]>
Commit | Line | Data |
---|---|---|
1 | /* | |
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 | * Notifications support | |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
11 | * Copyright notices from the original cpuset code: | |
12 | * -------------------------------------------------- | |
13 | * Copyright (C) 2003 BULL SA. | |
14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
15 | * | |
16 | * Portions derived from Patrick Mochel's sysfs code. | |
17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
18 | * | |
19 | * 2003-10-10 Written by Simon Derr. | |
20 | * 2003-10-22 Updates by Stephen Hemminger. | |
21 | * 2004 May-July Rework by Paul Jackson. | |
22 | * --------------------------------------------------- | |
23 | * | |
24 | * This file is subject to the terms and conditions of the GNU General Public | |
25 | * License. See the file COPYING in the main directory of the Linux | |
26 | * distribution for more details. | |
27 | */ | |
28 | ||
29 | #include <linux/cgroup.h> | |
30 | #include <linux/ctype.h> | |
31 | #include <linux/errno.h> | |
32 | #include <linux/fs.h> | |
33 | #include <linux/kernel.h> | |
34 | #include <linux/list.h> | |
35 | #include <linux/mm.h> | |
36 | #include <linux/mutex.h> | |
37 | #include <linux/mount.h> | |
38 | #include <linux/pagemap.h> | |
39 | #include <linux/proc_fs.h> | |
40 | #include <linux/rcupdate.h> | |
41 | #include <linux/sched.h> | |
42 | #include <linux/backing-dev.h> | |
43 | #include <linux/seq_file.h> | |
44 | #include <linux/slab.h> | |
45 | #include <linux/magic.h> | |
46 | #include <linux/spinlock.h> | |
47 | #include <linux/string.h> | |
48 | #include <linux/sort.h> | |
49 | #include <linux/kmod.h> | |
50 | #include <linux/module.h> | |
51 | #include <linux/delayacct.h> | |
52 | #include <linux/cgroupstats.h> | |
53 | #include <linux/hash.h> | |
54 | #include <linux/namei.h> | |
55 | #include <linux/smp_lock.h> | |
56 | #include <linux/pid_namespace.h> | |
57 | #include <linux/idr.h> | |
58 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ | |
59 | #include <linux/eventfd.h> | |
60 | #include <linux/poll.h> | |
61 | ||
62 | #include <asm/atomic.h> | |
63 | ||
64 | static DEFINE_MUTEX(cgroup_mutex); | |
65 | ||
66 | /* | |
67 | * Generate an array of cgroup subsystem pointers. At boot time, this is | |
68 | * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are | |
69 | * registered after that. The mutable section of this array is protected by | |
70 | * cgroup_mutex. | |
71 | */ | |
72 | #define SUBSYS(_x) &_x ## _subsys, | |
73 | static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = { | |
74 | #include <linux/cgroup_subsys.h> | |
75 | }; | |
76 | ||
77 | #define MAX_CGROUP_ROOT_NAMELEN 64 | |
78 | ||
79 | /* | |
80 | * A cgroupfs_root represents the root of a cgroup hierarchy, | |
81 | * and may be associated with a superblock to form an active | |
82 | * hierarchy | |
83 | */ | |
84 | struct cgroupfs_root { | |
85 | struct super_block *sb; | |
86 | ||
87 | /* | |
88 | * The bitmask of subsystems intended to be attached to this | |
89 | * hierarchy | |
90 | */ | |
91 | unsigned long subsys_bits; | |
92 | ||
93 | /* Unique id for this hierarchy. */ | |
94 | int hierarchy_id; | |
95 | ||
96 | /* The bitmask of subsystems currently attached to this hierarchy */ | |
97 | unsigned long actual_subsys_bits; | |
98 | ||
99 | /* A list running through the attached subsystems */ | |
100 | struct list_head subsys_list; | |
101 | ||
102 | /* The root cgroup for this hierarchy */ | |
103 | struct cgroup top_cgroup; | |
104 | ||
105 | /* Tracks how many cgroups are currently defined in hierarchy.*/ | |
106 | int number_of_cgroups; | |
107 | ||
108 | /* A list running through the active hierarchies */ | |
109 | struct list_head root_list; | |
110 | ||
111 | /* Hierarchy-specific flags */ | |
112 | unsigned long flags; | |
113 | ||
114 | /* The path to use for release notifications. */ | |
115 | char release_agent_path[PATH_MAX]; | |
116 | ||
117 | /* The name for this hierarchy - may be empty */ | |
118 | char name[MAX_CGROUP_ROOT_NAMELEN]; | |
119 | }; | |
120 | ||
121 | /* | |
122 | * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the | |
123 | * subsystems that are otherwise unattached - it never has more than a | |
124 | * single cgroup, and all tasks are part of that cgroup. | |
125 | */ | |
126 | static struct cgroupfs_root rootnode; | |
127 | ||
128 | /* | |
129 | * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when | |
130 | * cgroup_subsys->use_id != 0. | |
131 | */ | |
132 | #define CSS_ID_MAX (65535) | |
133 | struct css_id { | |
134 | /* | |
135 | * The css to which this ID points. This pointer is set to valid value | |
136 | * after cgroup is populated. If cgroup is removed, this will be NULL. | |
137 | * This pointer is expected to be RCU-safe because destroy() | |
138 | * is called after synchronize_rcu(). But for safe use, css_is_removed() | |
139 | * css_tryget() should be used for avoiding race. | |
140 | */ | |
141 | struct cgroup_subsys_state *css; | |
142 | /* | |
143 | * ID of this css. | |
144 | */ | |
145 | unsigned short id; | |
146 | /* | |
147 | * Depth in hierarchy which this ID belongs to. | |
148 | */ | |
149 | unsigned short depth; | |
150 | /* | |
151 | * ID is freed by RCU. (and lookup routine is RCU safe.) | |
152 | */ | |
153 | struct rcu_head rcu_head; | |
154 | /* | |
155 | * Hierarchy of CSS ID belongs to. | |
156 | */ | |
157 | unsigned short stack[0]; /* Array of Length (depth+1) */ | |
158 | }; | |
159 | ||
160 | /* | |
161 | * cgroup_event represents events which userspace want to recieve. | |
162 | */ | |
163 | struct cgroup_event { | |
164 | /* | |
165 | * Cgroup which the event belongs to. | |
166 | */ | |
167 | struct cgroup *cgrp; | |
168 | /* | |
169 | * Control file which the event associated. | |
170 | */ | |
171 | struct cftype *cft; | |
172 | /* | |
173 | * eventfd to signal userspace about the event. | |
174 | */ | |
175 | struct eventfd_ctx *eventfd; | |
176 | /* | |
177 | * Each of these stored in a list by the cgroup. | |
178 | */ | |
179 | struct list_head list; | |
180 | /* | |
181 | * All fields below needed to unregister event when | |
182 | * userspace closes eventfd. | |
183 | */ | |
184 | poll_table pt; | |
185 | wait_queue_head_t *wqh; | |
186 | wait_queue_t wait; | |
187 | struct work_struct remove; | |
188 | }; | |
189 | ||
190 | /* The list of hierarchy roots */ | |
191 | ||
192 | static LIST_HEAD(roots); | |
193 | static int root_count; | |
194 | ||
195 | static DEFINE_IDA(hierarchy_ida); | |
196 | static int next_hierarchy_id; | |
197 | static DEFINE_SPINLOCK(hierarchy_id_lock); | |
198 | ||
199 | /* dummytop is a shorthand for the dummy hierarchy's top cgroup */ | |
200 | #define dummytop (&rootnode.top_cgroup) | |
201 | ||
202 | /* This flag indicates whether tasks in the fork and exit paths should | |
203 | * check for fork/exit handlers to call. This avoids us having to do | |
204 | * extra work in the fork/exit path if none of the subsystems need to | |
205 | * be called. | |
206 | */ | |
207 | static int need_forkexit_callback __read_mostly; | |
208 | ||
209 | #ifdef CONFIG_PROVE_LOCKING | |
210 | int cgroup_lock_is_held(void) | |
211 | { | |
212 | return lockdep_is_held(&cgroup_mutex); | |
213 | } | |
214 | #else /* #ifdef CONFIG_PROVE_LOCKING */ | |
215 | int cgroup_lock_is_held(void) | |
216 | { | |
217 | return mutex_is_locked(&cgroup_mutex); | |
218 | } | |
219 | #endif /* #else #ifdef CONFIG_PROVE_LOCKING */ | |
220 | ||
221 | EXPORT_SYMBOL_GPL(cgroup_lock_is_held); | |
222 | ||
223 | /* convenient tests for these bits */ | |
224 | inline int cgroup_is_removed(const struct cgroup *cgrp) | |
225 | { | |
226 | return test_bit(CGRP_REMOVED, &cgrp->flags); | |
227 | } | |
228 | ||
229 | /* bits in struct cgroupfs_root flags field */ | |
230 | enum { | |
231 | ROOT_NOPREFIX, /* mounted subsystems have no named prefix */ | |
232 | }; | |
233 | ||
234 | static int cgroup_is_releasable(const struct cgroup *cgrp) | |
235 | { | |
236 | const int bits = | |
237 | (1 << CGRP_RELEASABLE) | | |
238 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
239 | return (cgrp->flags & bits) == bits; | |
240 | } | |
241 | ||
242 | static int notify_on_release(const struct cgroup *cgrp) | |
243 | { | |
244 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
245 | } | |
246 | ||
247 | /* | |
248 | * for_each_subsys() allows you to iterate on each subsystem attached to | |
249 | * an active hierarchy | |
250 | */ | |
251 | #define for_each_subsys(_root, _ss) \ | |
252 | list_for_each_entry(_ss, &_root->subsys_list, sibling) | |
253 | ||
254 | /* for_each_active_root() allows you to iterate across the active hierarchies */ | |
255 | #define for_each_active_root(_root) \ | |
256 | list_for_each_entry(_root, &roots, root_list) | |
257 | ||
258 | /* the list of cgroups eligible for automatic release. Protected by | |
259 | * release_list_lock */ | |
260 | static LIST_HEAD(release_list); | |
261 | static DEFINE_SPINLOCK(release_list_lock); | |
262 | static void cgroup_release_agent(struct work_struct *work); | |
263 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
264 | static void check_for_release(struct cgroup *cgrp); | |
265 | ||
266 | /* Link structure for associating css_set objects with cgroups */ | |
267 | struct cg_cgroup_link { | |
268 | /* | |
269 | * List running through cg_cgroup_links associated with a | |
270 | * cgroup, anchored on cgroup->css_sets | |
271 | */ | |
272 | struct list_head cgrp_link_list; | |
273 | struct cgroup *cgrp; | |
274 | /* | |
275 | * List running through cg_cgroup_links pointing at a | |
276 | * single css_set object, anchored on css_set->cg_links | |
277 | */ | |
278 | struct list_head cg_link_list; | |
279 | struct css_set *cg; | |
280 | }; | |
281 | ||
282 | /* The default css_set - used by init and its children prior to any | |
283 | * hierarchies being mounted. It contains a pointer to the root state | |
284 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
285 | * reference-counted, to improve performance when child cgroups | |
286 | * haven't been created. | |
287 | */ | |
288 | ||
289 | static struct css_set init_css_set; | |
290 | static struct cg_cgroup_link init_css_set_link; | |
291 | ||
292 | static int cgroup_init_idr(struct cgroup_subsys *ss, | |
293 | struct cgroup_subsys_state *css); | |
294 | ||
295 | /* css_set_lock protects the list of css_set objects, and the | |
296 | * chain of tasks off each css_set. Nests outside task->alloc_lock | |
297 | * due to cgroup_iter_start() */ | |
298 | static DEFINE_RWLOCK(css_set_lock); | |
299 | static int css_set_count; | |
300 | ||
301 | /* | |
302 | * hash table for cgroup groups. This improves the performance to find | |
303 | * an existing css_set. This hash doesn't (currently) take into | |
304 | * account cgroups in empty hierarchies. | |
305 | */ | |
306 | #define CSS_SET_HASH_BITS 7 | |
307 | #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS) | |
308 | static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE]; | |
309 | ||
310 | static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[]) | |
311 | { | |
312 | int i; | |
313 | int index; | |
314 | unsigned long tmp = 0UL; | |
315 | ||
316 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) | |
317 | tmp += (unsigned long)css[i]; | |
318 | tmp = (tmp >> 16) ^ tmp; | |
319 | ||
320 | index = hash_long(tmp, CSS_SET_HASH_BITS); | |
321 | ||
322 | return &css_set_table[index]; | |
323 | } | |
324 | ||
325 | static void free_css_set_rcu(struct rcu_head *obj) | |
326 | { | |
327 | struct css_set *cg = container_of(obj, struct css_set, rcu_head); | |
328 | kfree(cg); | |
329 | } | |
330 | ||
331 | /* We don't maintain the lists running through each css_set to its | |
332 | * task until after the first call to cgroup_iter_start(). This | |
333 | * reduces the fork()/exit() overhead for people who have cgroups | |
334 | * compiled into their kernel but not actually in use */ | |
335 | static int use_task_css_set_links __read_mostly; | |
336 | ||
337 | static void __put_css_set(struct css_set *cg, int taskexit) | |
338 | { | |
339 | struct cg_cgroup_link *link; | |
340 | struct cg_cgroup_link *saved_link; | |
341 | /* | |
342 | * Ensure that the refcount doesn't hit zero while any readers | |
343 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
344 | * rwlock | |
345 | */ | |
346 | if (atomic_add_unless(&cg->refcount, -1, 1)) | |
347 | return; | |
348 | write_lock(&css_set_lock); | |
349 | if (!atomic_dec_and_test(&cg->refcount)) { | |
350 | write_unlock(&css_set_lock); | |
351 | return; | |
352 | } | |
353 | ||
354 | /* This css_set is dead. unlink it and release cgroup refcounts */ | |
355 | hlist_del(&cg->hlist); | |
356 | css_set_count--; | |
357 | ||
358 | list_for_each_entry_safe(link, saved_link, &cg->cg_links, | |
359 | cg_link_list) { | |
360 | struct cgroup *cgrp = link->cgrp; | |
361 | list_del(&link->cg_link_list); | |
362 | list_del(&link->cgrp_link_list); | |
363 | if (atomic_dec_and_test(&cgrp->count) && | |
364 | notify_on_release(cgrp)) { | |
365 | if (taskexit) | |
366 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
367 | check_for_release(cgrp); | |
368 | } | |
369 | ||
370 | kfree(link); | |
371 | } | |
372 | ||
373 | write_unlock(&css_set_lock); | |
374 | call_rcu(&cg->rcu_head, free_css_set_rcu); | |
375 | } | |
376 | ||
377 | /* | |
378 | * refcounted get/put for css_set objects | |
379 | */ | |
380 | static inline void get_css_set(struct css_set *cg) | |
381 | { | |
382 | atomic_inc(&cg->refcount); | |
383 | } | |
384 | ||
385 | static inline void put_css_set(struct css_set *cg) | |
386 | { | |
387 | __put_css_set(cg, 0); | |
388 | } | |
389 | ||
390 | static inline void put_css_set_taskexit(struct css_set *cg) | |
391 | { | |
392 | __put_css_set(cg, 1); | |
393 | } | |
394 | ||
395 | /* | |
396 | * compare_css_sets - helper function for find_existing_css_set(). | |
397 | * @cg: candidate css_set being tested | |
398 | * @old_cg: existing css_set for a task | |
399 | * @new_cgrp: cgroup that's being entered by the task | |
400 | * @template: desired set of css pointers in css_set (pre-calculated) | |
401 | * | |
402 | * Returns true if "cg" matches "old_cg" except for the hierarchy | |
403 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". | |
404 | */ | |
405 | static bool compare_css_sets(struct css_set *cg, | |
406 | struct css_set *old_cg, | |
407 | struct cgroup *new_cgrp, | |
408 | struct cgroup_subsys_state *template[]) | |
409 | { | |
410 | struct list_head *l1, *l2; | |
411 | ||
412 | if (memcmp(template, cg->subsys, sizeof(cg->subsys))) { | |
413 | /* Not all subsystems matched */ | |
414 | return false; | |
415 | } | |
416 | ||
417 | /* | |
418 | * Compare cgroup pointers in order to distinguish between | |
419 | * different cgroups in heirarchies with no subsystems. We | |
420 | * could get by with just this check alone (and skip the | |
421 | * memcmp above) but on most setups the memcmp check will | |
422 | * avoid the need for this more expensive check on almost all | |
423 | * candidates. | |
424 | */ | |
425 | ||
426 | l1 = &cg->cg_links; | |
427 | l2 = &old_cg->cg_links; | |
428 | while (1) { | |
429 | struct cg_cgroup_link *cgl1, *cgl2; | |
430 | struct cgroup *cg1, *cg2; | |
431 | ||
432 | l1 = l1->next; | |
433 | l2 = l2->next; | |
434 | /* See if we reached the end - both lists are equal length. */ | |
435 | if (l1 == &cg->cg_links) { | |
436 | BUG_ON(l2 != &old_cg->cg_links); | |
437 | break; | |
438 | } else { | |
439 | BUG_ON(l2 == &old_cg->cg_links); | |
440 | } | |
441 | /* Locate the cgroups associated with these links. */ | |
442 | cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list); | |
443 | cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list); | |
444 | cg1 = cgl1->cgrp; | |
445 | cg2 = cgl2->cgrp; | |
446 | /* Hierarchies should be linked in the same order. */ | |
447 | BUG_ON(cg1->root != cg2->root); | |
448 | ||
449 | /* | |
450 | * If this hierarchy is the hierarchy of the cgroup | |
451 | * that's changing, then we need to check that this | |
452 | * css_set points to the new cgroup; if it's any other | |
453 | * hierarchy, then this css_set should point to the | |
454 | * same cgroup as the old css_set. | |
455 | */ | |
456 | if (cg1->root == new_cgrp->root) { | |
457 | if (cg1 != new_cgrp) | |
458 | return false; | |
459 | } else { | |
460 | if (cg1 != cg2) | |
461 | return false; | |
462 | } | |
463 | } | |
464 | return true; | |
465 | } | |
466 | ||
467 | /* | |
468 | * find_existing_css_set() is a helper for | |
469 | * find_css_set(), and checks to see whether an existing | |
470 | * css_set is suitable. | |
471 | * | |
472 | * oldcg: the cgroup group that we're using before the cgroup | |
473 | * transition | |
474 | * | |
475 | * cgrp: the cgroup that we're moving into | |
476 | * | |
477 | * template: location in which to build the desired set of subsystem | |
478 | * state objects for the new cgroup group | |
479 | */ | |
480 | static struct css_set *find_existing_css_set( | |
481 | struct css_set *oldcg, | |
482 | struct cgroup *cgrp, | |
483 | struct cgroup_subsys_state *template[]) | |
484 | { | |
485 | int i; | |
486 | struct cgroupfs_root *root = cgrp->root; | |
487 | struct hlist_head *hhead; | |
488 | struct hlist_node *node; | |
489 | struct css_set *cg; | |
490 | ||
491 | /* | |
492 | * Build the set of subsystem state objects that we want to see in the | |
493 | * new css_set. while subsystems can change globally, the entries here | |
494 | * won't change, so no need for locking. | |
495 | */ | |
496 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
497 | if (root->subsys_bits & (1UL << i)) { | |
498 | /* Subsystem is in this hierarchy. So we want | |
499 | * the subsystem state from the new | |
500 | * cgroup */ | |
501 | template[i] = cgrp->subsys[i]; | |
502 | } else { | |
503 | /* Subsystem is not in this hierarchy, so we | |
504 | * don't want to change the subsystem state */ | |
505 | template[i] = oldcg->subsys[i]; | |
506 | } | |
507 | } | |
508 | ||
509 | hhead = css_set_hash(template); | |
510 | hlist_for_each_entry(cg, node, hhead, hlist) { | |
511 | if (!compare_css_sets(cg, oldcg, cgrp, template)) | |
512 | continue; | |
513 | ||
514 | /* This css_set matches what we need */ | |
515 | return cg; | |
516 | } | |
517 | ||
518 | /* No existing cgroup group matched */ | |
519 | return NULL; | |
520 | } | |
521 | ||
522 | static void free_cg_links(struct list_head *tmp) | |
523 | { | |
524 | struct cg_cgroup_link *link; | |
525 | struct cg_cgroup_link *saved_link; | |
526 | ||
527 | list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) { | |
528 | list_del(&link->cgrp_link_list); | |
529 | kfree(link); | |
530 | } | |
531 | } | |
532 | ||
533 | /* | |
534 | * allocate_cg_links() allocates "count" cg_cgroup_link structures | |
535 | * and chains them on tmp through their cgrp_link_list fields. Returns 0 on | |
536 | * success or a negative error | |
537 | */ | |
538 | static int allocate_cg_links(int count, struct list_head *tmp) | |
539 | { | |
540 | struct cg_cgroup_link *link; | |
541 | int i; | |
542 | INIT_LIST_HEAD(tmp); | |
543 | for (i = 0; i < count; i++) { | |
544 | link = kmalloc(sizeof(*link), GFP_KERNEL); | |
545 | if (!link) { | |
546 | free_cg_links(tmp); | |
547 | return -ENOMEM; | |
548 | } | |
549 | list_add(&link->cgrp_link_list, tmp); | |
550 | } | |
551 | return 0; | |
552 | } | |
553 | ||
554 | /** | |
555 | * link_css_set - a helper function to link a css_set to a cgroup | |
556 | * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links() | |
557 | * @cg: the css_set to be linked | |
558 | * @cgrp: the destination cgroup | |
559 | */ | |
560 | static void link_css_set(struct list_head *tmp_cg_links, | |
561 | struct css_set *cg, struct cgroup *cgrp) | |
562 | { | |
563 | struct cg_cgroup_link *link; | |
564 | ||
565 | BUG_ON(list_empty(tmp_cg_links)); | |
566 | link = list_first_entry(tmp_cg_links, struct cg_cgroup_link, | |
567 | cgrp_link_list); | |
568 | link->cg = cg; | |
569 | link->cgrp = cgrp; | |
570 | atomic_inc(&cgrp->count); | |
571 | list_move(&link->cgrp_link_list, &cgrp->css_sets); | |
572 | /* | |
573 | * Always add links to the tail of the list so that the list | |
574 | * is sorted by order of hierarchy creation | |
575 | */ | |
576 | list_add_tail(&link->cg_link_list, &cg->cg_links); | |
577 | } | |
578 | ||
579 | /* | |
580 | * find_css_set() takes an existing cgroup group and a | |
581 | * cgroup object, and returns a css_set object that's | |
582 | * equivalent to the old group, but with the given cgroup | |
583 | * substituted into the appropriate hierarchy. Must be called with | |
584 | * cgroup_mutex held | |
585 | */ | |
586 | static struct css_set *find_css_set( | |
587 | struct css_set *oldcg, struct cgroup *cgrp) | |
588 | { | |
589 | struct css_set *res; | |
590 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT]; | |
591 | ||
592 | struct list_head tmp_cg_links; | |
593 | ||
594 | struct hlist_head *hhead; | |
595 | struct cg_cgroup_link *link; | |
596 | ||
597 | /* First see if we already have a cgroup group that matches | |
598 | * the desired set */ | |
599 | read_lock(&css_set_lock); | |
600 | res = find_existing_css_set(oldcg, cgrp, template); | |
601 | if (res) | |
602 | get_css_set(res); | |
603 | read_unlock(&css_set_lock); | |
604 | ||
605 | if (res) | |
606 | return res; | |
607 | ||
608 | res = kmalloc(sizeof(*res), GFP_KERNEL); | |
609 | if (!res) | |
610 | return NULL; | |
611 | ||
612 | /* Allocate all the cg_cgroup_link objects that we'll need */ | |
613 | if (allocate_cg_links(root_count, &tmp_cg_links) < 0) { | |
614 | kfree(res); | |
615 | return NULL; | |
616 | } | |
617 | ||
618 | atomic_set(&res->refcount, 1); | |
619 | INIT_LIST_HEAD(&res->cg_links); | |
620 | INIT_LIST_HEAD(&res->tasks); | |
621 | INIT_HLIST_NODE(&res->hlist); | |
622 | ||
623 | /* Copy the set of subsystem state objects generated in | |
624 | * find_existing_css_set() */ | |
625 | memcpy(res->subsys, template, sizeof(res->subsys)); | |
626 | ||
627 | write_lock(&css_set_lock); | |
628 | /* Add reference counts and links from the new css_set. */ | |
629 | list_for_each_entry(link, &oldcg->cg_links, cg_link_list) { | |
630 | struct cgroup *c = link->cgrp; | |
631 | if (c->root == cgrp->root) | |
632 | c = cgrp; | |
633 | link_css_set(&tmp_cg_links, res, c); | |
634 | } | |
635 | ||
636 | BUG_ON(!list_empty(&tmp_cg_links)); | |
637 | ||
638 | css_set_count++; | |
639 | ||
640 | /* Add this cgroup group to the hash table */ | |
641 | hhead = css_set_hash(res->subsys); | |
642 | hlist_add_head(&res->hlist, hhead); | |
643 | ||
644 | write_unlock(&css_set_lock); | |
645 | ||
646 | return res; | |
647 | } | |
648 | ||
649 | /* | |
650 | * Return the cgroup for "task" from the given hierarchy. Must be | |
651 | * called with cgroup_mutex held. | |
652 | */ | |
653 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
654 | struct cgroupfs_root *root) | |
655 | { | |
656 | struct css_set *css; | |
657 | struct cgroup *res = NULL; | |
658 | ||
659 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
660 | read_lock(&css_set_lock); | |
661 | /* | |
662 | * No need to lock the task - since we hold cgroup_mutex the | |
663 | * task can't change groups, so the only thing that can happen | |
664 | * is that it exits and its css is set back to init_css_set. | |
665 | */ | |
666 | css = task->cgroups; | |
667 | if (css == &init_css_set) { | |
668 | res = &root->top_cgroup; | |
669 | } else { | |
670 | struct cg_cgroup_link *link; | |
671 | list_for_each_entry(link, &css->cg_links, cg_link_list) { | |
672 | struct cgroup *c = link->cgrp; | |
673 | if (c->root == root) { | |
674 | res = c; | |
675 | break; | |
676 | } | |
677 | } | |
678 | } | |
679 | read_unlock(&css_set_lock); | |
680 | BUG_ON(!res); | |
681 | return res; | |
682 | } | |
683 | ||
684 | /* | |
685 | * There is one global cgroup mutex. We also require taking | |
686 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
687 | * See "The task_lock() exception", at the end of this comment. | |
688 | * | |
689 | * A task must hold cgroup_mutex to modify cgroups. | |
690 | * | |
691 | * Any task can increment and decrement the count field without lock. | |
692 | * So in general, code holding cgroup_mutex can't rely on the count | |
693 | * field not changing. However, if the count goes to zero, then only | |
694 | * cgroup_attach_task() can increment it again. Because a count of zero | |
695 | * means that no tasks are currently attached, therefore there is no | |
696 | * way a task attached to that cgroup can fork (the other way to | |
697 | * increment the count). So code holding cgroup_mutex can safely | |
698 | * assume that if the count is zero, it will stay zero. Similarly, if | |
699 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
700 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
701 | * needs that mutex. | |
702 | * | |
703 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't | |
704 | * (usually) take cgroup_mutex. These are the two most performance | |
705 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
706 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
707 | * is taken, and if the cgroup count is zero, a usermode call made | |
708 | * to the release agent with the name of the cgroup (path relative to | |
709 | * the root of cgroup file system) as the argument. | |
710 | * | |
711 | * A cgroup can only be deleted if both its 'count' of using tasks | |
712 | * is zero, and its list of 'children' cgroups is empty. Since all | |
713 | * tasks in the system use _some_ cgroup, and since there is always at | |
714 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
715 | * always has either children cgroups and/or using tasks. So we don't | |
716 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
717 | * | |
718 | * The task_lock() exception | |
719 | * | |
720 | * The need for this exception arises from the action of | |
721 | * cgroup_attach_task(), which overwrites one tasks cgroup pointer with | |
722 | * another. It does so using cgroup_mutex, however there are | |
723 | * several performance critical places that need to reference | |
724 | * task->cgroup without the expense of grabbing a system global | |
725 | * mutex. Therefore except as noted below, when dereferencing or, as | |
726 | * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use | |
727 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in | |
728 | * the task_struct routinely used for such matters. | |
729 | * | |
730 | * P.S. One more locking exception. RCU is used to guard the | |
731 | * update of a tasks cgroup pointer by cgroup_attach_task() | |
732 | */ | |
733 | ||
734 | /** | |
735 | * cgroup_lock - lock out any changes to cgroup structures | |
736 | * | |
737 | */ | |
738 | void cgroup_lock(void) | |
739 | { | |
740 | mutex_lock(&cgroup_mutex); | |
741 | } | |
742 | EXPORT_SYMBOL_GPL(cgroup_lock); | |
743 | ||
744 | /** | |
745 | * cgroup_unlock - release lock on cgroup changes | |
746 | * | |
747 | * Undo the lock taken in a previous cgroup_lock() call. | |
748 | */ | |
749 | void cgroup_unlock(void) | |
750 | { | |
751 | mutex_unlock(&cgroup_mutex); | |
752 | } | |
753 | EXPORT_SYMBOL_GPL(cgroup_unlock); | |
754 | ||
755 | /* | |
756 | * A couple of forward declarations required, due to cyclic reference loop: | |
757 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> | |
758 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations | |
759 | * -> cgroup_mkdir. | |
760 | */ | |
761 | ||
762 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode); | |
763 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); | |
764 | static int cgroup_populate_dir(struct cgroup *cgrp); | |
765 | static const struct inode_operations cgroup_dir_inode_operations; | |
766 | static const struct file_operations proc_cgroupstats_operations; | |
767 | ||
768 | static struct backing_dev_info cgroup_backing_dev_info = { | |
769 | .name = "cgroup", | |
770 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, | |
771 | }; | |
772 | ||
773 | static int alloc_css_id(struct cgroup_subsys *ss, | |
774 | struct cgroup *parent, struct cgroup *child); | |
775 | ||
776 | static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb) | |
777 | { | |
778 | struct inode *inode = new_inode(sb); | |
779 | ||
780 | if (inode) { | |
781 | inode->i_mode = mode; | |
782 | inode->i_uid = current_fsuid(); | |
783 | inode->i_gid = current_fsgid(); | |
784 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
785 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; | |
786 | } | |
787 | return inode; | |
788 | } | |
789 | ||
790 | /* | |
791 | * Call subsys's pre_destroy handler. | |
792 | * This is called before css refcnt check. | |
793 | */ | |
794 | static int cgroup_call_pre_destroy(struct cgroup *cgrp) | |
795 | { | |
796 | struct cgroup_subsys *ss; | |
797 | int ret = 0; | |
798 | ||
799 | for_each_subsys(cgrp->root, ss) | |
800 | if (ss->pre_destroy) { | |
801 | ret = ss->pre_destroy(ss, cgrp); | |
802 | if (ret) | |
803 | break; | |
804 | } | |
805 | ||
806 | return ret; | |
807 | } | |
808 | ||
809 | static void free_cgroup_rcu(struct rcu_head *obj) | |
810 | { | |
811 | struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head); | |
812 | ||
813 | kfree(cgrp); | |
814 | } | |
815 | ||
816 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) | |
817 | { | |
818 | /* is dentry a directory ? if so, kfree() associated cgroup */ | |
819 | if (S_ISDIR(inode->i_mode)) { | |
820 | struct cgroup *cgrp = dentry->d_fsdata; | |
821 | struct cgroup_subsys *ss; | |
822 | BUG_ON(!(cgroup_is_removed(cgrp))); | |
823 | /* It's possible for external users to be holding css | |
824 | * reference counts on a cgroup; css_put() needs to | |
825 | * be able to access the cgroup after decrementing | |
826 | * the reference count in order to know if it needs to | |
827 | * queue the cgroup to be handled by the release | |
828 | * agent */ | |
829 | synchronize_rcu(); | |
830 | ||
831 | mutex_lock(&cgroup_mutex); | |
832 | /* | |
833 | * Release the subsystem state objects. | |
834 | */ | |
835 | for_each_subsys(cgrp->root, ss) | |
836 | ss->destroy(ss, cgrp); | |
837 | ||
838 | cgrp->root->number_of_cgroups--; | |
839 | mutex_unlock(&cgroup_mutex); | |
840 | ||
841 | /* | |
842 | * Drop the active superblock reference that we took when we | |
843 | * created the cgroup | |
844 | */ | |
845 | deactivate_super(cgrp->root->sb); | |
846 | ||
847 | /* | |
848 | * if we're getting rid of the cgroup, refcount should ensure | |
849 | * that there are no pidlists left. | |
850 | */ | |
851 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
852 | ||
853 | call_rcu(&cgrp->rcu_head, free_cgroup_rcu); | |
854 | } | |
855 | iput(inode); | |
856 | } | |
857 | ||
858 | static void remove_dir(struct dentry *d) | |
859 | { | |
860 | struct dentry *parent = dget(d->d_parent); | |
861 | ||
862 | d_delete(d); | |
863 | simple_rmdir(parent->d_inode, d); | |
864 | dput(parent); | |
865 | } | |
866 | ||
867 | static void cgroup_clear_directory(struct dentry *dentry) | |
868 | { | |
869 | struct list_head *node; | |
870 | ||
871 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
872 | spin_lock(&dcache_lock); | |
873 | node = dentry->d_subdirs.next; | |
874 | while (node != &dentry->d_subdirs) { | |
875 | struct dentry *d = list_entry(node, struct dentry, d_u.d_child); | |
876 | list_del_init(node); | |
877 | if (d->d_inode) { | |
878 | /* This should never be called on a cgroup | |
879 | * directory with child cgroups */ | |
880 | BUG_ON(d->d_inode->i_mode & S_IFDIR); | |
881 | d = dget_locked(d); | |
882 | spin_unlock(&dcache_lock); | |
883 | d_delete(d); | |
884 | simple_unlink(dentry->d_inode, d); | |
885 | dput(d); | |
886 | spin_lock(&dcache_lock); | |
887 | } | |
888 | node = dentry->d_subdirs.next; | |
889 | } | |
890 | spin_unlock(&dcache_lock); | |
891 | } | |
892 | ||
893 | /* | |
894 | * NOTE : the dentry must have been dget()'ed | |
895 | */ | |
896 | static void cgroup_d_remove_dir(struct dentry *dentry) | |
897 | { | |
898 | cgroup_clear_directory(dentry); | |
899 | ||
900 | spin_lock(&dcache_lock); | |
901 | list_del_init(&dentry->d_u.d_child); | |
902 | spin_unlock(&dcache_lock); | |
903 | remove_dir(dentry); | |
904 | } | |
905 | ||
906 | /* | |
907 | * A queue for waiters to do rmdir() cgroup. A tasks will sleep when | |
908 | * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some | |
909 | * reference to css->refcnt. In general, this refcnt is expected to goes down | |
910 | * to zero, soon. | |
911 | * | |
912 | * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex; | |
913 | */ | |
914 | DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq); | |
915 | ||
916 | static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp) | |
917 | { | |
918 | if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))) | |
919 | wake_up_all(&cgroup_rmdir_waitq); | |
920 | } | |
921 | ||
922 | void cgroup_exclude_rmdir(struct cgroup_subsys_state *css) | |
923 | { | |
924 | css_get(css); | |
925 | } | |
926 | ||
927 | void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css) | |
928 | { | |
929 | cgroup_wakeup_rmdir_waiter(css->cgroup); | |
930 | css_put(css); | |
931 | } | |
932 | ||
933 | /* | |
934 | * Call with cgroup_mutex held. Drops reference counts on modules, including | |
935 | * any duplicate ones that parse_cgroupfs_options took. If this function | |
936 | * returns an error, no reference counts are touched. | |
937 | */ | |
938 | static int rebind_subsystems(struct cgroupfs_root *root, | |
939 | unsigned long final_bits) | |
940 | { | |
941 | unsigned long added_bits, removed_bits; | |
942 | struct cgroup *cgrp = &root->top_cgroup; | |
943 | int i; | |
944 | ||
945 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
946 | ||
947 | removed_bits = root->actual_subsys_bits & ~final_bits; | |
948 | added_bits = final_bits & ~root->actual_subsys_bits; | |
949 | /* Check that any added subsystems are currently free */ | |
950 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
951 | unsigned long bit = 1UL << i; | |
952 | struct cgroup_subsys *ss = subsys[i]; | |
953 | if (!(bit & added_bits)) | |
954 | continue; | |
955 | /* | |
956 | * Nobody should tell us to do a subsys that doesn't exist: | |
957 | * parse_cgroupfs_options should catch that case and refcounts | |
958 | * ensure that subsystems won't disappear once selected. | |
959 | */ | |
960 | BUG_ON(ss == NULL); | |
961 | if (ss->root != &rootnode) { | |
962 | /* Subsystem isn't free */ | |
963 | return -EBUSY; | |
964 | } | |
965 | } | |
966 | ||
967 | /* Currently we don't handle adding/removing subsystems when | |
968 | * any child cgroups exist. This is theoretically supportable | |
969 | * but involves complex error handling, so it's being left until | |
970 | * later */ | |
971 | if (root->number_of_cgroups > 1) | |
972 | return -EBUSY; | |
973 | ||
974 | /* Process each subsystem */ | |
975 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
976 | struct cgroup_subsys *ss = subsys[i]; | |
977 | unsigned long bit = 1UL << i; | |
978 | if (bit & added_bits) { | |
979 | /* We're binding this subsystem to this hierarchy */ | |
980 | BUG_ON(ss == NULL); | |
981 | BUG_ON(cgrp->subsys[i]); | |
982 | BUG_ON(!dummytop->subsys[i]); | |
983 | BUG_ON(dummytop->subsys[i]->cgroup != dummytop); | |
984 | mutex_lock(&ss->hierarchy_mutex); | |
985 | cgrp->subsys[i] = dummytop->subsys[i]; | |
986 | cgrp->subsys[i]->cgroup = cgrp; | |
987 | list_move(&ss->sibling, &root->subsys_list); | |
988 | ss->root = root; | |
989 | if (ss->bind) | |
990 | ss->bind(ss, cgrp); | |
991 | mutex_unlock(&ss->hierarchy_mutex); | |
992 | /* refcount was already taken, and we're keeping it */ | |
993 | } else if (bit & removed_bits) { | |
994 | /* We're removing this subsystem */ | |
995 | BUG_ON(ss == NULL); | |
996 | BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]); | |
997 | BUG_ON(cgrp->subsys[i]->cgroup != cgrp); | |
998 | mutex_lock(&ss->hierarchy_mutex); | |
999 | if (ss->bind) | |
1000 | ss->bind(ss, dummytop); | |
1001 | dummytop->subsys[i]->cgroup = dummytop; | |
1002 | cgrp->subsys[i] = NULL; | |
1003 | subsys[i]->root = &rootnode; | |
1004 | list_move(&ss->sibling, &rootnode.subsys_list); | |
1005 | mutex_unlock(&ss->hierarchy_mutex); | |
1006 | /* subsystem is now free - drop reference on module */ | |
1007 | module_put(ss->module); | |
1008 | } else if (bit & final_bits) { | |
1009 | /* Subsystem state should already exist */ | |
1010 | BUG_ON(ss == NULL); | |
1011 | BUG_ON(!cgrp->subsys[i]); | |
1012 | /* | |
1013 | * a refcount was taken, but we already had one, so | |
1014 | * drop the extra reference. | |
1015 | */ | |
1016 | module_put(ss->module); | |
1017 | #ifdef CONFIG_MODULE_UNLOAD | |
1018 | BUG_ON(ss->module && !module_refcount(ss->module)); | |
1019 | #endif | |
1020 | } else { | |
1021 | /* Subsystem state shouldn't exist */ | |
1022 | BUG_ON(cgrp->subsys[i]); | |
1023 | } | |
1024 | } | |
1025 | root->subsys_bits = root->actual_subsys_bits = final_bits; | |
1026 | synchronize_rcu(); | |
1027 | ||
1028 | return 0; | |
1029 | } | |
1030 | ||
1031 | static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs) | |
1032 | { | |
1033 | struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info; | |
1034 | struct cgroup_subsys *ss; | |
1035 | ||
1036 | mutex_lock(&cgroup_mutex); | |
1037 | for_each_subsys(root, ss) | |
1038 | seq_printf(seq, ",%s", ss->name); | |
1039 | if (test_bit(ROOT_NOPREFIX, &root->flags)) | |
1040 | seq_puts(seq, ",noprefix"); | |
1041 | if (strlen(root->release_agent_path)) | |
1042 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
1043 | if (strlen(root->name)) | |
1044 | seq_printf(seq, ",name=%s", root->name); | |
1045 | mutex_unlock(&cgroup_mutex); | |
1046 | return 0; | |
1047 | } | |
1048 | ||
1049 | struct cgroup_sb_opts { | |
1050 | unsigned long subsys_bits; | |
1051 | unsigned long flags; | |
1052 | char *release_agent; | |
1053 | char *name; | |
1054 | /* User explicitly requested empty subsystem */ | |
1055 | bool none; | |
1056 | ||
1057 | struct cgroupfs_root *new_root; | |
1058 | ||
1059 | }; | |
1060 | ||
1061 | /* | |
1062 | * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call | |
1063 | * with cgroup_mutex held to protect the subsys[] array. This function takes | |
1064 | * refcounts on subsystems to be used, unless it returns error, in which case | |
1065 | * no refcounts are taken. | |
1066 | */ | |
1067 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) | |
1068 | { | |
1069 | char *token, *o = data ?: "all"; | |
1070 | unsigned long mask = (unsigned long)-1; | |
1071 | int i; | |
1072 | bool module_pin_failed = false; | |
1073 | ||
1074 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
1075 | ||
1076 | #ifdef CONFIG_CPUSETS | |
1077 | mask = ~(1UL << cpuset_subsys_id); | |
1078 | #endif | |
1079 | ||
1080 | memset(opts, 0, sizeof(*opts)); | |
1081 | ||
1082 | while ((token = strsep(&o, ",")) != NULL) { | |
1083 | if (!*token) | |
1084 | return -EINVAL; | |
1085 | if (!strcmp(token, "all")) { | |
1086 | /* Add all non-disabled subsystems */ | |
1087 | opts->subsys_bits = 0; | |
1088 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1089 | struct cgroup_subsys *ss = subsys[i]; | |
1090 | if (ss == NULL) | |
1091 | continue; | |
1092 | if (!ss->disabled) | |
1093 | opts->subsys_bits |= 1ul << i; | |
1094 | } | |
1095 | } else if (!strcmp(token, "none")) { | |
1096 | /* Explicitly have no subsystems */ | |
1097 | opts->none = true; | |
1098 | } else if (!strcmp(token, "noprefix")) { | |
1099 | set_bit(ROOT_NOPREFIX, &opts->flags); | |
1100 | } else if (!strncmp(token, "release_agent=", 14)) { | |
1101 | /* Specifying two release agents is forbidden */ | |
1102 | if (opts->release_agent) | |
1103 | return -EINVAL; | |
1104 | opts->release_agent = | |
1105 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); | |
1106 | if (!opts->release_agent) | |
1107 | return -ENOMEM; | |
1108 | } else if (!strncmp(token, "name=", 5)) { | |
1109 | const char *name = token + 5; | |
1110 | /* Can't specify an empty name */ | |
1111 | if (!strlen(name)) | |
1112 | return -EINVAL; | |
1113 | /* Must match [\w.-]+ */ | |
1114 | for (i = 0; i < strlen(name); i++) { | |
1115 | char c = name[i]; | |
1116 | if (isalnum(c)) | |
1117 | continue; | |
1118 | if ((c == '.') || (c == '-') || (c == '_')) | |
1119 | continue; | |
1120 | return -EINVAL; | |
1121 | } | |
1122 | /* Specifying two names is forbidden */ | |
1123 | if (opts->name) | |
1124 | return -EINVAL; | |
1125 | opts->name = kstrndup(name, | |
1126 | MAX_CGROUP_ROOT_NAMELEN - 1, | |
1127 | GFP_KERNEL); | |
1128 | if (!opts->name) | |
1129 | return -ENOMEM; | |
1130 | } else { | |
1131 | struct cgroup_subsys *ss; | |
1132 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1133 | ss = subsys[i]; | |
1134 | if (ss == NULL) | |
1135 | continue; | |
1136 | if (!strcmp(token, ss->name)) { | |
1137 | if (!ss->disabled) | |
1138 | set_bit(i, &opts->subsys_bits); | |
1139 | break; | |
1140 | } | |
1141 | } | |
1142 | if (i == CGROUP_SUBSYS_COUNT) | |
1143 | return -ENOENT; | |
1144 | } | |
1145 | } | |
1146 | ||
1147 | /* Consistency checks */ | |
1148 | ||
1149 | /* | |
1150 | * Option noprefix was introduced just for backward compatibility | |
1151 | * with the old cpuset, so we allow noprefix only if mounting just | |
1152 | * the cpuset subsystem. | |
1153 | */ | |
1154 | if (test_bit(ROOT_NOPREFIX, &opts->flags) && | |
1155 | (opts->subsys_bits & mask)) | |
1156 | return -EINVAL; | |
1157 | ||
1158 | ||
1159 | /* Can't specify "none" and some subsystems */ | |
1160 | if (opts->subsys_bits && opts->none) | |
1161 | return -EINVAL; | |
1162 | ||
1163 | /* | |
1164 | * We either have to specify by name or by subsystems. (So all | |
1165 | * empty hierarchies must have a name). | |
1166 | */ | |
1167 | if (!opts->subsys_bits && !opts->name) | |
1168 | return -EINVAL; | |
1169 | ||
1170 | /* | |
1171 | * Grab references on all the modules we'll need, so the subsystems | |
1172 | * don't dance around before rebind_subsystems attaches them. This may | |
1173 | * take duplicate reference counts on a subsystem that's already used, | |
1174 | * but rebind_subsystems handles this case. | |
1175 | */ | |
1176 | for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) { | |
1177 | unsigned long bit = 1UL << i; | |
1178 | ||
1179 | if (!(bit & opts->subsys_bits)) | |
1180 | continue; | |
1181 | if (!try_module_get(subsys[i]->module)) { | |
1182 | module_pin_failed = true; | |
1183 | break; | |
1184 | } | |
1185 | } | |
1186 | if (module_pin_failed) { | |
1187 | /* | |
1188 | * oops, one of the modules was going away. this means that we | |
1189 | * raced with a module_delete call, and to the user this is | |
1190 | * essentially a "subsystem doesn't exist" case. | |
1191 | */ | |
1192 | for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) { | |
1193 | /* drop refcounts only on the ones we took */ | |
1194 | unsigned long bit = 1UL << i; | |
1195 | ||
1196 | if (!(bit & opts->subsys_bits)) | |
1197 | continue; | |
1198 | module_put(subsys[i]->module); | |
1199 | } | |
1200 | return -ENOENT; | |
1201 | } | |
1202 | ||
1203 | return 0; | |
1204 | } | |
1205 | ||
1206 | static void drop_parsed_module_refcounts(unsigned long subsys_bits) | |
1207 | { | |
1208 | int i; | |
1209 | for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) { | |
1210 | unsigned long bit = 1UL << i; | |
1211 | ||
1212 | if (!(bit & subsys_bits)) | |
1213 | continue; | |
1214 | module_put(subsys[i]->module); | |
1215 | } | |
1216 | } | |
1217 | ||
1218 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) | |
1219 | { | |
1220 | int ret = 0; | |
1221 | struct cgroupfs_root *root = sb->s_fs_info; | |
1222 | struct cgroup *cgrp = &root->top_cgroup; | |
1223 | struct cgroup_sb_opts opts; | |
1224 | ||
1225 | lock_kernel(); | |
1226 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); | |
1227 | mutex_lock(&cgroup_mutex); | |
1228 | ||
1229 | /* See what subsystems are wanted */ | |
1230 | ret = parse_cgroupfs_options(data, &opts); | |
1231 | if (ret) | |
1232 | goto out_unlock; | |
1233 | ||
1234 | /* Don't allow flags or name to change at remount */ | |
1235 | if (opts.flags != root->flags || | |
1236 | (opts.name && strcmp(opts.name, root->name))) { | |
1237 | ret = -EINVAL; | |
1238 | drop_parsed_module_refcounts(opts.subsys_bits); | |
1239 | goto out_unlock; | |
1240 | } | |
1241 | ||
1242 | ret = rebind_subsystems(root, opts.subsys_bits); | |
1243 | if (ret) { | |
1244 | drop_parsed_module_refcounts(opts.subsys_bits); | |
1245 | goto out_unlock; | |
1246 | } | |
1247 | ||
1248 | /* (re)populate subsystem files */ | |
1249 | cgroup_populate_dir(cgrp); | |
1250 | ||
1251 | if (opts.release_agent) | |
1252 | strcpy(root->release_agent_path, opts.release_agent); | |
1253 | out_unlock: | |
1254 | kfree(opts.release_agent); | |
1255 | kfree(opts.name); | |
1256 | mutex_unlock(&cgroup_mutex); | |
1257 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); | |
1258 | unlock_kernel(); | |
1259 | return ret; | |
1260 | } | |
1261 | ||
1262 | static const struct super_operations cgroup_ops = { | |
1263 | .statfs = simple_statfs, | |
1264 | .drop_inode = generic_delete_inode, | |
1265 | .show_options = cgroup_show_options, | |
1266 | .remount_fs = cgroup_remount, | |
1267 | }; | |
1268 | ||
1269 | static void init_cgroup_housekeeping(struct cgroup *cgrp) | |
1270 | { | |
1271 | INIT_LIST_HEAD(&cgrp->sibling); | |
1272 | INIT_LIST_HEAD(&cgrp->children); | |
1273 | INIT_LIST_HEAD(&cgrp->css_sets); | |
1274 | INIT_LIST_HEAD(&cgrp->release_list); | |
1275 | INIT_LIST_HEAD(&cgrp->pidlists); | |
1276 | mutex_init(&cgrp->pidlist_mutex); | |
1277 | INIT_LIST_HEAD(&cgrp->event_list); | |
1278 | spin_lock_init(&cgrp->event_list_lock); | |
1279 | } | |
1280 | ||
1281 | static void init_cgroup_root(struct cgroupfs_root *root) | |
1282 | { | |
1283 | struct cgroup *cgrp = &root->top_cgroup; | |
1284 | INIT_LIST_HEAD(&root->subsys_list); | |
1285 | INIT_LIST_HEAD(&root->root_list); | |
1286 | root->number_of_cgroups = 1; | |
1287 | cgrp->root = root; | |
1288 | cgrp->top_cgroup = cgrp; | |
1289 | init_cgroup_housekeeping(cgrp); | |
1290 | } | |
1291 | ||
1292 | static bool init_root_id(struct cgroupfs_root *root) | |
1293 | { | |
1294 | int ret = 0; | |
1295 | ||
1296 | do { | |
1297 | if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL)) | |
1298 | return false; | |
1299 | spin_lock(&hierarchy_id_lock); | |
1300 | /* Try to allocate the next unused ID */ | |
1301 | ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id, | |
1302 | &root->hierarchy_id); | |
1303 | if (ret == -ENOSPC) | |
1304 | /* Try again starting from 0 */ | |
1305 | ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id); | |
1306 | if (!ret) { | |
1307 | next_hierarchy_id = root->hierarchy_id + 1; | |
1308 | } else if (ret != -EAGAIN) { | |
1309 | /* Can only get here if the 31-bit IDR is full ... */ | |
1310 | BUG_ON(ret); | |
1311 | } | |
1312 | spin_unlock(&hierarchy_id_lock); | |
1313 | } while (ret); | |
1314 | return true; | |
1315 | } | |
1316 | ||
1317 | static int cgroup_test_super(struct super_block *sb, void *data) | |
1318 | { | |
1319 | struct cgroup_sb_opts *opts = data; | |
1320 | struct cgroupfs_root *root = sb->s_fs_info; | |
1321 | ||
1322 | /* If we asked for a name then it must match */ | |
1323 | if (opts->name && strcmp(opts->name, root->name)) | |
1324 | return 0; | |
1325 | ||
1326 | /* | |
1327 | * If we asked for subsystems (or explicitly for no | |
1328 | * subsystems) then they must match | |
1329 | */ | |
1330 | if ((opts->subsys_bits || opts->none) | |
1331 | && (opts->subsys_bits != root->subsys_bits)) | |
1332 | return 0; | |
1333 | ||
1334 | return 1; | |
1335 | } | |
1336 | ||
1337 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) | |
1338 | { | |
1339 | struct cgroupfs_root *root; | |
1340 | ||
1341 | if (!opts->subsys_bits && !opts->none) | |
1342 | return NULL; | |
1343 | ||
1344 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1345 | if (!root) | |
1346 | return ERR_PTR(-ENOMEM); | |
1347 | ||
1348 | if (!init_root_id(root)) { | |
1349 | kfree(root); | |
1350 | return ERR_PTR(-ENOMEM); | |
1351 | } | |
1352 | init_cgroup_root(root); | |
1353 | ||
1354 | root->subsys_bits = opts->subsys_bits; | |
1355 | root->flags = opts->flags; | |
1356 | if (opts->release_agent) | |
1357 | strcpy(root->release_agent_path, opts->release_agent); | |
1358 | if (opts->name) | |
1359 | strcpy(root->name, opts->name); | |
1360 | return root; | |
1361 | } | |
1362 | ||
1363 | static void cgroup_drop_root(struct cgroupfs_root *root) | |
1364 | { | |
1365 | if (!root) | |
1366 | return; | |
1367 | ||
1368 | BUG_ON(!root->hierarchy_id); | |
1369 | spin_lock(&hierarchy_id_lock); | |
1370 | ida_remove(&hierarchy_ida, root->hierarchy_id); | |
1371 | spin_unlock(&hierarchy_id_lock); | |
1372 | kfree(root); | |
1373 | } | |
1374 | ||
1375 | static int cgroup_set_super(struct super_block *sb, void *data) | |
1376 | { | |
1377 | int ret; | |
1378 | struct cgroup_sb_opts *opts = data; | |
1379 | ||
1380 | /* If we don't have a new root, we can't set up a new sb */ | |
1381 | if (!opts->new_root) | |
1382 | return -EINVAL; | |
1383 | ||
1384 | BUG_ON(!opts->subsys_bits && !opts->none); | |
1385 | ||
1386 | ret = set_anon_super(sb, NULL); | |
1387 | if (ret) | |
1388 | return ret; | |
1389 | ||
1390 | sb->s_fs_info = opts->new_root; | |
1391 | opts->new_root->sb = sb; | |
1392 | ||
1393 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
1394 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
1395 | sb->s_magic = CGROUP_SUPER_MAGIC; | |
1396 | sb->s_op = &cgroup_ops; | |
1397 | ||
1398 | return 0; | |
1399 | } | |
1400 | ||
1401 | static int cgroup_get_rootdir(struct super_block *sb) | |
1402 | { | |
1403 | struct inode *inode = | |
1404 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); | |
1405 | struct dentry *dentry; | |
1406 | ||
1407 | if (!inode) | |
1408 | return -ENOMEM; | |
1409 | ||
1410 | inode->i_fop = &simple_dir_operations; | |
1411 | inode->i_op = &cgroup_dir_inode_operations; | |
1412 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
1413 | inc_nlink(inode); | |
1414 | dentry = d_alloc_root(inode); | |
1415 | if (!dentry) { | |
1416 | iput(inode); | |
1417 | return -ENOMEM; | |
1418 | } | |
1419 | sb->s_root = dentry; | |
1420 | return 0; | |
1421 | } | |
1422 | ||
1423 | static int cgroup_get_sb(struct file_system_type *fs_type, | |
1424 | int flags, const char *unused_dev_name, | |
1425 | void *data, struct vfsmount *mnt) | |
1426 | { | |
1427 | struct cgroup_sb_opts opts; | |
1428 | struct cgroupfs_root *root; | |
1429 | int ret = 0; | |
1430 | struct super_block *sb; | |
1431 | struct cgroupfs_root *new_root; | |
1432 | ||
1433 | /* First find the desired set of subsystems */ | |
1434 | mutex_lock(&cgroup_mutex); | |
1435 | ret = parse_cgroupfs_options(data, &opts); | |
1436 | mutex_unlock(&cgroup_mutex); | |
1437 | if (ret) | |
1438 | goto out_err; | |
1439 | ||
1440 | /* | |
1441 | * Allocate a new cgroup root. We may not need it if we're | |
1442 | * reusing an existing hierarchy. | |
1443 | */ | |
1444 | new_root = cgroup_root_from_opts(&opts); | |
1445 | if (IS_ERR(new_root)) { | |
1446 | ret = PTR_ERR(new_root); | |
1447 | goto drop_modules; | |
1448 | } | |
1449 | opts.new_root = new_root; | |
1450 | ||
1451 | /* Locate an existing or new sb for this hierarchy */ | |
1452 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts); | |
1453 | if (IS_ERR(sb)) { | |
1454 | ret = PTR_ERR(sb); | |
1455 | cgroup_drop_root(opts.new_root); | |
1456 | goto drop_modules; | |
1457 | } | |
1458 | ||
1459 | root = sb->s_fs_info; | |
1460 | BUG_ON(!root); | |
1461 | if (root == opts.new_root) { | |
1462 | /* We used the new root structure, so this is a new hierarchy */ | |
1463 | struct list_head tmp_cg_links; | |
1464 | struct cgroup *root_cgrp = &root->top_cgroup; | |
1465 | struct inode *inode; | |
1466 | struct cgroupfs_root *existing_root; | |
1467 | int i; | |
1468 | ||
1469 | BUG_ON(sb->s_root != NULL); | |
1470 | ||
1471 | ret = cgroup_get_rootdir(sb); | |
1472 | if (ret) | |
1473 | goto drop_new_super; | |
1474 | inode = sb->s_root->d_inode; | |
1475 | ||
1476 | mutex_lock(&inode->i_mutex); | |
1477 | mutex_lock(&cgroup_mutex); | |
1478 | ||
1479 | if (strlen(root->name)) { | |
1480 | /* Check for name clashes with existing mounts */ | |
1481 | for_each_active_root(existing_root) { | |
1482 | if (!strcmp(existing_root->name, root->name)) { | |
1483 | ret = -EBUSY; | |
1484 | mutex_unlock(&cgroup_mutex); | |
1485 | mutex_unlock(&inode->i_mutex); | |
1486 | goto drop_new_super; | |
1487 | } | |
1488 | } | |
1489 | } | |
1490 | ||
1491 | /* | |
1492 | * We're accessing css_set_count without locking | |
1493 | * css_set_lock here, but that's OK - it can only be | |
1494 | * increased by someone holding cgroup_lock, and | |
1495 | * that's us. The worst that can happen is that we | |
1496 | * have some link structures left over | |
1497 | */ | |
1498 | ret = allocate_cg_links(css_set_count, &tmp_cg_links); | |
1499 | if (ret) { | |
1500 | mutex_unlock(&cgroup_mutex); | |
1501 | mutex_unlock(&inode->i_mutex); | |
1502 | goto drop_new_super; | |
1503 | } | |
1504 | ||
1505 | ret = rebind_subsystems(root, root->subsys_bits); | |
1506 | if (ret == -EBUSY) { | |
1507 | mutex_unlock(&cgroup_mutex); | |
1508 | mutex_unlock(&inode->i_mutex); | |
1509 | free_cg_links(&tmp_cg_links); | |
1510 | goto drop_new_super; | |
1511 | } | |
1512 | /* | |
1513 | * There must be no failure case after here, since rebinding | |
1514 | * takes care of subsystems' refcounts, which are explicitly | |
1515 | * dropped in the failure exit path. | |
1516 | */ | |
1517 | ||
1518 | /* EBUSY should be the only error here */ | |
1519 | BUG_ON(ret); | |
1520 | ||
1521 | list_add(&root->root_list, &roots); | |
1522 | root_count++; | |
1523 | ||
1524 | sb->s_root->d_fsdata = root_cgrp; | |
1525 | root->top_cgroup.dentry = sb->s_root; | |
1526 | ||
1527 | /* Link the top cgroup in this hierarchy into all | |
1528 | * the css_set objects */ | |
1529 | write_lock(&css_set_lock); | |
1530 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { | |
1531 | struct hlist_head *hhead = &css_set_table[i]; | |
1532 | struct hlist_node *node; | |
1533 | struct css_set *cg; | |
1534 | ||
1535 | hlist_for_each_entry(cg, node, hhead, hlist) | |
1536 | link_css_set(&tmp_cg_links, cg, root_cgrp); | |
1537 | } | |
1538 | write_unlock(&css_set_lock); | |
1539 | ||
1540 | free_cg_links(&tmp_cg_links); | |
1541 | ||
1542 | BUG_ON(!list_empty(&root_cgrp->sibling)); | |
1543 | BUG_ON(!list_empty(&root_cgrp->children)); | |
1544 | BUG_ON(root->number_of_cgroups != 1); | |
1545 | ||
1546 | cgroup_populate_dir(root_cgrp); | |
1547 | mutex_unlock(&cgroup_mutex); | |
1548 | mutex_unlock(&inode->i_mutex); | |
1549 | } else { | |
1550 | /* | |
1551 | * We re-used an existing hierarchy - the new root (if | |
1552 | * any) is not needed | |
1553 | */ | |
1554 | cgroup_drop_root(opts.new_root); | |
1555 | /* no subsys rebinding, so refcounts don't change */ | |
1556 | drop_parsed_module_refcounts(opts.subsys_bits); | |
1557 | } | |
1558 | ||
1559 | simple_set_mnt(mnt, sb); | |
1560 | kfree(opts.release_agent); | |
1561 | kfree(opts.name); | |
1562 | return 0; | |
1563 | ||
1564 | drop_new_super: | |
1565 | deactivate_locked_super(sb); | |
1566 | drop_modules: | |
1567 | drop_parsed_module_refcounts(opts.subsys_bits); | |
1568 | out_err: | |
1569 | kfree(opts.release_agent); | |
1570 | kfree(opts.name); | |
1571 | ||
1572 | return ret; | |
1573 | } | |
1574 | ||
1575 | static void cgroup_kill_sb(struct super_block *sb) { | |
1576 | struct cgroupfs_root *root = sb->s_fs_info; | |
1577 | struct cgroup *cgrp = &root->top_cgroup; | |
1578 | int ret; | |
1579 | struct cg_cgroup_link *link; | |
1580 | struct cg_cgroup_link *saved_link; | |
1581 | ||
1582 | BUG_ON(!root); | |
1583 | ||
1584 | BUG_ON(root->number_of_cgroups != 1); | |
1585 | BUG_ON(!list_empty(&cgrp->children)); | |
1586 | BUG_ON(!list_empty(&cgrp->sibling)); | |
1587 | ||
1588 | mutex_lock(&cgroup_mutex); | |
1589 | ||
1590 | /* Rebind all subsystems back to the default hierarchy */ | |
1591 | ret = rebind_subsystems(root, 0); | |
1592 | /* Shouldn't be able to fail ... */ | |
1593 | BUG_ON(ret); | |
1594 | ||
1595 | /* | |
1596 | * Release all the links from css_sets to this hierarchy's | |
1597 | * root cgroup | |
1598 | */ | |
1599 | write_lock(&css_set_lock); | |
1600 | ||
1601 | list_for_each_entry_safe(link, saved_link, &cgrp->css_sets, | |
1602 | cgrp_link_list) { | |
1603 | list_del(&link->cg_link_list); | |
1604 | list_del(&link->cgrp_link_list); | |
1605 | kfree(link); | |
1606 | } | |
1607 | write_unlock(&css_set_lock); | |
1608 | ||
1609 | if (!list_empty(&root->root_list)) { | |
1610 | list_del(&root->root_list); | |
1611 | root_count--; | |
1612 | } | |
1613 | ||
1614 | mutex_unlock(&cgroup_mutex); | |
1615 | ||
1616 | kill_litter_super(sb); | |
1617 | cgroup_drop_root(root); | |
1618 | } | |
1619 | ||
1620 | static struct file_system_type cgroup_fs_type = { | |
1621 | .name = "cgroup", | |
1622 | .get_sb = cgroup_get_sb, | |
1623 | .kill_sb = cgroup_kill_sb, | |
1624 | }; | |
1625 | ||
1626 | static struct kobject *cgroup_kobj; | |
1627 | ||
1628 | static inline struct cgroup *__d_cgrp(struct dentry *dentry) | |
1629 | { | |
1630 | return dentry->d_fsdata; | |
1631 | } | |
1632 | ||
1633 | static inline struct cftype *__d_cft(struct dentry *dentry) | |
1634 | { | |
1635 | return dentry->d_fsdata; | |
1636 | } | |
1637 | ||
1638 | /** | |
1639 | * cgroup_path - generate the path of a cgroup | |
1640 | * @cgrp: the cgroup in question | |
1641 | * @buf: the buffer to write the path into | |
1642 | * @buflen: the length of the buffer | |
1643 | * | |
1644 | * Called with cgroup_mutex held or else with an RCU-protected cgroup | |
1645 | * reference. Writes path of cgroup into buf. Returns 0 on success, | |
1646 | * -errno on error. | |
1647 | */ | |
1648 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) | |
1649 | { | |
1650 | char *start; | |
1651 | struct dentry *dentry = rcu_dereference_check(cgrp->dentry, | |
1652 | rcu_read_lock_held() || | |
1653 | cgroup_lock_is_held()); | |
1654 | ||
1655 | if (!dentry || cgrp == dummytop) { | |
1656 | /* | |
1657 | * Inactive subsystems have no dentry for their root | |
1658 | * cgroup | |
1659 | */ | |
1660 | strcpy(buf, "/"); | |
1661 | return 0; | |
1662 | } | |
1663 | ||
1664 | start = buf + buflen; | |
1665 | ||
1666 | *--start = '\0'; | |
1667 | for (;;) { | |
1668 | int len = dentry->d_name.len; | |
1669 | ||
1670 | if ((start -= len) < buf) | |
1671 | return -ENAMETOOLONG; | |
1672 | memcpy(start, dentry->d_name.name, len); | |
1673 | cgrp = cgrp->parent; | |
1674 | if (!cgrp) | |
1675 | break; | |
1676 | ||
1677 | dentry = rcu_dereference_check(cgrp->dentry, | |
1678 | rcu_read_lock_held() || | |
1679 | cgroup_lock_is_held()); | |
1680 | if (!cgrp->parent) | |
1681 | continue; | |
1682 | if (--start < buf) | |
1683 | return -ENAMETOOLONG; | |
1684 | *start = '/'; | |
1685 | } | |
1686 | memmove(buf, start, buf + buflen - start); | |
1687 | return 0; | |
1688 | } | |
1689 | EXPORT_SYMBOL_GPL(cgroup_path); | |
1690 | ||
1691 | /** | |
1692 | * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp' | |
1693 | * @cgrp: the cgroup the task is attaching to | |
1694 | * @tsk: the task to be attached | |
1695 | * | |
1696 | * Call holding cgroup_mutex. May take task_lock of | |
1697 | * the task 'tsk' during call. | |
1698 | */ | |
1699 | int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) | |
1700 | { | |
1701 | int retval = 0; | |
1702 | struct cgroup_subsys *ss, *failed_ss = NULL; | |
1703 | struct cgroup *oldcgrp; | |
1704 | struct css_set *cg; | |
1705 | struct css_set *newcg; | |
1706 | struct cgroupfs_root *root = cgrp->root; | |
1707 | ||
1708 | /* Nothing to do if the task is already in that cgroup */ | |
1709 | oldcgrp = task_cgroup_from_root(tsk, root); | |
1710 | if (cgrp == oldcgrp) | |
1711 | return 0; | |
1712 | ||
1713 | for_each_subsys(root, ss) { | |
1714 | if (ss->can_attach) { | |
1715 | retval = ss->can_attach(ss, cgrp, tsk, false); | |
1716 | if (retval) { | |
1717 | /* | |
1718 | * Remember on which subsystem the can_attach() | |
1719 | * failed, so that we only call cancel_attach() | |
1720 | * against the subsystems whose can_attach() | |
1721 | * succeeded. (See below) | |
1722 | */ | |
1723 | failed_ss = ss; | |
1724 | goto out; | |
1725 | } | |
1726 | } | |
1727 | } | |
1728 | ||
1729 | task_lock(tsk); | |
1730 | cg = tsk->cgroups; | |
1731 | get_css_set(cg); | |
1732 | task_unlock(tsk); | |
1733 | /* | |
1734 | * Locate or allocate a new css_set for this task, | |
1735 | * based on its final set of cgroups | |
1736 | */ | |
1737 | newcg = find_css_set(cg, cgrp); | |
1738 | put_css_set(cg); | |
1739 | if (!newcg) { | |
1740 | retval = -ENOMEM; | |
1741 | goto out; | |
1742 | } | |
1743 | ||
1744 | task_lock(tsk); | |
1745 | if (tsk->flags & PF_EXITING) { | |
1746 | task_unlock(tsk); | |
1747 | put_css_set(newcg); | |
1748 | retval = -ESRCH; | |
1749 | goto out; | |
1750 | } | |
1751 | rcu_assign_pointer(tsk->cgroups, newcg); | |
1752 | task_unlock(tsk); | |
1753 | ||
1754 | /* Update the css_set linked lists if we're using them */ | |
1755 | write_lock(&css_set_lock); | |
1756 | if (!list_empty(&tsk->cg_list)) { | |
1757 | list_del(&tsk->cg_list); | |
1758 | list_add(&tsk->cg_list, &newcg->tasks); | |
1759 | } | |
1760 | write_unlock(&css_set_lock); | |
1761 | ||
1762 | for_each_subsys(root, ss) { | |
1763 | if (ss->attach) | |
1764 | ss->attach(ss, cgrp, oldcgrp, tsk, false); | |
1765 | } | |
1766 | set_bit(CGRP_RELEASABLE, &oldcgrp->flags); | |
1767 | synchronize_rcu(); | |
1768 | put_css_set(cg); | |
1769 | ||
1770 | /* | |
1771 | * wake up rmdir() waiter. the rmdir should fail since the cgroup | |
1772 | * is no longer empty. | |
1773 | */ | |
1774 | cgroup_wakeup_rmdir_waiter(cgrp); | |
1775 | out: | |
1776 | if (retval) { | |
1777 | for_each_subsys(root, ss) { | |
1778 | if (ss == failed_ss) | |
1779 | /* | |
1780 | * This subsystem was the one that failed the | |
1781 | * can_attach() check earlier, so we don't need | |
1782 | * to call cancel_attach() against it or any | |
1783 | * remaining subsystems. | |
1784 | */ | |
1785 | break; | |
1786 | if (ss->cancel_attach) | |
1787 | ss->cancel_attach(ss, cgrp, tsk, false); | |
1788 | } | |
1789 | } | |
1790 | return retval; | |
1791 | } | |
1792 | ||
1793 | /** | |
1794 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
1795 | * @from: attach to all cgroups of a given task | |
1796 | * @tsk: the task to be attached | |
1797 | */ | |
1798 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
1799 | { | |
1800 | struct cgroupfs_root *root; | |
1801 | struct cgroup *cur_cg; | |
1802 | int retval = 0; | |
1803 | ||
1804 | cgroup_lock(); | |
1805 | for_each_active_root(root) { | |
1806 | cur_cg = task_cgroup_from_root(from, root); | |
1807 | retval = cgroup_attach_task(cur_cg, tsk); | |
1808 | if (retval) | |
1809 | break; | |
1810 | } | |
1811 | cgroup_unlock(); | |
1812 | ||
1813 | return retval; | |
1814 | } | |
1815 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
1816 | ||
1817 | /* | |
1818 | * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex | |
1819 | * held. May take task_lock of task | |
1820 | */ | |
1821 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid) | |
1822 | { | |
1823 | struct task_struct *tsk; | |
1824 | const struct cred *cred = current_cred(), *tcred; | |
1825 | int ret; | |
1826 | ||
1827 | if (pid) { | |
1828 | rcu_read_lock(); | |
1829 | tsk = find_task_by_vpid(pid); | |
1830 | if (!tsk || tsk->flags & PF_EXITING) { | |
1831 | rcu_read_unlock(); | |
1832 | return -ESRCH; | |
1833 | } | |
1834 | ||
1835 | tcred = __task_cred(tsk); | |
1836 | if (cred->euid && | |
1837 | cred->euid != tcred->uid && | |
1838 | cred->euid != tcred->suid) { | |
1839 | rcu_read_unlock(); | |
1840 | return -EACCES; | |
1841 | } | |
1842 | get_task_struct(tsk); | |
1843 | rcu_read_unlock(); | |
1844 | } else { | |
1845 | tsk = current; | |
1846 | get_task_struct(tsk); | |
1847 | } | |
1848 | ||
1849 | ret = cgroup_attach_task(cgrp, tsk); | |
1850 | put_task_struct(tsk); | |
1851 | return ret; | |
1852 | } | |
1853 | ||
1854 | static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid) | |
1855 | { | |
1856 | int ret; | |
1857 | if (!cgroup_lock_live_group(cgrp)) | |
1858 | return -ENODEV; | |
1859 | ret = attach_task_by_pid(cgrp, pid); | |
1860 | cgroup_unlock(); | |
1861 | return ret; | |
1862 | } | |
1863 | ||
1864 | /** | |
1865 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
1866 | * @cgrp: the cgroup to be checked for liveness | |
1867 | * | |
1868 | * On success, returns true; the lock should be later released with | |
1869 | * cgroup_unlock(). On failure returns false with no lock held. | |
1870 | */ | |
1871 | bool cgroup_lock_live_group(struct cgroup *cgrp) | |
1872 | { | |
1873 | mutex_lock(&cgroup_mutex); | |
1874 | if (cgroup_is_removed(cgrp)) { | |
1875 | mutex_unlock(&cgroup_mutex); | |
1876 | return false; | |
1877 | } | |
1878 | return true; | |
1879 | } | |
1880 | EXPORT_SYMBOL_GPL(cgroup_lock_live_group); | |
1881 | ||
1882 | static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft, | |
1883 | const char *buffer) | |
1884 | { | |
1885 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
1886 | if (!cgroup_lock_live_group(cgrp)) | |
1887 | return -ENODEV; | |
1888 | strcpy(cgrp->root->release_agent_path, buffer); | |
1889 | cgroup_unlock(); | |
1890 | return 0; | |
1891 | } | |
1892 | ||
1893 | static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft, | |
1894 | struct seq_file *seq) | |
1895 | { | |
1896 | if (!cgroup_lock_live_group(cgrp)) | |
1897 | return -ENODEV; | |
1898 | seq_puts(seq, cgrp->root->release_agent_path); | |
1899 | seq_putc(seq, '\n'); | |
1900 | cgroup_unlock(); | |
1901 | return 0; | |
1902 | } | |
1903 | ||
1904 | /* A buffer size big enough for numbers or short strings */ | |
1905 | #define CGROUP_LOCAL_BUFFER_SIZE 64 | |
1906 | ||
1907 | static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft, | |
1908 | struct file *file, | |
1909 | const char __user *userbuf, | |
1910 | size_t nbytes, loff_t *unused_ppos) | |
1911 | { | |
1912 | char buffer[CGROUP_LOCAL_BUFFER_SIZE]; | |
1913 | int retval = 0; | |
1914 | char *end; | |
1915 | ||
1916 | if (!nbytes) | |
1917 | return -EINVAL; | |
1918 | if (nbytes >= sizeof(buffer)) | |
1919 | return -E2BIG; | |
1920 | if (copy_from_user(buffer, userbuf, nbytes)) | |
1921 | return -EFAULT; | |
1922 | ||
1923 | buffer[nbytes] = 0; /* nul-terminate */ | |
1924 | if (cft->write_u64) { | |
1925 | u64 val = simple_strtoull(strstrip(buffer), &end, 0); | |
1926 | if (*end) | |
1927 | return -EINVAL; | |
1928 | retval = cft->write_u64(cgrp, cft, val); | |
1929 | } else { | |
1930 | s64 val = simple_strtoll(strstrip(buffer), &end, 0); | |
1931 | if (*end) | |
1932 | return -EINVAL; | |
1933 | retval = cft->write_s64(cgrp, cft, val); | |
1934 | } | |
1935 | if (!retval) | |
1936 | retval = nbytes; | |
1937 | return retval; | |
1938 | } | |
1939 | ||
1940 | static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft, | |
1941 | struct file *file, | |
1942 | const char __user *userbuf, | |
1943 | size_t nbytes, loff_t *unused_ppos) | |
1944 | { | |
1945 | char local_buffer[CGROUP_LOCAL_BUFFER_SIZE]; | |
1946 | int retval = 0; | |
1947 | size_t max_bytes = cft->max_write_len; | |
1948 | char *buffer = local_buffer; | |
1949 | ||
1950 | if (!max_bytes) | |
1951 | max_bytes = sizeof(local_buffer) - 1; | |
1952 | if (nbytes >= max_bytes) | |
1953 | return -E2BIG; | |
1954 | /* Allocate a dynamic buffer if we need one */ | |
1955 | if (nbytes >= sizeof(local_buffer)) { | |
1956 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); | |
1957 | if (buffer == NULL) | |
1958 | return -ENOMEM; | |
1959 | } | |
1960 | if (nbytes && copy_from_user(buffer, userbuf, nbytes)) { | |
1961 | retval = -EFAULT; | |
1962 | goto out; | |
1963 | } | |
1964 | ||
1965 | buffer[nbytes] = 0; /* nul-terminate */ | |
1966 | retval = cft->write_string(cgrp, cft, strstrip(buffer)); | |
1967 | if (!retval) | |
1968 | retval = nbytes; | |
1969 | out: | |
1970 | if (buffer != local_buffer) | |
1971 | kfree(buffer); | |
1972 | return retval; | |
1973 | } | |
1974 | ||
1975 | static ssize_t cgroup_file_write(struct file *file, const char __user *buf, | |
1976 | size_t nbytes, loff_t *ppos) | |
1977 | { | |
1978 | struct cftype *cft = __d_cft(file->f_dentry); | |
1979 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); | |
1980 | ||
1981 | if (cgroup_is_removed(cgrp)) | |
1982 | return -ENODEV; | |
1983 | if (cft->write) | |
1984 | return cft->write(cgrp, cft, file, buf, nbytes, ppos); | |
1985 | if (cft->write_u64 || cft->write_s64) | |
1986 | return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos); | |
1987 | if (cft->write_string) | |
1988 | return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos); | |
1989 | if (cft->trigger) { | |
1990 | int ret = cft->trigger(cgrp, (unsigned int)cft->private); | |
1991 | return ret ? ret : nbytes; | |
1992 | } | |
1993 | return -EINVAL; | |
1994 | } | |
1995 | ||
1996 | static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft, | |
1997 | struct file *file, | |
1998 | char __user *buf, size_t nbytes, | |
1999 | loff_t *ppos) | |
2000 | { | |
2001 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; | |
2002 | u64 val = cft->read_u64(cgrp, cft); | |
2003 | int len = sprintf(tmp, "%llu\n", (unsigned long long) val); | |
2004 | ||
2005 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
2006 | } | |
2007 | ||
2008 | static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft, | |
2009 | struct file *file, | |
2010 | char __user *buf, size_t nbytes, | |
2011 | loff_t *ppos) | |
2012 | { | |
2013 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; | |
2014 | s64 val = cft->read_s64(cgrp, cft); | |
2015 | int len = sprintf(tmp, "%lld\n", (long long) val); | |
2016 | ||
2017 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
2018 | } | |
2019 | ||
2020 | static ssize_t cgroup_file_read(struct file *file, char __user *buf, | |
2021 | size_t nbytes, loff_t *ppos) | |
2022 | { | |
2023 | struct cftype *cft = __d_cft(file->f_dentry); | |
2024 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); | |
2025 | ||
2026 | if (cgroup_is_removed(cgrp)) | |
2027 | return -ENODEV; | |
2028 | ||
2029 | if (cft->read) | |
2030 | return cft->read(cgrp, cft, file, buf, nbytes, ppos); | |
2031 | if (cft->read_u64) | |
2032 | return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos); | |
2033 | if (cft->read_s64) | |
2034 | return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos); | |
2035 | return -EINVAL; | |
2036 | } | |
2037 | ||
2038 | /* | |
2039 | * seqfile ops/methods for returning structured data. Currently just | |
2040 | * supports string->u64 maps, but can be extended in future. | |
2041 | */ | |
2042 | ||
2043 | struct cgroup_seqfile_state { | |
2044 | struct cftype *cft; | |
2045 | struct cgroup *cgroup; | |
2046 | }; | |
2047 | ||
2048 | static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value) | |
2049 | { | |
2050 | struct seq_file *sf = cb->state; | |
2051 | return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value); | |
2052 | } | |
2053 | ||
2054 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) | |
2055 | { | |
2056 | struct cgroup_seqfile_state *state = m->private; | |
2057 | struct cftype *cft = state->cft; | |
2058 | if (cft->read_map) { | |
2059 | struct cgroup_map_cb cb = { | |
2060 | .fill = cgroup_map_add, | |
2061 | .state = m, | |
2062 | }; | |
2063 | return cft->read_map(state->cgroup, cft, &cb); | |
2064 | } | |
2065 | return cft->read_seq_string(state->cgroup, cft, m); | |
2066 | } | |
2067 | ||
2068 | static int cgroup_seqfile_release(struct inode *inode, struct file *file) | |
2069 | { | |
2070 | struct seq_file *seq = file->private_data; | |
2071 | kfree(seq->private); | |
2072 | return single_release(inode, file); | |
2073 | } | |
2074 | ||
2075 | static const struct file_operations cgroup_seqfile_operations = { | |
2076 | .read = seq_read, | |
2077 | .write = cgroup_file_write, | |
2078 | .llseek = seq_lseek, | |
2079 | .release = cgroup_seqfile_release, | |
2080 | }; | |
2081 | ||
2082 | static int cgroup_file_open(struct inode *inode, struct file *file) | |
2083 | { | |
2084 | int err; | |
2085 | struct cftype *cft; | |
2086 | ||
2087 | err = generic_file_open(inode, file); | |
2088 | if (err) | |
2089 | return err; | |
2090 | cft = __d_cft(file->f_dentry); | |
2091 | ||
2092 | if (cft->read_map || cft->read_seq_string) { | |
2093 | struct cgroup_seqfile_state *state = | |
2094 | kzalloc(sizeof(*state), GFP_USER); | |
2095 | if (!state) | |
2096 | return -ENOMEM; | |
2097 | state->cft = cft; | |
2098 | state->cgroup = __d_cgrp(file->f_dentry->d_parent); | |
2099 | file->f_op = &cgroup_seqfile_operations; | |
2100 | err = single_open(file, cgroup_seqfile_show, state); | |
2101 | if (err < 0) | |
2102 | kfree(state); | |
2103 | } else if (cft->open) | |
2104 | err = cft->open(inode, file); | |
2105 | else | |
2106 | err = 0; | |
2107 | ||
2108 | return err; | |
2109 | } | |
2110 | ||
2111 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
2112 | { | |
2113 | struct cftype *cft = __d_cft(file->f_dentry); | |
2114 | if (cft->release) | |
2115 | return cft->release(inode, file); | |
2116 | return 0; | |
2117 | } | |
2118 | ||
2119 | /* | |
2120 | * cgroup_rename - Only allow simple rename of directories in place. | |
2121 | */ | |
2122 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
2123 | struct inode *new_dir, struct dentry *new_dentry) | |
2124 | { | |
2125 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) | |
2126 | return -ENOTDIR; | |
2127 | if (new_dentry->d_inode) | |
2128 | return -EEXIST; | |
2129 | if (old_dir != new_dir) | |
2130 | return -EIO; | |
2131 | return simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
2132 | } | |
2133 | ||
2134 | static const struct file_operations cgroup_file_operations = { | |
2135 | .read = cgroup_file_read, | |
2136 | .write = cgroup_file_write, | |
2137 | .llseek = generic_file_llseek, | |
2138 | .open = cgroup_file_open, | |
2139 | .release = cgroup_file_release, | |
2140 | }; | |
2141 | ||
2142 | static const struct inode_operations cgroup_dir_inode_operations = { | |
2143 | .lookup = simple_lookup, | |
2144 | .mkdir = cgroup_mkdir, | |
2145 | .rmdir = cgroup_rmdir, | |
2146 | .rename = cgroup_rename, | |
2147 | }; | |
2148 | ||
2149 | /* | |
2150 | * Check if a file is a control file | |
2151 | */ | |
2152 | static inline struct cftype *__file_cft(struct file *file) | |
2153 | { | |
2154 | if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations) | |
2155 | return ERR_PTR(-EINVAL); | |
2156 | return __d_cft(file->f_dentry); | |
2157 | } | |
2158 | ||
2159 | static int cgroup_create_file(struct dentry *dentry, mode_t mode, | |
2160 | struct super_block *sb) | |
2161 | { | |
2162 | static const struct dentry_operations cgroup_dops = { | |
2163 | .d_iput = cgroup_diput, | |
2164 | }; | |
2165 | ||
2166 | struct inode *inode; | |
2167 | ||
2168 | if (!dentry) | |
2169 | return -ENOENT; | |
2170 | if (dentry->d_inode) | |
2171 | return -EEXIST; | |
2172 | ||
2173 | inode = cgroup_new_inode(mode, sb); | |
2174 | if (!inode) | |
2175 | return -ENOMEM; | |
2176 | ||
2177 | if (S_ISDIR(mode)) { | |
2178 | inode->i_op = &cgroup_dir_inode_operations; | |
2179 | inode->i_fop = &simple_dir_operations; | |
2180 | ||
2181 | /* start off with i_nlink == 2 (for "." entry) */ | |
2182 | inc_nlink(inode); | |
2183 | ||
2184 | /* start with the directory inode held, so that we can | |
2185 | * populate it without racing with another mkdir */ | |
2186 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); | |
2187 | } else if (S_ISREG(mode)) { | |
2188 | inode->i_size = 0; | |
2189 | inode->i_fop = &cgroup_file_operations; | |
2190 | } | |
2191 | dentry->d_op = &cgroup_dops; | |
2192 | d_instantiate(dentry, inode); | |
2193 | dget(dentry); /* Extra count - pin the dentry in core */ | |
2194 | return 0; | |
2195 | } | |
2196 | ||
2197 | /* | |
2198 | * cgroup_create_dir - create a directory for an object. | |
2199 | * @cgrp: the cgroup we create the directory for. It must have a valid | |
2200 | * ->parent field. And we are going to fill its ->dentry field. | |
2201 | * @dentry: dentry of the new cgroup | |
2202 | * @mode: mode to set on new directory. | |
2203 | */ | |
2204 | static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry, | |
2205 | mode_t mode) | |
2206 | { | |
2207 | struct dentry *parent; | |
2208 | int error = 0; | |
2209 | ||
2210 | parent = cgrp->parent->dentry; | |
2211 | error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb); | |
2212 | if (!error) { | |
2213 | dentry->d_fsdata = cgrp; | |
2214 | inc_nlink(parent->d_inode); | |
2215 | rcu_assign_pointer(cgrp->dentry, dentry); | |
2216 | dget(dentry); | |
2217 | } | |
2218 | dput(dentry); | |
2219 | ||
2220 | return error; | |
2221 | } | |
2222 | ||
2223 | /** | |
2224 | * cgroup_file_mode - deduce file mode of a control file | |
2225 | * @cft: the control file in question | |
2226 | * | |
2227 | * returns cft->mode if ->mode is not 0 | |
2228 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
2229 | * returns S_IRUGO if it has only a read handler | |
2230 | * returns S_IWUSR if it has only a write hander | |
2231 | */ | |
2232 | static mode_t cgroup_file_mode(const struct cftype *cft) | |
2233 | { | |
2234 | mode_t mode = 0; | |
2235 | ||
2236 | if (cft->mode) | |
2237 | return cft->mode; | |
2238 | ||
2239 | if (cft->read || cft->read_u64 || cft->read_s64 || | |
2240 | cft->read_map || cft->read_seq_string) | |
2241 | mode |= S_IRUGO; | |
2242 | ||
2243 | if (cft->write || cft->write_u64 || cft->write_s64 || | |
2244 | cft->write_string || cft->trigger) | |
2245 | mode |= S_IWUSR; | |
2246 | ||
2247 | return mode; | |
2248 | } | |
2249 | ||
2250 | int cgroup_add_file(struct cgroup *cgrp, | |
2251 | struct cgroup_subsys *subsys, | |
2252 | const struct cftype *cft) | |
2253 | { | |
2254 | struct dentry *dir = cgrp->dentry; | |
2255 | struct dentry *dentry; | |
2256 | int error; | |
2257 | mode_t mode; | |
2258 | ||
2259 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; | |
2260 | if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) { | |
2261 | strcpy(name, subsys->name); | |
2262 | strcat(name, "."); | |
2263 | } | |
2264 | strcat(name, cft->name); | |
2265 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); | |
2266 | dentry = lookup_one_len(name, dir, strlen(name)); | |
2267 | if (!IS_ERR(dentry)) { | |
2268 | mode = cgroup_file_mode(cft); | |
2269 | error = cgroup_create_file(dentry, mode | S_IFREG, | |
2270 | cgrp->root->sb); | |
2271 | if (!error) | |
2272 | dentry->d_fsdata = (void *)cft; | |
2273 | dput(dentry); | |
2274 | } else | |
2275 | error = PTR_ERR(dentry); | |
2276 | return error; | |
2277 | } | |
2278 | EXPORT_SYMBOL_GPL(cgroup_add_file); | |
2279 | ||
2280 | int cgroup_add_files(struct cgroup *cgrp, | |
2281 | struct cgroup_subsys *subsys, | |
2282 | const struct cftype cft[], | |
2283 | int count) | |
2284 | { | |
2285 | int i, err; | |
2286 | for (i = 0; i < count; i++) { | |
2287 | err = cgroup_add_file(cgrp, subsys, &cft[i]); | |
2288 | if (err) | |
2289 | return err; | |
2290 | } | |
2291 | return 0; | |
2292 | } | |
2293 | EXPORT_SYMBOL_GPL(cgroup_add_files); | |
2294 | ||
2295 | /** | |
2296 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2297 | * @cgrp: the cgroup in question | |
2298 | * | |
2299 | * Return the number of tasks in the cgroup. | |
2300 | */ | |
2301 | int cgroup_task_count(const struct cgroup *cgrp) | |
2302 | { | |
2303 | int count = 0; | |
2304 | struct cg_cgroup_link *link; | |
2305 | ||
2306 | read_lock(&css_set_lock); | |
2307 | list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) { | |
2308 | count += atomic_read(&link->cg->refcount); | |
2309 | } | |
2310 | read_unlock(&css_set_lock); | |
2311 | return count; | |
2312 | } | |
2313 | ||
2314 | /* | |
2315 | * Advance a list_head iterator. The iterator should be positioned at | |
2316 | * the start of a css_set | |
2317 | */ | |
2318 | static void cgroup_advance_iter(struct cgroup *cgrp, | |
2319 | struct cgroup_iter *it) | |
2320 | { | |
2321 | struct list_head *l = it->cg_link; | |
2322 | struct cg_cgroup_link *link; | |
2323 | struct css_set *cg; | |
2324 | ||
2325 | /* Advance to the next non-empty css_set */ | |
2326 | do { | |
2327 | l = l->next; | |
2328 | if (l == &cgrp->css_sets) { | |
2329 | it->cg_link = NULL; | |
2330 | return; | |
2331 | } | |
2332 | link = list_entry(l, struct cg_cgroup_link, cgrp_link_list); | |
2333 | cg = link->cg; | |
2334 | } while (list_empty(&cg->tasks)); | |
2335 | it->cg_link = l; | |
2336 | it->task = cg->tasks.next; | |
2337 | } | |
2338 | ||
2339 | /* | |
2340 | * To reduce the fork() overhead for systems that are not actually | |
2341 | * using their cgroups capability, we don't maintain the lists running | |
2342 | * through each css_set to its tasks until we see the list actually | |
2343 | * used - in other words after the first call to cgroup_iter_start(). | |
2344 | * | |
2345 | * The tasklist_lock is not held here, as do_each_thread() and | |
2346 | * while_each_thread() are protected by RCU. | |
2347 | */ | |
2348 | static void cgroup_enable_task_cg_lists(void) | |
2349 | { | |
2350 | struct task_struct *p, *g; | |
2351 | write_lock(&css_set_lock); | |
2352 | use_task_css_set_links = 1; | |
2353 | do_each_thread(g, p) { | |
2354 | task_lock(p); | |
2355 | /* | |
2356 | * We should check if the process is exiting, otherwise | |
2357 | * it will race with cgroup_exit() in that the list | |
2358 | * entry won't be deleted though the process has exited. | |
2359 | */ | |
2360 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
2361 | list_add(&p->cg_list, &p->cgroups->tasks); | |
2362 | task_unlock(p); | |
2363 | } while_each_thread(g, p); | |
2364 | write_unlock(&css_set_lock); | |
2365 | } | |
2366 | ||
2367 | void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it) | |
2368 | { | |
2369 | /* | |
2370 | * The first time anyone tries to iterate across a cgroup, | |
2371 | * we need to enable the list linking each css_set to its | |
2372 | * tasks, and fix up all existing tasks. | |
2373 | */ | |
2374 | if (!use_task_css_set_links) | |
2375 | cgroup_enable_task_cg_lists(); | |
2376 | ||
2377 | read_lock(&css_set_lock); | |
2378 | it->cg_link = &cgrp->css_sets; | |
2379 | cgroup_advance_iter(cgrp, it); | |
2380 | } | |
2381 | ||
2382 | struct task_struct *cgroup_iter_next(struct cgroup *cgrp, | |
2383 | struct cgroup_iter *it) | |
2384 | { | |
2385 | struct task_struct *res; | |
2386 | struct list_head *l = it->task; | |
2387 | struct cg_cgroup_link *link; | |
2388 | ||
2389 | /* If the iterator cg is NULL, we have no tasks */ | |
2390 | if (!it->cg_link) | |
2391 | return NULL; | |
2392 | res = list_entry(l, struct task_struct, cg_list); | |
2393 | /* Advance iterator to find next entry */ | |
2394 | l = l->next; | |
2395 | link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list); | |
2396 | if (l == &link->cg->tasks) { | |
2397 | /* We reached the end of this task list - move on to | |
2398 | * the next cg_cgroup_link */ | |
2399 | cgroup_advance_iter(cgrp, it); | |
2400 | } else { | |
2401 | it->task = l; | |
2402 | } | |
2403 | return res; | |
2404 | } | |
2405 | ||
2406 | void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it) | |
2407 | { | |
2408 | read_unlock(&css_set_lock); | |
2409 | } | |
2410 | ||
2411 | static inline int started_after_time(struct task_struct *t1, | |
2412 | struct timespec *time, | |
2413 | struct task_struct *t2) | |
2414 | { | |
2415 | int start_diff = timespec_compare(&t1->start_time, time); | |
2416 | if (start_diff > 0) { | |
2417 | return 1; | |
2418 | } else if (start_diff < 0) { | |
2419 | return 0; | |
2420 | } else { | |
2421 | /* | |
2422 | * Arbitrarily, if two processes started at the same | |
2423 | * time, we'll say that the lower pointer value | |
2424 | * started first. Note that t2 may have exited by now | |
2425 | * so this may not be a valid pointer any longer, but | |
2426 | * that's fine - it still serves to distinguish | |
2427 | * between two tasks started (effectively) simultaneously. | |
2428 | */ | |
2429 | return t1 > t2; | |
2430 | } | |
2431 | } | |
2432 | ||
2433 | /* | |
2434 | * This function is a callback from heap_insert() and is used to order | |
2435 | * the heap. | |
2436 | * In this case we order the heap in descending task start time. | |
2437 | */ | |
2438 | static inline int started_after(void *p1, void *p2) | |
2439 | { | |
2440 | struct task_struct *t1 = p1; | |
2441 | struct task_struct *t2 = p2; | |
2442 | return started_after_time(t1, &t2->start_time, t2); | |
2443 | } | |
2444 | ||
2445 | /** | |
2446 | * cgroup_scan_tasks - iterate though all the tasks in a cgroup | |
2447 | * @scan: struct cgroup_scanner containing arguments for the scan | |
2448 | * | |
2449 | * Arguments include pointers to callback functions test_task() and | |
2450 | * process_task(). | |
2451 | * Iterate through all the tasks in a cgroup, calling test_task() for each, | |
2452 | * and if it returns true, call process_task() for it also. | |
2453 | * The test_task pointer may be NULL, meaning always true (select all tasks). | |
2454 | * Effectively duplicates cgroup_iter_{start,next,end}() | |
2455 | * but does not lock css_set_lock for the call to process_task(). | |
2456 | * The struct cgroup_scanner may be embedded in any structure of the caller's | |
2457 | * creation. | |
2458 | * It is guaranteed that process_task() will act on every task that | |
2459 | * is a member of the cgroup for the duration of this call. This | |
2460 | * function may or may not call process_task() for tasks that exit | |
2461 | * or move to a different cgroup during the call, or are forked or | |
2462 | * move into the cgroup during the call. | |
2463 | * | |
2464 | * Note that test_task() may be called with locks held, and may in some | |
2465 | * situations be called multiple times for the same task, so it should | |
2466 | * be cheap. | |
2467 | * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been | |
2468 | * pre-allocated and will be used for heap operations (and its "gt" member will | |
2469 | * be overwritten), else a temporary heap will be used (allocation of which | |
2470 | * may cause this function to fail). | |
2471 | */ | |
2472 | int cgroup_scan_tasks(struct cgroup_scanner *scan) | |
2473 | { | |
2474 | int retval, i; | |
2475 | struct cgroup_iter it; | |
2476 | struct task_struct *p, *dropped; | |
2477 | /* Never dereference latest_task, since it's not refcounted */ | |
2478 | struct task_struct *latest_task = NULL; | |
2479 | struct ptr_heap tmp_heap; | |
2480 | struct ptr_heap *heap; | |
2481 | struct timespec latest_time = { 0, 0 }; | |
2482 | ||
2483 | if (scan->heap) { | |
2484 | /* The caller supplied our heap and pre-allocated its memory */ | |
2485 | heap = scan->heap; | |
2486 | heap->gt = &started_after; | |
2487 | } else { | |
2488 | /* We need to allocate our own heap memory */ | |
2489 | heap = &tmp_heap; | |
2490 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
2491 | if (retval) | |
2492 | /* cannot allocate the heap */ | |
2493 | return retval; | |
2494 | } | |
2495 | ||
2496 | again: | |
2497 | /* | |
2498 | * Scan tasks in the cgroup, using the scanner's "test_task" callback | |
2499 | * to determine which are of interest, and using the scanner's | |
2500 | * "process_task" callback to process any of them that need an update. | |
2501 | * Since we don't want to hold any locks during the task updates, | |
2502 | * gather tasks to be processed in a heap structure. | |
2503 | * The heap is sorted by descending task start time. | |
2504 | * If the statically-sized heap fills up, we overflow tasks that | |
2505 | * started later, and in future iterations only consider tasks that | |
2506 | * started after the latest task in the previous pass. This | |
2507 | * guarantees forward progress and that we don't miss any tasks. | |
2508 | */ | |
2509 | heap->size = 0; | |
2510 | cgroup_iter_start(scan->cg, &it); | |
2511 | while ((p = cgroup_iter_next(scan->cg, &it))) { | |
2512 | /* | |
2513 | * Only affect tasks that qualify per the caller's callback, | |
2514 | * if he provided one | |
2515 | */ | |
2516 | if (scan->test_task && !scan->test_task(p, scan)) | |
2517 | continue; | |
2518 | /* | |
2519 | * Only process tasks that started after the last task | |
2520 | * we processed | |
2521 | */ | |
2522 | if (!started_after_time(p, &latest_time, latest_task)) | |
2523 | continue; | |
2524 | dropped = heap_insert(heap, p); | |
2525 | if (dropped == NULL) { | |
2526 | /* | |
2527 | * The new task was inserted; the heap wasn't | |
2528 | * previously full | |
2529 | */ | |
2530 | get_task_struct(p); | |
2531 | } else if (dropped != p) { | |
2532 | /* | |
2533 | * The new task was inserted, and pushed out a | |
2534 | * different task | |
2535 | */ | |
2536 | get_task_struct(p); | |
2537 | put_task_struct(dropped); | |
2538 | } | |
2539 | /* | |
2540 | * Else the new task was newer than anything already in | |
2541 | * the heap and wasn't inserted | |
2542 | */ | |
2543 | } | |
2544 | cgroup_iter_end(scan->cg, &it); | |
2545 | ||
2546 | if (heap->size) { | |
2547 | for (i = 0; i < heap->size; i++) { | |
2548 | struct task_struct *q = heap->ptrs[i]; | |
2549 | if (i == 0) { | |
2550 | latest_time = q->start_time; | |
2551 | latest_task = q; | |
2552 | } | |
2553 | /* Process the task per the caller's callback */ | |
2554 | scan->process_task(q, scan); | |
2555 | put_task_struct(q); | |
2556 | } | |
2557 | /* | |
2558 | * If we had to process any tasks at all, scan again | |
2559 | * in case some of them were in the middle of forking | |
2560 | * children that didn't get processed. | |
2561 | * Not the most efficient way to do it, but it avoids | |
2562 | * having to take callback_mutex in the fork path | |
2563 | */ | |
2564 | goto again; | |
2565 | } | |
2566 | if (heap == &tmp_heap) | |
2567 | heap_free(&tmp_heap); | |
2568 | return 0; | |
2569 | } | |
2570 | ||
2571 | /* | |
2572 | * Stuff for reading the 'tasks'/'procs' files. | |
2573 | * | |
2574 | * Reading this file can return large amounts of data if a cgroup has | |
2575 | * *lots* of attached tasks. So it may need several calls to read(), | |
2576 | * but we cannot guarantee that the information we produce is correct | |
2577 | * unless we produce it entirely atomically. | |
2578 | * | |
2579 | */ | |
2580 | ||
2581 | /* | |
2582 | * The following two functions "fix" the issue where there are more pids | |
2583 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
2584 | * TODO: replace with a kernel-wide solution to this problem | |
2585 | */ | |
2586 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
2587 | static void *pidlist_allocate(int count) | |
2588 | { | |
2589 | if (PIDLIST_TOO_LARGE(count)) | |
2590 | return vmalloc(count * sizeof(pid_t)); | |
2591 | else | |
2592 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
2593 | } | |
2594 | static void pidlist_free(void *p) | |
2595 | { | |
2596 | if (is_vmalloc_addr(p)) | |
2597 | vfree(p); | |
2598 | else | |
2599 | kfree(p); | |
2600 | } | |
2601 | static void *pidlist_resize(void *p, int newcount) | |
2602 | { | |
2603 | void *newlist; | |
2604 | /* note: if new alloc fails, old p will still be valid either way */ | |
2605 | if (is_vmalloc_addr(p)) { | |
2606 | newlist = vmalloc(newcount * sizeof(pid_t)); | |
2607 | if (!newlist) | |
2608 | return NULL; | |
2609 | memcpy(newlist, p, newcount * sizeof(pid_t)); | |
2610 | vfree(p); | |
2611 | } else { | |
2612 | newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL); | |
2613 | } | |
2614 | return newlist; | |
2615 | } | |
2616 | ||
2617 | /* | |
2618 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries | |
2619 | * If the new stripped list is sufficiently smaller and there's enough memory | |
2620 | * to allocate a new buffer, will let go of the unneeded memory. Returns the | |
2621 | * number of unique elements. | |
2622 | */ | |
2623 | /* is the size difference enough that we should re-allocate the array? */ | |
2624 | #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new)) | |
2625 | static int pidlist_uniq(pid_t **p, int length) | |
2626 | { | |
2627 | int src, dest = 1; | |
2628 | pid_t *list = *p; | |
2629 | pid_t *newlist; | |
2630 | ||
2631 | /* | |
2632 | * we presume the 0th element is unique, so i starts at 1. trivial | |
2633 | * edge cases first; no work needs to be done for either | |
2634 | */ | |
2635 | if (length == 0 || length == 1) | |
2636 | return length; | |
2637 | /* src and dest walk down the list; dest counts unique elements */ | |
2638 | for (src = 1; src < length; src++) { | |
2639 | /* find next unique element */ | |
2640 | while (list[src] == list[src-1]) { | |
2641 | src++; | |
2642 | if (src == length) | |
2643 | goto after; | |
2644 | } | |
2645 | /* dest always points to where the next unique element goes */ | |
2646 | list[dest] = list[src]; | |
2647 | dest++; | |
2648 | } | |
2649 | after: | |
2650 | /* | |
2651 | * if the length difference is large enough, we want to allocate a | |
2652 | * smaller buffer to save memory. if this fails due to out of memory, | |
2653 | * we'll just stay with what we've got. | |
2654 | */ | |
2655 | if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) { | |
2656 | newlist = pidlist_resize(list, dest); | |
2657 | if (newlist) | |
2658 | *p = newlist; | |
2659 | } | |
2660 | return dest; | |
2661 | } | |
2662 | ||
2663 | static int cmppid(const void *a, const void *b) | |
2664 | { | |
2665 | return *(pid_t *)a - *(pid_t *)b; | |
2666 | } | |
2667 | ||
2668 | /* | |
2669 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
2670 | * returns with the lock on that pidlist already held, and takes care | |
2671 | * of the use count, or returns NULL with no locks held if we're out of | |
2672 | * memory. | |
2673 | */ | |
2674 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
2675 | enum cgroup_filetype type) | |
2676 | { | |
2677 | struct cgroup_pidlist *l; | |
2678 | /* don't need task_nsproxy() if we're looking at ourself */ | |
2679 | struct pid_namespace *ns = current->nsproxy->pid_ns; | |
2680 | ||
2681 | /* | |
2682 | * We can't drop the pidlist_mutex before taking the l->mutex in case | |
2683 | * the last ref-holder is trying to remove l from the list at the same | |
2684 | * time. Holding the pidlist_mutex precludes somebody taking whichever | |
2685 | * list we find out from under us - compare release_pid_array(). | |
2686 | */ | |
2687 | mutex_lock(&cgrp->pidlist_mutex); | |
2688 | list_for_each_entry(l, &cgrp->pidlists, links) { | |
2689 | if (l->key.type == type && l->key.ns == ns) { | |
2690 | /* make sure l doesn't vanish out from under us */ | |
2691 | down_write(&l->mutex); | |
2692 | mutex_unlock(&cgrp->pidlist_mutex); | |
2693 | return l; | |
2694 | } | |
2695 | } | |
2696 | /* entry not found; create a new one */ | |
2697 | l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); | |
2698 | if (!l) { | |
2699 | mutex_unlock(&cgrp->pidlist_mutex); | |
2700 | return l; | |
2701 | } | |
2702 | init_rwsem(&l->mutex); | |
2703 | down_write(&l->mutex); | |
2704 | l->key.type = type; | |
2705 | l->key.ns = get_pid_ns(ns); | |
2706 | l->use_count = 0; /* don't increment here */ | |
2707 | l->list = NULL; | |
2708 | l->owner = cgrp; | |
2709 | list_add(&l->links, &cgrp->pidlists); | |
2710 | mutex_unlock(&cgrp->pidlist_mutex); | |
2711 | return l; | |
2712 | } | |
2713 | ||
2714 | /* | |
2715 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
2716 | */ | |
2717 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, | |
2718 | struct cgroup_pidlist **lp) | |
2719 | { | |
2720 | pid_t *array; | |
2721 | int length; | |
2722 | int pid, n = 0; /* used for populating the array */ | |
2723 | struct cgroup_iter it; | |
2724 | struct task_struct *tsk; | |
2725 | struct cgroup_pidlist *l; | |
2726 | ||
2727 | /* | |
2728 | * If cgroup gets more users after we read count, we won't have | |
2729 | * enough space - tough. This race is indistinguishable to the | |
2730 | * caller from the case that the additional cgroup users didn't | |
2731 | * show up until sometime later on. | |
2732 | */ | |
2733 | length = cgroup_task_count(cgrp); | |
2734 | array = pidlist_allocate(length); | |
2735 | if (!array) | |
2736 | return -ENOMEM; | |
2737 | /* now, populate the array */ | |
2738 | cgroup_iter_start(cgrp, &it); | |
2739 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
2740 | if (unlikely(n == length)) | |
2741 | break; | |
2742 | /* get tgid or pid for procs or tasks file respectively */ | |
2743 | if (type == CGROUP_FILE_PROCS) | |
2744 | pid = task_tgid_vnr(tsk); | |
2745 | else | |
2746 | pid = task_pid_vnr(tsk); | |
2747 | if (pid > 0) /* make sure to only use valid results */ | |
2748 | array[n++] = pid; | |
2749 | } | |
2750 | cgroup_iter_end(cgrp, &it); | |
2751 | length = n; | |
2752 | /* now sort & (if procs) strip out duplicates */ | |
2753 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
2754 | if (type == CGROUP_FILE_PROCS) | |
2755 | length = pidlist_uniq(&array, length); | |
2756 | l = cgroup_pidlist_find(cgrp, type); | |
2757 | if (!l) { | |
2758 | pidlist_free(array); | |
2759 | return -ENOMEM; | |
2760 | } | |
2761 | /* store array, freeing old if necessary - lock already held */ | |
2762 | pidlist_free(l->list); | |
2763 | l->list = array; | |
2764 | l->length = length; | |
2765 | l->use_count++; | |
2766 | up_write(&l->mutex); | |
2767 | *lp = l; | |
2768 | return 0; | |
2769 | } | |
2770 | ||
2771 | /** | |
2772 | * cgroupstats_build - build and fill cgroupstats | |
2773 | * @stats: cgroupstats to fill information into | |
2774 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
2775 | * been requested. | |
2776 | * | |
2777 | * Build and fill cgroupstats so that taskstats can export it to user | |
2778 | * space. | |
2779 | */ | |
2780 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
2781 | { | |
2782 | int ret = -EINVAL; | |
2783 | struct cgroup *cgrp; | |
2784 | struct cgroup_iter it; | |
2785 | struct task_struct *tsk; | |
2786 | ||
2787 | /* | |
2788 | * Validate dentry by checking the superblock operations, | |
2789 | * and make sure it's a directory. | |
2790 | */ | |
2791 | if (dentry->d_sb->s_op != &cgroup_ops || | |
2792 | !S_ISDIR(dentry->d_inode->i_mode)) | |
2793 | goto err; | |
2794 | ||
2795 | ret = 0; | |
2796 | cgrp = dentry->d_fsdata; | |
2797 | ||
2798 | cgroup_iter_start(cgrp, &it); | |
2799 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
2800 | switch (tsk->state) { | |
2801 | case TASK_RUNNING: | |
2802 | stats->nr_running++; | |
2803 | break; | |
2804 | case TASK_INTERRUPTIBLE: | |
2805 | stats->nr_sleeping++; | |
2806 | break; | |
2807 | case TASK_UNINTERRUPTIBLE: | |
2808 | stats->nr_uninterruptible++; | |
2809 | break; | |
2810 | case TASK_STOPPED: | |
2811 | stats->nr_stopped++; | |
2812 | break; | |
2813 | default: | |
2814 | if (delayacct_is_task_waiting_on_io(tsk)) | |
2815 | stats->nr_io_wait++; | |
2816 | break; | |
2817 | } | |
2818 | } | |
2819 | cgroup_iter_end(cgrp, &it); | |
2820 | ||
2821 | err: | |
2822 | return ret; | |
2823 | } | |
2824 | ||
2825 | ||
2826 | /* | |
2827 | * seq_file methods for the tasks/procs files. The seq_file position is the | |
2828 | * next pid to display; the seq_file iterator is a pointer to the pid | |
2829 | * in the cgroup->l->list array. | |
2830 | */ | |
2831 | ||
2832 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) | |
2833 | { | |
2834 | /* | |
2835 | * Initially we receive a position value that corresponds to | |
2836 | * one more than the last pid shown (or 0 on the first call or | |
2837 | * after a seek to the start). Use a binary-search to find the | |
2838 | * next pid to display, if any | |
2839 | */ | |
2840 | struct cgroup_pidlist *l = s->private; | |
2841 | int index = 0, pid = *pos; | |
2842 | int *iter; | |
2843 | ||
2844 | down_read(&l->mutex); | |
2845 | if (pid) { | |
2846 | int end = l->length; | |
2847 | ||
2848 | while (index < end) { | |
2849 | int mid = (index + end) / 2; | |
2850 | if (l->list[mid] == pid) { | |
2851 | index = mid; | |
2852 | break; | |
2853 | } else if (l->list[mid] <= pid) | |
2854 | index = mid + 1; | |
2855 | else | |
2856 | end = mid; | |
2857 | } | |
2858 | } | |
2859 | /* If we're off the end of the array, we're done */ | |
2860 | if (index >= l->length) | |
2861 | return NULL; | |
2862 | /* Update the abstract position to be the actual pid that we found */ | |
2863 | iter = l->list + index; | |
2864 | *pos = *iter; | |
2865 | return iter; | |
2866 | } | |
2867 | ||
2868 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) | |
2869 | { | |
2870 | struct cgroup_pidlist *l = s->private; | |
2871 | up_read(&l->mutex); | |
2872 | } | |
2873 | ||
2874 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) | |
2875 | { | |
2876 | struct cgroup_pidlist *l = s->private; | |
2877 | pid_t *p = v; | |
2878 | pid_t *end = l->list + l->length; | |
2879 | /* | |
2880 | * Advance to the next pid in the array. If this goes off the | |
2881 | * end, we're done | |
2882 | */ | |
2883 | p++; | |
2884 | if (p >= end) { | |
2885 | return NULL; | |
2886 | } else { | |
2887 | *pos = *p; | |
2888 | return p; | |
2889 | } | |
2890 | } | |
2891 | ||
2892 | static int cgroup_pidlist_show(struct seq_file *s, void *v) | |
2893 | { | |
2894 | return seq_printf(s, "%d\n", *(int *)v); | |
2895 | } | |
2896 | ||
2897 | /* | |
2898 | * seq_operations functions for iterating on pidlists through seq_file - | |
2899 | * independent of whether it's tasks or procs | |
2900 | */ | |
2901 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
2902 | .start = cgroup_pidlist_start, | |
2903 | .stop = cgroup_pidlist_stop, | |
2904 | .next = cgroup_pidlist_next, | |
2905 | .show = cgroup_pidlist_show, | |
2906 | }; | |
2907 | ||
2908 | static void cgroup_release_pid_array(struct cgroup_pidlist *l) | |
2909 | { | |
2910 | /* | |
2911 | * the case where we're the last user of this particular pidlist will | |
2912 | * have us remove it from the cgroup's list, which entails taking the | |
2913 | * mutex. since in pidlist_find the pidlist->lock depends on cgroup-> | |
2914 | * pidlist_mutex, we have to take pidlist_mutex first. | |
2915 | */ | |
2916 | mutex_lock(&l->owner->pidlist_mutex); | |
2917 | down_write(&l->mutex); | |
2918 | BUG_ON(!l->use_count); | |
2919 | if (!--l->use_count) { | |
2920 | /* we're the last user if refcount is 0; remove and free */ | |
2921 | list_del(&l->links); | |
2922 | mutex_unlock(&l->owner->pidlist_mutex); | |
2923 | pidlist_free(l->list); | |
2924 | put_pid_ns(l->key.ns); | |
2925 | up_write(&l->mutex); | |
2926 | kfree(l); | |
2927 | return; | |
2928 | } | |
2929 | mutex_unlock(&l->owner->pidlist_mutex); | |
2930 | up_write(&l->mutex); | |
2931 | } | |
2932 | ||
2933 | static int cgroup_pidlist_release(struct inode *inode, struct file *file) | |
2934 | { | |
2935 | struct cgroup_pidlist *l; | |
2936 | if (!(file->f_mode & FMODE_READ)) | |
2937 | return 0; | |
2938 | /* | |
2939 | * the seq_file will only be initialized if the file was opened for | |
2940 | * reading; hence we check if it's not null only in that case. | |
2941 | */ | |
2942 | l = ((struct seq_file *)file->private_data)->private; | |
2943 | cgroup_release_pid_array(l); | |
2944 | return seq_release(inode, file); | |
2945 | } | |
2946 | ||
2947 | static const struct file_operations cgroup_pidlist_operations = { | |
2948 | .read = seq_read, | |
2949 | .llseek = seq_lseek, | |
2950 | .write = cgroup_file_write, | |
2951 | .release = cgroup_pidlist_release, | |
2952 | }; | |
2953 | ||
2954 | /* | |
2955 | * The following functions handle opens on a file that displays a pidlist | |
2956 | * (tasks or procs). Prepare an array of the process/thread IDs of whoever's | |
2957 | * in the cgroup. | |
2958 | */ | |
2959 | /* helper function for the two below it */ | |
2960 | static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type) | |
2961 | { | |
2962 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); | |
2963 | struct cgroup_pidlist *l; | |
2964 | int retval; | |
2965 | ||
2966 | /* Nothing to do for write-only files */ | |
2967 | if (!(file->f_mode & FMODE_READ)) | |
2968 | return 0; | |
2969 | ||
2970 | /* have the array populated */ | |
2971 | retval = pidlist_array_load(cgrp, type, &l); | |
2972 | if (retval) | |
2973 | return retval; | |
2974 | /* configure file information */ | |
2975 | file->f_op = &cgroup_pidlist_operations; | |
2976 | ||
2977 | retval = seq_open(file, &cgroup_pidlist_seq_operations); | |
2978 | if (retval) { | |
2979 | cgroup_release_pid_array(l); | |
2980 | return retval; | |
2981 | } | |
2982 | ((struct seq_file *)file->private_data)->private = l; | |
2983 | return 0; | |
2984 | } | |
2985 | static int cgroup_tasks_open(struct inode *unused, struct file *file) | |
2986 | { | |
2987 | return cgroup_pidlist_open(file, CGROUP_FILE_TASKS); | |
2988 | } | |
2989 | static int cgroup_procs_open(struct inode *unused, struct file *file) | |
2990 | { | |
2991 | return cgroup_pidlist_open(file, CGROUP_FILE_PROCS); | |
2992 | } | |
2993 | ||
2994 | static u64 cgroup_read_notify_on_release(struct cgroup *cgrp, | |
2995 | struct cftype *cft) | |
2996 | { | |
2997 | return notify_on_release(cgrp); | |
2998 | } | |
2999 | ||
3000 | static int cgroup_write_notify_on_release(struct cgroup *cgrp, | |
3001 | struct cftype *cft, | |
3002 | u64 val) | |
3003 | { | |
3004 | clear_bit(CGRP_RELEASABLE, &cgrp->flags); | |
3005 | if (val) | |
3006 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3007 | else | |
3008 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3009 | return 0; | |
3010 | } | |
3011 | ||
3012 | /* | |
3013 | * Unregister event and free resources. | |
3014 | * | |
3015 | * Gets called from workqueue. | |
3016 | */ | |
3017 | static void cgroup_event_remove(struct work_struct *work) | |
3018 | { | |
3019 | struct cgroup_event *event = container_of(work, struct cgroup_event, | |
3020 | remove); | |
3021 | struct cgroup *cgrp = event->cgrp; | |
3022 | ||
3023 | event->cft->unregister_event(cgrp, event->cft, event->eventfd); | |
3024 | ||
3025 | eventfd_ctx_put(event->eventfd); | |
3026 | kfree(event); | |
3027 | dput(cgrp->dentry); | |
3028 | } | |
3029 | ||
3030 | /* | |
3031 | * Gets called on POLLHUP on eventfd when user closes it. | |
3032 | * | |
3033 | * Called with wqh->lock held and interrupts disabled. | |
3034 | */ | |
3035 | static int cgroup_event_wake(wait_queue_t *wait, unsigned mode, | |
3036 | int sync, void *key) | |
3037 | { | |
3038 | struct cgroup_event *event = container_of(wait, | |
3039 | struct cgroup_event, wait); | |
3040 | struct cgroup *cgrp = event->cgrp; | |
3041 | unsigned long flags = (unsigned long)key; | |
3042 | ||
3043 | if (flags & POLLHUP) { | |
3044 | __remove_wait_queue(event->wqh, &event->wait); | |
3045 | spin_lock(&cgrp->event_list_lock); | |
3046 | list_del(&event->list); | |
3047 | spin_unlock(&cgrp->event_list_lock); | |
3048 | /* | |
3049 | * We are in atomic context, but cgroup_event_remove() may | |
3050 | * sleep, so we have to call it in workqueue. | |
3051 | */ | |
3052 | schedule_work(&event->remove); | |
3053 | } | |
3054 | ||
3055 | return 0; | |
3056 | } | |
3057 | ||
3058 | static void cgroup_event_ptable_queue_proc(struct file *file, | |
3059 | wait_queue_head_t *wqh, poll_table *pt) | |
3060 | { | |
3061 | struct cgroup_event *event = container_of(pt, | |
3062 | struct cgroup_event, pt); | |
3063 | ||
3064 | event->wqh = wqh; | |
3065 | add_wait_queue(wqh, &event->wait); | |
3066 | } | |
3067 | ||
3068 | /* | |
3069 | * Parse input and register new cgroup event handler. | |
3070 | * | |
3071 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3072 | * Interpretation of args is defined by control file implementation. | |
3073 | */ | |
3074 | static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft, | |
3075 | const char *buffer) | |
3076 | { | |
3077 | struct cgroup_event *event = NULL; | |
3078 | unsigned int efd, cfd; | |
3079 | struct file *efile = NULL; | |
3080 | struct file *cfile = NULL; | |
3081 | char *endp; | |
3082 | int ret; | |
3083 | ||
3084 | efd = simple_strtoul(buffer, &endp, 10); | |
3085 | if (*endp != ' ') | |
3086 | return -EINVAL; | |
3087 | buffer = endp + 1; | |
3088 | ||
3089 | cfd = simple_strtoul(buffer, &endp, 10); | |
3090 | if ((*endp != ' ') && (*endp != '\0')) | |
3091 | return -EINVAL; | |
3092 | buffer = endp + 1; | |
3093 | ||
3094 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3095 | if (!event) | |
3096 | return -ENOMEM; | |
3097 | event->cgrp = cgrp; | |
3098 | INIT_LIST_HEAD(&event->list); | |
3099 | init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc); | |
3100 | init_waitqueue_func_entry(&event->wait, cgroup_event_wake); | |
3101 | INIT_WORK(&event->remove, cgroup_event_remove); | |
3102 | ||
3103 | efile = eventfd_fget(efd); | |
3104 | if (IS_ERR(efile)) { | |
3105 | ret = PTR_ERR(efile); | |
3106 | goto fail; | |
3107 | } | |
3108 | ||
3109 | event->eventfd = eventfd_ctx_fileget(efile); | |
3110 | if (IS_ERR(event->eventfd)) { | |
3111 | ret = PTR_ERR(event->eventfd); | |
3112 | goto fail; | |
3113 | } | |
3114 | ||
3115 | cfile = fget(cfd); | |
3116 | if (!cfile) { | |
3117 | ret = -EBADF; | |
3118 | goto fail; | |
3119 | } | |
3120 | ||
3121 | /* the process need read permission on control file */ | |
3122 | ret = file_permission(cfile, MAY_READ); | |
3123 | if (ret < 0) | |
3124 | goto fail; | |
3125 | ||
3126 | event->cft = __file_cft(cfile); | |
3127 | if (IS_ERR(event->cft)) { | |
3128 | ret = PTR_ERR(event->cft); | |
3129 | goto fail; | |
3130 | } | |
3131 | ||
3132 | if (!event->cft->register_event || !event->cft->unregister_event) { | |
3133 | ret = -EINVAL; | |
3134 | goto fail; | |
3135 | } | |
3136 | ||
3137 | ret = event->cft->register_event(cgrp, event->cft, | |
3138 | event->eventfd, buffer); | |
3139 | if (ret) | |
3140 | goto fail; | |
3141 | ||
3142 | if (efile->f_op->poll(efile, &event->pt) & POLLHUP) { | |
3143 | event->cft->unregister_event(cgrp, event->cft, event->eventfd); | |
3144 | ret = 0; | |
3145 | goto fail; | |
3146 | } | |
3147 | ||
3148 | /* | |
3149 | * Events should be removed after rmdir of cgroup directory, but before | |
3150 | * destroying subsystem state objects. Let's take reference to cgroup | |
3151 | * directory dentry to do that. | |
3152 | */ | |
3153 | dget(cgrp->dentry); | |
3154 | ||
3155 | spin_lock(&cgrp->event_list_lock); | |
3156 | list_add(&event->list, &cgrp->event_list); | |
3157 | spin_unlock(&cgrp->event_list_lock); | |
3158 | ||
3159 | fput(cfile); | |
3160 | fput(efile); | |
3161 | ||
3162 | return 0; | |
3163 | ||
3164 | fail: | |
3165 | if (cfile) | |
3166 | fput(cfile); | |
3167 | ||
3168 | if (event && event->eventfd && !IS_ERR(event->eventfd)) | |
3169 | eventfd_ctx_put(event->eventfd); | |
3170 | ||
3171 | if (!IS_ERR_OR_NULL(efile)) | |
3172 | fput(efile); | |
3173 | ||
3174 | kfree(event); | |
3175 | ||
3176 | return ret; | |
3177 | } | |
3178 | ||
3179 | /* | |
3180 | * for the common functions, 'private' gives the type of file | |
3181 | */ | |
3182 | /* for hysterical raisins, we can't put this on the older files */ | |
3183 | #define CGROUP_FILE_GENERIC_PREFIX "cgroup." | |
3184 | static struct cftype files[] = { | |
3185 | { | |
3186 | .name = "tasks", | |
3187 | .open = cgroup_tasks_open, | |
3188 | .write_u64 = cgroup_tasks_write, | |
3189 | .release = cgroup_pidlist_release, | |
3190 | .mode = S_IRUGO | S_IWUSR, | |
3191 | }, | |
3192 | { | |
3193 | .name = CGROUP_FILE_GENERIC_PREFIX "procs", | |
3194 | .open = cgroup_procs_open, | |
3195 | /* .write_u64 = cgroup_procs_write, TODO */ | |
3196 | .release = cgroup_pidlist_release, | |
3197 | .mode = S_IRUGO, | |
3198 | }, | |
3199 | { | |
3200 | .name = "notify_on_release", | |
3201 | .read_u64 = cgroup_read_notify_on_release, | |
3202 | .write_u64 = cgroup_write_notify_on_release, | |
3203 | }, | |
3204 | { | |
3205 | .name = CGROUP_FILE_GENERIC_PREFIX "event_control", | |
3206 | .write_string = cgroup_write_event_control, | |
3207 | .mode = S_IWUGO, | |
3208 | }, | |
3209 | }; | |
3210 | ||
3211 | static struct cftype cft_release_agent = { | |
3212 | .name = "release_agent", | |
3213 | .read_seq_string = cgroup_release_agent_show, | |
3214 | .write_string = cgroup_release_agent_write, | |
3215 | .max_write_len = PATH_MAX, | |
3216 | }; | |
3217 | ||
3218 | static int cgroup_populate_dir(struct cgroup *cgrp) | |
3219 | { | |
3220 | int err; | |
3221 | struct cgroup_subsys *ss; | |
3222 | ||
3223 | /* First clear out any existing files */ | |
3224 | cgroup_clear_directory(cgrp->dentry); | |
3225 | ||
3226 | err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files)); | |
3227 | if (err < 0) | |
3228 | return err; | |
3229 | ||
3230 | if (cgrp == cgrp->top_cgroup) { | |
3231 | if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0) | |
3232 | return err; | |
3233 | } | |
3234 | ||
3235 | for_each_subsys(cgrp->root, ss) { | |
3236 | if (ss->populate && (err = ss->populate(ss, cgrp)) < 0) | |
3237 | return err; | |
3238 | } | |
3239 | /* This cgroup is ready now */ | |
3240 | for_each_subsys(cgrp->root, ss) { | |
3241 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3242 | /* | |
3243 | * Update id->css pointer and make this css visible from | |
3244 | * CSS ID functions. This pointer will be dereferened | |
3245 | * from RCU-read-side without locks. | |
3246 | */ | |
3247 | if (css->id) | |
3248 | rcu_assign_pointer(css->id->css, css); | |
3249 | } | |
3250 | ||
3251 | return 0; | |
3252 | } | |
3253 | ||
3254 | static void init_cgroup_css(struct cgroup_subsys_state *css, | |
3255 | struct cgroup_subsys *ss, | |
3256 | struct cgroup *cgrp) | |
3257 | { | |
3258 | css->cgroup = cgrp; | |
3259 | atomic_set(&css->refcnt, 1); | |
3260 | css->flags = 0; | |
3261 | css->id = NULL; | |
3262 | if (cgrp == dummytop) | |
3263 | set_bit(CSS_ROOT, &css->flags); | |
3264 | BUG_ON(cgrp->subsys[ss->subsys_id]); | |
3265 | cgrp->subsys[ss->subsys_id] = css; | |
3266 | } | |
3267 | ||
3268 | static void cgroup_lock_hierarchy(struct cgroupfs_root *root) | |
3269 | { | |
3270 | /* We need to take each hierarchy_mutex in a consistent order */ | |
3271 | int i; | |
3272 | ||
3273 | /* | |
3274 | * No worry about a race with rebind_subsystems that might mess up the | |
3275 | * locking order, since both parties are under cgroup_mutex. | |
3276 | */ | |
3277 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
3278 | struct cgroup_subsys *ss = subsys[i]; | |
3279 | if (ss == NULL) | |
3280 | continue; | |
3281 | if (ss->root == root) | |
3282 | mutex_lock(&ss->hierarchy_mutex); | |
3283 | } | |
3284 | } | |
3285 | ||
3286 | static void cgroup_unlock_hierarchy(struct cgroupfs_root *root) | |
3287 | { | |
3288 | int i; | |
3289 | ||
3290 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
3291 | struct cgroup_subsys *ss = subsys[i]; | |
3292 | if (ss == NULL) | |
3293 | continue; | |
3294 | if (ss->root == root) | |
3295 | mutex_unlock(&ss->hierarchy_mutex); | |
3296 | } | |
3297 | } | |
3298 | ||
3299 | /* | |
3300 | * cgroup_create - create a cgroup | |
3301 | * @parent: cgroup that will be parent of the new cgroup | |
3302 | * @dentry: dentry of the new cgroup | |
3303 | * @mode: mode to set on new inode | |
3304 | * | |
3305 | * Must be called with the mutex on the parent inode held | |
3306 | */ | |
3307 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, | |
3308 | mode_t mode) | |
3309 | { | |
3310 | struct cgroup *cgrp; | |
3311 | struct cgroupfs_root *root = parent->root; | |
3312 | int err = 0; | |
3313 | struct cgroup_subsys *ss; | |
3314 | struct super_block *sb = root->sb; | |
3315 | ||
3316 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); | |
3317 | if (!cgrp) | |
3318 | return -ENOMEM; | |
3319 | ||
3320 | /* Grab a reference on the superblock so the hierarchy doesn't | |
3321 | * get deleted on unmount if there are child cgroups. This | |
3322 | * can be done outside cgroup_mutex, since the sb can't | |
3323 | * disappear while someone has an open control file on the | |
3324 | * fs */ | |
3325 | atomic_inc(&sb->s_active); | |
3326 | ||
3327 | mutex_lock(&cgroup_mutex); | |
3328 | ||
3329 | init_cgroup_housekeeping(cgrp); | |
3330 | ||
3331 | cgrp->parent = parent; | |
3332 | cgrp->root = parent->root; | |
3333 | cgrp->top_cgroup = parent->top_cgroup; | |
3334 | ||
3335 | if (notify_on_release(parent)) | |
3336 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3337 | ||
3338 | for_each_subsys(root, ss) { | |
3339 | struct cgroup_subsys_state *css = ss->create(ss, cgrp); | |
3340 | ||
3341 | if (IS_ERR(css)) { | |
3342 | err = PTR_ERR(css); | |
3343 | goto err_destroy; | |
3344 | } | |
3345 | init_cgroup_css(css, ss, cgrp); | |
3346 | if (ss->use_id) { | |
3347 | err = alloc_css_id(ss, parent, cgrp); | |
3348 | if (err) | |
3349 | goto err_destroy; | |
3350 | } | |
3351 | /* At error, ->destroy() callback has to free assigned ID. */ | |
3352 | } | |
3353 | ||
3354 | cgroup_lock_hierarchy(root); | |
3355 | list_add(&cgrp->sibling, &cgrp->parent->children); | |
3356 | cgroup_unlock_hierarchy(root); | |
3357 | root->number_of_cgroups++; | |
3358 | ||
3359 | err = cgroup_create_dir(cgrp, dentry, mode); | |
3360 | if (err < 0) | |
3361 | goto err_remove; | |
3362 | ||
3363 | /* The cgroup directory was pre-locked for us */ | |
3364 | BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex)); | |
3365 | ||
3366 | err = cgroup_populate_dir(cgrp); | |
3367 | /* If err < 0, we have a half-filled directory - oh well ;) */ | |
3368 | ||
3369 | mutex_unlock(&cgroup_mutex); | |
3370 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); | |
3371 | ||
3372 | return 0; | |
3373 | ||
3374 | err_remove: | |
3375 | ||
3376 | cgroup_lock_hierarchy(root); | |
3377 | list_del(&cgrp->sibling); | |
3378 | cgroup_unlock_hierarchy(root); | |
3379 | root->number_of_cgroups--; | |
3380 | ||
3381 | err_destroy: | |
3382 | ||
3383 | for_each_subsys(root, ss) { | |
3384 | if (cgrp->subsys[ss->subsys_id]) | |
3385 | ss->destroy(ss, cgrp); | |
3386 | } | |
3387 | ||
3388 | mutex_unlock(&cgroup_mutex); | |
3389 | ||
3390 | /* Release the reference count that we took on the superblock */ | |
3391 | deactivate_super(sb); | |
3392 | ||
3393 | kfree(cgrp); | |
3394 | return err; | |
3395 | } | |
3396 | ||
3397 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
3398 | { | |
3399 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
3400 | ||
3401 | /* the vfs holds inode->i_mutex already */ | |
3402 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
3403 | } | |
3404 | ||
3405 | static int cgroup_has_css_refs(struct cgroup *cgrp) | |
3406 | { | |
3407 | /* Check the reference count on each subsystem. Since we | |
3408 | * already established that there are no tasks in the | |
3409 | * cgroup, if the css refcount is also 1, then there should | |
3410 | * be no outstanding references, so the subsystem is safe to | |
3411 | * destroy. We scan across all subsystems rather than using | |
3412 | * the per-hierarchy linked list of mounted subsystems since | |
3413 | * we can be called via check_for_release() with no | |
3414 | * synchronization other than RCU, and the subsystem linked | |
3415 | * list isn't RCU-safe */ | |
3416 | int i; | |
3417 | /* | |
3418 | * We won't need to lock the subsys array, because the subsystems | |
3419 | * we're concerned about aren't going anywhere since our cgroup root | |
3420 | * has a reference on them. | |
3421 | */ | |
3422 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
3423 | struct cgroup_subsys *ss = subsys[i]; | |
3424 | struct cgroup_subsys_state *css; | |
3425 | /* Skip subsystems not present or not in this hierarchy */ | |
3426 | if (ss == NULL || ss->root != cgrp->root) | |
3427 | continue; | |
3428 | css = cgrp->subsys[ss->subsys_id]; | |
3429 | /* When called from check_for_release() it's possible | |
3430 | * that by this point the cgroup has been removed | |
3431 | * and the css deleted. But a false-positive doesn't | |
3432 | * matter, since it can only happen if the cgroup | |
3433 | * has been deleted and hence no longer needs the | |
3434 | * release agent to be called anyway. */ | |
3435 | if (css && (atomic_read(&css->refcnt) > 1)) | |
3436 | return 1; | |
3437 | } | |
3438 | return 0; | |
3439 | } | |
3440 | ||
3441 | /* | |
3442 | * Atomically mark all (or else none) of the cgroup's CSS objects as | |
3443 | * CSS_REMOVED. Return true on success, or false if the cgroup has | |
3444 | * busy subsystems. Call with cgroup_mutex held | |
3445 | */ | |
3446 | ||
3447 | static int cgroup_clear_css_refs(struct cgroup *cgrp) | |
3448 | { | |
3449 | struct cgroup_subsys *ss; | |
3450 | unsigned long flags; | |
3451 | bool failed = false; | |
3452 | local_irq_save(flags); | |
3453 | for_each_subsys(cgrp->root, ss) { | |
3454 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3455 | int refcnt; | |
3456 | while (1) { | |
3457 | /* We can only remove a CSS with a refcnt==1 */ | |
3458 | refcnt = atomic_read(&css->refcnt); | |
3459 | if (refcnt > 1) { | |
3460 | failed = true; | |
3461 | goto done; | |
3462 | } | |
3463 | BUG_ON(!refcnt); | |
3464 | /* | |
3465 | * Drop the refcnt to 0 while we check other | |
3466 | * subsystems. This will cause any racing | |
3467 | * css_tryget() to spin until we set the | |
3468 | * CSS_REMOVED bits or abort | |
3469 | */ | |
3470 | if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt) | |
3471 | break; | |
3472 | cpu_relax(); | |
3473 | } | |
3474 | } | |
3475 | done: | |
3476 | for_each_subsys(cgrp->root, ss) { | |
3477 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3478 | if (failed) { | |
3479 | /* | |
3480 | * Restore old refcnt if we previously managed | |
3481 | * to clear it from 1 to 0 | |
3482 | */ | |
3483 | if (!atomic_read(&css->refcnt)) | |
3484 | atomic_set(&css->refcnt, 1); | |
3485 | } else { | |
3486 | /* Commit the fact that the CSS is removed */ | |
3487 | set_bit(CSS_REMOVED, &css->flags); | |
3488 | } | |
3489 | } | |
3490 | local_irq_restore(flags); | |
3491 | return !failed; | |
3492 | } | |
3493 | ||
3494 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) | |
3495 | { | |
3496 | struct cgroup *cgrp = dentry->d_fsdata; | |
3497 | struct dentry *d; | |
3498 | struct cgroup *parent; | |
3499 | DEFINE_WAIT(wait); | |
3500 | struct cgroup_event *event, *tmp; | |
3501 | int ret; | |
3502 | ||
3503 | /* the vfs holds both inode->i_mutex already */ | |
3504 | again: | |
3505 | mutex_lock(&cgroup_mutex); | |
3506 | if (atomic_read(&cgrp->count) != 0) { | |
3507 | mutex_unlock(&cgroup_mutex); | |
3508 | return -EBUSY; | |
3509 | } | |
3510 | if (!list_empty(&cgrp->children)) { | |
3511 | mutex_unlock(&cgroup_mutex); | |
3512 | return -EBUSY; | |
3513 | } | |
3514 | mutex_unlock(&cgroup_mutex); | |
3515 | ||
3516 | /* | |
3517 | * In general, subsystem has no css->refcnt after pre_destroy(). But | |
3518 | * in racy cases, subsystem may have to get css->refcnt after | |
3519 | * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes | |
3520 | * make rmdir return -EBUSY too often. To avoid that, we use waitqueue | |
3521 | * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir | |
3522 | * and subsystem's reference count handling. Please see css_get/put | |
3523 | * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation. | |
3524 | */ | |
3525 | set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
3526 | ||
3527 | /* | |
3528 | * Call pre_destroy handlers of subsys. Notify subsystems | |
3529 | * that rmdir() request comes. | |
3530 | */ | |
3531 | ret = cgroup_call_pre_destroy(cgrp); | |
3532 | if (ret) { | |
3533 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
3534 | return ret; | |
3535 | } | |
3536 | ||
3537 | mutex_lock(&cgroup_mutex); | |
3538 | parent = cgrp->parent; | |
3539 | if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) { | |
3540 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
3541 | mutex_unlock(&cgroup_mutex); | |
3542 | return -EBUSY; | |
3543 | } | |
3544 | prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE); | |
3545 | if (!cgroup_clear_css_refs(cgrp)) { | |
3546 | mutex_unlock(&cgroup_mutex); | |
3547 | /* | |
3548 | * Because someone may call cgroup_wakeup_rmdir_waiter() before | |
3549 | * prepare_to_wait(), we need to check this flag. | |
3550 | */ | |
3551 | if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)) | |
3552 | schedule(); | |
3553 | finish_wait(&cgroup_rmdir_waitq, &wait); | |
3554 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
3555 | if (signal_pending(current)) | |
3556 | return -EINTR; | |
3557 | goto again; | |
3558 | } | |
3559 | /* NO css_tryget() can success after here. */ | |
3560 | finish_wait(&cgroup_rmdir_waitq, &wait); | |
3561 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
3562 | ||
3563 | spin_lock(&release_list_lock); | |
3564 | set_bit(CGRP_REMOVED, &cgrp->flags); | |
3565 | if (!list_empty(&cgrp->release_list)) | |
3566 | list_del(&cgrp->release_list); | |
3567 | spin_unlock(&release_list_lock); | |
3568 | ||
3569 | cgroup_lock_hierarchy(cgrp->root); | |
3570 | /* delete this cgroup from parent->children */ | |
3571 | list_del(&cgrp->sibling); | |
3572 | cgroup_unlock_hierarchy(cgrp->root); | |
3573 | ||
3574 | spin_lock(&cgrp->dentry->d_lock); | |
3575 | d = dget(cgrp->dentry); | |
3576 | spin_unlock(&d->d_lock); | |
3577 | ||
3578 | cgroup_d_remove_dir(d); | |
3579 | dput(d); | |
3580 | ||
3581 | set_bit(CGRP_RELEASABLE, &parent->flags); | |
3582 | check_for_release(parent); | |
3583 | ||
3584 | /* | |
3585 | * Unregister events and notify userspace. | |
3586 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
3587 | * directory to avoid race between userspace and kernelspace | |
3588 | */ | |
3589 | spin_lock(&cgrp->event_list_lock); | |
3590 | list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) { | |
3591 | list_del(&event->list); | |
3592 | remove_wait_queue(event->wqh, &event->wait); | |
3593 | eventfd_signal(event->eventfd, 1); | |
3594 | schedule_work(&event->remove); | |
3595 | } | |
3596 | spin_unlock(&cgrp->event_list_lock); | |
3597 | ||
3598 | mutex_unlock(&cgroup_mutex); | |
3599 | return 0; | |
3600 | } | |
3601 | ||
3602 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) | |
3603 | { | |
3604 | struct cgroup_subsys_state *css; | |
3605 | ||
3606 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
3607 | ||
3608 | /* Create the top cgroup state for this subsystem */ | |
3609 | list_add(&ss->sibling, &rootnode.subsys_list); | |
3610 | ss->root = &rootnode; | |
3611 | css = ss->create(ss, dummytop); | |
3612 | /* We don't handle early failures gracefully */ | |
3613 | BUG_ON(IS_ERR(css)); | |
3614 | init_cgroup_css(css, ss, dummytop); | |
3615 | ||
3616 | /* Update the init_css_set to contain a subsys | |
3617 | * pointer to this state - since the subsystem is | |
3618 | * newly registered, all tasks and hence the | |
3619 | * init_css_set is in the subsystem's top cgroup. */ | |
3620 | init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id]; | |
3621 | ||
3622 | need_forkexit_callback |= ss->fork || ss->exit; | |
3623 | ||
3624 | /* At system boot, before all subsystems have been | |
3625 | * registered, no tasks have been forked, so we don't | |
3626 | * need to invoke fork callbacks here. */ | |
3627 | BUG_ON(!list_empty(&init_task.tasks)); | |
3628 | ||
3629 | mutex_init(&ss->hierarchy_mutex); | |
3630 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); | |
3631 | ss->active = 1; | |
3632 | ||
3633 | /* this function shouldn't be used with modular subsystems, since they | |
3634 | * need to register a subsys_id, among other things */ | |
3635 | BUG_ON(ss->module); | |
3636 | } | |
3637 | ||
3638 | /** | |
3639 | * cgroup_load_subsys: load and register a modular subsystem at runtime | |
3640 | * @ss: the subsystem to load | |
3641 | * | |
3642 | * This function should be called in a modular subsystem's initcall. If the | |
3643 | * subsystem is built as a module, it will be assigned a new subsys_id and set | |
3644 | * up for use. If the subsystem is built-in anyway, work is delegated to the | |
3645 | * simpler cgroup_init_subsys. | |
3646 | */ | |
3647 | int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss) | |
3648 | { | |
3649 | int i; | |
3650 | struct cgroup_subsys_state *css; | |
3651 | ||
3652 | /* check name and function validity */ | |
3653 | if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN || | |
3654 | ss->create == NULL || ss->destroy == NULL) | |
3655 | return -EINVAL; | |
3656 | ||
3657 | /* | |
3658 | * we don't support callbacks in modular subsystems. this check is | |
3659 | * before the ss->module check for consistency; a subsystem that could | |
3660 | * be a module should still have no callbacks even if the user isn't | |
3661 | * compiling it as one. | |
3662 | */ | |
3663 | if (ss->fork || ss->exit) | |
3664 | return -EINVAL; | |
3665 | ||
3666 | /* | |
3667 | * an optionally modular subsystem is built-in: we want to do nothing, | |
3668 | * since cgroup_init_subsys will have already taken care of it. | |
3669 | */ | |
3670 | if (ss->module == NULL) { | |
3671 | /* a few sanity checks */ | |
3672 | BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT); | |
3673 | BUG_ON(subsys[ss->subsys_id] != ss); | |
3674 | return 0; | |
3675 | } | |
3676 | ||
3677 | /* | |
3678 | * need to register a subsys id before anything else - for example, | |
3679 | * init_cgroup_css needs it. | |
3680 | */ | |
3681 | mutex_lock(&cgroup_mutex); | |
3682 | /* find the first empty slot in the array */ | |
3683 | for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) { | |
3684 | if (subsys[i] == NULL) | |
3685 | break; | |
3686 | } | |
3687 | if (i == CGROUP_SUBSYS_COUNT) { | |
3688 | /* maximum number of subsystems already registered! */ | |
3689 | mutex_unlock(&cgroup_mutex); | |
3690 | return -EBUSY; | |
3691 | } | |
3692 | /* assign ourselves the subsys_id */ | |
3693 | ss->subsys_id = i; | |
3694 | subsys[i] = ss; | |
3695 | ||
3696 | /* | |
3697 | * no ss->create seems to need anything important in the ss struct, so | |
3698 | * this can happen first (i.e. before the rootnode attachment). | |
3699 | */ | |
3700 | css = ss->create(ss, dummytop); | |
3701 | if (IS_ERR(css)) { | |
3702 | /* failure case - need to deassign the subsys[] slot. */ | |
3703 | subsys[i] = NULL; | |
3704 | mutex_unlock(&cgroup_mutex); | |
3705 | return PTR_ERR(css); | |
3706 | } | |
3707 | ||
3708 | list_add(&ss->sibling, &rootnode.subsys_list); | |
3709 | ss->root = &rootnode; | |
3710 | ||
3711 | /* our new subsystem will be attached to the dummy hierarchy. */ | |
3712 | init_cgroup_css(css, ss, dummytop); | |
3713 | /* init_idr must be after init_cgroup_css because it sets css->id. */ | |
3714 | if (ss->use_id) { | |
3715 | int ret = cgroup_init_idr(ss, css); | |
3716 | if (ret) { | |
3717 | dummytop->subsys[ss->subsys_id] = NULL; | |
3718 | ss->destroy(ss, dummytop); | |
3719 | subsys[i] = NULL; | |
3720 | mutex_unlock(&cgroup_mutex); | |
3721 | return ret; | |
3722 | } | |
3723 | } | |
3724 | ||
3725 | /* | |
3726 | * Now we need to entangle the css into the existing css_sets. unlike | |
3727 | * in cgroup_init_subsys, there are now multiple css_sets, so each one | |
3728 | * will need a new pointer to it; done by iterating the css_set_table. | |
3729 | * furthermore, modifying the existing css_sets will corrupt the hash | |
3730 | * table state, so each changed css_set will need its hash recomputed. | |
3731 | * this is all done under the css_set_lock. | |
3732 | */ | |
3733 | write_lock(&css_set_lock); | |
3734 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { | |
3735 | struct css_set *cg; | |
3736 | struct hlist_node *node, *tmp; | |
3737 | struct hlist_head *bucket = &css_set_table[i], *new_bucket; | |
3738 | ||
3739 | hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) { | |
3740 | /* skip entries that we already rehashed */ | |
3741 | if (cg->subsys[ss->subsys_id]) | |
3742 | continue; | |
3743 | /* remove existing entry */ | |
3744 | hlist_del(&cg->hlist); | |
3745 | /* set new value */ | |
3746 | cg->subsys[ss->subsys_id] = css; | |
3747 | /* recompute hash and restore entry */ | |
3748 | new_bucket = css_set_hash(cg->subsys); | |
3749 | hlist_add_head(&cg->hlist, new_bucket); | |
3750 | } | |
3751 | } | |
3752 | write_unlock(&css_set_lock); | |
3753 | ||
3754 | mutex_init(&ss->hierarchy_mutex); | |
3755 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); | |
3756 | ss->active = 1; | |
3757 | ||
3758 | /* success! */ | |
3759 | mutex_unlock(&cgroup_mutex); | |
3760 | return 0; | |
3761 | } | |
3762 | EXPORT_SYMBOL_GPL(cgroup_load_subsys); | |
3763 | ||
3764 | /** | |
3765 | * cgroup_unload_subsys: unload a modular subsystem | |
3766 | * @ss: the subsystem to unload | |
3767 | * | |
3768 | * This function should be called in a modular subsystem's exitcall. When this | |
3769 | * function is invoked, the refcount on the subsystem's module will be 0, so | |
3770 | * the subsystem will not be attached to any hierarchy. | |
3771 | */ | |
3772 | void cgroup_unload_subsys(struct cgroup_subsys *ss) | |
3773 | { | |
3774 | struct cg_cgroup_link *link; | |
3775 | struct hlist_head *hhead; | |
3776 | ||
3777 | BUG_ON(ss->module == NULL); | |
3778 | ||
3779 | /* | |
3780 | * we shouldn't be called if the subsystem is in use, and the use of | |
3781 | * try_module_get in parse_cgroupfs_options should ensure that it | |
3782 | * doesn't start being used while we're killing it off. | |
3783 | */ | |
3784 | BUG_ON(ss->root != &rootnode); | |
3785 | ||
3786 | mutex_lock(&cgroup_mutex); | |
3787 | /* deassign the subsys_id */ | |
3788 | BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT); | |
3789 | subsys[ss->subsys_id] = NULL; | |
3790 | ||
3791 | /* remove subsystem from rootnode's list of subsystems */ | |
3792 | list_del(&ss->sibling); | |
3793 | ||
3794 | /* | |
3795 | * disentangle the css from all css_sets attached to the dummytop. as | |
3796 | * in loading, we need to pay our respects to the hashtable gods. | |
3797 | */ | |
3798 | write_lock(&css_set_lock); | |
3799 | list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) { | |
3800 | struct css_set *cg = link->cg; | |
3801 | ||
3802 | hlist_del(&cg->hlist); | |
3803 | BUG_ON(!cg->subsys[ss->subsys_id]); | |
3804 | cg->subsys[ss->subsys_id] = NULL; | |
3805 | hhead = css_set_hash(cg->subsys); | |
3806 | hlist_add_head(&cg->hlist, hhead); | |
3807 | } | |
3808 | write_unlock(&css_set_lock); | |
3809 | ||
3810 | /* | |
3811 | * remove subsystem's css from the dummytop and free it - need to free | |
3812 | * before marking as null because ss->destroy needs the cgrp->subsys | |
3813 | * pointer to find their state. note that this also takes care of | |
3814 | * freeing the css_id. | |
3815 | */ | |
3816 | ss->destroy(ss, dummytop); | |
3817 | dummytop->subsys[ss->subsys_id] = NULL; | |
3818 | ||
3819 | mutex_unlock(&cgroup_mutex); | |
3820 | } | |
3821 | EXPORT_SYMBOL_GPL(cgroup_unload_subsys); | |
3822 | ||
3823 | /** | |
3824 | * cgroup_init_early - cgroup initialization at system boot | |
3825 | * | |
3826 | * Initialize cgroups at system boot, and initialize any | |
3827 | * subsystems that request early init. | |
3828 | */ | |
3829 | int __init cgroup_init_early(void) | |
3830 | { | |
3831 | int i; | |
3832 | atomic_set(&init_css_set.refcount, 1); | |
3833 | INIT_LIST_HEAD(&init_css_set.cg_links); | |
3834 | INIT_LIST_HEAD(&init_css_set.tasks); | |
3835 | INIT_HLIST_NODE(&init_css_set.hlist); | |
3836 | css_set_count = 1; | |
3837 | init_cgroup_root(&rootnode); | |
3838 | root_count = 1; | |
3839 | init_task.cgroups = &init_css_set; | |
3840 | ||
3841 | init_css_set_link.cg = &init_css_set; | |
3842 | init_css_set_link.cgrp = dummytop; | |
3843 | list_add(&init_css_set_link.cgrp_link_list, | |
3844 | &rootnode.top_cgroup.css_sets); | |
3845 | list_add(&init_css_set_link.cg_link_list, | |
3846 | &init_css_set.cg_links); | |
3847 | ||
3848 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) | |
3849 | INIT_HLIST_HEAD(&css_set_table[i]); | |
3850 | ||
3851 | /* at bootup time, we don't worry about modular subsystems */ | |
3852 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
3853 | struct cgroup_subsys *ss = subsys[i]; | |
3854 | ||
3855 | BUG_ON(!ss->name); | |
3856 | BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); | |
3857 | BUG_ON(!ss->create); | |
3858 | BUG_ON(!ss->destroy); | |
3859 | if (ss->subsys_id != i) { | |
3860 | printk(KERN_ERR "cgroup: Subsys %s id == %d\n", | |
3861 | ss->name, ss->subsys_id); | |
3862 | BUG(); | |
3863 | } | |
3864 | ||
3865 | if (ss->early_init) | |
3866 | cgroup_init_subsys(ss); | |
3867 | } | |
3868 | return 0; | |
3869 | } | |
3870 | ||
3871 | /** | |
3872 | * cgroup_init - cgroup initialization | |
3873 | * | |
3874 | * Register cgroup filesystem and /proc file, and initialize | |
3875 | * any subsystems that didn't request early init. | |
3876 | */ | |
3877 | int __init cgroup_init(void) | |
3878 | { | |
3879 | int err; | |
3880 | int i; | |
3881 | struct hlist_head *hhead; | |
3882 | ||
3883 | err = bdi_init(&cgroup_backing_dev_info); | |
3884 | if (err) | |
3885 | return err; | |
3886 | ||
3887 | /* at bootup time, we don't worry about modular subsystems */ | |
3888 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
3889 | struct cgroup_subsys *ss = subsys[i]; | |
3890 | if (!ss->early_init) | |
3891 | cgroup_init_subsys(ss); | |
3892 | if (ss->use_id) | |
3893 | cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]); | |
3894 | } | |
3895 | ||
3896 | /* Add init_css_set to the hash table */ | |
3897 | hhead = css_set_hash(init_css_set.subsys); | |
3898 | hlist_add_head(&init_css_set.hlist, hhead); | |
3899 | BUG_ON(!init_root_id(&rootnode)); | |
3900 | ||
3901 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); | |
3902 | if (!cgroup_kobj) { | |
3903 | err = -ENOMEM; | |
3904 | goto out; | |
3905 | } | |
3906 | ||
3907 | err = register_filesystem(&cgroup_fs_type); | |
3908 | if (err < 0) { | |
3909 | kobject_put(cgroup_kobj); | |
3910 | goto out; | |
3911 | } | |
3912 | ||
3913 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); | |
3914 | ||
3915 | out: | |
3916 | if (err) | |
3917 | bdi_destroy(&cgroup_backing_dev_info); | |
3918 | ||
3919 | return err; | |
3920 | } | |
3921 | ||
3922 | /* | |
3923 | * proc_cgroup_show() | |
3924 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
3925 | * - Used for /proc/<pid>/cgroup. | |
3926 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
3927 | * doesn't really matter if tsk->cgroup changes after we read it, | |
3928 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it | |
3929 | * anyway. No need to check that tsk->cgroup != NULL, thanks to | |
3930 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
3931 | * cgroup to top_cgroup. | |
3932 | */ | |
3933 | ||
3934 | /* TODO: Use a proper seq_file iterator */ | |
3935 | static int proc_cgroup_show(struct seq_file *m, void *v) | |
3936 | { | |
3937 | struct pid *pid; | |
3938 | struct task_struct *tsk; | |
3939 | char *buf; | |
3940 | int retval; | |
3941 | struct cgroupfs_root *root; | |
3942 | ||
3943 | retval = -ENOMEM; | |
3944 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
3945 | if (!buf) | |
3946 | goto out; | |
3947 | ||
3948 | retval = -ESRCH; | |
3949 | pid = m->private; | |
3950 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
3951 | if (!tsk) | |
3952 | goto out_free; | |
3953 | ||
3954 | retval = 0; | |
3955 | ||
3956 | mutex_lock(&cgroup_mutex); | |
3957 | ||
3958 | for_each_active_root(root) { | |
3959 | struct cgroup_subsys *ss; | |
3960 | struct cgroup *cgrp; | |
3961 | int count = 0; | |
3962 | ||
3963 | seq_printf(m, "%d:", root->hierarchy_id); | |
3964 | for_each_subsys(root, ss) | |
3965 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
3966 | if (strlen(root->name)) | |
3967 | seq_printf(m, "%sname=%s", count ? "," : "", | |
3968 | root->name); | |
3969 | seq_putc(m, ':'); | |
3970 | cgrp = task_cgroup_from_root(tsk, root); | |
3971 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); | |
3972 | if (retval < 0) | |
3973 | goto out_unlock; | |
3974 | seq_puts(m, buf); | |
3975 | seq_putc(m, '\n'); | |
3976 | } | |
3977 | ||
3978 | out_unlock: | |
3979 | mutex_unlock(&cgroup_mutex); | |
3980 | put_task_struct(tsk); | |
3981 | out_free: | |
3982 | kfree(buf); | |
3983 | out: | |
3984 | return retval; | |
3985 | } | |
3986 | ||
3987 | static int cgroup_open(struct inode *inode, struct file *file) | |
3988 | { | |
3989 | struct pid *pid = PROC_I(inode)->pid; | |
3990 | return single_open(file, proc_cgroup_show, pid); | |
3991 | } | |
3992 | ||
3993 | const struct file_operations proc_cgroup_operations = { | |
3994 | .open = cgroup_open, | |
3995 | .read = seq_read, | |
3996 | .llseek = seq_lseek, | |
3997 | .release = single_release, | |
3998 | }; | |
3999 | ||
4000 | /* Display information about each subsystem and each hierarchy */ | |
4001 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4002 | { | |
4003 | int i; | |
4004 | ||
4005 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); | |
4006 | /* | |
4007 | * ideally we don't want subsystems moving around while we do this. | |
4008 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4009 | * subsys/hierarchy state. | |
4010 | */ | |
4011 | mutex_lock(&cgroup_mutex); | |
4012 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
4013 | struct cgroup_subsys *ss = subsys[i]; | |
4014 | if (ss == NULL) | |
4015 | continue; | |
4016 | seq_printf(m, "%s\t%d\t%d\t%d\n", | |
4017 | ss->name, ss->root->hierarchy_id, | |
4018 | ss->root->number_of_cgroups, !ss->disabled); | |
4019 | } | |
4020 | mutex_unlock(&cgroup_mutex); | |
4021 | return 0; | |
4022 | } | |
4023 | ||
4024 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4025 | { | |
4026 | return single_open(file, proc_cgroupstats_show, NULL); | |
4027 | } | |
4028 | ||
4029 | static const struct file_operations proc_cgroupstats_operations = { | |
4030 | .open = cgroupstats_open, | |
4031 | .read = seq_read, | |
4032 | .llseek = seq_lseek, | |
4033 | .release = single_release, | |
4034 | }; | |
4035 | ||
4036 | /** | |
4037 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
4038 | * @child: pointer to task_struct of forking parent process. | |
4039 | * | |
4040 | * Description: A task inherits its parent's cgroup at fork(). | |
4041 | * | |
4042 | * A pointer to the shared css_set was automatically copied in | |
4043 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
4044 | * it was not made under the protection of RCU or cgroup_mutex, so | |
4045 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
4046 | * have already changed current->cgroups, allowing the previously | |
4047 | * referenced cgroup group to be removed and freed. | |
4048 | * | |
4049 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4050 | * task, and the passed argument 'child' points to the child task. | |
4051 | */ | |
4052 | void cgroup_fork(struct task_struct *child) | |
4053 | { | |
4054 | task_lock(current); | |
4055 | child->cgroups = current->cgroups; | |
4056 | get_css_set(child->cgroups); | |
4057 | task_unlock(current); | |
4058 | INIT_LIST_HEAD(&child->cg_list); | |
4059 | } | |
4060 | ||
4061 | /** | |
4062 | * cgroup_fork_callbacks - run fork callbacks | |
4063 | * @child: the new task | |
4064 | * | |
4065 | * Called on a new task very soon before adding it to the | |
4066 | * tasklist. No need to take any locks since no-one can | |
4067 | * be operating on this task. | |
4068 | */ | |
4069 | void cgroup_fork_callbacks(struct task_struct *child) | |
4070 | { | |
4071 | if (need_forkexit_callback) { | |
4072 | int i; | |
4073 | /* | |
4074 | * forkexit callbacks are only supported for builtin | |
4075 | * subsystems, and the builtin section of the subsys array is | |
4076 | * immutable, so we don't need to lock the subsys array here. | |
4077 | */ | |
4078 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
4079 | struct cgroup_subsys *ss = subsys[i]; | |
4080 | if (ss->fork) | |
4081 | ss->fork(ss, child); | |
4082 | } | |
4083 | } | |
4084 | } | |
4085 | ||
4086 | /** | |
4087 | * cgroup_post_fork - called on a new task after adding it to the task list | |
4088 | * @child: the task in question | |
4089 | * | |
4090 | * Adds the task to the list running through its css_set if necessary. | |
4091 | * Has to be after the task is visible on the task list in case we race | |
4092 | * with the first call to cgroup_iter_start() - to guarantee that the | |
4093 | * new task ends up on its list. | |
4094 | */ | |
4095 | void cgroup_post_fork(struct task_struct *child) | |
4096 | { | |
4097 | if (use_task_css_set_links) { | |
4098 | write_lock(&css_set_lock); | |
4099 | task_lock(child); | |
4100 | if (list_empty(&child->cg_list)) | |
4101 | list_add(&child->cg_list, &child->cgroups->tasks); | |
4102 | task_unlock(child); | |
4103 | write_unlock(&css_set_lock); | |
4104 | } | |
4105 | } | |
4106 | /** | |
4107 | * cgroup_exit - detach cgroup from exiting task | |
4108 | * @tsk: pointer to task_struct of exiting process | |
4109 | * @run_callback: run exit callbacks? | |
4110 | * | |
4111 | * Description: Detach cgroup from @tsk and release it. | |
4112 | * | |
4113 | * Note that cgroups marked notify_on_release force every task in | |
4114 | * them to take the global cgroup_mutex mutex when exiting. | |
4115 | * This could impact scaling on very large systems. Be reluctant to | |
4116 | * use notify_on_release cgroups where very high task exit scaling | |
4117 | * is required on large systems. | |
4118 | * | |
4119 | * the_top_cgroup_hack: | |
4120 | * | |
4121 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4122 | * | |
4123 | * We call cgroup_exit() while the task is still competent to | |
4124 | * handle notify_on_release(), then leave the task attached to the | |
4125 | * root cgroup in each hierarchy for the remainder of its exit. | |
4126 | * | |
4127 | * To do this properly, we would increment the reference count on | |
4128 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4129 | * code we would add a second cgroup function call, to drop that | |
4130 | * reference. This would just create an unnecessary hot spot on | |
4131 | * the top_cgroup reference count, to no avail. | |
4132 | * | |
4133 | * Normally, holding a reference to a cgroup without bumping its | |
4134 | * count is unsafe. The cgroup could go away, or someone could | |
4135 | * attach us to a different cgroup, decrementing the count on | |
4136 | * the first cgroup that we never incremented. But in this case, | |
4137 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
4138 | * which wards off any cgroup_attach_task() attempts, or task is a failed | |
4139 | * fork, never visible to cgroup_attach_task. | |
4140 | */ | |
4141 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4142 | { | |
4143 | int i; | |
4144 | struct css_set *cg; | |
4145 | ||
4146 | if (run_callbacks && need_forkexit_callback) { | |
4147 | /* | |
4148 | * modular subsystems can't use callbacks, so no need to lock | |
4149 | * the subsys array | |
4150 | */ | |
4151 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
4152 | struct cgroup_subsys *ss = subsys[i]; | |
4153 | if (ss->exit) | |
4154 | ss->exit(ss, tsk); | |
4155 | } | |
4156 | } | |
4157 | ||
4158 | /* | |
4159 | * Unlink from the css_set task list if necessary. | |
4160 | * Optimistically check cg_list before taking | |
4161 | * css_set_lock | |
4162 | */ | |
4163 | if (!list_empty(&tsk->cg_list)) { | |
4164 | write_lock(&css_set_lock); | |
4165 | if (!list_empty(&tsk->cg_list)) | |
4166 | list_del(&tsk->cg_list); | |
4167 | write_unlock(&css_set_lock); | |
4168 | } | |
4169 | ||
4170 | /* Reassign the task to the init_css_set. */ | |
4171 | task_lock(tsk); | |
4172 | cg = tsk->cgroups; | |
4173 | tsk->cgroups = &init_css_set; | |
4174 | task_unlock(tsk); | |
4175 | if (cg) | |
4176 | put_css_set_taskexit(cg); | |
4177 | } | |
4178 | ||
4179 | /** | |
4180 | * cgroup_clone - clone the cgroup the given subsystem is attached to | |
4181 | * @tsk: the task to be moved | |
4182 | * @subsys: the given subsystem | |
4183 | * @nodename: the name for the new cgroup | |
4184 | * | |
4185 | * Duplicate the current cgroup in the hierarchy that the given | |
4186 | * subsystem is attached to, and move this task into the new | |
4187 | * child. | |
4188 | */ | |
4189 | int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys, | |
4190 | char *nodename) | |
4191 | { | |
4192 | struct dentry *dentry; | |
4193 | int ret = 0; | |
4194 | struct cgroup *parent, *child; | |
4195 | struct inode *inode; | |
4196 | struct css_set *cg; | |
4197 | struct cgroupfs_root *root; | |
4198 | struct cgroup_subsys *ss; | |
4199 | ||
4200 | /* We shouldn't be called by an unregistered subsystem */ | |
4201 | BUG_ON(!subsys->active); | |
4202 | ||
4203 | /* First figure out what hierarchy and cgroup we're dealing | |
4204 | * with, and pin them so we can drop cgroup_mutex */ | |
4205 | mutex_lock(&cgroup_mutex); | |
4206 | again: | |
4207 | root = subsys->root; | |
4208 | if (root == &rootnode) { | |
4209 | mutex_unlock(&cgroup_mutex); | |
4210 | return 0; | |
4211 | } | |
4212 | ||
4213 | /* Pin the hierarchy */ | |
4214 | if (!atomic_inc_not_zero(&root->sb->s_active)) { | |
4215 | /* We race with the final deactivate_super() */ | |
4216 | mutex_unlock(&cgroup_mutex); | |
4217 | return 0; | |
4218 | } | |
4219 | ||
4220 | /* Keep the cgroup alive */ | |
4221 | task_lock(tsk); | |
4222 | parent = task_cgroup(tsk, subsys->subsys_id); | |
4223 | cg = tsk->cgroups; | |
4224 | get_css_set(cg); | |
4225 | task_unlock(tsk); | |
4226 | ||
4227 | mutex_unlock(&cgroup_mutex); | |
4228 | ||
4229 | /* Now do the VFS work to create a cgroup */ | |
4230 | inode = parent->dentry->d_inode; | |
4231 | ||
4232 | /* Hold the parent directory mutex across this operation to | |
4233 | * stop anyone else deleting the new cgroup */ | |
4234 | mutex_lock(&inode->i_mutex); | |
4235 | dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename)); | |
4236 | if (IS_ERR(dentry)) { | |
4237 | printk(KERN_INFO | |
4238 | "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename, | |
4239 | PTR_ERR(dentry)); | |
4240 | ret = PTR_ERR(dentry); | |
4241 | goto out_release; | |
4242 | } | |
4243 | ||
4244 | /* Create the cgroup directory, which also creates the cgroup */ | |
4245 | ret = vfs_mkdir(inode, dentry, 0755); | |
4246 | child = __d_cgrp(dentry); | |
4247 | dput(dentry); | |
4248 | if (ret) { | |
4249 | printk(KERN_INFO | |
4250 | "Failed to create cgroup %s: %d\n", nodename, | |
4251 | ret); | |
4252 | goto out_release; | |
4253 | } | |
4254 | ||
4255 | /* The cgroup now exists. Retake cgroup_mutex and check | |
4256 | * that we're still in the same state that we thought we | |
4257 | * were. */ | |
4258 | mutex_lock(&cgroup_mutex); | |
4259 | if ((root != subsys->root) || | |
4260 | (parent != task_cgroup(tsk, subsys->subsys_id))) { | |
4261 | /* Aargh, we raced ... */ | |
4262 | mutex_unlock(&inode->i_mutex); | |
4263 | put_css_set(cg); | |
4264 | ||
4265 | deactivate_super(root->sb); | |
4266 | /* The cgroup is still accessible in the VFS, but | |
4267 | * we're not going to try to rmdir() it at this | |
4268 | * point. */ | |
4269 | printk(KERN_INFO | |
4270 | "Race in cgroup_clone() - leaking cgroup %s\n", | |
4271 | nodename); | |
4272 | goto again; | |
4273 | } | |
4274 | ||
4275 | /* do any required auto-setup */ | |
4276 | for_each_subsys(root, ss) { | |
4277 | if (ss->post_clone) | |
4278 | ss->post_clone(ss, child); | |
4279 | } | |
4280 | ||
4281 | /* All seems fine. Finish by moving the task into the new cgroup */ | |
4282 | ret = cgroup_attach_task(child, tsk); | |
4283 | mutex_unlock(&cgroup_mutex); | |
4284 | ||
4285 | out_release: | |
4286 | mutex_unlock(&inode->i_mutex); | |
4287 | ||
4288 | mutex_lock(&cgroup_mutex); | |
4289 | put_css_set(cg); | |
4290 | mutex_unlock(&cgroup_mutex); | |
4291 | deactivate_super(root->sb); | |
4292 | return ret; | |
4293 | } | |
4294 | ||
4295 | /** | |
4296 | * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp | |
4297 | * @cgrp: the cgroup in question | |
4298 | * @task: the task in question | |
4299 | * | |
4300 | * See if @cgrp is a descendant of @task's cgroup in the appropriate | |
4301 | * hierarchy. | |
4302 | * | |
4303 | * If we are sending in dummytop, then presumably we are creating | |
4304 | * the top cgroup in the subsystem. | |
4305 | * | |
4306 | * Called only by the ns (nsproxy) cgroup. | |
4307 | */ | |
4308 | int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task) | |
4309 | { | |
4310 | int ret; | |
4311 | struct cgroup *target; | |
4312 | ||
4313 | if (cgrp == dummytop) | |
4314 | return 1; | |
4315 | ||
4316 | target = task_cgroup_from_root(task, cgrp->root); | |
4317 | while (cgrp != target && cgrp!= cgrp->top_cgroup) | |
4318 | cgrp = cgrp->parent; | |
4319 | ret = (cgrp == target); | |
4320 | return ret; | |
4321 | } | |
4322 | ||
4323 | static void check_for_release(struct cgroup *cgrp) | |
4324 | { | |
4325 | /* All of these checks rely on RCU to keep the cgroup | |
4326 | * structure alive */ | |
4327 | if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count) | |
4328 | && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) { | |
4329 | /* Control Group is currently removeable. If it's not | |
4330 | * already queued for a userspace notification, queue | |
4331 | * it now */ | |
4332 | int need_schedule_work = 0; | |
4333 | spin_lock(&release_list_lock); | |
4334 | if (!cgroup_is_removed(cgrp) && | |
4335 | list_empty(&cgrp->release_list)) { | |
4336 | list_add(&cgrp->release_list, &release_list); | |
4337 | need_schedule_work = 1; | |
4338 | } | |
4339 | spin_unlock(&release_list_lock); | |
4340 | if (need_schedule_work) | |
4341 | schedule_work(&release_agent_work); | |
4342 | } | |
4343 | } | |
4344 | ||
4345 | /* Caller must verify that the css is not for root cgroup */ | |
4346 | void __css_put(struct cgroup_subsys_state *css, int count) | |
4347 | { | |
4348 | struct cgroup *cgrp = css->cgroup; | |
4349 | int val; | |
4350 | rcu_read_lock(); | |
4351 | val = atomic_sub_return(count, &css->refcnt); | |
4352 | if (val == 1) { | |
4353 | if (notify_on_release(cgrp)) { | |
4354 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
4355 | check_for_release(cgrp); | |
4356 | } | |
4357 | cgroup_wakeup_rmdir_waiter(cgrp); | |
4358 | } | |
4359 | rcu_read_unlock(); | |
4360 | WARN_ON_ONCE(val < 1); | |
4361 | } | |
4362 | EXPORT_SYMBOL_GPL(__css_put); | |
4363 | ||
4364 | /* | |
4365 | * Notify userspace when a cgroup is released, by running the | |
4366 | * configured release agent with the name of the cgroup (path | |
4367 | * relative to the root of cgroup file system) as the argument. | |
4368 | * | |
4369 | * Most likely, this user command will try to rmdir this cgroup. | |
4370 | * | |
4371 | * This races with the possibility that some other task will be | |
4372 | * attached to this cgroup before it is removed, or that some other | |
4373 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4374 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4375 | * unused, and this cgroup will be reprieved from its death sentence, | |
4376 | * to continue to serve a useful existence. Next time it's released, | |
4377 | * we will get notified again, if it still has 'notify_on_release' set. | |
4378 | * | |
4379 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4380 | * means only wait until the task is successfully execve()'d. The | |
4381 | * separate release agent task is forked by call_usermodehelper(), | |
4382 | * then control in this thread returns here, without waiting for the | |
4383 | * release agent task. We don't bother to wait because the caller of | |
4384 | * this routine has no use for the exit status of the release agent | |
4385 | * task, so no sense holding our caller up for that. | |
4386 | */ | |
4387 | static void cgroup_release_agent(struct work_struct *work) | |
4388 | { | |
4389 | BUG_ON(work != &release_agent_work); | |
4390 | mutex_lock(&cgroup_mutex); | |
4391 | spin_lock(&release_list_lock); | |
4392 | while (!list_empty(&release_list)) { | |
4393 | char *argv[3], *envp[3]; | |
4394 | int i; | |
4395 | char *pathbuf = NULL, *agentbuf = NULL; | |
4396 | struct cgroup *cgrp = list_entry(release_list.next, | |
4397 | struct cgroup, | |
4398 | release_list); | |
4399 | list_del_init(&cgrp->release_list); | |
4400 | spin_unlock(&release_list_lock); | |
4401 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4402 | if (!pathbuf) | |
4403 | goto continue_free; | |
4404 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
4405 | goto continue_free; | |
4406 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4407 | if (!agentbuf) | |
4408 | goto continue_free; | |
4409 | ||
4410 | i = 0; | |
4411 | argv[i++] = agentbuf; | |
4412 | argv[i++] = pathbuf; | |
4413 | argv[i] = NULL; | |
4414 | ||
4415 | i = 0; | |
4416 | /* minimal command environment */ | |
4417 | envp[i++] = "HOME=/"; | |
4418 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4419 | envp[i] = NULL; | |
4420 | ||
4421 | /* Drop the lock while we invoke the usermode helper, | |
4422 | * since the exec could involve hitting disk and hence | |
4423 | * be a slow process */ | |
4424 | mutex_unlock(&cgroup_mutex); | |
4425 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
4426 | mutex_lock(&cgroup_mutex); | |
4427 | continue_free: | |
4428 | kfree(pathbuf); | |
4429 | kfree(agentbuf); | |
4430 | spin_lock(&release_list_lock); | |
4431 | } | |
4432 | spin_unlock(&release_list_lock); | |
4433 | mutex_unlock(&cgroup_mutex); | |
4434 | } | |
4435 | ||
4436 | static int __init cgroup_disable(char *str) | |
4437 | { | |
4438 | int i; | |
4439 | char *token; | |
4440 | ||
4441 | while ((token = strsep(&str, ",")) != NULL) { | |
4442 | if (!*token) | |
4443 | continue; | |
4444 | /* | |
4445 | * cgroup_disable, being at boot time, can't know about module | |
4446 | * subsystems, so we don't worry about them. | |
4447 | */ | |
4448 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
4449 | struct cgroup_subsys *ss = subsys[i]; | |
4450 | ||
4451 | if (!strcmp(token, ss->name)) { | |
4452 | ss->disabled = 1; | |
4453 | printk(KERN_INFO "Disabling %s control group" | |
4454 | " subsystem\n", ss->name); | |
4455 | break; | |
4456 | } | |
4457 | } | |
4458 | } | |
4459 | return 1; | |
4460 | } | |
4461 | __setup("cgroup_disable=", cgroup_disable); | |
4462 | ||
4463 | /* | |
4464 | * Functons for CSS ID. | |
4465 | */ | |
4466 | ||
4467 | /* | |
4468 | *To get ID other than 0, this should be called when !cgroup_is_removed(). | |
4469 | */ | |
4470 | unsigned short css_id(struct cgroup_subsys_state *css) | |
4471 | { | |
4472 | struct css_id *cssid; | |
4473 | ||
4474 | /* | |
4475 | * This css_id() can return correct value when somone has refcnt | |
4476 | * on this or this is under rcu_read_lock(). Once css->id is allocated, | |
4477 | * it's unchanged until freed. | |
4478 | */ | |
4479 | cssid = rcu_dereference_check(css->id, | |
4480 | rcu_read_lock_held() || atomic_read(&css->refcnt)); | |
4481 | ||
4482 | if (cssid) | |
4483 | return cssid->id; | |
4484 | return 0; | |
4485 | } | |
4486 | EXPORT_SYMBOL_GPL(css_id); | |
4487 | ||
4488 | unsigned short css_depth(struct cgroup_subsys_state *css) | |
4489 | { | |
4490 | struct css_id *cssid; | |
4491 | ||
4492 | cssid = rcu_dereference_check(css->id, | |
4493 | rcu_read_lock_held() || atomic_read(&css->refcnt)); | |
4494 | ||
4495 | if (cssid) | |
4496 | return cssid->depth; | |
4497 | return 0; | |
4498 | } | |
4499 | EXPORT_SYMBOL_GPL(css_depth); | |
4500 | ||
4501 | /** | |
4502 | * css_is_ancestor - test "root" css is an ancestor of "child" | |
4503 | * @child: the css to be tested. | |
4504 | * @root: the css supporsed to be an ancestor of the child. | |
4505 | * | |
4506 | * Returns true if "root" is an ancestor of "child" in its hierarchy. Because | |
4507 | * this function reads css->id, this use rcu_dereference() and rcu_read_lock(). | |
4508 | * But, considering usual usage, the csses should be valid objects after test. | |
4509 | * Assuming that the caller will do some action to the child if this returns | |
4510 | * returns true, the caller must take "child";s reference count. | |
4511 | * If "child" is valid object and this returns true, "root" is valid, too. | |
4512 | */ | |
4513 | ||
4514 | bool css_is_ancestor(struct cgroup_subsys_state *child, | |
4515 | const struct cgroup_subsys_state *root) | |
4516 | { | |
4517 | struct css_id *child_id; | |
4518 | struct css_id *root_id; | |
4519 | bool ret = true; | |
4520 | ||
4521 | rcu_read_lock(); | |
4522 | child_id = rcu_dereference(child->id); | |
4523 | root_id = rcu_dereference(root->id); | |
4524 | if (!child_id | |
4525 | || !root_id | |
4526 | || (child_id->depth < root_id->depth) | |
4527 | || (child_id->stack[root_id->depth] != root_id->id)) | |
4528 | ret = false; | |
4529 | rcu_read_unlock(); | |
4530 | return ret; | |
4531 | } | |
4532 | ||
4533 | static void __free_css_id_cb(struct rcu_head *head) | |
4534 | { | |
4535 | struct css_id *id; | |
4536 | ||
4537 | id = container_of(head, struct css_id, rcu_head); | |
4538 | kfree(id); | |
4539 | } | |
4540 | ||
4541 | void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css) | |
4542 | { | |
4543 | struct css_id *id = css->id; | |
4544 | /* When this is called before css_id initialization, id can be NULL */ | |
4545 | if (!id) | |
4546 | return; | |
4547 | ||
4548 | BUG_ON(!ss->use_id); | |
4549 | ||
4550 | rcu_assign_pointer(id->css, NULL); | |
4551 | rcu_assign_pointer(css->id, NULL); | |
4552 | spin_lock(&ss->id_lock); | |
4553 | idr_remove(&ss->idr, id->id); | |
4554 | spin_unlock(&ss->id_lock); | |
4555 | call_rcu(&id->rcu_head, __free_css_id_cb); | |
4556 | } | |
4557 | EXPORT_SYMBOL_GPL(free_css_id); | |
4558 | ||
4559 | /* | |
4560 | * This is called by init or create(). Then, calls to this function are | |
4561 | * always serialized (By cgroup_mutex() at create()). | |
4562 | */ | |
4563 | ||
4564 | static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth) | |
4565 | { | |
4566 | struct css_id *newid; | |
4567 | int myid, error, size; | |
4568 | ||
4569 | BUG_ON(!ss->use_id); | |
4570 | ||
4571 | size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1); | |
4572 | newid = kzalloc(size, GFP_KERNEL); | |
4573 | if (!newid) | |
4574 | return ERR_PTR(-ENOMEM); | |
4575 | /* get id */ | |
4576 | if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) { | |
4577 | error = -ENOMEM; | |
4578 | goto err_out; | |
4579 | } | |
4580 | spin_lock(&ss->id_lock); | |
4581 | /* Don't use 0. allocates an ID of 1-65535 */ | |
4582 | error = idr_get_new_above(&ss->idr, newid, 1, &myid); | |
4583 | spin_unlock(&ss->id_lock); | |
4584 | ||
4585 | /* Returns error when there are no free spaces for new ID.*/ | |
4586 | if (error) { | |
4587 | error = -ENOSPC; | |
4588 | goto err_out; | |
4589 | } | |
4590 | if (myid > CSS_ID_MAX) | |
4591 | goto remove_idr; | |
4592 | ||
4593 | newid->id = myid; | |
4594 | newid->depth = depth; | |
4595 | return newid; | |
4596 | remove_idr: | |
4597 | error = -ENOSPC; | |
4598 | spin_lock(&ss->id_lock); | |
4599 | idr_remove(&ss->idr, myid); | |
4600 | spin_unlock(&ss->id_lock); | |
4601 | err_out: | |
4602 | kfree(newid); | |
4603 | return ERR_PTR(error); | |
4604 | ||
4605 | } | |
4606 | ||
4607 | static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss, | |
4608 | struct cgroup_subsys_state *rootcss) | |
4609 | { | |
4610 | struct css_id *newid; | |
4611 | ||
4612 | spin_lock_init(&ss->id_lock); | |
4613 | idr_init(&ss->idr); | |
4614 | ||
4615 | newid = get_new_cssid(ss, 0); | |
4616 | if (IS_ERR(newid)) | |
4617 | return PTR_ERR(newid); | |
4618 | ||
4619 | newid->stack[0] = newid->id; | |
4620 | newid->css = rootcss; | |
4621 | rootcss->id = newid; | |
4622 | return 0; | |
4623 | } | |
4624 | ||
4625 | static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent, | |
4626 | struct cgroup *child) | |
4627 | { | |
4628 | int subsys_id, i, depth = 0; | |
4629 | struct cgroup_subsys_state *parent_css, *child_css; | |
4630 | struct css_id *child_id, *parent_id; | |
4631 | ||
4632 | subsys_id = ss->subsys_id; | |
4633 | parent_css = parent->subsys[subsys_id]; | |
4634 | child_css = child->subsys[subsys_id]; | |
4635 | parent_id = parent_css->id; | |
4636 | depth = parent_id->depth + 1; | |
4637 | ||
4638 | child_id = get_new_cssid(ss, depth); | |
4639 | if (IS_ERR(child_id)) | |
4640 | return PTR_ERR(child_id); | |
4641 | ||
4642 | for (i = 0; i < depth; i++) | |
4643 | child_id->stack[i] = parent_id->stack[i]; | |
4644 | child_id->stack[depth] = child_id->id; | |
4645 | /* | |
4646 | * child_id->css pointer will be set after this cgroup is available | |
4647 | * see cgroup_populate_dir() | |
4648 | */ | |
4649 | rcu_assign_pointer(child_css->id, child_id); | |
4650 | ||
4651 | return 0; | |
4652 | } | |
4653 | ||
4654 | /** | |
4655 | * css_lookup - lookup css by id | |
4656 | * @ss: cgroup subsys to be looked into. | |
4657 | * @id: the id | |
4658 | * | |
4659 | * Returns pointer to cgroup_subsys_state if there is valid one with id. | |
4660 | * NULL if not. Should be called under rcu_read_lock() | |
4661 | */ | |
4662 | struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id) | |
4663 | { | |
4664 | struct css_id *cssid = NULL; | |
4665 | ||
4666 | BUG_ON(!ss->use_id); | |
4667 | cssid = idr_find(&ss->idr, id); | |
4668 | ||
4669 | if (unlikely(!cssid)) | |
4670 | return NULL; | |
4671 | ||
4672 | return rcu_dereference(cssid->css); | |
4673 | } | |
4674 | EXPORT_SYMBOL_GPL(css_lookup); | |
4675 | ||
4676 | /** | |
4677 | * css_get_next - lookup next cgroup under specified hierarchy. | |
4678 | * @ss: pointer to subsystem | |
4679 | * @id: current position of iteration. | |
4680 | * @root: pointer to css. search tree under this. | |
4681 | * @foundid: position of found object. | |
4682 | * | |
4683 | * Search next css under the specified hierarchy of rootid. Calling under | |
4684 | * rcu_read_lock() is necessary. Returns NULL if it reaches the end. | |
4685 | */ | |
4686 | struct cgroup_subsys_state * | |
4687 | css_get_next(struct cgroup_subsys *ss, int id, | |
4688 | struct cgroup_subsys_state *root, int *foundid) | |
4689 | { | |
4690 | struct cgroup_subsys_state *ret = NULL; | |
4691 | struct css_id *tmp; | |
4692 | int tmpid; | |
4693 | int rootid = css_id(root); | |
4694 | int depth = css_depth(root); | |
4695 | ||
4696 | if (!rootid) | |
4697 | return NULL; | |
4698 | ||
4699 | BUG_ON(!ss->use_id); | |
4700 | /* fill start point for scan */ | |
4701 | tmpid = id; | |
4702 | while (1) { | |
4703 | /* | |
4704 | * scan next entry from bitmap(tree), tmpid is updated after | |
4705 | * idr_get_next(). | |
4706 | */ | |
4707 | spin_lock(&ss->id_lock); | |
4708 | tmp = idr_get_next(&ss->idr, &tmpid); | |
4709 | spin_unlock(&ss->id_lock); | |
4710 | ||
4711 | if (!tmp) | |
4712 | break; | |
4713 | if (tmp->depth >= depth && tmp->stack[depth] == rootid) { | |
4714 | ret = rcu_dereference(tmp->css); | |
4715 | if (ret) { | |
4716 | *foundid = tmpid; | |
4717 | break; | |
4718 | } | |
4719 | } | |
4720 | /* continue to scan from next id */ | |
4721 | tmpid = tmpid + 1; | |
4722 | } | |
4723 | return ret; | |
4724 | } | |
4725 | ||
4726 | #ifdef CONFIG_CGROUP_DEBUG | |
4727 | static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss, | |
4728 | struct cgroup *cont) | |
4729 | { | |
4730 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
4731 | ||
4732 | if (!css) | |
4733 | return ERR_PTR(-ENOMEM); | |
4734 | ||
4735 | return css; | |
4736 | } | |
4737 | ||
4738 | static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
4739 | { | |
4740 | kfree(cont->subsys[debug_subsys_id]); | |
4741 | } | |
4742 | ||
4743 | static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft) | |
4744 | { | |
4745 | return atomic_read(&cont->count); | |
4746 | } | |
4747 | ||
4748 | static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft) | |
4749 | { | |
4750 | return cgroup_task_count(cont); | |
4751 | } | |
4752 | ||
4753 | static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft) | |
4754 | { | |
4755 | return (u64)(unsigned long)current->cgroups; | |
4756 | } | |
4757 | ||
4758 | static u64 current_css_set_refcount_read(struct cgroup *cont, | |
4759 | struct cftype *cft) | |
4760 | { | |
4761 | u64 count; | |
4762 | ||
4763 | rcu_read_lock(); | |
4764 | count = atomic_read(¤t->cgroups->refcount); | |
4765 | rcu_read_unlock(); | |
4766 | return count; | |
4767 | } | |
4768 | ||
4769 | static int current_css_set_cg_links_read(struct cgroup *cont, | |
4770 | struct cftype *cft, | |
4771 | struct seq_file *seq) | |
4772 | { | |
4773 | struct cg_cgroup_link *link; | |
4774 | struct css_set *cg; | |
4775 | ||
4776 | read_lock(&css_set_lock); | |
4777 | rcu_read_lock(); | |
4778 | cg = rcu_dereference(current->cgroups); | |
4779 | list_for_each_entry(link, &cg->cg_links, cg_link_list) { | |
4780 | struct cgroup *c = link->cgrp; | |
4781 | const char *name; | |
4782 | ||
4783 | if (c->dentry) | |
4784 | name = c->dentry->d_name.name; | |
4785 | else | |
4786 | name = "?"; | |
4787 | seq_printf(seq, "Root %d group %s\n", | |
4788 | c->root->hierarchy_id, name); | |
4789 | } | |
4790 | rcu_read_unlock(); | |
4791 | read_unlock(&css_set_lock); | |
4792 | return 0; | |
4793 | } | |
4794 | ||
4795 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
4796 | static int cgroup_css_links_read(struct cgroup *cont, | |
4797 | struct cftype *cft, | |
4798 | struct seq_file *seq) | |
4799 | { | |
4800 | struct cg_cgroup_link *link; | |
4801 | ||
4802 | read_lock(&css_set_lock); | |
4803 | list_for_each_entry(link, &cont->css_sets, cgrp_link_list) { | |
4804 | struct css_set *cg = link->cg; | |
4805 | struct task_struct *task; | |
4806 | int count = 0; | |
4807 | seq_printf(seq, "css_set %p\n", cg); | |
4808 | list_for_each_entry(task, &cg->tasks, cg_list) { | |
4809 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { | |
4810 | seq_puts(seq, " ...\n"); | |
4811 | break; | |
4812 | } else { | |
4813 | seq_printf(seq, " task %d\n", | |
4814 | task_pid_vnr(task)); | |
4815 | } | |
4816 | } | |
4817 | } | |
4818 | read_unlock(&css_set_lock); | |
4819 | return 0; | |
4820 | } | |
4821 | ||
4822 | static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft) | |
4823 | { | |
4824 | return test_bit(CGRP_RELEASABLE, &cgrp->flags); | |
4825 | } | |
4826 | ||
4827 | static struct cftype debug_files[] = { | |
4828 | { | |
4829 | .name = "cgroup_refcount", | |
4830 | .read_u64 = cgroup_refcount_read, | |
4831 | }, | |
4832 | { | |
4833 | .name = "taskcount", | |
4834 | .read_u64 = debug_taskcount_read, | |
4835 | }, | |
4836 | ||
4837 | { | |
4838 | .name = "current_css_set", | |
4839 | .read_u64 = current_css_set_read, | |
4840 | }, | |
4841 | ||
4842 | { | |
4843 | .name = "current_css_set_refcount", | |
4844 | .read_u64 = current_css_set_refcount_read, | |
4845 | }, | |
4846 | ||
4847 | { | |
4848 | .name = "current_css_set_cg_links", | |
4849 | .read_seq_string = current_css_set_cg_links_read, | |
4850 | }, | |
4851 | ||
4852 | { | |
4853 | .name = "cgroup_css_links", | |
4854 | .read_seq_string = cgroup_css_links_read, | |
4855 | }, | |
4856 | ||
4857 | { | |
4858 | .name = "releasable", | |
4859 | .read_u64 = releasable_read, | |
4860 | }, | |
4861 | }; | |
4862 | ||
4863 | static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
4864 | { | |
4865 | return cgroup_add_files(cont, ss, debug_files, | |
4866 | ARRAY_SIZE(debug_files)); | |
4867 | } | |
4868 | ||
4869 | struct cgroup_subsys debug_subsys = { | |
4870 | .name = "debug", | |
4871 | .create = debug_create, | |
4872 | .destroy = debug_destroy, | |
4873 | .populate = debug_populate, | |
4874 | .subsys_id = debug_subsys_id, | |
4875 | }; | |
4876 | #endif /* CONFIG_CGROUP_DEBUG */ |