397 | s64 val = simple_strtoll(buffer, &end, 0); | |
1398 | if (*end) | |
1399 | return -EINVAL; | |
1400 | retval = cft->write_s64(cgrp, cft, val); | |
1401 | } | |
355e0c48 PM |
1402 | if (!retval) |
1403 | retval = nbytes; | |
1404 | return retval; | |
1405 | } | |
1406 | ||
db3b1497 PM |
1407 | static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft, |
1408 | struct file *file, | |
1409 | const char __user *userbuf, | |
1410 | size_t nbytes, loff_t *unused_ppos) | |
1411 | { | |
84eea842 | 1412 | char local_buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
db3b1497 PM |
1413 | int retval = 0; |
1414 | size_t max_bytes = cft->max_write_len; | |
1415 | char *buffer = local_buffer; | |
1416 | ||
1417 | if (!max_bytes) | |
1418 | max_bytes = sizeof(local_buffer) - 1; | |
1419 | if (nbytes >= max_bytes) | |
1420 | return -E2BIG; | |
1421 | /* Allocate a dynamic buffer if we need one */ | |
1422 | if (nbytes >= sizeof(local_buffer)) { | |
1423 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); | |
1424 | if (buffer == NULL) | |
1425 | return -ENOMEM; | |
1426 | } | |
5a3eb9f6 LZ |
1427 | if (nbytes && copy_from_user(buffer, userbuf, nbytes)) { |
1428 | retval = -EFAULT; | |
1429 | goto out; | |
1430 | } | |
db3b1497 PM |
1431 | |
1432 | buffer[nbytes] = 0; /* nul-terminate */ | |
1433 | strstrip(buffer); | |
1434 | retval = cft->write_string(cgrp, cft, buffer); | |
1435 | if (!retval) | |
1436 | retval = nbytes; | |
5a3eb9f6 | 1437 | out: |
db3b1497 PM |
1438 | if (buffer != local_buffer) |
1439 | kfree(buffer); | |
1440 | return retval; | |
1441 | } | |
1442 | ||
ddbcc7e8 PM |
1443 | static ssize_t cgroup_file_write(struct file *file, const char __user *buf, |
1444 | size_t nbytes, loff_t *ppos) | |
1445 | { | |
1446 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 1447 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 1448 | |
8dc4f3e1 | 1449 | if (!cft || cgroup_is_removed(cgrp)) |
ddbcc7e8 | 1450 | return -ENODEV; |
355e0c48 | 1451 | if (cft->write) |
bd89aabc | 1452 | return cft->write(cgrp, cft, file, buf, nbytes, ppos); |
e73d2c61 PM |
1453 | if (cft->write_u64 || cft->write_s64) |
1454 | return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos); | |
db3b1497 PM |
1455 | if (cft->write_string) |
1456 | return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos); | |
d447ea2f PE |
1457 | if (cft->trigger) { |
1458 | int ret = cft->trigger(cgrp, (unsigned int)cft->private); | |
1459 | return ret ? ret : nbytes; | |
1460 | } | |
355e0c48 | 1461 | return -EINVAL; |
ddbcc7e8 PM |
1462 | } |
1463 | ||
f4c753b7 PM |
1464 | static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft, |
1465 | struct file *file, | |
1466 | char __user *buf, size_t nbytes, | |
1467 | loff_t *ppos) | |
ddbcc7e8 | 1468 | { |
84eea842 | 1469 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
f4c753b7 | 1470 | u64 val = cft->read_u64(cgrp, cft); |
ddbcc7e8 PM |
1471 | int len = sprintf(tmp, "%llu\n", (unsigned long long) val); |
1472 | ||
1473 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
1474 | } | |
1475 | ||
e73d2c61 PM |
1476 | static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft, |
1477 | struct file *file, | |
1478 | char __user *buf, size_t nbytes, | |
1479 | loff_t *ppos) | |
1480 | { | |
84eea842 | 1481 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
e73d2c61 PM |
1482 | s64 val = cft->read_s64(cgrp, cft); |
1483 | int len = sprintf(tmp, "%lld\n", (long long) val); | |
1484 | ||
1485 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
1486 | } | |
1487 | ||
ddbcc7e8 PM |
1488 | static ssize_t cgroup_file_read(struct file *file, char __user *buf, |
1489 | size_t nbytes, loff_t *ppos) | |
1490 | { | |
1491 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 1492 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 1493 | |
8dc4f3e1 | 1494 | if (!cft || cgroup_is_removed(cgrp)) |
ddbcc7e8 PM |
1495 | return -ENODEV; |
1496 | ||
1497 | if (cft->read) | |
bd89aabc | 1498 | return cft->read(cgrp, cft, file, buf, nbytes, ppos); |
f4c753b7 PM |
1499 | if (cft->read_u64) |
1500 | return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos); | |
e73d2c61 PM |
1501 | if (cft->read_s64) |
1502 | return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos); | |
ddbcc7e8 PM |
1503 | return -EINVAL; |
1504 | } | |
1505 | ||
91796569 PM |
1506 | /* |
1507 | * seqfile ops/methods for returning structured data. Currently just | |
1508 | * supports string->u64 maps, but can be extended in future. | |
1509 | */ | |
1510 | ||
1511 | struct cgroup_seqfile_state { | |
1512 | struct cftype *cft; | |
1513 | struct cgroup *cgroup; | |
1514 | }; | |
1515 | ||
1516 | static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value) | |
1517 | { | |
1518 | struct seq_file *sf = cb->state; | |
1519 | return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value); | |
1520 | } | |
1521 | ||
1522 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) | |
1523 | { | |
1524 | struct cgroup_seqfile_state *state = m->private; | |
1525 | struct cftype *cft = state->cft; | |
29486df3 SH |
1526 | if (cft->read_map) { |
1527 | struct cgroup_map_cb cb = { | |
1528 | .fill = cgroup_map_add, | |
1529 | .state = m, | |
1530 | }; | |
1531 | return cft->read_map(state->cgroup, cft, &cb); | |
1532 | } | |
1533 | return cft->read_seq_string(state->cgroup, cft, m); | |
91796569 PM |
1534 | } |
1535 | ||
96930a63 | 1536 | static int cgroup_seqfile_release(struct inode *inode, struct file *file) |
91796569 PM |
1537 | { |
1538 | struct seq_file *seq = file->private_data; | |
1539 | kfree(seq->private); | |
1540 | return single_release(inode, file); | |
1541 | } | |
1542 | ||
1543 | static struct file_operations cgroup_seqfile_operations = { | |
1544 | .read = seq_read, | |
e788e066 | 1545 | .write = cgroup_file_write, |
91796569 PM |
1546 | .llseek = seq_lseek, |
1547 | .release = cgroup_seqfile_release, | |
1548 | }; | |
1549 | ||
ddbcc7e8 PM |
1550 | static int cgroup_file_open(struct inode *inode, struct file *file) |
1551 | { | |
1552 | int err; | |
1553 | struct cftype *cft; | |
1554 | ||
1555 | err = generic_file_open(inode, file); | |
1556 | if (err) | |
1557 | return err; | |
1558 | ||
1559 | cft = __d_cft(file->f_dentry); | |
1560 | if (!cft) | |
1561 | return -ENODEV; | |
29486df3 | 1562 | if (cft->read_map || cft->read_seq_string) { |
91796569 PM |
1563 | struct cgroup_seqfile_state *state = |
1564 | kzalloc(sizeof(*state), GFP_USER); | |
1565 | if (!state) | |
1566 | return -ENOMEM; | |
1567 | state->cft = cft; | |
1568 | state->cgroup = __d_cgrp(file->f_dentry->d_parent); | |
1569 | file->f_op = &cgroup_seqfile_operations; | |
1570 | err = single_open(file, cgroup_seqfile_show, state); | |
1571 | if (err < 0) | |
1572 | kfree(state); | |
1573 | } else if (cft->open) | |
ddbcc7e8 PM |
1574 | err = cft->open(inode, file); |
1575 | else | |
1576 | err = 0; | |
1577 | ||
1578 | return err; | |
1579 | } | |
1580 | ||
1581 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
1582 | { | |
1583 | struct cftype *cft = __d_cft(file->f_dentry); | |
1584 | if (cft->release) | |
1585 | return cft->release(inode, file); | |
1586 | return 0; | |
1587 | } | |
1588 | ||
1589 | /* | |
1590 | * cgroup_rename - Only allow simple rename of directories in place. | |
1591 | */ | |
1592 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1593 | struct inode *new_dir, struct dentry *new_dentry) | |
1594 | { | |
1595 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) | |
1596 | return -ENOTDIR; | |
1597 | if (new_dentry->d_inode) | |
1598 | return -EEXIST; | |
1599 | if (old_dir != new_dir) | |
1600 | return -EIO; | |
1601 | return simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
1602 | } | |
1603 | ||
1604 | static struct file_operations cgroup_file_operations = { | |
1605 | .read = cgroup_file_read, | |
1606 | .write = cgroup_file_write, | |
1607 | .llseek = generic_file_llseek, | |
1608 | .open = cgroup_file_open, | |
1609 | .release = cgroup_file_release, | |
1610 | }; | |
1611 | ||
1612 | static struct inode_operations cgroup_dir_inode_operations = { | |
1613 | .lookup = simple_lookup, | |
1614 | .mkdir = cgroup_mkdir, | |
1615 | .rmdir = cgroup_rmdir, | |
1616 | .rename = cgroup_rename, | |
1617 | }; | |
1618 | ||
1619 | static int cgroup_create_file(struct dentry *dentry, int mode, | |
1620 | struct super_block *sb) | |
1621 | { | |
1622 | static struct dentry_operations cgroup_dops = { | |
1623 | .d_iput = cgroup_diput, | |
1624 | }; | |
1625 | ||
1626 | struct inode *inode; | |
1627 | ||
1628 | if (!dentry) | |
1629 | return -ENOENT; | |
1630 | if (dentry->d_inode) | |
1631 | return -EEXIST; | |
1632 | ||
1633 | inode = cgroup_new_inode(mode, sb); | |
1634 | if (!inode) | |
1635 | return -ENOMEM; | |
1636 | ||
1637 | if (S_ISDIR(mode)) { | |
1638 | inode->i_op = &cgroup_dir_inode_operations; | |
1639 | inode->i_fop = &simple_dir_operations; | |
1640 | ||
1641 | /* start off with i_nlink == 2 (for "." entry) */ | |
1642 | inc_nlink(inode); | |
1643 | ||
1644 | /* start with the directory inode held, so that we can | |
1645 | * populate it without racing with another mkdir */ | |
817929ec | 1646 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); |
ddbcc7e8 PM |
1647 | } else if (S_ISREG(mode)) { |
1648 | inode->i_size = 0; | |
1649 | inode->i_fop = &cgroup_file_operations; | |
1650 | } | |
1651 | dentry->d_op = &cgroup_dops; | |
1652 | d_instantiate(dentry, inode); | |
1653 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1654 | return 0; | |
1655 | } | |
1656 | ||
1657 | /* | |
a043e3b2 LZ |
1658 | * cgroup_create_dir - create a directory for an object. |
1659 | * @cgrp: the cgroup we create the directory for. It must have a valid | |
1660 | * ->parent field. And we are going to fill its ->dentry field. | |
1661 | * @dentry: dentry of the new cgroup | |
1662 | * @mode: mode to set on new directory. | |
ddbcc7e8 | 1663 | */ |
bd89aabc | 1664 | static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry, |
ddbcc7e8 PM |
1665 | int mode) |
1666 | { | |
1667 | struct dentry *parent; | |
1668 | int error = 0; | |
1669 | ||
bd89aabc PM |
1670 | parent = cgrp->parent->dentry; |
1671 | error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb); | |
ddbcc7e8 | 1672 | if (!error) { |
bd89aabc | 1673 | dentry->d_fsdata = cgrp; |
ddbcc7e8 | 1674 | inc_nlink(parent->d_inode); |
bd89aabc | 1675 | cgrp->dentry = dentry; |
ddbcc7e8 PM |
1676 | dget(dentry); |
1677 | } | |
1678 | dput(dentry); | |
1679 | ||
1680 | return error; | |
1681 | } | |
1682 | ||
bd89aabc | 1683 | int cgroup_add_file(struct cgroup *cgrp, |
ddbcc7e8 PM |
1684 | struct cgroup_subsys *subsys, |
1685 | const struct cftype *cft) | |
1686 | { | |
bd89aabc | 1687 | struct dentry *dir = cgrp->dentry; |
ddbcc7e8 PM |
1688 | struct dentry *dentry; |
1689 | int error; | |
1690 | ||
1691 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; | |
bd89aabc | 1692 | if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) { |
ddbcc7e8 PM |
1693 | strcpy(name, subsys->name); |
1694 | strcat(name, "."); | |
1695 | } | |
1696 | strcat(name, cft->name); | |
1697 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); | |
1698 | dentry = lookup_one_len(name, dir, strlen(name)); | |
1699 | if (!IS_ERR(dentry)) { | |
1700 | error = cgroup_create_file(dentry, 0644 | S_IFREG, | |
bd89aabc | 1701 | cgrp->root->sb); |
ddbcc7e8 PM |
1702 | if (!error) |
1703 | dentry->d_fsdata = (void *)cft; | |
1704 | dput(dentry); | |
1705 | } else | |
1706 | error = PTR_ERR(dentry); | |
1707 | return error; | |
1708 | } | |
1709 | ||
bd89aabc | 1710 | int cgroup_add_files(struct cgroup *cgrp, |
ddbcc7e8 PM |
1711 | struct cgroup_subsys *subsys, |
1712 | const struct cftype cft[], | |
1713 | int count) | |
1714 | { | |
1715 | int i, err; | |
1716 | for (i = 0; i < count; i++) { | |
bd89aabc | 1717 | err = cgroup_add_file(cgrp, subsys, &cft[i]); |
ddbcc7e8 PM |
1718 | if (err) |
1719 | return err; | |
1720 | } | |
1721 | return 0; | |
1722 | } | |
1723 | ||
a043e3b2 LZ |
1724 | /** |
1725 | * cgroup_task_count - count the number of tasks in a cgroup. | |
1726 | * @cgrp: the cgroup in question | |
1727 | * | |
1728 | * Return the number of tasks in the cgroup. | |
1729 | */ | |
bd89aabc | 1730 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
1731 | { |
1732 | int count = 0; | |
71cbb949 | 1733 | struct cg_cgroup_link *link; |
817929ec PM |
1734 | |
1735 | read_lock(&css_set_lock); | |
71cbb949 | 1736 | list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) { |
146aa1bd | 1737 | count += atomic_read(&link->cg->refcount); |
817929ec PM |
1738 | } |
1739 | read_unlock(&css_set_lock); | |
bbcb81d0 PM |
1740 | return count; |
1741 | } | |
1742 | ||
817929ec PM |
1743 | /* |
1744 | * Advance a list_head iterator. The iterator should be positioned at | |
1745 | * the start of a css_set | |
1746 | */ | |
bd89aabc | 1747 | static void cgroup_advance_iter(struct cgroup *cgrp, |
817929ec PM |
1748 | struct cgroup_iter *it) |
1749 | { | |
1750 | struct list_head *l = it->cg_link; | |
1751 | struct cg_cgroup_link *link; | |
1752 | struct css_set *cg; | |
1753 | ||
1754 | /* Advance to the next non-empty css_set */ | |
1755 | do { | |
1756 | l = l->next; | |
bd89aabc | 1757 | if (l == &cgrp->css_sets) { |
817929ec PM |
1758 | it->cg_link = NULL; |
1759 | return; | |
1760 | } | |
bd89aabc | 1761 | link = list_entry(l, struct cg_cgroup_link, cgrp_link_list); |
817929ec PM |
1762 | cg = link->cg; |
1763 | } while (list_empty(&cg->tasks)); | |
1764 | it->cg_link = l; | |
1765 | it->task = cg->tasks.next; | |
1766 | } | |
1767 | ||
31a7df01 CW |
1768 | /* |
1769 | * To reduce the fork() overhead for systems that are not actually | |
1770 | * using their cgroups capability, we don't maintain the lists running | |
1771 | * through each css_set to its tasks until we see the list actually | |
1772 | * used - in other words after the first call to cgroup_iter_start(). | |
1773 | * | |
1774 | * The tasklist_lock is not held here, as do_each_thread() and | |
1775 | * while_each_thread() are protected by RCU. | |
1776 | */ | |
3df91fe3 | 1777 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
1778 | { |
1779 | struct task_struct *p, *g; | |
1780 | write_lock(&css_set_lock); | |
1781 | use_task_css_set_links = 1; | |
1782 | do_each_thread(g, p) { | |
1783 | task_lock(p); | |
0e04388f LZ |
1784 | /* |
1785 | * We should check if the process is exiting, otherwise | |
1786 | * it will race with cgroup_exit() in that the list | |
1787 | * entry won't be deleted though the process has exited. | |
1788 | */ | |
1789 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
31a7df01 CW |
1790 | list_add(&p->cg_list, &p->cgroups->tasks); |
1791 | task_unlock(p); | |
1792 | } while_each_thread(g, p); | |
1793 | write_unlock(&css_set_lock); | |
1794 | } | |
1795 | ||
bd89aabc | 1796 | void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it) |
817929ec PM |
1797 | { |
1798 | /* | |
1799 | * The first time anyone tries to iterate across a cgroup, | |
1800 | * we need to enable the list linking each css_set to its | |
1801 | * tasks, and fix up all existing tasks. | |
1802 | */ | |
31a7df01 CW |
1803 | if (!use_task_css_set_links) |
1804 | cgroup_enable_task_cg_lists(); | |
1805 | ||
817929ec | 1806 | read_lock(&css_set_lock); |
bd89aabc PM |
1807 | it->cg_link = &cgrp->css_sets; |
1808 | cgroup_advance_iter(cgrp, it); | |
817929ec PM |
1809 | } |
1810 | ||
bd89aabc | 1811 | struct task_struct *cgroup_iter_next(struct cgroup *cgrp, |
817929ec PM |
1812 | struct cgroup_iter *it) |
1813 | { | |
1814 | struct task_struct *res; | |
1815 | struct list_head *l = it->task; | |
1816 | ||
1817 | /* If the iterator cg is NULL, we have no tasks */ | |
1818 | if (!it->cg_link) | |
1819 | return NULL; | |
1820 | res = list_entry(l, struct task_struct, cg_list); | |
1821 | /* Advance iterator to find next entry */ | |
1822 | l = l->next; | |
1823 | if (l == &res->cgroups->tasks) { | |
1824 | /* We reached the end of this task list - move on to | |
1825 | * the next cg_cgroup_link */ | |
bd89aabc | 1826 | cgroup_advance_iter(cgrp, it); |
817929ec PM |
1827 | } else { |
1828 | it->task = l; | |
1829 | } | |
1830 | return res; | |
1831 | } | |
1832 | ||
bd89aabc | 1833 | void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it) |
817929ec PM |
1834 | { |
1835 | read_unlock(&css_set_lock); | |
1836 | } | |
1837 | ||
31a7df01 CW |
1838 | static inline int started_after_time(struct task_struct *t1, |
1839 | struct timespec *time, | |
1840 | struct task_struct *t2) | |
1841 | { | |
1842 | int start_diff = timespec_compare(&t1->start_time, time); | |
1843 | if (start_diff > 0) { | |
1844 | return 1; | |
1845 | } else if (start_diff < 0) { | |
1846 | return 0; | |
1847 | } else { | |
1848 | /* | |
1849 | * Arbitrarily, if two processes started at the same | |
1850 | * time, we'll say that the lower pointer value | |
1851 | * started first. Note that t2 may have exited by now | |
1852 | * so this may not be a valid pointer any longer, but | |
1853 | * that's fine - it still serves to distinguish | |
1854 | * between two tasks started (effectively) simultaneously. | |
1855 | */ | |
1856 | return t1 > t2; | |
1857 | } | |
1858 | } | |
1859 | ||
1860 | /* | |
1861 | * This function is a callback from heap_insert() and is used to order | |
1862 | * the heap. | |
1863 | * In this case we order the heap in descending task start time. | |
1864 | */ | |
1865 | static inline int started_after(void *p1, void *p2) | |
1866 | { | |
1867 | struct task_struct *t1 = p1; | |
1868 | struct task_struct *t2 = p2; | |
1869 | return started_after_time(t1, &t2->start_time, t2); | |
1870 | } | |
1871 | ||
1872 | /** | |
1873 | * cgroup_scan_tasks - iterate though all the tasks in a cgroup | |
1874 | * @scan: struct cgroup_scanner containing arguments for the scan | |
1875 | * | |
1876 | * Arguments include pointers to callback functions test_task() and | |
1877 | * process_task(). | |
1878 | * Iterate through all the tasks in a cgroup, calling test_task() for each, | |
1879 | * and if it returns true, call process_task() for it also. | |
1880 | * The test_task pointer may be NULL, meaning always true (select all tasks). | |
1881 | * Effectively duplicates cgroup_iter_{start,next,end}() | |
1882 | * but does not lock css_set_lock for the call to process_task(). | |
1883 | * The struct cgroup_scanner may be embedded in any structure of the caller's | |
1884 | * creation. | |
1885 | * It is guaranteed that process_task() will act on every task that | |
1886 | * is a member of the cgroup for the duration of this call. This | |
1887 | * function may or may not call process_task() for tasks that exit | |
1888 | * or move to a different cgroup during the call, or are forked or | |
1889 | * move into the cgroup during the call. | |
1890 | * | |
1891 | * Note that test_task() may be called with locks held, and may in some | |
1892 | * situations be called multiple times for the same task, so it should | |
1893 | * be cheap. | |
1894 | * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been | |
1895 | * pre-allocated and will be used for heap operations (and its "gt" member will | |
1896 | * be overwritten), else a temporary heap will be used (allocation of which | |
1897 | * may cause this function to fail). | |
1898 | */ | |
1899 | int cgroup_scan_tasks(struct cgroup_scanner *scan) | |
1900 | { | |
1901 | int retval, i; | |
1902 | struct cgroup_iter it; | |
1903 | struct task_struct *p, *dropped; | |
1904 | /* Never dereference latest_task, since it's not refcounted */ | |
1905 | struct task_struct *latest_task = NULL; | |
1906 | struct ptr_heap tmp_heap; | |
1907 | struct ptr_heap *heap; | |
1908 | struct timespec latest_time = { 0, 0 }; | |
1909 | ||
1910 | if (scan->heap) { | |
1911 | /* The caller supplied our heap and pre-allocated its memory */ | |
1912 | heap = scan->heap; | |
1913 | heap->gt = &started_after; | |
1914 | } else { | |
1915 | /* We need to allocate our own heap memory */ | |
1916 | heap = &tmp_heap; | |
1917 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
1918 | if (retval) | |
1919 | /* cannot allocate the heap */ | |
1920 | return retval; | |
1921 | } | |
1922 | ||
1923 | again: | |
1924 | /* | |
1925 | * Scan tasks in the cgroup, using the scanner's "test_task" callback | |
1926 | * to determine which are of interest, and using the scanner's | |
1927 | * "process_task" callback to process any of them that need an update. | |
1928 | * Since we don't want to hold any locks during the task updates, | |
1929 | * gather tasks to be processed in a heap structure. | |
1930 | * The heap is sorted by descending task start time. | |
1931 | * If the statically-sized heap fills up, we overflow tasks that | |
1932 | * started later, and in future iterations only consider tasks that | |
1933 | * started after the latest task in the previous pass. This | |
1934 | * guarantees forward progress and that we don't miss any tasks. | |
1935 | */ | |
1936 | heap->size = 0; | |
1937 | cgroup_iter_start(scan->cg, &it); | |
1938 | while ((p = cgroup_iter_next(scan->cg, &it))) { | |
1939 | /* | |
1940 | * Only affect tasks that qualify per the caller's callback, | |
1941 | * if he provided one | |
1942 | */ | |
1943 | if (scan->test_task && !scan->test_task(p, scan)) | |
1944 | continue; | |
1945 | /* | |
1946 | * Only process tasks that started after the last task | |
1947 | * we processed | |
1948 | */ | |
1949 | if (!started_after_time(p, &latest_time, latest_task)) | |
1950 | continue; | |
1951 | dropped = heap_insert(heap, p); | |
1952 | if (dropped == NULL) { | |
1953 | /* | |
1954 | * The new task was inserted; the heap wasn't | |
1955 | * previously full | |
1956 | */ | |
1957 | get_task_struct(p); | |
1958 | } else if (dropped != p) { | |
1959 | /* | |
1960 | * The new task was inserted, and pushed out a | |
1961 | * different task | |
1962 | */ | |
1963 | get_task_struct(p); | |
1964 | put_task_struct(dropped); | |
1965 | } | |
1966 | /* | |
1967 | * Else the new task was newer than anything already in | |
1968 | * the heap and wasn't inserted | |
1969 | */ | |
1970 | } | |
1971 | cgroup_iter_end(scan->cg, &it); | |
1972 | ||
1973 | if (heap->size) { | |
1974 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 1975 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 1976 | if (i == 0) { |
4fe91d51 PJ |
1977 | latest_time = q->start_time; |
1978 | latest_task = q; | |
31a7df01 CW |
1979 | } |
1980 | /* Process the task per the caller's callback */ | |
4fe91d51 PJ |
1981 | scan->process_task(q, scan); |
1982 | put_task_struct(q); | |
31a7df01 CW |
1983 | } |
1984 | /* | |
1985 | * If we had to process any tasks at all, scan again | |
1986 | * in case some of them were in the middle of forking | |
1987 | * children that didn't get processed. | |
1988 | * Not the most efficient way to do it, but it avoids | |
1989 | * having to take callback_mutex in the fork path | |
1990 | */ | |
1991 | goto again; | |
1992 | } | |
1993 | if (heap == &tmp_heap) | |
1994 | heap_free(&tmp_heap); | |
1995 | return 0; | |
1996 | } | |
1997 | ||
bbcb81d0 PM |
1998 | /* |
1999 | * Stuff for reading the 'tasks' file. | |
2000 | * | |
2001 | * Reading this file can return large amounts of data if a cgroup has | |
2002 | * *lots* of attached tasks. So it may need several calls to read(), | |
2003 | * but we cannot guarantee that the information we produce is correct | |
2004 | * unless we produce it entirely atomically. | |
2005 | * | |
bbcb81d0 | 2006 | */ |
bbcb81d0 PM |
2007 | |
2008 | /* | |
2009 | * Load into 'pidarray' up to 'npids' of the tasks using cgroup | |
bd89aabc | 2010 | * 'cgrp'. Return actual number of pids loaded. No need to |
bbcb81d0 PM |
2011 | * task_lock(p) when reading out p->cgroup, since we're in an RCU |
2012 | * read section, so the css_set can't go away, and is | |
2013 | * immutable after creation. | |
2014 | */ | |
bd89aabc | 2015 | static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cgrp) |
bbcb81d0 PM |
2016 | { |
2017 | int n = 0; | |
817929ec PM |
2018 | struct cgroup_iter it; |
2019 | struct task_struct *tsk; | |
bd89aabc PM |
2020 | cgroup_iter_start(cgrp, &it); |
2021 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
817929ec PM |
2022 | if (unlikely(n == npids)) |
2023 | break; | |
73507f33 | 2024 | pidarray[n++] = task_pid_vnr(tsk); |
817929ec | 2025 | } |
bd89aabc | 2026 | cgroup_iter_end(cgrp, &it); |
bbcb81d0 PM |
2027 | return n; |
2028 | } | |
2029 | ||
846c7bb0 | 2030 | /** |
a043e3b2 | 2031 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
2032 | * @stats: cgroupstats to fill information into |
2033 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
2034 | * been requested. | |
a043e3b2 LZ |
2035 | * |
2036 | * Build and fill cgroupstats so that taskstats can export it to user | |
2037 | * space. | |
846c7bb0 BS |
2038 | */ |
2039 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
2040 | { | |
2041 | int ret = -EINVAL; | |
bd89aabc | 2042 | struct cgroup *cgrp; |
846c7bb0 BS |
2043 | struct cgroup_iter it; |
2044 | struct task_struct *tsk; | |
33d283be | 2045 | |
846c7bb0 | 2046 | /* |
33d283be LZ |
2047 | * Validate dentry by checking the superblock operations, |
2048 | * and make sure it's a directory. | |
846c7bb0 | 2049 | */ |
33d283be LZ |
2050 | if (dentry->d_sb->s_op != &cgroup_ops || |
2051 | !S_ISDIR(dentry->d_inode->i_mode)) | |
846c7bb0 BS |
2052 | goto err; |
2053 | ||
2054 | ret = 0; | |
bd89aabc | 2055 | cgrp = dentry->d_fsdata; |
846c7bb0 BS |
2056 | rcu_read_lock(); |
2057 | ||
bd89aabc PM |
2058 | cgroup_iter_start(cgrp, &it); |
2059 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
846c7bb0 BS |
2060 | switch (tsk->state) { |
2061 | case TASK_RUNNING: | |
2062 | stats->nr_running++; | |
2063 | break; | |
2064 | case TASK_INTERRUPTIBLE: | |
2065 | stats->nr_sleeping++; | |
2066 | break; | |
2067 | case TASK_UNINTERRUPTIBLE: | |
2068 | stats->nr_uninterruptible++; | |
2069 | break; | |
2070 | case TASK_STOPPED: | |
2071 | stats->nr_stopped++; | |
2072 | break; | |
2073 | default: | |
2074 | if (delayacct_is_task_waiting_on_io(tsk)) | |
2075 | stats->nr_io_wait++; | |
2076 | break; | |
2077 | } | |
2078 | } | |
bd89aabc | 2079 | cgroup_iter_end(cgrp, &it); |
846c7bb0 BS |
2080 | |
2081 | rcu_read_unlock(); | |
2082 | err: | |
2083 | return ret; | |
2084 | } | |
2085 | ||
bbcb81d0 PM |
2086 | static int cmppid(const void *a, const void *b) |
2087 | { | |
2088 | return *(pid_t *)a - *(pid_t *)b; | |
2089 | } | |
2090 | ||
cc31edce | 2091 | |
bbcb81d0 | 2092 | /* |
cc31edce PM |
2093 | * seq_file methods for the "tasks" file. The seq_file position is the |
2094 | * next pid to display; the seq_file iterator is a pointer to the pid | |
2095 | * in the cgroup->tasks_pids array. | |
bbcb81d0 | 2096 | */ |
cc31edce PM |
2097 | |
2098 | static void *cgroup_tasks_start(struct seq_file *s, loff_t *pos) | |
bbcb81d0 | 2099 | { |
cc31edce PM |
2100 | /* |
2101 | * Initially we receive a position value that corresponds to | |
2102 | * one more than the last pid shown (or 0 on the first call or | |
2103 | * after a seek to the start). Use a binary-search to find the | |
2104 | * next pid to display, if any | |
2105 | */ | |
2106 | struct cgroup *cgrp = s->private; | |
2107 | int index = 0, pid = *pos; | |
2108 | int *iter; | |
2109 | ||
2110 | down_read(&cgrp->pids_mutex); | |
2111 | if (pid) { | |
2112 | int end = cgrp->pids_length; | |
20777766 | 2113 | |
cc31edce PM |
2114 | while (index < end) { |
2115 | int mid = (index + end) / 2; | |
2116 | if (cgrp->tasks_pids[mid] == pid) { | |
2117 | index = mid; | |
2118 | break; | |
2119 | } else if (cgrp->tasks_pids[mid] <= pid) | |
2120 | index = mid + 1; | |
2121 | else | |
2122 | end = mid; | |
2123 | } | |
2124 | } | |
2125 | /* If we're off the end of the array, we're done */ | |
2126 | if (index >= cgrp->pids_length) | |
2127 | return NULL; | |
2128 | /* Update the abstract position to be the actual pid that we found */ | |
2129 | iter = cgrp->tasks_pids + index; | |
2130 | *pos = *iter; | |
2131 | return iter; | |
2132 | } | |
2133 | ||
2134 | static void cgroup_tasks_stop(struct seq_file *s, void *v) | |
2135 | { | |
2136 | struct cgroup *cgrp = s->private; | |
2137 | up_read(&cgrp->pids_mutex); | |
2138 | } | |
2139 | ||
2140 | static void *cgroup_tasks_next(struct seq_file *s, void *v, loff_t *pos) | |
2141 | { | |
2142 | struct cgroup *cgrp = s->private; | |
2143 | int *p = v; | |
2144 | int *end = cgrp->tasks_pids + cgrp->pids_length; | |
2145 | ||
2146 | /* | |
2147 | * Advance to the next pid in the array. If this goes off the | |
2148 | * end, we're done | |
2149 | */ | |
2150 | p++; | |
2151 | if (p >= end) { | |
2152 | return NULL; | |
2153 | } else { | |
2154 | *pos = *p; | |
2155 | return p; | |
2156 | } | |
2157 | } | |
2158 | ||
2159 | static int cgroup_tasks_show(struct seq_file *s, void *v) | |
2160 | { | |
2161 | return seq_printf(s, "%d\n", *(int *)v); | |
2162 | } | |
bbcb81d0 | 2163 | |
cc31edce PM |
2164 | static struct seq_operations cgroup_tasks_seq_operations = { |
2165 | .start = cgroup_tasks_start, | |
2166 | .stop = cgroup_tasks_stop, | |
2167 | .next = cgroup_tasks_next, | |
2168 | .show = cgroup_tasks_show, | |
2169 | }; | |
2170 | ||
2171 | static void release_cgroup_pid_array(struct cgroup *cgrp) | |
2172 | { | |
2173 | down_write(&cgrp->pids_mutex); | |
2174 | BUG_ON(!cgrp->pids_use_count); | |
2175 | if (!--cgrp->pids_use_count) { | |
2176 | kfree(cgrp->tasks_pids); | |
2177 | cgrp->tasks_pids = NULL; | |
2178 | cgrp->pids_length = 0; | |
2179 | } | |
2180 | up_write(&cgrp->pids_mutex); | |
bbcb81d0 PM |
2181 | } |
2182 | ||
cc31edce PM |
2183 | static int cgroup_tasks_release(struct inode *inode, struct file *file) |
2184 | { | |
2185 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); | |
2186 | ||
2187 | if (!(file->f_mode & FMODE_READ)) | |
2188 | return 0; | |
2189 | ||
2190 | release_cgroup_pid_array(cgrp); | |
2191 | return seq_release(inode, file); | |
2192 | } | |
2193 | ||
2194 | static struct file_operations cgroup_tasks_operations = { | |
2195 | .read = seq_read, | |
2196 | .llseek = seq_lseek, | |
2197 | .write = cgroup_file_write, | |
2198 | .release = cgroup_tasks_release, | |
2199 | }; | |
2200 | ||
bbcb81d0 | 2201 | /* |
cc31edce | 2202 | * Handle an open on 'tasks' file. Prepare an array containing the |
bbcb81d0 | 2203 | * process id's of tasks currently attached to the cgroup being opened. |
bbcb81d0 | 2204 | */ |
cc31edce | 2205 | |
bbcb81d0 PM |
2206 | static int cgroup_tasks_open(struct inode *unused, struct file *file) |
2207 | { | |
bd89aabc | 2208 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
bbcb81d0 PM |
2209 | pid_t *pidarray; |
2210 | int npids; | |
cc31edce | 2211 | int retval; |
bbcb81d0 | 2212 | |
cc31edce | 2213 | /* Nothing to do for write-only files */ |
bbcb81d0 PM |
2214 | if (!(file->f_mode & FMODE_READ)) |
2215 | return 0; | |
2216 | ||
bbcb81d0 PM |
2217 | /* |
2218 | * If cgroup gets more users after we read count, we won't have | |
2219 | * enough space - tough. This race is indistinguishable to the | |
2220 | * caller from the case that the additional cgroup users didn't | |
2221 | * show up until sometime later on. | |
2222 | */ | |
bd89aabc | 2223 | npids = cgroup_task_count(cgrp); |
cc31edce PM |
2224 | pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL); |
2225 | if (!pidarray) | |
2226 | return -ENOMEM; | |
2227 | npids = pid_array_load(pidarray, npids, cgrp); | |
2228 | sort(pidarray, npids, sizeof(pid_t), cmppid, NULL); | |
bbcb81d0 | 2229 | |
cc31edce PM |
2230 | /* |
2231 | * Store the array in the cgroup, freeing the old | |
2232 | * array if necessary | |
2233 | */ | |
2234 | down_write(&cgrp->pids_mutex); | |
2235 | kfree(cgrp->tasks_pids); | |
2236 | cgrp->tasks_pids = pidarray; | |
2237 | cgrp->pids_length = npids; | |
2238 | cgrp->pids_use_count++; | |
2239 | up_write(&cgrp->pids_mutex); | |
2240 | ||
2241 | file->f_op = &cgroup_tasks_operations; | |
2242 | ||
2243 | retval = seq_open(file, &cgroup_tasks_seq_operations); | |
2244 | if (retval) { | |
2245 | release_cgroup_pid_array(cgrp); | |
2246 | return retval; | |
bbcb81d0 | 2247 | } |
cc31edce | 2248 | ((struct seq_file *)file->private_data)->private = cgrp; |
bbcb81d0 PM |
2249 | return 0; |
2250 | } | |
2251 | ||
bd89aabc | 2252 | static u64 cgroup_read_notify_on_release(struct cgroup *cgrp, |
81a6a5cd PM |
2253 | struct cftype *cft) |
2254 | { | |
bd89aabc | 2255 | return notify_on_release(cgrp); |
81a6a5cd PM |
2256 | } |
2257 | ||
6379c106 PM |
2258 | static int cgroup_write_notify_on_release(struct cgroup *cgrp, |
2259 | struct cftype *cft, | |
2260 | u64 val) | |
2261 | { | |
2262 | clear_bit(CGRP_RELEASABLE, &cgrp->flags); | |
2263 | if (val) | |
2264 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
2265 | else | |
2266 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
2267 | return 0; | |
2268 | } | |
2269 | ||
bbcb81d0 PM |
2270 | /* |
2271 | * for the common functions, 'private' gives the type of file | |
2272 | */ | |
81a6a5cd PM |
2273 | static struct cftype files[] = { |
2274 | { | |
2275 | .name = "tasks", | |
2276 | .open = cgroup_tasks_open, | |
af351026 | 2277 | .write_u64 = cgroup_tasks_write, |
81a6a5cd PM |
2278 | .release = cgroup_tasks_release, |
2279 | .private = FILE_TASKLIST, | |
2280 | }, | |
2281 | ||
2282 | { | |
2283 | .name = "notify_on_release", | |
f4c753b7 | 2284 | .read_u64 = cgroup_read_notify_on_release, |
6379c106 | 2285 | .write_u64 = cgroup_write_notify_on_release, |
81a6a5cd PM |
2286 | .private = FILE_NOTIFY_ON_RELEASE, |
2287 | }, | |
81a6a5cd PM |
2288 | }; |
2289 | ||
2290 | static struct cftype cft_release_agent = { | |
2291 | .name = "release_agent", | |
e788e066 PM |
2292 | .read_seq_string = cgroup_release_agent_show, |
2293 | .write_string = cgroup_release_agent_write, | |
2294 | .max_write_len = PATH_MAX, | |
81a6a5cd | 2295 | .private = FILE_RELEASE_AGENT, |
bbcb81d0 PM |
2296 | }; |
2297 | ||
bd89aabc | 2298 | static int cgroup_populate_dir(struct cgroup *cgrp) |
ddbcc7e8 PM |
2299 | { |
2300 | int err; | |
2301 | struct cgroup_subsys *ss; | |
2302 | ||
2303 | /* First clear out any existing files */ | |
bd89aabc | 2304 | cgroup_clear_directory(cgrp->dentry); |
ddbcc7e8 | 2305 | |
bd89aabc | 2306 | err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files)); |
bbcb81d0 PM |
2307 | if (err < 0) |
2308 | return err; | |
2309 | ||
bd89aabc PM |
2310 | if (cgrp == cgrp->top_cgroup) { |
2311 | if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0) | |
81a6a5cd PM |
2312 | return err; |
2313 | } | |
2314 | ||
bd89aabc PM |
2315 | for_each_subsys(cgrp->root, ss) { |
2316 | if (ss->populate && (err = ss->populate(ss, cgrp)) < 0) | |
ddbcc7e8 PM |
2317 | return err; |
2318 | } | |
2319 | ||
2320 | return 0; | |
2321 | } | |
2322 | ||
2323 | static void init_cgroup_css(struct cgroup_subsys_state *css, | |
2324 | struct cgroup_subsys *ss, | |
bd89aabc | 2325 | struct cgroup *cgrp) |
ddbcc7e8 | 2326 | { |
bd89aabc | 2327 | css->cgroup = cgrp; |
ddbcc7e8 PM |
2328 | atomic_set(&css->refcnt, 0); |
2329 | css->flags = 0; | |
bd89aabc | 2330 | if (cgrp == dummytop) |
ddbcc7e8 | 2331 | set_bit(CSS_ROOT, &css->flags); |
bd89aabc PM |
2332 | BUG_ON(cgrp->subsys[ss->subsys_id]); |
2333 | cgrp->subsys[ss->subsys_id] = css; | |
ddbcc7e8 PM |
2334 | } |
2335 | ||
2336 | /* | |
a043e3b2 LZ |
2337 | * cgroup_create - create a cgroup |
2338 | * @parent: cgroup that will be parent of the new cgroup | |
2339 | * @dentry: dentry of the new cgroup | |
2340 | * @mode: mode to set on new inode | |
ddbcc7e8 | 2341 | * |
a043e3b2 | 2342 | * Must be called with the mutex on the parent inode held |
ddbcc7e8 | 2343 | */ |
ddbcc7e8 PM |
2344 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, |
2345 | int mode) | |
2346 | { | |
bd89aabc | 2347 | struct cgroup *cgrp; |
ddbcc7e8 PM |
2348 | struct cgroupfs_root *root = parent->root; |
2349 | int err = 0; | |
2350 | struct cgroup_subsys *ss; | |
2351 | struct super_block *sb = root->sb; | |
2352 | ||
bd89aabc PM |
2353 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
2354 | if (!cgrp) | |
ddbcc7e8 PM |
2355 | return -ENOMEM; |
2356 | ||
2357 | /* Grab a reference on the superblock so the hierarchy doesn't | |
2358 | * get deleted on unmount if there are child cgroups. This | |
2359 | * can be done outside cgroup_mutex, since the sb can't | |
2360 | * disappear while someone has an open control file on the | |
2361 | * fs */ | |
2362 | atomic_inc(&sb->s_active); | |
2363 | ||
2364 | mutex_lock(&cgroup_mutex); | |
2365 | ||
cc31edce | 2366 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 2367 | |
bd89aabc PM |
2368 | cgrp->parent = parent; |
2369 | cgrp->root = parent->root; | |
2370 | cgrp->top_cgroup = parent->top_cgroup; | |
ddbcc7e8 | 2371 | |
b6abdb0e LZ |
2372 | if (notify_on_release(parent)) |
2373 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
2374 | ||
ddbcc7e8 | 2375 | for_each_subsys(root, ss) { |
bd89aabc | 2376 | struct cgroup_subsys_state *css = ss->create(ss, cgrp); |
ddbcc7e8 PM |
2377 | if (IS_ERR(css)) { |
2378 | err = PTR_ERR(css); | |
2379 | goto err_destroy; | |
2380 | } | |
bd89aabc | 2381 | init_cgroup_css(css, ss, cgrp); |
ddbcc7e8 PM |
2382 | } |
2383 | ||
bd89aabc | 2384 | list_add(&cgrp->sibling, &cgrp->parent->children); |
ddbcc7e8 PM |
2385 | root->number_of_cgroups++; |
2386 | ||
bd89aabc | 2387 | err = cgroup_create_dir(cgrp, dentry, mode); |
ddbcc7e8 PM |
2388 | if (err < 0) |
2389 | goto err_remove; | |
2390 | ||
2391 | /* The cgroup directory was pre-locked for us */ | |
bd89aabc | 2392 | BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex)); |
ddbcc7e8 | 2393 | |
bd89aabc | 2394 | err = cgroup_populate_dir(cgrp); |
ddbcc7e8 PM |
2395 | /* If err < 0, we have a half-filled directory - oh well ;) */ |
2396 | ||
2397 | mutex_unlock(&cgroup_mutex); | |
bd89aabc | 2398 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
2399 | |
2400 | return 0; | |
2401 | ||
2402 | err_remove: | |
2403 | ||
bd89aabc | 2404 | list_del(&cgrp->sibling); |
ddbcc7e8 PM |
2405 | root->number_of_cgroups--; |
2406 | ||
2407 | err_destroy: | |
2408 | ||
2409 | for_each_subsys(root, ss) { | |
bd89aabc PM |
2410 | if (cgrp->subsys[ss->subsys_id]) |
2411 | ss->destroy(ss, cgrp); | |
ddbcc7e8 PM |
2412 | } |
2413 | ||
2414 | mutex_unlock(&cgroup_mutex); | |
2415 | ||
2416 | /* Release the reference count that we took on the superblock */ | |
2417 | deactivate_super(sb); | |
2418 | ||
bd89aabc | 2419 | kfree(cgrp); |
ddbcc7e8 PM |
2420 | return err; |
2421 | } | |
2422 | ||
2423 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
2424 | { | |
2425 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
2426 | ||
2427 | /* the vfs holds inode->i_mutex already */ | |
2428 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
2429 | } | |
2430 | ||
55b6fd01 | 2431 | static int cgroup_has_css_refs(struct cgroup *cgrp) |
81a6a5cd PM |
2432 | { |
2433 | /* Check the reference count on each subsystem. Since we | |
2434 | * already established that there are no tasks in the | |
2435 | * cgroup, if the css refcount is also 0, then there should | |
2436 | * be no outstanding references, so the subsystem is safe to | |
2437 | * destroy. We scan across all subsystems rather than using | |
2438 | * the per-hierarchy linked list of mounted subsystems since | |
2439 | * we can be called via check_for_release() with no | |
2440 | * synchronization other than RCU, and the subsystem linked | |
2441 | * list isn't RCU-safe */ | |
2442 | int i; | |
2443 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2444 | struct cgroup_subsys *ss = subsys[i]; | |
2445 | struct cgroup_subsys_state *css; | |
2446 | /* Skip subsystems not in this hierarchy */ | |
bd89aabc | 2447 | if (ss->root != cgrp->root) |
81a6a5cd | 2448 | continue; |
bd89aabc | 2449 | css = cgrp->subsys[ss->subsys_id]; |
81a6a5cd PM |
2450 | /* When called from check_for_release() it's possible |
2451 | * that by this point the cgroup has been removed | |
2452 | * and the css deleted. But a false-positive doesn't | |
2453 | * matter, since it can only happen if the cgroup | |
2454 | * has been deleted and hence no longer needs the | |
2455 | * release agent to be called anyway. */ | |
e18f6318 | 2456 | if (css && atomic_read(&css->refcnt)) |
81a6a5cd | 2457 | return 1; |
81a6a5cd PM |
2458 | } |
2459 | return 0; | |
2460 | } | |
2461 | ||
ddbcc7e8 PM |
2462 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) |
2463 | { | |
bd89aabc | 2464 | struct cgroup *cgrp = dentry->d_fsdata; |
ddbcc7e8 PM |
2465 | struct dentry *d; |
2466 | struct cgroup *parent; | |
ddbcc7e8 PM |
2467 | struct super_block *sb; |
2468 | struct cgroupfs_root *root; | |
ddbcc7e8 PM |
2469 | |
2470 | /* the vfs holds both inode->i_mutex already */ | |
2471 | ||
2472 | mutex_lock(&cgroup_mutex); | |
bd89aabc | 2473 | if (atomic_read(&cgrp->count) != 0) { |
ddbcc7e8 PM |
2474 | mutex_unlock(&cgroup_mutex); |
2475 | return -EBUSY; | |
2476 | } | |
bd89aabc | 2477 | if (!list_empty(&cgrp->children)) { |
ddbcc7e8 PM |
2478 | mutex_unlock(&cgroup_mutex); |
2479 | return -EBUSY; | |
2480 | } | |
3fa59dfb | 2481 | mutex_unlock(&cgroup_mutex); |
a043e3b2 | 2482 | |
4fca88c8 | 2483 | /* |
a043e3b2 LZ |
2484 | * Call pre_destroy handlers of subsys. Notify subsystems |
2485 | * that rmdir() request comes. | |
4fca88c8 KH |
2486 | */ |
2487 | cgroup_call_pre_destroy(cgrp); | |
ddbcc7e8 | 2488 | |
3fa59dfb KH |
2489 | mutex_lock(&cgroup_mutex); |
2490 | parent = cgrp->parent; | |
2491 | root = cgrp->root; | |
2492 | sb = root->sb; | |
2493 | ||
2494 | if (atomic_read(&cgrp->count) | |
2495 | || !list_empty(&cgrp->children) | |
2496 | || cgroup_has_css_refs(cgrp)) { | |
ddbcc7e8 PM |
2497 | mutex_unlock(&cgroup_mutex); |
2498 | return -EBUSY; | |
2499 | } | |
2500 | ||
81a6a5cd | 2501 | spin_lock(&release_list_lock); |
bd89aabc PM |
2502 | set_bit(CGRP_REMOVED, &cgrp->flags); |
2503 | if (!list_empty(&cgrp->release_list)) | |
2504 | list_del(&cgrp->release_list); | |
81a6a5cd | 2505 | spin_unlock(&release_list_lock); |
ddbcc7e8 | 2506 | /* delete my sibling from parent->children */ |
bd89aabc PM |
2507 | list_del(&cgrp->sibling); |
2508 | spin_lock(&cgrp->dentry->d_lock); | |
2509 | d = dget(cgrp->dentry); | |
ddbcc7e8 PM |
2510 | spin_unlock(&d->d_lock); |
2511 | ||
2512 | cgroup_d_remove_dir(d); | |
2513 | dput(d); | |
ddbcc7e8 | 2514 | |
bd89aabc | 2515 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd PM |
2516 | check_for_release(parent); |
2517 | ||
ddbcc7e8 | 2518 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
2519 | return 0; |
2520 | } | |
2521 | ||
06a11920 | 2522 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 2523 | { |
ddbcc7e8 | 2524 | struct cgroup_subsys_state *css; |
cfe36bde DC |
2525 | |
2526 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 PM |
2527 | |
2528 | /* Create the top cgroup state for this subsystem */ | |
2529 | ss->root = &rootnode; | |
2530 | css = ss->create(ss, dummytop); | |
2531 | /* We don't handle early failures gracefully */ | |
2532 | BUG_ON(IS_ERR(css)); | |
2533 | init_cgroup_css(css, ss, dummytop); | |
2534 | ||
e8d55fde | 2535 | /* Update the init_css_set to contain a subsys |
817929ec | 2536 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
2537 | * newly registered, all tasks and hence the |
2538 | * init_css_set is in the subsystem's top cgroup. */ | |
2539 | init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id]; | |
ddbcc7e8 PM |
2540 | |
2541 | need_forkexit_callback |= ss->fork || ss->exit; | |
cf475ad2 | 2542 | need_mm_owner_callback |= !!ss->mm_owner_changed; |
ddbcc7e8 | 2543 | |
e8d55fde LZ |
2544 | /* At system boot, before all subsystems have been |
2545 | * registered, no tasks have been forked, so we don't | |
2546 | * need to invoke fork callbacks here. */ | |
2547 | BUG_ON(!list_empty(&init_task.tasks)); | |
2548 | ||
ddbcc7e8 PM |
2549 | ss->active = 1; |
2550 | } | |
2551 | ||
2552 | /** | |
a043e3b2 LZ |
2553 | * cgroup_init_early - cgroup initialization at system boot |
2554 | * | |
2555 | * Initialize cgroups at system boot, and initialize any | |
2556 | * subsystems that request early init. | |
ddbcc7e8 PM |
2557 | */ |
2558 | int __init cgroup_init_early(void) | |
2559 | { | |
2560 | int i; | |
146aa1bd | 2561 | atomic_set(&init_css_set.refcount, 1); |
817929ec PM |
2562 | INIT_LIST_HEAD(&init_css_set.cg_links); |
2563 | INIT_LIST_HEAD(&init_css_set.tasks); | |
472b1053 | 2564 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 2565 | css_set_count = 1; |
ddbcc7e8 PM |
2566 | init_cgroup_root(&rootnode); |
2567 | list_add(&rootnode.root_list, &roots); | |
817929ec PM |
2568 | root_count = 1; |
2569 | init_task.cgroups = &init_css_set; | |
2570 | ||
2571 | init_css_set_link.cg = &init_css_set; | |
bd89aabc | 2572 | list_add(&init_css_set_link.cgrp_link_list, |
817929ec PM |
2573 | &rootnode.top_cgroup.css_sets); |
2574 | list_add(&init_css_set_link.cg_link_list, | |
2575 | &init_css_set.cg_links); | |
ddbcc7e8 | 2576 | |
472b1053 LZ |
2577 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) |
2578 | INIT_HLIST_HEAD(&css_set_table[i]); | |
2579 | ||
ddbcc7e8 PM |
2580 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
2581 | struct cgroup_subsys *ss = subsys[i]; | |
2582 | ||
2583 | BUG_ON(!ss->name); | |
2584 | BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); | |
2585 | BUG_ON(!ss->create); | |
2586 | BUG_ON(!ss->destroy); | |
2587 | if (ss->subsys_id != i) { | |
cfe36bde | 2588 | printk(KERN_ERR "cgroup: Subsys %s id == %d\n", |
ddbcc7e8 PM |
2589 | ss->name, ss->subsys_id); |
2590 | BUG(); | |
2591 | } | |
2592 | ||
2593 | if (ss->early_init) | |
2594 | cgroup_init_subsys(ss); | |
2595 | } | |
2596 | return 0; | |
2597 | } | |
2598 | ||
2599 | /** | |
a043e3b2 LZ |
2600 | * cgroup_init - cgroup initialization |
2601 | * | |
2602 | * Register cgroup filesystem and /proc file, and initialize | |
2603 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
2604 | */ |
2605 | int __init cgroup_init(void) | |
2606 | { | |
2607 | int err; | |
2608 | int i; | |
472b1053 | 2609 | struct hlist_head *hhead; |
a424316c PM |
2610 | |
2611 | err = bdi_init(&cgroup_backing_dev_info); | |
2612 | if (err) | |
2613 | return err; | |
ddbcc7e8 PM |
2614 | |
2615 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2616 | struct cgroup_subsys *ss = subsys[i]; | |
2617 | if (!ss->early_init) | |
2618 | cgroup_init_subsys(ss); | |
2619 | } | |
2620 | ||
472b1053 LZ |
2621 | /* Add init_css_set to the hash table */ |
2622 | hhead = css_set_hash(init_css_set.subsys); | |
2623 | hlist_add_head(&init_css_set.hlist, hhead); | |
2624 | ||
ddbcc7e8 PM |
2625 | err = register_filesystem(&cgroup_fs_type); |
2626 | if (err < 0) | |
2627 | goto out; | |
2628 | ||
46ae220b | 2629 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
a424316c | 2630 | |
ddbcc7e8 | 2631 | out: |
a424316c PM |
2632 | if (err) |
2633 | bdi_destroy(&cgroup_backing_dev_info); | |
2634 | ||
ddbcc7e8 PM |
2635 | return err; |
2636 | } | |
b4f48b63 | 2637 | |
a424316c PM |
2638 | /* |
2639 | * proc_cgroup_show() | |
2640 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
2641 | * - Used for /proc/<pid>/cgroup. | |
2642 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
2643 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 2644 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
2645 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
2646 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
2647 | * cgroup to top_cgroup. | |
2648 | */ | |
2649 | ||
2650 | /* TODO: Use a proper seq_file iterator */ | |
2651 | static int proc_cgroup_show(struct seq_file *m, void *v) | |
2652 | { | |
2653 | struct pid *pid; | |
2654 | struct task_struct *tsk; | |
2655 | char *buf; | |
2656 | int retval; | |
2657 | struct cgroupfs_root *root; | |
2658 | ||
2659 | retval = -ENOMEM; | |
2660 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
2661 | if (!buf) | |
2662 | goto out; | |
2663 | ||
2664 | retval = -ESRCH; | |
2665 | pid = m->private; | |
2666 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
2667 | if (!tsk) | |
2668 | goto out_free; | |
2669 | ||
2670 | retval = 0; | |
2671 | ||
2672 | mutex_lock(&cgroup_mutex); | |
2673 | ||
2674 | for_each_root(root) { | |
2675 | struct cgroup_subsys *ss; | |
bd89aabc | 2676 | struct cgroup *cgrp; |
a424316c PM |
2677 | int subsys_id; |
2678 | int count = 0; | |
2679 | ||
2680 | /* Skip this hierarchy if it has no active subsystems */ | |
2681 | if (!root->actual_subsys_bits) | |
2682 | continue; | |
b6c3006d | 2683 | seq_printf(m, "%lu:", root->subsys_bits); |
a424316c PM |
2684 | for_each_subsys(root, ss) |
2685 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
2686 | seq_putc(m, ':'); | |
2687 | get_first_subsys(&root->top_cgroup, NULL, &subsys_id); | |
bd89aabc PM |
2688 | cgrp = task_cgroup(tsk, subsys_id); |
2689 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); | |
a424316c PM |
2690 | if (retval < 0) |
2691 | goto out_unlock; | |
2692 | seq_puts(m, buf); | |
2693 | seq_putc(m, '\n'); | |
2694 | } | |
2695 | ||
2696 | out_unlock: | |
2697 | mutex_unlock(&cgroup_mutex); | |
2698 | put_task_struct(tsk); | |
2699 | out_free: | |
2700 | kfree(buf); | |
2701 | out: | |
2702 | return retval; | |
2703 | } | |
2704 | ||
2705 | static int cgroup_open(struct inode *inode, struct file *file) | |
2706 | { | |
2707 | struct pid *pid = PROC_I(inode)->pid; | |
2708 | return single_open(file, proc_cgroup_show, pid); | |
2709 | } | |
2710 | ||
2711 | struct file_operations proc_cgroup_operations = { | |
2712 | .open = cgroup_open, | |
2713 | .read = seq_read, | |
2714 | .llseek = seq_lseek, | |
2715 | .release = single_release, | |
2716 | }; | |
2717 | ||
2718 | /* Display information about each subsystem and each hierarchy */ | |
2719 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
2720 | { | |
2721 | int i; | |
a424316c | 2722 | |
8bab8dde | 2723 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
a424316c | 2724 | mutex_lock(&cgroup_mutex); |
a424316c PM |
2725 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
2726 | struct cgroup_subsys *ss = subsys[i]; | |
8bab8dde | 2727 | seq_printf(m, "%s\t%lu\t%d\t%d\n", |
817929ec | 2728 | ss->name, ss->root->subsys_bits, |
8bab8dde | 2729 | ss->root->number_of_cgroups, !ss->disabled); |
a424316c PM |
2730 | } |
2731 | mutex_unlock(&cgroup_mutex); | |
2732 | return 0; | |
2733 | } | |
2734 | ||
2735 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
2736 | { | |
9dce07f1 | 2737 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
2738 | } |
2739 | ||
2740 | static struct file_operations proc_cgroupstats_operations = { | |
2741 | .open = cgroupstats_open, | |
2742 | .read = seq_read, | |
2743 | .llseek = seq_lseek, | |
2744 | .release = single_release, | |
2745 | }; | |
2746 | ||
b4f48b63 PM |
2747 | /** |
2748 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 2749 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
2750 | * |
2751 | * Description: A task inherits its parent's cgroup at fork(). | |
2752 | * | |
2753 | * A pointer to the shared css_set was automatically copied in | |
2754 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
2755 | * it was not made under the protection of RCU or cgroup_mutex, so | |
956db3ca | 2756 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might |
817929ec PM |
2757 | * have already changed current->cgroups, allowing the previously |
2758 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
2759 | * |
2760 | * At the point that cgroup_fork() is called, 'current' is the parent | |
2761 | * task, and the passed argument 'child' points to the child task. | |
2762 | */ | |
2763 | void cgroup_fork(struct task_struct *child) | |
2764 | { | |
817929ec PM |
2765 | task_lock(current); |
2766 | child->cgroups = current->cgroups; | |
2767 | get_css_set(child->cgroups); | |
2768 | task_unlock(current); | |
2769 | INIT_LIST_HEAD(&child->cg_list); | |
b4f48b63 PM |
2770 | } |
2771 | ||
2772 | /** | |
a043e3b2 LZ |
2773 | * cgroup_fork_callbacks - run fork callbacks |
2774 | * @child: the new task | |
2775 | * | |
2776 | * Called on a new task very soon before adding it to the | |
2777 | * tasklist. No need to take any locks since no-one can | |
2778 | * be operating on this task. | |
b4f48b63 PM |
2779 | */ |
2780 | void cgroup_fork_callbacks(struct task_struct *child) | |
2781 | { | |
2782 | if (need_forkexit_callback) { | |
2783 | int i; | |
2784 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2785 | struct cgroup_subsys *ss = subsys[i]; | |
2786 | if (ss->fork) | |
2787 | ss->fork(ss, child); | |
2788 | } | |
2789 | } | |
2790 | } | |
2791 | ||
cf475ad2 BS |
2792 | #ifdef CONFIG_MM_OWNER |
2793 | /** | |
2794 | * cgroup_mm_owner_callbacks - run callbacks when the mm->owner changes | |
2795 | * @p: the new owner | |
2796 | * | |
2797 | * Called on every change to mm->owner. mm_init_owner() does not | |
2798 | * invoke this routine, since it assigns the mm->owner the first time | |
2799 | * and does not change it. | |
9363b9f2 BS |
2800 | * |
2801 | * The callbacks are invoked with mmap_sem held in read mode. | |
cf475ad2 BS |
2802 | */ |
2803 | void cgroup_mm_owner_callbacks(struct task_struct *old, struct task_struct *new) | |
2804 | { | |
31a78f23 | 2805 | struct cgroup *oldcgrp, *newcgrp = NULL; |
cf475ad2 BS |
2806 | |
2807 | if (need_mm_owner_callback) { | |
2808 | int i; | |
2809 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2810 | struct cgroup_subsys *ss = subsys[i]; | |
2811 | oldcgrp = task_cgroup(old, ss->subsys_id); | |
31a78f23 BS |
2812 | if (new) |
2813 | newcgrp = task_cgroup(new, ss->subsys_id); | |
cf475ad2 BS |
2814 | if (oldcgrp == newcgrp) |
2815 | continue; | |
2816 | if (ss->mm_owner_changed) | |
9363b9f2 | 2817 | ss->mm_owner_changed(ss, oldcgrp, newcgrp, new); |
cf475ad2 BS |
2818 | } |
2819 | } | |
2820 | } | |
2821 | #endif /* CONFIG_MM_OWNER */ | |
2822 | ||
817929ec | 2823 | /** |
a043e3b2 LZ |
2824 | * cgroup_post_fork - called on a new task after adding it to the task list |
2825 | * @child: the task in question | |
2826 | * | |
2827 | * Adds the task to the list running through its css_set if necessary. | |
2828 | * Has to be after the task is visible on the task list in case we race | |
2829 | * with the first call to cgroup_iter_start() - to guarantee that the | |
2830 | * new task ends up on its list. | |
2831 | */ | |
817929ec PM |
2832 | void cgroup_post_fork(struct task_struct *child) |
2833 | { | |
2834 | if (use_task_css_set_links) { | |
2835 | write_lock(&css_set_lock); | |
2836 | if (list_empty(&child->cg_list)) | |
2837 | list_add(&child->cg_list, &child->cgroups->tasks); | |
2838 | write_unlock(&css_set_lock); | |
2839 | } | |
2840 | } | |
b4f48b63 PM |
2841 | /** |
2842 | * cgroup_exit - detach cgroup from exiting task | |
2843 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 2844 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
2845 | * |
2846 | * Description: Detach cgroup from @tsk and release it. | |
2847 | * | |
2848 | * Note that cgroups marked notify_on_release force every task in | |
2849 | * them to take the global cgroup_mutex mutex when exiting. | |
2850 | * This could impact scaling on very large systems. Be reluctant to | |
2851 | * use notify_on_release cgroups where very high task exit scaling | |
2852 | * is required on large systems. | |
2853 | * | |
2854 | * the_top_cgroup_hack: | |
2855 | * | |
2856 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
2857 | * | |
2858 | * We call cgroup_exit() while the task is still competent to | |
2859 | * handle notify_on_release(), then leave the task attached to the | |
2860 | * root cgroup in each hierarchy for the remainder of its exit. | |
2861 | * | |
2862 | * To do this properly, we would increment the reference count on | |
2863 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
2864 | * code we would add a second cgroup function call, to drop that | |
2865 | * reference. This would just create an unnecessary hot spot on | |
2866 | * the top_cgroup reference count, to no avail. | |
2867 | * | |
2868 | * Normally, holding a reference to a cgroup without bumping its | |
2869 | * count is unsafe. The cgroup could go away, or someone could | |
2870 | * attach us to a different cgroup, decrementing the count on | |
2871 | * the first cgroup that we never incremented. But in this case, | |
2872 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
2873 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
2874 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
2875 | */ |
2876 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
2877 | { | |
2878 | int i; | |
817929ec | 2879 | struct css_set *cg; |
b4f48b63 PM |
2880 | |
2881 | if (run_callbacks && need_forkexit_callback) { | |
2882 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2883 | struct cgroup_subsys *ss = subsys[i]; | |
2884 | if (ss->exit) | |
2885 | ss->exit(ss, tsk); | |
2886 | } | |
2887 | } | |
817929ec PM |
2888 | |
2889 | /* | |
2890 | * Unlink from the css_set task list if necessary. | |
2891 | * Optimistically check cg_list before taking | |
2892 | * css_set_lock | |
2893 | */ | |
2894 | if (!list_empty(&tsk->cg_list)) { | |
2895 | write_lock(&css_set_lock); | |
2896 | if (!list_empty(&tsk->cg_list)) | |
2897 | list_del(&tsk->cg_list); | |
2898 | write_unlock(&css_set_lock); | |
2899 | } | |
2900 | ||
b4f48b63 PM |
2901 | /* Reassign the task to the init_css_set. */ |
2902 | task_lock(tsk); | |
817929ec PM |
2903 | cg = tsk->cgroups; |
2904 | tsk->cgroups = &init_css_set; | |
b4f48b63 | 2905 | task_unlock(tsk); |
817929ec | 2906 | if (cg) |
81a6a5cd | 2907 | put_css_set_taskexit(cg); |
b4f48b63 | 2908 | } |
697f4161 PM |
2909 | |
2910 | /** | |
a043e3b2 LZ |
2911 | * cgroup_clone - clone the cgroup the given subsystem is attached to |
2912 | * @tsk: the task to be moved | |
2913 | * @subsys: the given subsystem | |
e885dcde | 2914 | * @nodename: the name for the new cgroup |
a043e3b2 LZ |
2915 | * |
2916 | * Duplicate the current cgroup in the hierarchy that the given | |
2917 | * subsystem is attached to, and move this task into the new | |
2918 | * child. | |
697f4161 | 2919 | */ |
e885dcde SH |
2920 | int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys, |
2921 | char *nodename) | |
697f4161 PM |
2922 | { |
2923 | struct dentry *dentry; | |
2924 | int ret = 0; | |
697f4161 PM |
2925 | struct cgroup *parent, *child; |
2926 | struct inode *inode; | |
2927 | struct css_set *cg; | |
2928 | struct cgroupfs_root *root; | |
2929 | struct cgroup_subsys *ss; | |
2930 | ||
2931 | /* We shouldn't be called by an unregistered subsystem */ | |
2932 | BUG_ON(!subsys->active); | |
2933 | ||
2934 | /* First figure out what hierarchy and cgroup we're dealing | |
2935 | * with, and pin them so we can drop cgroup_mutex */ | |
2936 | mutex_lock(&cgroup_mutex); | |
2937 | again: | |
2938 | root = subsys->root; | |
2939 | if (root == &rootnode) { | |
697f4161 PM |
2940 | mutex_unlock(&cgroup_mutex); |
2941 | return 0; | |
2942 | } | |
817929ec | 2943 | cg = tsk->cgroups; |
697f4161 PM |
2944 | parent = task_cgroup(tsk, subsys->subsys_id); |
2945 | ||
697f4161 | 2946 | /* Pin the hierarchy */ |
7b574b7b LZ |
2947 | if (!atomic_inc_not_zero(&parent->root->sb->s_active)) { |
2948 | /* We race with the final deactivate_super() */ | |
2949 | mutex_unlock(&cgroup_mutex); | |
2950 | return 0; | |
2951 | } | |
697f4161 | 2952 | |
817929ec PM |
2953 | /* Keep the cgroup alive */ |
2954 | get_css_set(cg); | |
697f4161 PM |
2955 | mutex_unlock(&cgroup_mutex); |
2956 | ||
2957 | /* Now do the VFS work to create a cgroup */ | |
2958 | inode = parent->dentry->d_inode; | |
2959 | ||
2960 | /* Hold the parent directory mutex across this operation to | |
2961 | * stop anyone else deleting the new cgroup */ | |
2962 | mutex_lock(&inode->i_mutex); | |
2963 | dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename)); | |
2964 | if (IS_ERR(dentry)) { | |
2965 | printk(KERN_INFO | |
cfe36bde | 2966 | "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename, |
697f4161 PM |
2967 | PTR_ERR(dentry)); |
2968 | ret = PTR_ERR(dentry); | |
2969 | goto out_release; | |
2970 | } | |
2971 | ||
2972 | /* Create the cgroup directory, which also creates the cgroup */ | |
2973 | ret = vfs_mkdir(inode, dentry, S_IFDIR | 0755); | |
bd89aabc | 2974 | child = __d_cgrp(dentry); |
697f4161 PM |
2975 | dput(dentry); |
2976 | if (ret) { | |
2977 | printk(KERN_INFO | |
2978 | "Failed to create cgroup %s: %d\n", nodename, | |
2979 | ret); | |
2980 | goto out_release; | |
2981 | } | |
2982 | ||
2983 | if (!child) { | |
2984 | printk(KERN_INFO | |
2985 | "Couldn't find new cgroup %s\n", nodename); | |
2986 | ret = -ENOMEM; | |
2987 | goto out_release; | |
2988 | } | |
2989 | ||
2990 | /* The cgroup now exists. Retake cgroup_mutex and check | |
2991 | * that we're still in the same state that we thought we | |
2992 | * were. */ | |
2993 | mutex_lock(&cgroup_mutex); | |
2994 | if ((root != subsys->root) || | |
2995 | (parent != task_cgroup(tsk, subsys->subsys_id))) { | |
2996 | /* Aargh, we raced ... */ | |
2997 | mutex_unlock(&inode->i_mutex); | |
817929ec | 2998 | put_css_set(cg); |
697f4161 PM |
2999 | |
3000 | deactivate_super(parent->root->sb); | |
3001 | /* The cgroup is still accessible in the VFS, but | |
3002 | * we're not going to try to rmdir() it at this | |
3003 | * point. */ | |
3004 | printk(KERN_INFO | |
3005 | "Race in cgroup_clone() - leaking cgroup %s\n", | |
3006 | nodename); | |
3007 | goto again; | |
3008 | } | |
3009 | ||
3010 | /* do any required auto-setup */ | |
3011 | for_each_subsys(root, ss) { | |
3012 | if (ss->post_clone) | |
3013 | ss->post_clone(ss, child); | |
3014 | } | |
3015 | ||
3016 | /* All seems fine. Finish by moving the task into the new cgroup */ | |
956db3ca | 3017 | ret = cgroup_attach_task(child, tsk); |
697f4161 PM |
3018 | mutex_unlock(&cgroup_mutex); |
3019 | ||
3020 | out_release: | |
3021 | mutex_unlock(&inode->i_mutex); | |
81a6a5cd PM |
3022 | |
3023 | mutex_lock(&cgroup_mutex); | |
817929ec | 3024 | put_css_set(cg); |
81a6a5cd | 3025 | mutex_unlock(&cgroup_mutex); |
697f4161 PM |
3026 | deactivate_super(parent->root->sb); |
3027 | return ret; | |
3028 | } | |
3029 | ||
a043e3b2 LZ |
3030 | /** |
3031 | * cgroup_is_descendant - see if @cgrp is a descendant of current task's cgrp | |
3032 | * @cgrp: the cgroup in question | |
3033 | * | |
3034 | * See if @cgrp is a descendant of the current task's cgroup in | |
3035 | * the appropriate hierarchy. | |
697f4161 PM |
3036 | * |
3037 | * If we are sending in dummytop, then presumably we are creating | |
3038 | * the top cgroup in the subsystem. | |
3039 | * | |
3040 | * Called only by the ns (nsproxy) cgroup. | |
3041 | */ | |
bd89aabc | 3042 | int cgroup_is_descendant(const struct cgroup *cgrp) |
697f4161 PM |
3043 | { |
3044 | int ret; | |
3045 | struct cgroup *target; | |
3046 | int subsys_id; | |
3047 | ||
bd89aabc | 3048 | if (cgrp == dummytop) |
697f4161 PM |
3049 | return 1; |
3050 | ||
bd89aabc | 3051 | get_first_subsys(cgrp, NULL, &subsys_id); |
697f4161 | 3052 | target = task_cgroup(current, subsys_id); |
bd89aabc PM |
3053 | while (cgrp != target && cgrp!= cgrp->top_cgroup) |
3054 | cgrp = cgrp->parent; | |
3055 | ret = (cgrp == target); | |
697f4161 PM |
3056 | return ret; |
3057 | } | |
81a6a5cd | 3058 | |
bd89aabc | 3059 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd PM |
3060 | { |
3061 | /* All of these checks rely on RCU to keep the cgroup | |
3062 | * structure alive */ | |
bd89aabc PM |
3063 | if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count) |
3064 | && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) { | |
81a6a5cd PM |
3065 | /* Control Group is currently removeable. If it's not |
3066 | * already queued for a userspace notification, queue | |
3067 | * it now */ | |
3068 | int need_schedule_work = 0; | |
3069 | spin_lock(&release_list_lock); | |
bd89aabc PM |
3070 | if (!cgroup_is_removed(cgrp) && |
3071 | list_empty(&cgrp->release_list)) { | |
3072 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
3073 | need_schedule_work = 1; |
3074 | } | |
3075 | spin_unlock(&release_list_lock); | |
3076 | if (need_schedule_work) | |
3077 | schedule_work(&release_agent_work); | |
3078 | } | |
3079 | } | |
3080 | ||
3081 | void __css_put(struct cgroup_subsys_state *css) | |
3082 | { | |
bd89aabc | 3083 | struct cgroup *cgrp = css->cgroup; |
81a6a5cd | 3084 | rcu_read_lock(); |
bd89aabc PM |
3085 | if (atomic_dec_and_test(&css->refcnt) && notify_on_release(cgrp)) { |
3086 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
3087 | check_for_release(cgrp); | |
81a6a5cd PM |
3088 | } |
3089 | rcu_read_unlock(); | |
3090 | } | |
3091 | ||
3092 | /* | |
3093 | * Notify userspace when a cgroup is released, by running the | |
3094 | * configured release agent with the name of the cgroup (path | |
3095 | * relative to the root of cgroup file system) as the argument. | |
3096 | * | |
3097 | * Most likely, this user command will try to rmdir this cgroup. | |
3098 | * | |
3099 | * This races with the possibility that some other task will be | |
3100 | * attached to this cgroup before it is removed, or that some other | |
3101 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
3102 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
3103 | * unused, and this cgroup will be reprieved from its death sentence, | |
3104 | * to continue to serve a useful existence. Next time it's released, | |
3105 | * we will get notified again, if it still has 'notify_on_release' set. | |
3106 | * | |
3107 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
3108 | * means only wait until the task is successfully execve()'d. The | |
3109 | * separate release agent task is forked by call_usermodehelper(), | |
3110 | * then control in this thread returns here, without waiting for the | |
3111 | * release agent task. We don't bother to wait because the caller of | |
3112 | * this routine has no use for the exit status of the release agent | |
3113 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 3114 | */ |
81a6a5cd PM |
3115 | static void cgroup_release_agent(struct work_struct *work) |
3116 | { | |
3117 | BUG_ON(work != &release_agent_work); | |
3118 | mutex_lock(&cgroup_mutex); | |
3119 | spin_lock(&release_list_lock); | |
3120 | while (!list_empty(&release_list)) { | |
3121 | char *argv[3], *envp[3]; | |
3122 | int i; | |
e788e066 | 3123 | char *pathbuf = NULL, *agentbuf = NULL; |
bd89aabc | 3124 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
3125 | struct cgroup, |
3126 | release_list); | |
bd89aabc | 3127 | list_del_init(&cgrp->release_list); |
81a6a5cd PM |
3128 | spin_unlock(&release_list_lock); |
3129 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
e788e066 PM |
3130 | if (!pathbuf) |
3131 | goto continue_free; | |
3132 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
3133 | goto continue_free; | |
3134 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
3135 | if (!agentbuf) | |
3136 | goto continue_free; | |
81a6a5cd PM |
3137 | |
3138 | i = 0; | |
e788e066 PM |
3139 | argv[i++] = agentbuf; |
3140 | argv[i++] = pathbuf; | |
81a6a5cd PM |
3141 | argv[i] = NULL; |
3142 | ||
3143 | i = 0; | |
3144 | /* minimal command environment */ | |
3145 | envp[i++] = "HOME=/"; | |
3146 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
3147 | envp[i] = NULL; | |
3148 | ||
3149 | /* Drop the lock while we invoke the usermode helper, | |
3150 | * since the exec could involve hitting disk and hence | |
3151 | * be a slow process */ | |
3152 | mutex_unlock(&cgroup_mutex); | |
3153 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 3154 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
3155 | continue_free: |
3156 | kfree(pathbuf); | |
3157 | kfree(agentbuf); | |
81a6a5cd PM |
3158 | spin_lock(&release_list_lock); |
3159 | } | |
3160 | spin_unlock(&release_list_lock); | |
3161 | mutex_unlock(&cgroup_mutex); | |
3162 | } | |
8bab8dde PM |
3163 | |
3164 | static int __init cgroup_disable(char *str) | |
3165 | { | |
3166 | int i; | |
3167 | char *token; | |
3168 | ||
3169 | while ((token = strsep(&str, ",")) != NULL) { | |
3170 | if (!*token) | |
3171 | continue; | |
3172 | ||
3173 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
3174 | struct cgroup_subsys *ss = subsys[i]; | |
3175 | ||
3176 | if (!strcmp(token, ss->name)) { | |
3177 | ss->disabled = 1; | |
3178 | printk(KERN_INFO "Disabling %s control group" | |
3179 | " subsystem\n", ss->name); | |
3180 | break; | |
3181 | } | |
3182 | } | |
3183 | } | |
3184 | return 1; | |
3185 | } | |
3186 | __setup("cgroup_disable=", cgroup_disable); |