Commit | Line | Data |
---|---|---|
ddbcc7e8 | 1 | /* |
ddbcc7e8 PM |
2 | * Generic process-grouping system. |
3 | * | |
4 | * Based originally on the cpuset system, extracted by Paul Menage | |
5 | * Copyright (C) 2006 Google, Inc | |
6 | * | |
7 | * Copyright notices from the original cpuset code: | |
8 | * -------------------------------------------------- | |
9 | * Copyright (C) 2003 BULL SA. | |
10 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
11 | * | |
12 | * Portions derived from Patrick Mochel's sysfs code. | |
13 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
14 | * | |
15 | * 2003-10-10 Written by Simon Derr. | |
16 | * 2003-10-22 Updates by Stephen Hemminger. | |
17 | * 2004 May-July Rework by Paul Jackson. | |
18 | * --------------------------------------------------- | |
19 | * | |
20 | * This file is subject to the terms and conditions of the GNU General Public | |
21 | * License. See the file COPYING in the main directory of the Linux | |
22 | * distribution for more details. | |
23 | */ | |
24 | ||
25 | #include <linux/cgroup.h> | |
c50cc752 | 26 | #include <linux/module.h> |
c6d57f33 | 27 | #include <linux/ctype.h> |
ddbcc7e8 PM |
28 | #include <linux/errno.h> |
29 | #include <linux/fs.h> | |
30 | #include <linux/kernel.h> | |
31 | #include <linux/list.h> | |
32 | #include <linux/mm.h> | |
33 | #include <linux/mutex.h> | |
34 | #include <linux/mount.h> | |
35 | #include <linux/pagemap.h> | |
a424316c | 36 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
37 | #include <linux/rcupdate.h> |
38 | #include <linux/sched.h> | |
817929ec | 39 | #include <linux/backing-dev.h> |
ddbcc7e8 PM |
40 | #include <linux/seq_file.h> |
41 | #include <linux/slab.h> | |
42 | #include <linux/magic.h> | |
43 | #include <linux/spinlock.h> | |
44 | #include <linux/string.h> | |
bbcb81d0 | 45 | #include <linux/sort.h> |
81a6a5cd | 46 | #include <linux/kmod.h> |
e6a1105b | 47 | #include <linux/module.h> |
846c7bb0 BS |
48 | #include <linux/delayacct.h> |
49 | #include <linux/cgroupstats.h> | |
472b1053 | 50 | #include <linux/hash.h> |
3f8206d4 | 51 | #include <linux/namei.h> |
337eb00a | 52 | #include <linux/smp_lock.h> |
096b7fe0 | 53 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 54 | #include <linux/idr.h> |
d1d9fd33 | 55 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
846c7bb0 | 56 | |
ddbcc7e8 PM |
57 | #include <asm/atomic.h> |
58 | ||
81a6a5cd PM |
59 | static DEFINE_MUTEX(cgroup_mutex); |
60 | ||
aae8aab4 BB |
61 | /* |
62 | * Generate an array of cgroup subsystem pointers. At boot time, this is | |
63 | * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are | |
64 | * registered after that. The mutable section of this array is protected by | |
65 | * cgroup_mutex. | |
66 | */ | |
ddbcc7e8 | 67 | #define SUBSYS(_x) &_x ## _subsys, |
aae8aab4 | 68 | static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = { |
ddbcc7e8 PM |
69 | #include <linux/cgroup_subsys.h> |
70 | }; | |
71 | ||
c6d57f33 PM |
72 | #define MAX_CGROUP_ROOT_NAMELEN 64 |
73 | ||
ddbcc7e8 PM |
74 | /* |
75 | * A cgroupfs_root represents the root of a cgroup hierarchy, | |
76 | * and may be associated with a superblock to form an active | |
77 | * hierarchy | |
78 | */ | |
79 | struct cgroupfs_root { | |
80 | struct super_block *sb; | |
81 | ||
82 | /* | |
83 | * The bitmask of subsystems intended to be attached to this | |
84 | * hierarchy | |
85 | */ | |
86 | unsigned long subsys_bits; | |
87 | ||
2c6ab6d2 PM |
88 | /* Unique id for this hierarchy. */ |
89 | int hierarchy_id; | |
90 | ||
ddbcc7e8 PM |
91 | /* The bitmask of subsystems currently attached to this hierarchy */ |
92 | unsigned long actual_subsys_bits; | |
93 | ||
94 | /* A list running through the attached subsystems */ | |
95 | struct list_head subsys_list; | |
96 | ||
97 | /* The root cgroup for this hierarchy */ | |
98 | struct cgroup top_cgroup; | |
99 | ||
100 | /* Tracks how many cgroups are currently defined in hierarchy.*/ | |
101 | int number_of_cgroups; | |
102 | ||
e5f6a860 | 103 | /* A list running through the active hierarchies */ |
ddbcc7e8 PM |
104 | struct list_head root_list; |
105 | ||
106 | /* Hierarchy-specific flags */ | |
107 | unsigned long flags; | |
81a6a5cd | 108 | |
e788e066 | 109 | /* The path to use for release notifications. */ |
81a6a5cd | 110 | char release_agent_path[PATH_MAX]; |
c6d57f33 PM |
111 | |
112 | /* The name for this hierarchy - may be empty */ | |
113 | char name[MAX_CGROUP_ROOT_NAMELEN]; | |
ddbcc7e8 PM |
114 | }; |
115 | ||
ddbcc7e8 PM |
116 | /* |
117 | * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the | |
118 | * subsystems that are otherwise unattached - it never has more than a | |
119 | * single cgroup, and all tasks are part of that cgroup. | |
120 | */ | |
121 | static struct cgroupfs_root rootnode; | |
122 | ||
38460b48 KH |
123 | /* |
124 | * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when | |
125 | * cgroup_subsys->use_id != 0. | |
126 | */ | |
127 | #define CSS_ID_MAX (65535) | |
128 | struct css_id { | |
129 | /* | |
130 | * The css to which this ID points. This pointer is set to valid value | |
131 | * after cgroup is populated. If cgroup is removed, this will be NULL. | |
132 | * This pointer is expected to be RCU-safe because destroy() | |
133 | * is called after synchronize_rcu(). But for safe use, css_is_removed() | |
134 | * css_tryget() should be used for avoiding race. | |
135 | */ | |
136 | struct cgroup_subsys_state *css; | |
137 | /* | |
138 | * ID of this css. | |
139 | */ | |
140 | unsigned short id; | |
141 | /* | |
142 | * Depth in hierarchy which this ID belongs to. | |
143 | */ | |
144 | unsigned short depth; | |
145 | /* | |
146 | * ID is freed by RCU. (and lookup routine is RCU safe.) | |
147 | */ | |
148 | struct rcu_head rcu_head; | |
149 | /* | |
150 | * Hierarchy of CSS ID belongs to. | |
151 | */ | |
152 | unsigned short stack[0]; /* Array of Length (depth+1) */ | |
153 | }; | |
154 | ||
155 | ||
ddbcc7e8 PM |
156 | /* The list of hierarchy roots */ |
157 | ||
158 | static LIST_HEAD(roots); | |
817929ec | 159 | static int root_count; |
ddbcc7e8 | 160 | |
2c6ab6d2 PM |
161 | static DEFINE_IDA(hierarchy_ida); |
162 | static int next_hierarchy_id; | |
163 | static DEFINE_SPINLOCK(hierarchy_id_lock); | |
164 | ||
ddbcc7e8 PM |
165 | /* dummytop is a shorthand for the dummy hierarchy's top cgroup */ |
166 | #define dummytop (&rootnode.top_cgroup) | |
167 | ||
168 | /* This flag indicates whether tasks in the fork and exit paths should | |
a043e3b2 LZ |
169 | * check for fork/exit handlers to call. This avoids us having to do |
170 | * extra work in the fork/exit path if none of the subsystems need to | |
171 | * be called. | |
ddbcc7e8 | 172 | */ |
8947f9d5 | 173 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 174 | |
d11c563d PM |
175 | #ifdef CONFIG_PROVE_LOCKING |
176 | int cgroup_lock_is_held(void) | |
177 | { | |
178 | return lockdep_is_held(&cgroup_mutex); | |
179 | } | |
180 | #else /* #ifdef CONFIG_PROVE_LOCKING */ | |
181 | int cgroup_lock_is_held(void) | |
182 | { | |
183 | return mutex_is_locked(&cgroup_mutex); | |
184 | } | |
185 | #endif /* #else #ifdef CONFIG_PROVE_LOCKING */ | |
186 | ||
187 | EXPORT_SYMBOL_GPL(cgroup_lock_is_held); | |
188 | ||
ddbcc7e8 | 189 | /* convenient tests for these bits */ |
bd89aabc | 190 | inline int cgroup_is_removed(const struct cgroup *cgrp) |
ddbcc7e8 | 191 | { |
bd89aabc | 192 | return test_bit(CGRP_REMOVED, &cgrp->flags); |
ddbcc7e8 PM |
193 | } |
194 | ||
195 | /* bits in struct cgroupfs_root flags field */ | |
196 | enum { | |
197 | ROOT_NOPREFIX, /* mounted subsystems have no named prefix */ | |
198 | }; | |
199 | ||
e9685a03 | 200 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
201 | { |
202 | const int bits = | |
bd89aabc PM |
203 | (1 << CGRP_RELEASABLE) | |
204 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
205 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
206 | } |
207 | ||
e9685a03 | 208 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 209 | { |
bd89aabc | 210 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
211 | } |
212 | ||
ddbcc7e8 PM |
213 | /* |
214 | * for_each_subsys() allows you to iterate on each subsystem attached to | |
215 | * an active hierarchy | |
216 | */ | |
217 | #define for_each_subsys(_root, _ss) \ | |
218 | list_for_each_entry(_ss, &_root->subsys_list, sibling) | |
219 | ||
e5f6a860 LZ |
220 | /* for_each_active_root() allows you to iterate across the active hierarchies */ |
221 | #define for_each_active_root(_root) \ | |
ddbcc7e8 PM |
222 | list_for_each_entry(_root, &roots, root_list) |
223 | ||
81a6a5cd PM |
224 | /* the list of cgroups eligible for automatic release. Protected by |
225 | * release_list_lock */ | |
226 | static LIST_HEAD(release_list); | |
227 | static DEFINE_SPINLOCK(release_list_lock); | |
228 | static void cgroup_release_agent(struct work_struct *work); | |
229 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 230 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 231 | |
817929ec PM |
232 | /* Link structure for associating css_set objects with cgroups */ |
233 | struct cg_cgroup_link { | |
234 | /* | |
235 | * List running through cg_cgroup_links associated with a | |
236 | * cgroup, anchored on cgroup->css_sets | |
237 | */ | |
bd89aabc | 238 | struct list_head cgrp_link_list; |
7717f7ba | 239 | struct cgroup *cgrp; |
817929ec PM |
240 | /* |
241 | * List running through cg_cgroup_links pointing at a | |
242 | * single css_set object, anchored on css_set->cg_links | |
243 | */ | |
244 | struct list_head cg_link_list; | |
245 | struct css_set *cg; | |
246 | }; | |
247 | ||
248 | /* The default css_set - used by init and its children prior to any | |
249 | * hierarchies being mounted. It contains a pointer to the root state | |
250 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
251 | * reference-counted, to improve performance when child cgroups | |
252 | * haven't been created. | |
253 | */ | |
254 | ||
255 | static struct css_set init_css_set; | |
256 | static struct cg_cgroup_link init_css_set_link; | |
257 | ||
e6a1105b BB |
258 | static int cgroup_init_idr(struct cgroup_subsys *ss, |
259 | struct cgroup_subsys_state *css); | |
38460b48 | 260 | |
817929ec PM |
261 | /* css_set_lock protects the list of css_set objects, and the |
262 | * chain of tasks off each css_set. Nests outside task->alloc_lock | |
263 | * due to cgroup_iter_start() */ | |
264 | static DEFINE_RWLOCK(css_set_lock); | |
265 | static int css_set_count; | |
266 | ||
7717f7ba PM |
267 | /* |
268 | * hash table for cgroup groups. This improves the performance to find | |
269 | * an existing css_set. This hash doesn't (currently) take into | |
270 | * account cgroups in empty hierarchies. | |
271 | */ | |
472b1053 LZ |
272 | #define CSS_SET_HASH_BITS 7 |
273 | #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS) | |
274 | static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE]; | |
275 | ||
276 | static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[]) | |
277 | { | |
278 | int i; | |
279 | int index; | |
280 | unsigned long tmp = 0UL; | |
281 | ||
282 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) | |
283 | tmp += (unsigned long)css[i]; | |
284 | tmp = (tmp >> 16) ^ tmp; | |
285 | ||
286 | index = hash_long(tmp, CSS_SET_HASH_BITS); | |
287 | ||
288 | return &css_set_table[index]; | |
289 | } | |
290 | ||
c378369d BB |
291 | static void free_css_set_rcu(struct rcu_head *obj) |
292 | { | |
293 | struct css_set *cg = container_of(obj, struct css_set, rcu_head); | |
294 | kfree(cg); | |
295 | } | |
296 | ||
817929ec PM |
297 | /* We don't maintain the lists running through each css_set to its |
298 | * task until after the first call to cgroup_iter_start(). This | |
299 | * reduces the fork()/exit() overhead for people who have cgroups | |
300 | * compiled into their kernel but not actually in use */ | |
8947f9d5 | 301 | static int use_task_css_set_links __read_mostly; |
817929ec | 302 | |
2c6ab6d2 | 303 | static void __put_css_set(struct css_set *cg, int taskexit) |
b4f48b63 | 304 | { |
71cbb949 KM |
305 | struct cg_cgroup_link *link; |
306 | struct cg_cgroup_link *saved_link; | |
146aa1bd LJ |
307 | /* |
308 | * Ensure that the refcount doesn't hit zero while any readers | |
309 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
310 | * rwlock | |
311 | */ | |
312 | if (atomic_add_unless(&cg->refcount, -1, 1)) | |
313 | return; | |
314 | write_lock(&css_set_lock); | |
315 | if (!atomic_dec_and_test(&cg->refcount)) { | |
316 | write_unlock(&css_set_lock); | |
317 | return; | |
318 | } | |
81a6a5cd | 319 | |
2c6ab6d2 PM |
320 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
321 | hlist_del(&cg->hlist); | |
322 | css_set_count--; | |
323 | ||
324 | list_for_each_entry_safe(link, saved_link, &cg->cg_links, | |
325 | cg_link_list) { | |
326 | struct cgroup *cgrp = link->cgrp; | |
327 | list_del(&link->cg_link_list); | |
328 | list_del(&link->cgrp_link_list); | |
bd89aabc PM |
329 | if (atomic_dec_and_test(&cgrp->count) && |
330 | notify_on_release(cgrp)) { | |
81a6a5cd | 331 | if (taskexit) |
bd89aabc PM |
332 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
333 | check_for_release(cgrp); | |
81a6a5cd | 334 | } |
2c6ab6d2 PM |
335 | |
336 | kfree(link); | |
81a6a5cd | 337 | } |
2c6ab6d2 PM |
338 | |
339 | write_unlock(&css_set_lock); | |
c378369d | 340 | call_rcu(&cg->rcu_head, free_css_set_rcu); |
b4f48b63 PM |
341 | } |
342 | ||
817929ec PM |
343 | /* |
344 | * refcounted get/put for css_set objects | |
345 | */ | |
346 | static inline void get_css_set(struct css_set *cg) | |
347 | { | |
146aa1bd | 348 | atomic_inc(&cg->refcount); |
817929ec PM |
349 | } |
350 | ||
351 | static inline void put_css_set(struct css_set *cg) | |
352 | { | |
146aa1bd | 353 | __put_css_set(cg, 0); |
817929ec PM |
354 | } |
355 | ||
81a6a5cd PM |
356 | static inline void put_css_set_taskexit(struct css_set *cg) |
357 | { | |
146aa1bd | 358 | __put_css_set(cg, 1); |
81a6a5cd PM |
359 | } |
360 | ||
7717f7ba PM |
361 | /* |
362 | * compare_css_sets - helper function for find_existing_css_set(). | |
363 | * @cg: candidate css_set being tested | |
364 | * @old_cg: existing css_set for a task | |
365 | * @new_cgrp: cgroup that's being entered by the task | |
366 | * @template: desired set of css pointers in css_set (pre-calculated) | |
367 | * | |
368 | * Returns true if "cg" matches "old_cg" except for the hierarchy | |
369 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". | |
370 | */ | |
371 | static bool compare_css_sets(struct css_set *cg, | |
372 | struct css_set *old_cg, | |
373 | struct cgroup *new_cgrp, | |
374 | struct cgroup_subsys_state *template[]) | |
375 | { | |
376 | struct list_head *l1, *l2; | |
377 | ||
378 | if (memcmp(template, cg->subsys, sizeof(cg->subsys))) { | |
379 | /* Not all subsystems matched */ | |
380 | return false; | |
381 | } | |
382 | ||
383 | /* | |
384 | * Compare cgroup pointers in order to distinguish between | |
385 | * different cgroups in heirarchies with no subsystems. We | |
386 | * could get by with just this check alone (and skip the | |
387 | * memcmp above) but on most setups the memcmp check will | |
388 | * avoid the need for this more expensive check on almost all | |
389 | * candidates. | |
390 | */ | |
391 | ||
392 | l1 = &cg->cg_links; | |
393 | l2 = &old_cg->cg_links; | |
394 | while (1) { | |
395 | struct cg_cgroup_link *cgl1, *cgl2; | |
396 | struct cgroup *cg1, *cg2; | |
397 | ||
398 | l1 = l1->next; | |
399 | l2 = l2->next; | |
400 | /* See if we reached the end - both lists are equal length. */ | |
401 | if (l1 == &cg->cg_links) { | |
402 | BUG_ON(l2 != &old_cg->cg_links); | |
403 | break; | |
404 | } else { | |
405 | BUG_ON(l2 == &old_cg->cg_links); | |
406 | } | |
407 | /* Locate the cgroups associated with these links. */ | |
408 | cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list); | |
409 | cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list); | |
410 | cg1 = cgl1->cgrp; | |
411 | cg2 = cgl2->cgrp; | |
412 | /* Hierarchies should be linked in the same order. */ | |
413 | BUG_ON(cg1->root != cg2->root); | |
414 | ||
415 | /* | |
416 | * If this hierarchy is the hierarchy of the cgroup | |
417 | * that's changing, then we need to check that this | |
418 | * css_set points to the new cgroup; if it's any other | |
419 | * hierarchy, then this css_set should point to the | |
420 | * same cgroup as the old css_set. | |
421 | */ | |
422 | if (cg1->root == new_cgrp->root) { | |
423 | if (cg1 != new_cgrp) | |
424 | return false; | |
425 | } else { | |
426 | if (cg1 != cg2) | |
427 | return false; | |
428 | } | |
429 | } | |
430 | return true; | |
431 | } | |
432 | ||
817929ec PM |
433 | /* |
434 | * find_existing_css_set() is a helper for | |
435 | * find_css_set(), and checks to see whether an existing | |
472b1053 | 436 | * css_set is suitable. |
817929ec PM |
437 | * |
438 | * oldcg: the cgroup group that we're using before the cgroup | |
439 | * transition | |
440 | * | |
bd89aabc | 441 | * cgrp: the cgroup that we're moving into |
817929ec PM |
442 | * |
443 | * template: location in which to build the desired set of subsystem | |
444 | * state objects for the new cgroup group | |
445 | */ | |
817929ec PM |
446 | static struct css_set *find_existing_css_set( |
447 | struct css_set *oldcg, | |
bd89aabc | 448 | struct cgroup *cgrp, |
817929ec | 449 | struct cgroup_subsys_state *template[]) |
b4f48b63 PM |
450 | { |
451 | int i; | |
bd89aabc | 452 | struct cgroupfs_root *root = cgrp->root; |
472b1053 LZ |
453 | struct hlist_head *hhead; |
454 | struct hlist_node *node; | |
455 | struct css_set *cg; | |
817929ec | 456 | |
aae8aab4 BB |
457 | /* |
458 | * Build the set of subsystem state objects that we want to see in the | |
459 | * new css_set. while subsystems can change globally, the entries here | |
460 | * won't change, so no need for locking. | |
461 | */ | |
817929ec | 462 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
8d53d55d | 463 | if (root->subsys_bits & (1UL << i)) { |
817929ec PM |
464 | /* Subsystem is in this hierarchy. So we want |
465 | * the subsystem state from the new | |
466 | * cgroup */ | |
bd89aabc | 467 | template[i] = cgrp->subsys[i]; |
817929ec PM |
468 | } else { |
469 | /* Subsystem is not in this hierarchy, so we | |
470 | * don't want to change the subsystem state */ | |
471 | template[i] = oldcg->subsys[i]; | |
472 | } | |
473 | } | |
474 | ||
472b1053 LZ |
475 | hhead = css_set_hash(template); |
476 | hlist_for_each_entry(cg, node, hhead, hlist) { | |
7717f7ba PM |
477 | if (!compare_css_sets(cg, oldcg, cgrp, template)) |
478 | continue; | |
479 | ||
480 | /* This css_set matches what we need */ | |
481 | return cg; | |
472b1053 | 482 | } |
817929ec PM |
483 | |
484 | /* No existing cgroup group matched */ | |
485 | return NULL; | |
486 | } | |
487 | ||
36553434 LZ |
488 | static void free_cg_links(struct list_head *tmp) |
489 | { | |
490 | struct cg_cgroup_link *link; | |
491 | struct cg_cgroup_link *saved_link; | |
492 | ||
493 | list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) { | |
494 | list_del(&link->cgrp_link_list); | |
495 | kfree(link); | |
496 | } | |
497 | } | |
498 | ||
817929ec PM |
499 | /* |
500 | * allocate_cg_links() allocates "count" cg_cgroup_link structures | |
bd89aabc | 501 | * and chains them on tmp through their cgrp_link_list fields. Returns 0 on |
817929ec PM |
502 | * success or a negative error |
503 | */ | |
817929ec PM |
504 | static int allocate_cg_links(int count, struct list_head *tmp) |
505 | { | |
506 | struct cg_cgroup_link *link; | |
507 | int i; | |
508 | INIT_LIST_HEAD(tmp); | |
509 | for (i = 0; i < count; i++) { | |
510 | link = kmalloc(sizeof(*link), GFP_KERNEL); | |
511 | if (!link) { | |
36553434 | 512 | free_cg_links(tmp); |
817929ec PM |
513 | return -ENOMEM; |
514 | } | |
bd89aabc | 515 | list_add(&link->cgrp_link_list, tmp); |
817929ec PM |
516 | } |
517 | return 0; | |
518 | } | |
519 | ||
c12f65d4 LZ |
520 | /** |
521 | * link_css_set - a helper function to link a css_set to a cgroup | |
522 | * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links() | |
523 | * @cg: the css_set to be linked | |
524 | * @cgrp: the destination cgroup | |
525 | */ | |
526 | static void link_css_set(struct list_head *tmp_cg_links, | |
527 | struct css_set *cg, struct cgroup *cgrp) | |
528 | { | |
529 | struct cg_cgroup_link *link; | |
530 | ||
531 | BUG_ON(list_empty(tmp_cg_links)); | |
532 | link = list_first_entry(tmp_cg_links, struct cg_cgroup_link, | |
533 | cgrp_link_list); | |
534 | link->cg = cg; | |
7717f7ba | 535 | link->cgrp = cgrp; |
2c6ab6d2 | 536 | atomic_inc(&cgrp->count); |
c12f65d4 | 537 | list_move(&link->cgrp_link_list, &cgrp->css_sets); |
7717f7ba PM |
538 | /* |
539 | * Always add links to the tail of the list so that the list | |
540 | * is sorted by order of hierarchy creation | |
541 | */ | |
542 | list_add_tail(&link->cg_link_list, &cg->cg_links); | |
c12f65d4 LZ |
543 | } |
544 | ||
817929ec PM |
545 | /* |
546 | * find_css_set() takes an existing cgroup group and a | |
547 | * cgroup object, and returns a css_set object that's | |
548 | * equivalent to the old group, but with the given cgroup | |
549 | * substituted into the appropriate hierarchy. Must be called with | |
550 | * cgroup_mutex held | |
551 | */ | |
817929ec | 552 | static struct css_set *find_css_set( |
bd89aabc | 553 | struct css_set *oldcg, struct cgroup *cgrp) |
817929ec PM |
554 | { |
555 | struct css_set *res; | |
556 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT]; | |
817929ec PM |
557 | |
558 | struct list_head tmp_cg_links; | |
817929ec | 559 | |
472b1053 | 560 | struct hlist_head *hhead; |
7717f7ba | 561 | struct cg_cgroup_link *link; |
472b1053 | 562 | |
817929ec PM |
563 | /* First see if we already have a cgroup group that matches |
564 | * the desired set */ | |
7e9abd89 | 565 | read_lock(&css_set_lock); |
bd89aabc | 566 | res = find_existing_css_set(oldcg, cgrp, template); |
817929ec PM |
567 | if (res) |
568 | get_css_set(res); | |
7e9abd89 | 569 | read_unlock(&css_set_lock); |
817929ec PM |
570 | |
571 | if (res) | |
572 | return res; | |
573 | ||
574 | res = kmalloc(sizeof(*res), GFP_KERNEL); | |
575 | if (!res) | |
576 | return NULL; | |
577 | ||
578 | /* Allocate all the cg_cgroup_link objects that we'll need */ | |
579 | if (allocate_cg_links(root_count, &tmp_cg_links) < 0) { | |
580 | kfree(res); | |
581 | return NULL; | |
582 | } | |
583 | ||
146aa1bd | 584 | atomic_set(&res->refcount, 1); |
817929ec PM |
585 | INIT_LIST_HEAD(&res->cg_links); |
586 | INIT_LIST_HEAD(&res->tasks); | |
472b1053 | 587 | INIT_HLIST_NODE(&res->hlist); |
817929ec PM |
588 | |
589 | /* Copy the set of subsystem state objects generated in | |
590 | * find_existing_css_set() */ | |
591 | memcpy(res->subsys, template, sizeof(res->subsys)); | |
592 | ||
593 | write_lock(&css_set_lock); | |
594 | /* Add reference counts and links from the new css_set. */ | |
7717f7ba PM |
595 | list_for_each_entry(link, &oldcg->cg_links, cg_link_list) { |
596 | struct cgroup *c = link->cgrp; | |
597 | if (c->root == cgrp->root) | |
598 | c = cgrp; | |
599 | link_css_set(&tmp_cg_links, res, c); | |
600 | } | |
817929ec PM |
601 | |
602 | BUG_ON(!list_empty(&tmp_cg_links)); | |
603 | ||
817929ec | 604 | css_set_count++; |
472b1053 LZ |
605 | |
606 | /* Add this cgroup group to the hash table */ | |
607 | hhead = css_set_hash(res->subsys); | |
608 | hlist_add_head(&res->hlist, hhead); | |
609 | ||
817929ec PM |
610 | write_unlock(&css_set_lock); |
611 | ||
612 | return res; | |
b4f48b63 PM |
613 | } |
614 | ||
7717f7ba PM |
615 | /* |
616 | * Return the cgroup for "task" from the given hierarchy. Must be | |
617 | * called with cgroup_mutex held. | |
618 | */ | |
619 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
620 | struct cgroupfs_root *root) | |
621 | { | |
622 | struct css_set *css; | |
623 | struct cgroup *res = NULL; | |
624 | ||
625 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
626 | read_lock(&css_set_lock); | |
627 | /* | |
628 | * No need to lock the task - since we hold cgroup_mutex the | |
629 | * task can't change groups, so the only thing that can happen | |
630 | * is that it exits and its css is set back to init_css_set. | |
631 | */ | |
632 | css = task->cgroups; | |
633 | if (css == &init_css_set) { | |
634 | res = &root->top_cgroup; | |
635 | } else { | |
636 | struct cg_cgroup_link *link; | |
637 | list_for_each_entry(link, &css->cg_links, cg_link_list) { | |
638 | struct cgroup *c = link->cgrp; | |
639 | if (c->root == root) { | |
640 | res = c; | |
641 | break; | |
642 | } | |
643 | } | |
644 | } | |
645 | read_unlock(&css_set_lock); | |
646 | BUG_ON(!res); | |
647 | return res; | |
648 | } | |
649 | ||
ddbcc7e8 PM |
650 | /* |
651 | * There is one global cgroup mutex. We also require taking | |
652 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
653 | * See "The task_lock() exception", at the end of this comment. | |
654 | * | |
655 | * A task must hold cgroup_mutex to modify cgroups. | |
656 | * | |
657 | * Any task can increment and decrement the count field without lock. | |
658 | * So in general, code holding cgroup_mutex can't rely on the count | |
659 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 660 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
661 | * means that no tasks are currently attached, therefore there is no |
662 | * way a task attached to that cgroup can fork (the other way to | |
663 | * increment the count). So code holding cgroup_mutex can safely | |
664 | * assume that if the count is zero, it will stay zero. Similarly, if | |
665 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
666 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
667 | * needs that mutex. | |
668 | * | |
ddbcc7e8 PM |
669 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
670 | * (usually) take cgroup_mutex. These are the two most performance | |
671 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
672 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
673 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
674 | * to the release agent with the name of the cgroup (path relative to |
675 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
676 | * |
677 | * A cgroup can only be deleted if both its 'count' of using tasks | |
678 | * is zero, and its list of 'children' cgroups is empty. Since all | |
679 | * tasks in the system use _some_ cgroup, and since there is always at | |
680 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
681 | * always has either children cgroups and/or using tasks. So we don't | |
682 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
683 | * | |
684 | * The task_lock() exception | |
685 | * | |
686 | * The need for this exception arises from the action of | |
956db3ca | 687 | * cgroup_attach_task(), which overwrites one tasks cgroup pointer with |
a043e3b2 | 688 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
689 | * several performance critical places that need to reference |
690 | * task->cgroup without the expense of grabbing a system global | |
691 | * mutex. Therefore except as noted below, when dereferencing or, as | |
956db3ca | 692 | * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use |
ddbcc7e8 PM |
693 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
694 | * the task_struct routinely used for such matters. | |
695 | * | |
696 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 697 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
698 | */ |
699 | ||
ddbcc7e8 PM |
700 | /** |
701 | * cgroup_lock - lock out any changes to cgroup structures | |
702 | * | |
703 | */ | |
ddbcc7e8 PM |
704 | void cgroup_lock(void) |
705 | { | |
706 | mutex_lock(&cgroup_mutex); | |
707 | } | |
708 | ||
709 | /** | |
710 | * cgroup_unlock - release lock on cgroup changes | |
711 | * | |
712 | * Undo the lock taken in a previous cgroup_lock() call. | |
713 | */ | |
ddbcc7e8 PM |
714 | void cgroup_unlock(void) |
715 | { | |
716 | mutex_unlock(&cgroup_mutex); | |
717 | } | |
718 | ||
719 | /* | |
720 | * A couple of forward declarations required, due to cyclic reference loop: | |
721 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> | |
722 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations | |
723 | * -> cgroup_mkdir. | |
724 | */ | |
725 | ||
726 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode); | |
727 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); | |
bd89aabc | 728 | static int cgroup_populate_dir(struct cgroup *cgrp); |
6e1d5dcc | 729 | static const struct inode_operations cgroup_dir_inode_operations; |
828c0950 | 730 | static const struct file_operations proc_cgroupstats_operations; |
a424316c PM |
731 | |
732 | static struct backing_dev_info cgroup_backing_dev_info = { | |
d993831f | 733 | .name = "cgroup", |
e4ad08fe | 734 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
a424316c | 735 | }; |
ddbcc7e8 | 736 | |
38460b48 KH |
737 | static int alloc_css_id(struct cgroup_subsys *ss, |
738 | struct cgroup *parent, struct cgroup *child); | |
739 | ||
ddbcc7e8 PM |
740 | static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb) |
741 | { | |
742 | struct inode *inode = new_inode(sb); | |
ddbcc7e8 PM |
743 | |
744 | if (inode) { | |
745 | inode->i_mode = mode; | |
76aac0e9 DH |
746 | inode->i_uid = current_fsuid(); |
747 | inode->i_gid = current_fsgid(); | |
ddbcc7e8 PM |
748 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
749 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; | |
750 | } | |
751 | return inode; | |
752 | } | |
753 | ||
4fca88c8 KH |
754 | /* |
755 | * Call subsys's pre_destroy handler. | |
756 | * This is called before css refcnt check. | |
757 | */ | |
ec64f515 | 758 | static int cgroup_call_pre_destroy(struct cgroup *cgrp) |
4fca88c8 KH |
759 | { |
760 | struct cgroup_subsys *ss; | |
ec64f515 KH |
761 | int ret = 0; |
762 | ||
4fca88c8 | 763 | for_each_subsys(cgrp->root, ss) |
ec64f515 KH |
764 | if (ss->pre_destroy) { |
765 | ret = ss->pre_destroy(ss, cgrp); | |
766 | if (ret) | |
767 | break; | |
768 | } | |
769 | return ret; | |
4fca88c8 KH |
770 | } |
771 | ||
a47295e6 PM |
772 | static void free_cgroup_rcu(struct rcu_head *obj) |
773 | { | |
774 | struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head); | |
775 | ||
776 | kfree(cgrp); | |
777 | } | |
778 | ||
ddbcc7e8 PM |
779 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) |
780 | { | |
781 | /* is dentry a directory ? if so, kfree() associated cgroup */ | |
782 | if (S_ISDIR(inode->i_mode)) { | |
bd89aabc | 783 | struct cgroup *cgrp = dentry->d_fsdata; |
8dc4f3e1 | 784 | struct cgroup_subsys *ss; |
bd89aabc | 785 | BUG_ON(!(cgroup_is_removed(cgrp))); |
81a6a5cd PM |
786 | /* It's possible for external users to be holding css |
787 | * reference counts on a cgroup; css_put() needs to | |
788 | * be able to access the cgroup after decrementing | |
789 | * the reference count in order to know if it needs to | |
790 | * queue the cgroup to be handled by the release | |
791 | * agent */ | |
792 | synchronize_rcu(); | |
8dc4f3e1 PM |
793 | |
794 | mutex_lock(&cgroup_mutex); | |
795 | /* | |
796 | * Release the subsystem state objects. | |
797 | */ | |
75139b82 LZ |
798 | for_each_subsys(cgrp->root, ss) |
799 | ss->destroy(ss, cgrp); | |
8dc4f3e1 PM |
800 | |
801 | cgrp->root->number_of_cgroups--; | |
802 | mutex_unlock(&cgroup_mutex); | |
803 | ||
a47295e6 PM |
804 | /* |
805 | * Drop the active superblock reference that we took when we | |
806 | * created the cgroup | |
807 | */ | |
8dc4f3e1 PM |
808 | deactivate_super(cgrp->root->sb); |
809 | ||
72a8cb30 BB |
810 | /* |
811 | * if we're getting rid of the cgroup, refcount should ensure | |
812 | * that there are no pidlists left. | |
813 | */ | |
814 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
815 | ||
a47295e6 | 816 | call_rcu(&cgrp->rcu_head, free_cgroup_rcu); |
ddbcc7e8 PM |
817 | } |
818 | iput(inode); | |
819 | } | |
820 | ||
821 | static void remove_dir(struct dentry *d) | |
822 | { | |
823 | struct dentry *parent = dget(d->d_parent); | |
824 | ||
825 | d_delete(d); | |
826 | simple_rmdir(parent->d_inode, d); | |
827 | dput(parent); | |
828 | } | |
829 | ||
830 | static void cgroup_clear_directory(struct dentry *dentry) | |
831 | { | |
832 | struct list_head *node; | |
833 | ||
834 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
835 | spin_lock(&dcache_lock); | |
836 | node = dentry->d_subdirs.next; | |
837 | while (node != &dentry->d_subdirs) { | |
838 | struct dentry *d = list_entry(node, struct dentry, d_u.d_child); | |
839 | list_del_init(node); | |
840 | if (d->d_inode) { | |
841 | /* This should never be called on a cgroup | |
842 | * directory with child cgroups */ | |
843 | BUG_ON(d->d_inode->i_mode & S_IFDIR); | |
844 | d = dget_locked(d); | |
845 | spin_unlock(&dcache_lock); | |
846 | d_delete(d); | |
847 | simple_unlink(dentry->d_inode, d); | |
848 | dput(d); | |
849 | spin_lock(&dcache_lock); | |
850 | } | |
851 | node = dentry->d_subdirs.next; | |
852 | } | |
853 | spin_unlock(&dcache_lock); | |
854 | } | |
855 | ||
856 | /* | |
857 | * NOTE : the dentry must have been dget()'ed | |
858 | */ | |
859 | static void cgroup_d_remove_dir(struct dentry *dentry) | |
860 | { | |
861 | cgroup_clear_directory(dentry); | |
862 | ||
863 | spin_lock(&dcache_lock); | |
864 | list_del_init(&dentry->d_u.d_child); | |
865 | spin_unlock(&dcache_lock); | |
866 | remove_dir(dentry); | |
867 | } | |
868 | ||
ec64f515 KH |
869 | /* |
870 | * A queue for waiters to do rmdir() cgroup. A tasks will sleep when | |
871 | * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some | |
872 | * reference to css->refcnt. In general, this refcnt is expected to goes down | |
873 | * to zero, soon. | |
874 | * | |
88703267 | 875 | * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex; |
ec64f515 KH |
876 | */ |
877 | DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq); | |
878 | ||
88703267 | 879 | static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp) |
ec64f515 | 880 | { |
88703267 | 881 | if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))) |
ec64f515 KH |
882 | wake_up_all(&cgroup_rmdir_waitq); |
883 | } | |
884 | ||
88703267 KH |
885 | void cgroup_exclude_rmdir(struct cgroup_subsys_state *css) |
886 | { | |
887 | css_get(css); | |
888 | } | |
889 | ||
890 | void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css) | |
891 | { | |
892 | cgroup_wakeup_rmdir_waiter(css->cgroup); | |
893 | css_put(css); | |
894 | } | |
895 | ||
aae8aab4 BB |
896 | /* |
897 | * Call with cgroup_mutex held. | |
898 | */ | |
ddbcc7e8 PM |
899 | static int rebind_subsystems(struct cgroupfs_root *root, |
900 | unsigned long final_bits) | |
901 | { | |
902 | unsigned long added_bits, removed_bits; | |
bd89aabc | 903 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
904 | int i; |
905 | ||
aae8aab4 BB |
906 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
907 | ||
ddbcc7e8 PM |
908 | removed_bits = root->actual_subsys_bits & ~final_bits; |
909 | added_bits = final_bits & ~root->actual_subsys_bits; | |
910 | /* Check that any added subsystems are currently free */ | |
911 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
8d53d55d | 912 | unsigned long bit = 1UL << i; |
ddbcc7e8 PM |
913 | struct cgroup_subsys *ss = subsys[i]; |
914 | if (!(bit & added_bits)) | |
915 | continue; | |
aae8aab4 BB |
916 | /* |
917 | * Nobody should tell us to do a subsys that doesn't exist: | |
918 | * parse_cgroupfs_options should catch that case and refcounts | |
919 | * ensure that subsystems won't disappear once selected. | |
920 | */ | |
921 | BUG_ON(ss == NULL); | |
ddbcc7e8 PM |
922 | if (ss->root != &rootnode) { |
923 | /* Subsystem isn't free */ | |
924 | return -EBUSY; | |
925 | } | |
926 | } | |
927 | ||
928 | /* Currently we don't handle adding/removing subsystems when | |
929 | * any child cgroups exist. This is theoretically supportable | |
930 | * but involves complex error handling, so it's being left until | |
931 | * later */ | |
307257cf | 932 | if (root->number_of_cgroups > 1) |
ddbcc7e8 PM |
933 | return -EBUSY; |
934 | ||
935 | /* Process each subsystem */ | |
936 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
937 | struct cgroup_subsys *ss = subsys[i]; | |
938 | unsigned long bit = 1UL << i; | |
939 | if (bit & added_bits) { | |
940 | /* We're binding this subsystem to this hierarchy */ | |
aae8aab4 | 941 | BUG_ON(ss == NULL); |
bd89aabc | 942 | BUG_ON(cgrp->subsys[i]); |
ddbcc7e8 PM |
943 | BUG_ON(!dummytop->subsys[i]); |
944 | BUG_ON(dummytop->subsys[i]->cgroup != dummytop); | |
999cd8a4 | 945 | mutex_lock(&ss->hierarchy_mutex); |
bd89aabc PM |
946 | cgrp->subsys[i] = dummytop->subsys[i]; |
947 | cgrp->subsys[i]->cgroup = cgrp; | |
33a68ac1 | 948 | list_move(&ss->sibling, &root->subsys_list); |
b2aa30f7 | 949 | ss->root = root; |
ddbcc7e8 | 950 | if (ss->bind) |
bd89aabc | 951 | ss->bind(ss, cgrp); |
999cd8a4 | 952 | mutex_unlock(&ss->hierarchy_mutex); |
ddbcc7e8 PM |
953 | } else if (bit & removed_bits) { |
954 | /* We're removing this subsystem */ | |
aae8aab4 | 955 | BUG_ON(ss == NULL); |
bd89aabc PM |
956 | BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]); |
957 | BUG_ON(cgrp->subsys[i]->cgroup != cgrp); | |
999cd8a4 | 958 | mutex_lock(&ss->hierarchy_mutex); |
ddbcc7e8 PM |
959 | if (ss->bind) |
960 | ss->bind(ss, dummytop); | |
961 | dummytop->subsys[i]->cgroup = dummytop; | |
bd89aabc | 962 | cgrp->subsys[i] = NULL; |
b2aa30f7 | 963 | subsys[i]->root = &rootnode; |
33a68ac1 | 964 | list_move(&ss->sibling, &rootnode.subsys_list); |
999cd8a4 | 965 | mutex_unlock(&ss->hierarchy_mutex); |
ddbcc7e8 PM |
966 | } else if (bit & final_bits) { |
967 | /* Subsystem state should already exist */ | |
aae8aab4 | 968 | BUG_ON(ss == NULL); |
bd89aabc | 969 | BUG_ON(!cgrp->subsys[i]); |
ddbcc7e8 PM |
970 | } else { |
971 | /* Subsystem state shouldn't exist */ | |
bd89aabc | 972 | BUG_ON(cgrp->subsys[i]); |
ddbcc7e8 PM |
973 | } |
974 | } | |
975 | root->subsys_bits = root->actual_subsys_bits = final_bits; | |
976 | synchronize_rcu(); | |
977 | ||
978 | return 0; | |
979 | } | |
980 | ||
981 | static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs) | |
982 | { | |
983 | struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info; | |
984 | struct cgroup_subsys *ss; | |
985 | ||
986 | mutex_lock(&cgroup_mutex); | |
987 | for_each_subsys(root, ss) | |
988 | seq_printf(seq, ",%s", ss->name); | |
989 | if (test_bit(ROOT_NOPREFIX, &root->flags)) | |
990 | seq_puts(seq, ",noprefix"); | |
81a6a5cd PM |
991 | if (strlen(root->release_agent_path)) |
992 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
c6d57f33 PM |
993 | if (strlen(root->name)) |
994 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
995 | mutex_unlock(&cgroup_mutex); |
996 | return 0; | |
997 | } | |
998 | ||
999 | struct cgroup_sb_opts { | |
1000 | unsigned long subsys_bits; | |
1001 | unsigned long flags; | |
81a6a5cd | 1002 | char *release_agent; |
c6d57f33 | 1003 | char *name; |
2c6ab6d2 PM |
1004 | /* User explicitly requested empty subsystem */ |
1005 | bool none; | |
c6d57f33 PM |
1006 | |
1007 | struct cgroupfs_root *new_root; | |
2c6ab6d2 | 1008 | |
ddbcc7e8 PM |
1009 | }; |
1010 | ||
aae8aab4 BB |
1011 | /* |
1012 | * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call | |
1013 | * with cgroup_mutex held to protect the subsys[] array. | |
1014 | */ | |
ddbcc7e8 PM |
1015 | static int parse_cgroupfs_options(char *data, |
1016 | struct cgroup_sb_opts *opts) | |
1017 | { | |
1018 | char *token, *o = data ?: "all"; | |
f9ab5b5b LZ |
1019 | unsigned long mask = (unsigned long)-1; |
1020 | ||
aae8aab4 BB |
1021 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1022 | ||
f9ab5b5b LZ |
1023 | #ifdef CONFIG_CPUSETS |
1024 | mask = ~(1UL << cpuset_subsys_id); | |
1025 | #endif | |
ddbcc7e8 | 1026 | |
c6d57f33 | 1027 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1028 | |
1029 | while ((token = strsep(&o, ",")) != NULL) { | |
1030 | if (!*token) | |
1031 | return -EINVAL; | |
1032 | if (!strcmp(token, "all")) { | |
8bab8dde PM |
1033 | /* Add all non-disabled subsystems */ |
1034 | int i; | |
1035 | opts->subsys_bits = 0; | |
1036 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1037 | struct cgroup_subsys *ss = subsys[i]; | |
aae8aab4 BB |
1038 | if (ss == NULL) |
1039 | continue; | |
8bab8dde PM |
1040 | if (!ss->disabled) |
1041 | opts->subsys_bits |= 1ul << i; | |
1042 | } | |
2c6ab6d2 PM |
1043 | } else if (!strcmp(token, "none")) { |
1044 | /* Explicitly have no subsystems */ | |
1045 | opts->none = true; | |
ddbcc7e8 PM |
1046 | } else if (!strcmp(token, "noprefix")) { |
1047 | set_bit(ROOT_NOPREFIX, &opts->flags); | |
81a6a5cd PM |
1048 | } else if (!strncmp(token, "release_agent=", 14)) { |
1049 | /* Specifying two release agents is forbidden */ | |
1050 | if (opts->release_agent) | |
1051 | return -EINVAL; | |
c6d57f33 PM |
1052 | opts->release_agent = |
1053 | kstrndup(token + 14, PATH_MAX, GFP_KERNEL); | |
81a6a5cd PM |
1054 | if (!opts->release_agent) |
1055 | return -ENOMEM; | |
c6d57f33 PM |
1056 | } else if (!strncmp(token, "name=", 5)) { |
1057 | int i; | |
1058 | const char *name = token + 5; | |
1059 | /* Can't specify an empty name */ | |
1060 | if (!strlen(name)) | |
1061 | return -EINVAL; | |
1062 | /* Must match [\w.-]+ */ | |
1063 | for (i = 0; i < strlen(name); i++) { | |
1064 | char c = name[i]; | |
1065 | if (isalnum(c)) | |
1066 | continue; | |
1067 | if ((c == '.') || (c == '-') || (c == '_')) | |
1068 | continue; | |
1069 | return -EINVAL; | |
1070 | } | |
1071 | /* Specifying two names is forbidden */ | |
1072 | if (opts->name) | |
1073 | return -EINVAL; | |
1074 | opts->name = kstrndup(name, | |
1075 | MAX_CGROUP_ROOT_NAMELEN, | |
1076 | GFP_KERNEL); | |
1077 | if (!opts->name) | |
1078 | return -ENOMEM; | |
ddbcc7e8 PM |
1079 | } else { |
1080 | struct cgroup_subsys *ss; | |
1081 | int i; | |
1082 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1083 | ss = subsys[i]; | |
aae8aab4 BB |
1084 | if (ss == NULL) |
1085 | continue; | |
ddbcc7e8 | 1086 | if (!strcmp(token, ss->name)) { |
8bab8dde PM |
1087 | if (!ss->disabled) |
1088 | set_bit(i, &opts->subsys_bits); | |
ddbcc7e8 PM |
1089 | break; |
1090 | } | |
1091 | } | |
1092 | if (i == CGROUP_SUBSYS_COUNT) | |
1093 | return -ENOENT; | |
1094 | } | |
1095 | } | |
1096 | ||
2c6ab6d2 PM |
1097 | /* Consistency checks */ |
1098 | ||
f9ab5b5b LZ |
1099 | /* |
1100 | * Option noprefix was introduced just for backward compatibility | |
1101 | * with the old cpuset, so we allow noprefix only if mounting just | |
1102 | * the cpuset subsystem. | |
1103 | */ | |
1104 | if (test_bit(ROOT_NOPREFIX, &opts->flags) && | |
1105 | (opts->subsys_bits & mask)) | |
1106 | return -EINVAL; | |
1107 | ||
2c6ab6d2 PM |
1108 | |
1109 | /* Can't specify "none" and some subsystems */ | |
1110 | if (opts->subsys_bits && opts->none) | |
1111 | return -EINVAL; | |
1112 | ||
1113 | /* | |
1114 | * We either have to specify by name or by subsystems. (So all | |
1115 | * empty hierarchies must have a name). | |
1116 | */ | |
c6d57f33 | 1117 | if (!opts->subsys_bits && !opts->name) |
ddbcc7e8 PM |
1118 | return -EINVAL; |
1119 | ||
1120 | return 0; | |
1121 | } | |
1122 | ||
1123 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) | |
1124 | { | |
1125 | int ret = 0; | |
1126 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1127 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
1128 | struct cgroup_sb_opts opts; |
1129 | ||
337eb00a | 1130 | lock_kernel(); |
bd89aabc | 1131 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
1132 | mutex_lock(&cgroup_mutex); |
1133 | ||
1134 | /* See what subsystems are wanted */ | |
1135 | ret = parse_cgroupfs_options(data, &opts); | |
1136 | if (ret) | |
1137 | goto out_unlock; | |
1138 | ||
1139 | /* Don't allow flags to change at remount */ | |
1140 | if (opts.flags != root->flags) { | |
1141 | ret = -EINVAL; | |
1142 | goto out_unlock; | |
1143 | } | |
1144 | ||
c6d57f33 PM |
1145 | /* Don't allow name to change at remount */ |
1146 | if (opts.name && strcmp(opts.name, root->name)) { | |
1147 | ret = -EINVAL; | |
1148 | goto out_unlock; | |
1149 | } | |
1150 | ||
ddbcc7e8 | 1151 | ret = rebind_subsystems(root, opts.subsys_bits); |
0670e08b LZ |
1152 | if (ret) |
1153 | goto out_unlock; | |
ddbcc7e8 PM |
1154 | |
1155 | /* (re)populate subsystem files */ | |
0670e08b | 1156 | cgroup_populate_dir(cgrp); |
ddbcc7e8 | 1157 | |
81a6a5cd PM |
1158 | if (opts.release_agent) |
1159 | strcpy(root->release_agent_path, opts.release_agent); | |
ddbcc7e8 | 1160 | out_unlock: |
66bdc9cf | 1161 | kfree(opts.release_agent); |
c6d57f33 | 1162 | kfree(opts.name); |
ddbcc7e8 | 1163 | mutex_unlock(&cgroup_mutex); |
bd89aabc | 1164 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
337eb00a | 1165 | unlock_kernel(); |
ddbcc7e8 PM |
1166 | return ret; |
1167 | } | |
1168 | ||
b87221de | 1169 | static const struct super_operations cgroup_ops = { |
ddbcc7e8 PM |
1170 | .statfs = simple_statfs, |
1171 | .drop_inode = generic_delete_inode, | |
1172 | .show_options = cgroup_show_options, | |
1173 | .remount_fs = cgroup_remount, | |
1174 | }; | |
1175 | ||
cc31edce PM |
1176 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1177 | { | |
1178 | INIT_LIST_HEAD(&cgrp->sibling); | |
1179 | INIT_LIST_HEAD(&cgrp->children); | |
1180 | INIT_LIST_HEAD(&cgrp->css_sets); | |
1181 | INIT_LIST_HEAD(&cgrp->release_list); | |
72a8cb30 BB |
1182 | INIT_LIST_HEAD(&cgrp->pidlists); |
1183 | mutex_init(&cgrp->pidlist_mutex); | |
cc31edce | 1184 | } |
c6d57f33 | 1185 | |
ddbcc7e8 PM |
1186 | static void init_cgroup_root(struct cgroupfs_root *root) |
1187 | { | |
bd89aabc | 1188 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
1189 | INIT_LIST_HEAD(&root->subsys_list); |
1190 | INIT_LIST_HEAD(&root->root_list); | |
1191 | root->number_of_cgroups = 1; | |
bd89aabc PM |
1192 | cgrp->root = root; |
1193 | cgrp->top_cgroup = cgrp; | |
cc31edce | 1194 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 PM |
1195 | } |
1196 | ||
2c6ab6d2 PM |
1197 | static bool init_root_id(struct cgroupfs_root *root) |
1198 | { | |
1199 | int ret = 0; | |
1200 | ||
1201 | do { | |
1202 | if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL)) | |
1203 | return false; | |
1204 | spin_lock(&hierarchy_id_lock); | |
1205 | /* Try to allocate the next unused ID */ | |
1206 | ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id, | |
1207 | &root->hierarchy_id); | |
1208 | if (ret == -ENOSPC) | |
1209 | /* Try again starting from 0 */ | |
1210 | ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id); | |
1211 | if (!ret) { | |
1212 | next_hierarchy_id = root->hierarchy_id + 1; | |
1213 | } else if (ret != -EAGAIN) { | |
1214 | /* Can only get here if the 31-bit IDR is full ... */ | |
1215 | BUG_ON(ret); | |
1216 | } | |
1217 | spin_unlock(&hierarchy_id_lock); | |
1218 | } while (ret); | |
1219 | return true; | |
1220 | } | |
1221 | ||
ddbcc7e8 PM |
1222 | static int cgroup_test_super(struct super_block *sb, void *data) |
1223 | { | |
c6d57f33 | 1224 | struct cgroup_sb_opts *opts = data; |
ddbcc7e8 PM |
1225 | struct cgroupfs_root *root = sb->s_fs_info; |
1226 | ||
c6d57f33 PM |
1227 | /* If we asked for a name then it must match */ |
1228 | if (opts->name && strcmp(opts->name, root->name)) | |
1229 | return 0; | |
ddbcc7e8 | 1230 | |
2c6ab6d2 PM |
1231 | /* |
1232 | * If we asked for subsystems (or explicitly for no | |
1233 | * subsystems) then they must match | |
1234 | */ | |
1235 | if ((opts->subsys_bits || opts->none) | |
1236 | && (opts->subsys_bits != root->subsys_bits)) | |
ddbcc7e8 PM |
1237 | return 0; |
1238 | ||
1239 | return 1; | |
1240 | } | |
1241 | ||
c6d57f33 PM |
1242 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1243 | { | |
1244 | struct cgroupfs_root *root; | |
1245 | ||
2c6ab6d2 | 1246 | if (!opts->subsys_bits && !opts->none) |
c6d57f33 PM |
1247 | return NULL; |
1248 | ||
1249 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1250 | if (!root) | |
1251 | return ERR_PTR(-ENOMEM); | |
1252 | ||
2c6ab6d2 PM |
1253 | if (!init_root_id(root)) { |
1254 | kfree(root); | |
1255 | return ERR_PTR(-ENOMEM); | |
1256 | } | |
c6d57f33 | 1257 | init_cgroup_root(root); |
2c6ab6d2 | 1258 | |
c6d57f33 PM |
1259 | root->subsys_bits = opts->subsys_bits; |
1260 | root->flags = opts->flags; | |
1261 | if (opts->release_agent) | |
1262 | strcpy(root->release_agent_path, opts->release_agent); | |
1263 | if (opts->name) | |
1264 | strcpy(root->name, opts->name); | |
1265 | return root; | |
1266 | } | |
1267 | ||
2c6ab6d2 PM |
1268 | static void cgroup_drop_root(struct cgroupfs_root *root) |
1269 | { | |
1270 | if (!root) | |
1271 | return; | |
1272 | ||
1273 | BUG_ON(!root->hierarchy_id); | |
1274 | spin_lock(&hierarchy_id_lock); | |
1275 | ida_remove(&hierarchy_ida, root->hierarchy_id); | |
1276 | spin_unlock(&hierarchy_id_lock); | |
1277 | kfree(root); | |
1278 | } | |
1279 | ||
ddbcc7e8 PM |
1280 | static int cgroup_set_super(struct super_block *sb, void *data) |
1281 | { | |
1282 | int ret; | |
c6d57f33 PM |
1283 | struct cgroup_sb_opts *opts = data; |
1284 | ||
1285 | /* If we don't have a new root, we can't set up a new sb */ | |
1286 | if (!opts->new_root) | |
1287 | return -EINVAL; | |
1288 | ||
2c6ab6d2 | 1289 | BUG_ON(!opts->subsys_bits && !opts->none); |
ddbcc7e8 PM |
1290 | |
1291 | ret = set_anon_super(sb, NULL); | |
1292 | if (ret) | |
1293 | return ret; | |
1294 | ||
c6d57f33 PM |
1295 | sb->s_fs_info = opts->new_root; |
1296 | opts->new_root->sb = sb; | |
ddbcc7e8 PM |
1297 | |
1298 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
1299 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
1300 | sb->s_magic = CGROUP_SUPER_MAGIC; | |
1301 | sb->s_op = &cgroup_ops; | |
1302 | ||
1303 | return 0; | |
1304 | } | |
1305 | ||
1306 | static int cgroup_get_rootdir(struct super_block *sb) | |
1307 | { | |
1308 | struct inode *inode = | |
1309 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); | |
1310 | struct dentry *dentry; | |
1311 | ||
1312 | if (!inode) | |
1313 | return -ENOMEM; | |
1314 | ||
ddbcc7e8 PM |
1315 | inode->i_fop = &simple_dir_operations; |
1316 | inode->i_op = &cgroup_dir_inode_operations; | |
1317 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
1318 | inc_nlink(inode); | |
1319 | dentry = d_alloc_root(inode); | |
1320 | if (!dentry) { | |
1321 | iput(inode); | |
1322 | return -ENOMEM; | |
1323 | } | |
1324 | sb->s_root = dentry; | |
1325 | return 0; | |
1326 | } | |
1327 | ||
1328 | static int cgroup_get_sb(struct file_system_type *fs_type, | |
1329 | int flags, const char *unused_dev_name, | |
1330 | void *data, struct vfsmount *mnt) | |
1331 | { | |
1332 | struct cgroup_sb_opts opts; | |
c6d57f33 | 1333 | struct cgroupfs_root *root; |
ddbcc7e8 PM |
1334 | int ret = 0; |
1335 | struct super_block *sb; | |
c6d57f33 | 1336 | struct cgroupfs_root *new_root; |
ddbcc7e8 PM |
1337 | |
1338 | /* First find the desired set of subsystems */ | |
aae8aab4 | 1339 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1340 | ret = parse_cgroupfs_options(data, &opts); |
aae8aab4 | 1341 | mutex_unlock(&cgroup_mutex); |
c6d57f33 PM |
1342 | if (ret) |
1343 | goto out_err; | |
ddbcc7e8 | 1344 | |
c6d57f33 PM |
1345 | /* |
1346 | * Allocate a new cgroup root. We may not need it if we're | |
1347 | * reusing an existing hierarchy. | |
1348 | */ | |
1349 | new_root = cgroup_root_from_opts(&opts); | |
1350 | if (IS_ERR(new_root)) { | |
1351 | ret = PTR_ERR(new_root); | |
1352 | goto out_err; | |
81a6a5cd | 1353 | } |
c6d57f33 | 1354 | opts.new_root = new_root; |
ddbcc7e8 | 1355 | |
c6d57f33 PM |
1356 | /* Locate an existing or new sb for this hierarchy */ |
1357 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts); | |
ddbcc7e8 | 1358 | if (IS_ERR(sb)) { |
c6d57f33 | 1359 | ret = PTR_ERR(sb); |
2c6ab6d2 | 1360 | cgroup_drop_root(opts.new_root); |
c6d57f33 | 1361 | goto out_err; |
ddbcc7e8 PM |
1362 | } |
1363 | ||
c6d57f33 PM |
1364 | root = sb->s_fs_info; |
1365 | BUG_ON(!root); | |
1366 | if (root == opts.new_root) { | |
1367 | /* We used the new root structure, so this is a new hierarchy */ | |
1368 | struct list_head tmp_cg_links; | |
c12f65d4 | 1369 | struct cgroup *root_cgrp = &root->top_cgroup; |
817929ec | 1370 | struct inode *inode; |
c6d57f33 | 1371 | struct cgroupfs_root *existing_root; |
28fd5dfc | 1372 | int i; |
ddbcc7e8 PM |
1373 | |
1374 | BUG_ON(sb->s_root != NULL); | |
1375 | ||
1376 | ret = cgroup_get_rootdir(sb); | |
1377 | if (ret) | |
1378 | goto drop_new_super; | |
817929ec | 1379 | inode = sb->s_root->d_inode; |
ddbcc7e8 | 1380 | |
817929ec | 1381 | mutex_lock(&inode->i_mutex); |
ddbcc7e8 PM |
1382 | mutex_lock(&cgroup_mutex); |
1383 | ||
c6d57f33 PM |
1384 | if (strlen(root->name)) { |
1385 | /* Check for name clashes with existing mounts */ | |
1386 | for_each_active_root(existing_root) { | |
1387 | if (!strcmp(existing_root->name, root->name)) { | |
1388 | ret = -EBUSY; | |
1389 | mutex_unlock(&cgroup_mutex); | |
1390 | mutex_unlock(&inode->i_mutex); | |
1391 | goto drop_new_super; | |
1392 | } | |
1393 | } | |
1394 | } | |
1395 | ||
817929ec PM |
1396 | /* |
1397 | * We're accessing css_set_count without locking | |
1398 | * css_set_lock here, but that's OK - it can only be | |
1399 | * increased by someone holding cgroup_lock, and | |
1400 | * that's us. The worst that can happen is that we | |
1401 | * have some link structures left over | |
1402 | */ | |
1403 | ret = allocate_cg_links(css_set_count, &tmp_cg_links); | |
1404 | if (ret) { | |
1405 | mutex_unlock(&cgroup_mutex); | |
1406 | mutex_unlock(&inode->i_mutex); | |
1407 | goto drop_new_super; | |
1408 | } | |
1409 | ||
ddbcc7e8 PM |
1410 | ret = rebind_subsystems(root, root->subsys_bits); |
1411 | if (ret == -EBUSY) { | |
1412 | mutex_unlock(&cgroup_mutex); | |
817929ec | 1413 | mutex_unlock(&inode->i_mutex); |
c6d57f33 PM |
1414 | free_cg_links(&tmp_cg_links); |
1415 | goto drop_new_super; | |
ddbcc7e8 PM |
1416 | } |
1417 | ||
1418 | /* EBUSY should be the only error here */ | |
1419 | BUG_ON(ret); | |
1420 | ||
1421 | list_add(&root->root_list, &roots); | |
817929ec | 1422 | root_count++; |
ddbcc7e8 | 1423 | |
c12f65d4 | 1424 | sb->s_root->d_fsdata = root_cgrp; |
ddbcc7e8 PM |
1425 | root->top_cgroup.dentry = sb->s_root; |
1426 | ||
817929ec PM |
1427 | /* Link the top cgroup in this hierarchy into all |
1428 | * the css_set objects */ | |
1429 | write_lock(&css_set_lock); | |
28fd5dfc LZ |
1430 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { |
1431 | struct hlist_head *hhead = &css_set_table[i]; | |
1432 | struct hlist_node *node; | |
817929ec | 1433 | struct css_set *cg; |
28fd5dfc | 1434 | |
c12f65d4 LZ |
1435 | hlist_for_each_entry(cg, node, hhead, hlist) |
1436 | link_css_set(&tmp_cg_links, cg, root_cgrp); | |
28fd5dfc | 1437 | } |
817929ec PM |
1438 | write_unlock(&css_set_lock); |
1439 | ||
1440 | free_cg_links(&tmp_cg_links); | |
1441 | ||
c12f65d4 LZ |
1442 | BUG_ON(!list_empty(&root_cgrp->sibling)); |
1443 | BUG_ON(!list_empty(&root_cgrp->children)); | |
ddbcc7e8 PM |
1444 | BUG_ON(root->number_of_cgroups != 1); |
1445 | ||
c12f65d4 | 1446 | cgroup_populate_dir(root_cgrp); |
ddbcc7e8 | 1447 | mutex_unlock(&cgroup_mutex); |
34f77a90 | 1448 | mutex_unlock(&inode->i_mutex); |
c6d57f33 PM |
1449 | } else { |
1450 | /* | |
1451 | * We re-used an existing hierarchy - the new root (if | |
1452 | * any) is not needed | |
1453 | */ | |
2c6ab6d2 | 1454 | cgroup_drop_root(opts.new_root); |
ddbcc7e8 PM |
1455 | } |
1456 | ||
a3ec947c | 1457 | simple_set_mnt(mnt, sb); |
c6d57f33 PM |
1458 | kfree(opts.release_agent); |
1459 | kfree(opts.name); | |
a3ec947c | 1460 | return 0; |
ddbcc7e8 PM |
1461 | |
1462 | drop_new_super: | |
6f5bbff9 | 1463 | deactivate_locked_super(sb); |
c6d57f33 PM |
1464 | out_err: |
1465 | kfree(opts.release_agent); | |
1466 | kfree(opts.name); | |
1467 | ||
ddbcc7e8 PM |
1468 | return ret; |
1469 | } | |
1470 | ||
1471 | static void cgroup_kill_sb(struct super_block *sb) { | |
1472 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1473 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 | 1474 | int ret; |
71cbb949 KM |
1475 | struct cg_cgroup_link *link; |
1476 | struct cg_cgroup_link *saved_link; | |
ddbcc7e8 PM |
1477 | |
1478 | BUG_ON(!root); | |
1479 | ||
1480 | BUG_ON(root->number_of_cgroups != 1); | |
bd89aabc PM |
1481 | BUG_ON(!list_empty(&cgrp->children)); |
1482 | BUG_ON(!list_empty(&cgrp->sibling)); | |
ddbcc7e8 PM |
1483 | |
1484 | mutex_lock(&cgroup_mutex); | |
1485 | ||
1486 | /* Rebind all subsystems back to the default hierarchy */ | |
1487 | ret = rebind_subsystems(root, 0); | |
1488 | /* Shouldn't be able to fail ... */ | |
1489 | BUG_ON(ret); | |
1490 | ||
817929ec PM |
1491 | /* |
1492 | * Release all the links from css_sets to this hierarchy's | |
1493 | * root cgroup | |
1494 | */ | |
1495 | write_lock(&css_set_lock); | |
71cbb949 KM |
1496 | |
1497 | list_for_each_entry_safe(link, saved_link, &cgrp->css_sets, | |
1498 | cgrp_link_list) { | |
817929ec | 1499 | list_del(&link->cg_link_list); |
bd89aabc | 1500 | list_del(&link->cgrp_link_list); |
817929ec PM |
1501 | kfree(link); |
1502 | } | |
1503 | write_unlock(&css_set_lock); | |
1504 | ||
839ec545 PM |
1505 | if (!list_empty(&root->root_list)) { |
1506 | list_del(&root->root_list); | |
1507 | root_count--; | |
1508 | } | |
e5f6a860 | 1509 | |
ddbcc7e8 PM |
1510 | mutex_unlock(&cgroup_mutex); |
1511 | ||
ddbcc7e8 | 1512 | kill_litter_super(sb); |
2c6ab6d2 | 1513 | cgroup_drop_root(root); |
ddbcc7e8 PM |
1514 | } |
1515 | ||
1516 | static struct file_system_type cgroup_fs_type = { | |
1517 | .name = "cgroup", | |
1518 | .get_sb = cgroup_get_sb, | |
1519 | .kill_sb = cgroup_kill_sb, | |
1520 | }; | |
1521 | ||
bd89aabc | 1522 | static inline struct cgroup *__d_cgrp(struct dentry *dentry) |
ddbcc7e8 PM |
1523 | { |
1524 | return dentry->d_fsdata; | |
1525 | } | |
1526 | ||
1527 | static inline struct cftype *__d_cft(struct dentry *dentry) | |
1528 | { | |
1529 | return dentry->d_fsdata; | |
1530 | } | |
1531 | ||
a043e3b2 LZ |
1532 | /** |
1533 | * cgroup_path - generate the path of a cgroup | |
1534 | * @cgrp: the cgroup in question | |
1535 | * @buf: the buffer to write the path into | |
1536 | * @buflen: the length of the buffer | |
1537 | * | |
a47295e6 PM |
1538 | * Called with cgroup_mutex held or else with an RCU-protected cgroup |
1539 | * reference. Writes path of cgroup into buf. Returns 0 on success, | |
1540 | * -errno on error. | |
ddbcc7e8 | 1541 | */ |
bd89aabc | 1542 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
ddbcc7e8 PM |
1543 | { |
1544 | char *start; | |
a47295e6 | 1545 | struct dentry *dentry = rcu_dereference(cgrp->dentry); |
ddbcc7e8 | 1546 | |
a47295e6 | 1547 | if (!dentry || cgrp == dummytop) { |
ddbcc7e8 PM |
1548 | /* |
1549 | * Inactive subsystems have no dentry for their root | |
1550 | * cgroup | |
1551 | */ | |
1552 | strcpy(buf, "/"); | |
1553 | return 0; | |
1554 | } | |
1555 | ||
1556 | start = buf + buflen; | |
1557 | ||
1558 | *--start = '\0'; | |
1559 | for (;;) { | |
a47295e6 | 1560 | int len = dentry->d_name.len; |
ddbcc7e8 PM |
1561 | if ((start -= len) < buf) |
1562 | return -ENAMETOOLONG; | |
bd89aabc PM |
1563 | memcpy(start, cgrp->dentry->d_name.name, len); |
1564 | cgrp = cgrp->parent; | |
1565 | if (!cgrp) | |
ddbcc7e8 | 1566 | break; |
a47295e6 | 1567 | dentry = rcu_dereference(cgrp->dentry); |
bd89aabc | 1568 | if (!cgrp->parent) |
ddbcc7e8 PM |
1569 | continue; |
1570 | if (--start < buf) | |
1571 | return -ENAMETOOLONG; | |
1572 | *start = '/'; | |
1573 | } | |
1574 | memmove(buf, start, buf + buflen - start); | |
1575 | return 0; | |
1576 | } | |
1577 | ||
a043e3b2 LZ |
1578 | /** |
1579 | * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp' | |
1580 | * @cgrp: the cgroup the task is attaching to | |
1581 | * @tsk: the task to be attached | |
bbcb81d0 | 1582 | * |
a043e3b2 LZ |
1583 | * Call holding cgroup_mutex. May take task_lock of |
1584 | * the task 'tsk' during call. | |
bbcb81d0 | 1585 | */ |
956db3ca | 1586 | int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
bbcb81d0 PM |
1587 | { |
1588 | int retval = 0; | |
2468c723 | 1589 | struct cgroup_subsys *ss, *failed_ss = NULL; |
bd89aabc | 1590 | struct cgroup *oldcgrp; |
77efecd9 | 1591 | struct css_set *cg; |
817929ec | 1592 | struct css_set *newcg; |
bd89aabc | 1593 | struct cgroupfs_root *root = cgrp->root; |
bbcb81d0 PM |
1594 | |
1595 | /* Nothing to do if the task is already in that cgroup */ | |
7717f7ba | 1596 | oldcgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 1597 | if (cgrp == oldcgrp) |
bbcb81d0 PM |
1598 | return 0; |
1599 | ||
1600 | for_each_subsys(root, ss) { | |
1601 | if (ss->can_attach) { | |
be367d09 | 1602 | retval = ss->can_attach(ss, cgrp, tsk, false); |
2468c723 DN |
1603 | if (retval) { |
1604 | /* | |
1605 | * Remember on which subsystem the can_attach() | |
1606 | * failed, so that we only call cancel_attach() | |
1607 | * against the subsystems whose can_attach() | |
1608 | * succeeded. (See below) | |
1609 | */ | |
1610 | failed_ss = ss; | |
1611 | goto out; | |
1612 | } | |
bbcb81d0 PM |
1613 | } |
1614 | } | |
1615 | ||
77efecd9 LJ |
1616 | task_lock(tsk); |
1617 | cg = tsk->cgroups; | |
1618 | get_css_set(cg); | |
1619 | task_unlock(tsk); | |
817929ec PM |
1620 | /* |
1621 | * Locate or allocate a new css_set for this task, | |
1622 | * based on its final set of cgroups | |
1623 | */ | |
bd89aabc | 1624 | newcg = find_css_set(cg, cgrp); |
77efecd9 | 1625 | put_css_set(cg); |
2468c723 DN |
1626 | if (!newcg) { |
1627 | retval = -ENOMEM; | |
1628 | goto out; | |
1629 | } | |
817929ec | 1630 | |
bbcb81d0 PM |
1631 | task_lock(tsk); |
1632 | if (tsk->flags & PF_EXITING) { | |
1633 | task_unlock(tsk); | |
817929ec | 1634 | put_css_set(newcg); |
2468c723 DN |
1635 | retval = -ESRCH; |
1636 | goto out; | |
bbcb81d0 | 1637 | } |
817929ec | 1638 | rcu_assign_pointer(tsk->cgroups, newcg); |
bbcb81d0 PM |
1639 | task_unlock(tsk); |
1640 | ||
817929ec PM |
1641 | /* Update the css_set linked lists if we're using them */ |
1642 | write_lock(&css_set_lock); | |
1643 | if (!list_empty(&tsk->cg_list)) { | |
1644 | list_del(&tsk->cg_list); | |
1645 | list_add(&tsk->cg_list, &newcg->tasks); | |
1646 | } | |
1647 | write_unlock(&css_set_lock); | |
1648 | ||
bbcb81d0 | 1649 | for_each_subsys(root, ss) { |
e18f6318 | 1650 | if (ss->attach) |
be367d09 | 1651 | ss->attach(ss, cgrp, oldcgrp, tsk, false); |
bbcb81d0 | 1652 | } |
bd89aabc | 1653 | set_bit(CGRP_RELEASABLE, &oldcgrp->flags); |
bbcb81d0 | 1654 | synchronize_rcu(); |
817929ec | 1655 | put_css_set(cg); |
ec64f515 KH |
1656 | |
1657 | /* | |
1658 | * wake up rmdir() waiter. the rmdir should fail since the cgroup | |
1659 | * is no longer empty. | |
1660 | */ | |
88703267 | 1661 | cgroup_wakeup_rmdir_waiter(cgrp); |
2468c723 DN |
1662 | out: |
1663 | if (retval) { | |
1664 | for_each_subsys(root, ss) { | |
1665 | if (ss == failed_ss) | |
1666 | /* | |
1667 | * This subsystem was the one that failed the | |
1668 | * can_attach() check earlier, so we don't need | |
1669 | * to call cancel_attach() against it or any | |
1670 | * remaining subsystems. | |
1671 | */ | |
1672 | break; | |
1673 | if (ss->cancel_attach) | |
1674 | ss->cancel_attach(ss, cgrp, tsk, false); | |
1675 | } | |
1676 | } | |
1677 | return retval; | |
bbcb81d0 PM |
1678 | } |
1679 | ||
1680 | /* | |
af351026 PM |
1681 | * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex |
1682 | * held. May take task_lock of task | |
bbcb81d0 | 1683 | */ |
af351026 | 1684 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid) |
bbcb81d0 | 1685 | { |
bbcb81d0 | 1686 | struct task_struct *tsk; |
c69e8d9c | 1687 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
1688 | int ret; |
1689 | ||
bbcb81d0 PM |
1690 | if (pid) { |
1691 | rcu_read_lock(); | |
73507f33 | 1692 | tsk = find_task_by_vpid(pid); |
bbcb81d0 PM |
1693 | if (!tsk || tsk->flags & PF_EXITING) { |
1694 | rcu_read_unlock(); | |
1695 | return -ESRCH; | |
1696 | } | |
bbcb81d0 | 1697 | |
c69e8d9c DH |
1698 | tcred = __task_cred(tsk); |
1699 | if (cred->euid && | |
1700 | cred->euid != tcred->uid && | |
1701 | cred->euid != tcred->suid) { | |
1702 | rcu_read_unlock(); | |
bbcb81d0 PM |
1703 | return -EACCES; |
1704 | } | |
c69e8d9c DH |
1705 | get_task_struct(tsk); |
1706 | rcu_read_unlock(); | |
bbcb81d0 PM |
1707 | } else { |
1708 | tsk = current; | |
1709 | get_task_struct(tsk); | |
1710 | } | |
1711 | ||
956db3ca | 1712 | ret = cgroup_attach_task(cgrp, tsk); |
bbcb81d0 PM |
1713 | put_task_struct(tsk); |
1714 | return ret; | |
1715 | } | |
1716 | ||
af351026 PM |
1717 | static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid) |
1718 | { | |
1719 | int ret; | |
1720 | if (!cgroup_lock_live_group(cgrp)) | |
1721 | return -ENODEV; | |
1722 | ret = attach_task_by_pid(cgrp, pid); | |
1723 | cgroup_unlock(); | |
1724 | return ret; | |
1725 | } | |
1726 | ||
e788e066 PM |
1727 | /** |
1728 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
1729 | * @cgrp: the cgroup to be checked for liveness | |
1730 | * | |
84eea842 PM |
1731 | * On success, returns true; the lock should be later released with |
1732 | * cgroup_unlock(). On failure returns false with no lock held. | |
e788e066 | 1733 | */ |
84eea842 | 1734 | bool cgroup_lock_live_group(struct cgroup *cgrp) |
e788e066 PM |
1735 | { |
1736 | mutex_lock(&cgroup_mutex); | |
1737 | if (cgroup_is_removed(cgrp)) { | |
1738 | mutex_unlock(&cgroup_mutex); | |
1739 | return false; | |
1740 | } | |
1741 | return true; | |
1742 | } | |
1743 | ||
1744 | static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft, | |
1745 | const char *buffer) | |
1746 | { | |
1747 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
1748 | if (!cgroup_lock_live_group(cgrp)) | |
1749 | return -ENODEV; | |
1750 | strcpy(cgrp->root->release_agent_path, buffer); | |
84eea842 | 1751 | cgroup_unlock(); |
e788e066 PM |
1752 | return 0; |
1753 | } | |
1754 | ||
1755 | static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft, | |
1756 | struct seq_file *seq) | |
1757 | { | |
1758 | if (!cgroup_lock_live_group(cgrp)) | |
1759 | return -ENODEV; | |
1760 | seq_puts(seq, cgrp->root->release_agent_path); | |
1761 | seq_putc(seq, '\n'); | |
84eea842 | 1762 | cgroup_unlock(); |
e788e066 PM |
1763 | return 0; |
1764 | } | |
1765 | ||
84eea842 PM |
1766 | /* A buffer size big enough for numbers or short strings */ |
1767 | #define CGROUP_LOCAL_BUFFER_SIZE 64 | |
1768 | ||
e73d2c61 | 1769 | static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft, |
f4c753b7 PM |
1770 | struct file *file, |
1771 | const char __user *userbuf, | |
1772 | size_t nbytes, loff_t *unused_ppos) | |
355e0c48 | 1773 | { |
84eea842 | 1774 | char buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
355e0c48 | 1775 | int retval = 0; |
355e0c48 PM |
1776 | char *end; |
1777 | ||
1778 | if (!nbytes) | |
1779 | return -EINVAL; | |
1780 | if (nbytes >= sizeof(buffer)) | |
1781 | return -E2BIG; | |
1782 | if (copy_from_user(buffer, userbuf, nbytes)) | |
1783 | return -EFAULT; | |
1784 | ||
1785 | buffer[nbytes] = 0; /* nul-terminate */ | |
e73d2c61 | 1786 | if (cft->write_u64) { |
478988d3 | 1787 | u64 val = simple_strtoull(strstrip(buffer), &end, 0); |
e73d2c61 PM |
1788 | if (*end) |
1789 | return -EINVAL; | |
1790 | retval = cft->write_u64(cgrp, cft, val); | |
1791 | } else { | |
478988d3 | 1792 | s64 val = simple_strtoll(strstrip(buffer), &end, 0); |
e73d2c61 PM |
1793 | if (*end) |
1794 | return -EINVAL; | |
1795 | retval = cft->write_s64(cgrp, cft, val); | |
1796 | } | |
355e0c48 PM |
1797 | if (!retval) |
1798 | retval = nbytes; | |
1799 | return retval; | |
1800 | } | |
1801 | ||
db3b1497 PM |
1802 | static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft, |
1803 | struct file *file, | |
1804 | const char __user *userbuf, | |
1805 | size_t nbytes, loff_t *unused_ppos) | |
1806 | { | |
84eea842 | 1807 | char local_buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
db3b1497 PM |
1808 | int retval = 0; |
1809 | size_t max_bytes = cft->max_write_len; | |
1810 | char *buffer = local_buffer; | |
1811 | ||
1812 | if (!max_bytes) | |
1813 | max_bytes = sizeof(local_buffer) - 1; | |
1814 | if (nbytes >= max_bytes) | |
1815 | return -E2BIG; | |
1816 | /* Allocate a dynamic buffer if we need one */ | |
1817 | if (nbytes >= sizeof(local_buffer)) { | |
1818 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); | |
1819 | if (buffer == NULL) | |
1820 | return -ENOMEM; | |
1821 | } | |
5a3eb9f6 LZ |
1822 | if (nbytes && copy_from_user(buffer, userbuf, nbytes)) { |
1823 | retval = -EFAULT; | |
1824 | goto out; | |
1825 | } | |
db3b1497 PM |
1826 | |
1827 | buffer[nbytes] = 0; /* nul-terminate */ | |
478988d3 | 1828 | retval = cft->write_string(cgrp, cft, strstrip(buffer)); |
db3b1497 PM |
1829 | if (!retval) |
1830 | retval = nbytes; | |
5a3eb9f6 | 1831 | out: |
db3b1497 PM |
1832 | if (buffer != local_buffer) |
1833 | kfree(buffer); | |
1834 | return retval; | |
1835 | } | |
1836 | ||
ddbcc7e8 PM |
1837 | static ssize_t cgroup_file_write(struct file *file, const char __user *buf, |
1838 | size_t nbytes, loff_t *ppos) | |
1839 | { | |
1840 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 1841 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 1842 | |
75139b82 | 1843 | if (cgroup_is_removed(cgrp)) |
ddbcc7e8 | 1844 | return -ENODEV; |
355e0c48 | 1845 | if (cft->write) |
bd89aabc | 1846 | return cft->write(cgrp, cft, file, buf, nbytes, ppos); |
e73d2c61 PM |
1847 | if (cft->write_u64 || cft->write_s64) |
1848 | return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos); | |
db3b1497 PM |
1849 | if (cft->write_string) |
1850 | return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos); | |
d447ea2f PE |
1851 | if (cft->trigger) { |
1852 | int ret = cft->trigger(cgrp, (unsigned int)cft->private); | |
1853 | return ret ? ret : nbytes; | |
1854 | } | |
355e0c48 | 1855 | return -EINVAL; |
ddbcc7e8 PM |
1856 | } |
1857 | ||
f4c753b7 PM |
1858 | static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft, |
1859 | struct file *file, | |
1860 | char __user *buf, size_t nbytes, | |
1861 | loff_t *ppos) | |
ddbcc7e8 | 1862 | { |
84eea842 | 1863 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
f4c753b7 | 1864 | u64 val = cft->read_u64(cgrp, cft); |
ddbcc7e8 PM |
1865 | int len = sprintf(tmp, "%llu\n", (unsigned long long) val); |
1866 | ||
1867 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
1868 | } | |
1869 | ||
e73d2c61 PM |
1870 | static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft, |
1871 | struct file *file, | |
1872 | char __user *buf, size_t nbytes, | |
1873 | loff_t *ppos) | |
1874 | { | |
84eea842 | 1875 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
e73d2c61 PM |
1876 | s64 val = cft->read_s64(cgrp, cft); |
1877 | int len = sprintf(tmp, "%lld\n", (long long) val); | |
1878 | ||
1879 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
1880 | } | |
1881 | ||
ddbcc7e8 PM |
1882 | static ssize_t cgroup_file_read(struct file *file, char __user *buf, |
1883 | size_t nbytes, loff_t *ppos) | |
1884 | { | |
1885 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 1886 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 1887 | |
75139b82 | 1888 | if (cgroup_is_removed(cgrp)) |
ddbcc7e8 PM |
1889 | return -ENODEV; |
1890 | ||
1891 | if (cft->read) | |
bd89aabc | 1892 | return cft->read(cgrp, cft, file, buf, nbytes, ppos); |
f4c753b7 PM |
1893 | if (cft->read_u64) |
1894 | return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos); | |
e73d2c61 PM |
1895 | if (cft->read_s64) |
1896 | return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos); | |
ddbcc7e8 PM |
1897 | return -EINVAL; |
1898 | } | |
1899 | ||
91796569 PM |
1900 | /* |
1901 | * seqfile ops/methods for returning structured data. Currently just | |
1902 | * supports string->u64 maps, but can be extended in future. | |
1903 | */ | |
1904 | ||
1905 | struct cgroup_seqfile_state { | |
1906 | struct cftype *cft; | |
1907 | struct cgroup *cgroup; | |
1908 | }; | |
1909 | ||
1910 | static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value) | |
1911 | { | |
1912 | struct seq_file *sf = cb->state; | |
1913 | return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value); | |
1914 | } | |
1915 | ||
1916 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) | |
1917 | { | |
1918 | struct cgroup_seqfile_state *state = m->private; | |
1919 | struct cftype *cft = state->cft; | |
29486df3 SH |
1920 | if (cft->read_map) { |
1921 | struct cgroup_map_cb cb = { | |
1922 | .fill = cgroup_map_add, | |
1923 | .state = m, | |
1924 | }; | |
1925 | return cft->read_map(state->cgroup, cft, &cb); | |
1926 | } | |
1927 | return cft->read_seq_string(state->cgroup, cft, m); | |
91796569 PM |
1928 | } |
1929 | ||
96930a63 | 1930 | static int cgroup_seqfile_release(struct inode *inode, struct file *file) |
91796569 PM |
1931 | { |
1932 | struct seq_file *seq = file->private_data; | |
1933 | kfree(seq->private); | |
1934 | return single_release(inode, file); | |
1935 | } | |
1936 | ||
828c0950 | 1937 | static const struct file_operations cgroup_seqfile_operations = { |
91796569 | 1938 | .read = seq_read, |
e788e066 | 1939 | .write = cgroup_file_write, |
91796569 PM |
1940 | .llseek = seq_lseek, |
1941 | .release = cgroup_seqfile_release, | |
1942 | }; | |
1943 | ||
ddbcc7e8 PM |
1944 | static int cgroup_file_open(struct inode *inode, struct file *file) |
1945 | { | |
1946 | int err; | |
1947 | struct cftype *cft; | |
1948 | ||
1949 | err = generic_file_open(inode, file); | |
1950 | if (err) | |
1951 | return err; | |
ddbcc7e8 | 1952 | cft = __d_cft(file->f_dentry); |
75139b82 | 1953 | |
29486df3 | 1954 | if (cft->read_map || cft->read_seq_string) { |
91796569 PM |
1955 | struct cgroup_seqfile_state *state = |
1956 | kzalloc(sizeof(*state), GFP_USER); | |
1957 | if (!state) | |
1958 | return -ENOMEM; | |
1959 | state->cft = cft; | |
1960 | state->cgroup = __d_cgrp(file->f_dentry->d_parent); | |
1961 | file->f_op = &cgroup_seqfile_operations; | |
1962 | err = single_open(file, cgroup_seqfile_show, state); | |
1963 | if (err < 0) | |
1964 | kfree(state); | |
1965 | } else if (cft->open) | |
ddbcc7e8 PM |
1966 | err = cft->open(inode, file); |
1967 | else | |
1968 | err = 0; | |
1969 | ||
1970 | return err; | |
1971 | } | |
1972 | ||
1973 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
1974 | { | |
1975 | struct cftype *cft = __d_cft(file->f_dentry); | |
1976 | if (cft->release) | |
1977 | return cft->release(inode, file); | |
1978 | return 0; | |
1979 | } | |
1980 | ||
1981 | /* | |
1982 | * cgroup_rename - Only allow simple rename of directories in place. | |
1983 | */ | |
1984 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1985 | struct inode *new_dir, struct dentry *new_dentry) | |
1986 | { | |
1987 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) | |
1988 | return -ENOTDIR; | |
1989 | if (new_dentry->d_inode) | |
1990 | return -EEXIST; | |
1991 | if (old_dir != new_dir) | |
1992 | return -EIO; | |
1993 | return simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
1994 | } | |
1995 | ||
828c0950 | 1996 | static const struct file_operations cgroup_file_operations = { |
ddbcc7e8 PM |
1997 | .read = cgroup_file_read, |
1998 | .write = cgroup_file_write, | |
1999 | .llseek = generic_file_llseek, | |
2000 | .open = cgroup_file_open, | |
2001 | .release = cgroup_file_release, | |
2002 | }; | |
2003 | ||
6e1d5dcc | 2004 | static const struct inode_operations cgroup_dir_inode_operations = { |
ddbcc7e8 PM |
2005 | .lookup = simple_lookup, |
2006 | .mkdir = cgroup_mkdir, | |
2007 | .rmdir = cgroup_rmdir, | |
2008 | .rename = cgroup_rename, | |
2009 | }; | |
2010 | ||
099fca32 | 2011 | static int cgroup_create_file(struct dentry *dentry, mode_t mode, |
ddbcc7e8 PM |
2012 | struct super_block *sb) |
2013 | { | |
3ba13d17 | 2014 | static const struct dentry_operations cgroup_dops = { |
ddbcc7e8 PM |
2015 | .d_iput = cgroup_diput, |
2016 | }; | |
2017 | ||
2018 | struct inode *inode; | |
2019 | ||
2020 | if (!dentry) | |
2021 | return -ENOENT; | |
2022 | if (dentry->d_inode) | |
2023 | return -EEXIST; | |
2024 | ||
2025 | inode = cgroup_new_inode(mode, sb); | |
2026 | if (!inode) | |
2027 | return -ENOMEM; | |
2028 | ||
2029 | if (S_ISDIR(mode)) { | |
2030 | inode->i_op = &cgroup_dir_inode_operations; | |
2031 | inode->i_fop = &simple_dir_operations; | |
2032 | ||
2033 | /* start off with i_nlink == 2 (for "." entry) */ | |
2034 | inc_nlink(inode); | |
2035 | ||
2036 | /* start with the directory inode held, so that we can | |
2037 | * populate it without racing with another mkdir */ | |
817929ec | 2038 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); |
ddbcc7e8 PM |
2039 | } else if (S_ISREG(mode)) { |
2040 | inode->i_size = 0; | |
2041 | inode->i_fop = &cgroup_file_operations; | |
2042 | } | |
2043 | dentry->d_op = &cgroup_dops; | |
2044 | d_instantiate(dentry, inode); | |
2045 | dget(dentry); /* Extra count - pin the dentry in core */ | |
2046 | return 0; | |
2047 | } | |
2048 | ||
2049 | /* | |
a043e3b2 LZ |
2050 | * cgroup_create_dir - create a directory for an object. |
2051 | * @cgrp: the cgroup we create the directory for. It must have a valid | |
2052 | * ->parent field. And we are going to fill its ->dentry field. | |
2053 | * @dentry: dentry of the new cgroup | |
2054 | * @mode: mode to set on new directory. | |
ddbcc7e8 | 2055 | */ |
bd89aabc | 2056 | static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry, |
099fca32 | 2057 | mode_t mode) |
ddbcc7e8 PM |
2058 | { |
2059 | struct dentry *parent; | |
2060 | int error = 0; | |
2061 | ||
bd89aabc PM |
2062 | parent = cgrp->parent->dentry; |
2063 | error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb); | |
ddbcc7e8 | 2064 | if (!error) { |
bd89aabc | 2065 | dentry->d_fsdata = cgrp; |
ddbcc7e8 | 2066 | inc_nlink(parent->d_inode); |
a47295e6 | 2067 | rcu_assign_pointer(cgrp->dentry, dentry); |
ddbcc7e8 PM |
2068 | dget(dentry); |
2069 | } | |
2070 | dput(dentry); | |
2071 | ||
2072 | return error; | |
2073 | } | |
2074 | ||
099fca32 LZ |
2075 | /** |
2076 | * cgroup_file_mode - deduce file mode of a control file | |
2077 | * @cft: the control file in question | |
2078 | * | |
2079 | * returns cft->mode if ->mode is not 0 | |
2080 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
2081 | * returns S_IRUGO if it has only a read handler | |
2082 | * returns S_IWUSR if it has only a write hander | |
2083 | */ | |
2084 | static mode_t cgroup_file_mode(const struct cftype *cft) | |
2085 | { | |
2086 | mode_t mode = 0; | |
2087 | ||
2088 | if (cft->mode) | |
2089 | return cft->mode; | |
2090 | ||
2091 | if (cft->read || cft->read_u64 || cft->read_s64 || | |
2092 | cft->read_map || cft->read_seq_string) | |
2093 | mode |= S_IRUGO; | |
2094 | ||
2095 | if (cft->write || cft->write_u64 || cft->write_s64 || | |
2096 | cft->write_string || cft->trigger) | |
2097 | mode |= S_IWUSR; | |
2098 | ||
2099 | return mode; | |
2100 | } | |
2101 | ||
bd89aabc | 2102 | int cgroup_add_file(struct cgroup *cgrp, |
ddbcc7e8 PM |
2103 | struct cgroup_subsys *subsys, |
2104 | const struct cftype *cft) | |
2105 | { | |
bd89aabc | 2106 | struct dentry *dir = cgrp->dentry; |
ddbcc7e8 PM |
2107 | struct dentry *dentry; |
2108 | int error; | |
099fca32 | 2109 | mode_t mode; |
ddbcc7e8 PM |
2110 | |
2111 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; | |
bd89aabc | 2112 | if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) { |
ddbcc7e8 PM |
2113 | strcpy(name, subsys->name); |
2114 | strcat(name, "."); | |
2115 | } | |
2116 | strcat(name, cft->name); | |
2117 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); | |
2118 | dentry = lookup_one_len(name, dir, strlen(name)); | |
2119 | if (!IS_ERR(dentry)) { | |
099fca32 LZ |
2120 | mode = cgroup_file_mode(cft); |
2121 | error = cgroup_create_file(dentry, mode | S_IFREG, | |
bd89aabc | 2122 | cgrp->root->sb); |
ddbcc7e8 PM |
2123 | if (!error) |
2124 | dentry->d_fsdata = (void *)cft; | |
2125 | dput(dentry); | |
2126 | } else | |
2127 | error = PTR_ERR(dentry); | |
2128 | return error; | |
2129 | } | |
e6a1105b | 2130 | EXPORT_SYMBOL_GPL(cgroup_add_file); |
ddbcc7e8 | 2131 | |
bd89aabc | 2132 | int cgroup_add_files(struct cgroup *cgrp, |
ddbcc7e8 PM |
2133 | struct cgroup_subsys *subsys, |
2134 | const struct cftype cft[], | |
2135 | int count) | |
2136 | { | |
2137 | int i, err; | |
2138 | for (i = 0; i < count; i++) { | |
bd89aabc | 2139 | err = cgroup_add_file(cgrp, subsys, &cft[i]); |
ddbcc7e8 PM |
2140 | if (err) |
2141 | return err; | |
2142 | } | |
2143 | return 0; | |
2144 | } | |
e6a1105b | 2145 | EXPORT_SYMBOL_GPL(cgroup_add_files); |
ddbcc7e8 | 2146 | |
a043e3b2 LZ |
2147 | /** |
2148 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2149 | * @cgrp: the cgroup in question | |
2150 | * | |
2151 | * Return the number of tasks in the cgroup. | |
2152 | */ | |
bd89aabc | 2153 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2154 | { |
2155 | int count = 0; | |
71cbb949 | 2156 | struct cg_cgroup_link *link; |
817929ec PM |
2157 | |
2158 | read_lock(&css_set_lock); | |
71cbb949 | 2159 | list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) { |
146aa1bd | 2160 | count += atomic_read(&link->cg->refcount); |
817929ec PM |
2161 | } |
2162 | read_unlock(&css_set_lock); | |
bbcb81d0 PM |
2163 | return count; |
2164 | } | |
2165 | ||
817929ec PM |
2166 | /* |
2167 | * Advance a list_head iterator. The iterator should be positioned at | |
2168 | * the start of a css_set | |
2169 | */ | |
bd89aabc | 2170 | static void cgroup_advance_iter(struct cgroup *cgrp, |
7717f7ba | 2171 | struct cgroup_iter *it) |
817929ec PM |
2172 | { |
2173 | struct list_head *l = it->cg_link; | |
2174 | struct cg_cgroup_link *link; | |
2175 | struct css_set *cg; | |
2176 | ||
2177 | /* Advance to the next non-empty css_set */ | |
2178 | do { | |
2179 | l = l->next; | |
bd89aabc | 2180 | if (l == &cgrp->css_sets) { |
817929ec PM |
2181 | it->cg_link = NULL; |
2182 | return; | |
2183 | } | |
bd89aabc | 2184 | link = list_entry(l, struct cg_cgroup_link, cgrp_link_list); |
817929ec PM |
2185 | cg = link->cg; |
2186 | } while (list_empty(&cg->tasks)); | |
2187 | it->cg_link = l; | |
2188 | it->task = cg->tasks.next; | |
2189 | } | |
2190 | ||
31a7df01 CW |
2191 | /* |
2192 | * To reduce the fork() overhead for systems that are not actually | |
2193 | * using their cgroups capability, we don't maintain the lists running | |
2194 | * through each css_set to its tasks until we see the list actually | |
2195 | * used - in other words after the first call to cgroup_iter_start(). | |
2196 | * | |
2197 | * The tasklist_lock is not held here, as do_each_thread() and | |
2198 | * while_each_thread() are protected by RCU. | |
2199 | */ | |
3df91fe3 | 2200 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2201 | { |
2202 | struct task_struct *p, *g; | |
2203 | write_lock(&css_set_lock); | |
2204 | use_task_css_set_links = 1; | |
2205 | do_each_thread(g, p) { | |
2206 | task_lock(p); | |
0e04388f LZ |
2207 | /* |
2208 | * We should check if the process is exiting, otherwise | |
2209 | * it will race with cgroup_exit() in that the list | |
2210 | * entry won't be deleted though the process has exited. | |
2211 | */ | |
2212 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
31a7df01 CW |
2213 | list_add(&p->cg_list, &p->cgroups->tasks); |
2214 | task_unlock(p); | |
2215 | } while_each_thread(g, p); | |
2216 | write_unlock(&css_set_lock); | |
2217 | } | |
2218 | ||
bd89aabc | 2219 | void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it) |
817929ec PM |
2220 | { |
2221 | /* | |
2222 | * The first time anyone tries to iterate across a cgroup, | |
2223 | * we need to enable the list linking each css_set to its | |
2224 | * tasks, and fix up all existing tasks. | |
2225 | */ | |
31a7df01 CW |
2226 | if (!use_task_css_set_links) |
2227 | cgroup_enable_task_cg_lists(); | |
2228 | ||
817929ec | 2229 | read_lock(&css_set_lock); |
bd89aabc PM |
2230 | it->cg_link = &cgrp->css_sets; |
2231 | cgroup_advance_iter(cgrp, it); | |
817929ec PM |
2232 | } |
2233 | ||
bd89aabc | 2234 | struct task_struct *cgroup_iter_next(struct cgroup *cgrp, |
817929ec PM |
2235 | struct cgroup_iter *it) |
2236 | { | |
2237 | struct task_struct *res; | |
2238 | struct list_head *l = it->task; | |
2019f634 | 2239 | struct cg_cgroup_link *link; |
817929ec PM |
2240 | |
2241 | /* If the iterator cg is NULL, we have no tasks */ | |
2242 | if (!it->cg_link) | |
2243 | return NULL; | |
2244 | res = list_entry(l, struct task_struct, cg_list); | |
2245 | /* Advance iterator to find next entry */ | |
2246 | l = l->next; | |
2019f634 LJ |
2247 | link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list); |
2248 | if (l == &link->cg->tasks) { | |
817929ec PM |
2249 | /* We reached the end of this task list - move on to |
2250 | * the next cg_cgroup_link */ | |
bd89aabc | 2251 | cgroup_advance_iter(cgrp, it); |
817929ec PM |
2252 | } else { |
2253 | it->task = l; | |
2254 | } | |
2255 | return res; | |
2256 | } | |
2257 | ||
bd89aabc | 2258 | void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it) |
817929ec PM |
2259 | { |
2260 | read_unlock(&css_set_lock); | |
2261 | } | |
2262 | ||
31a7df01 CW |
2263 | static inline int started_after_time(struct task_struct *t1, |
2264 | struct timespec *time, | |
2265 | struct task_struct *t2) | |
2266 | { | |
2267 | int start_diff = timespec_compare(&t1->start_time, time); | |
2268 | if (start_diff > 0) { | |
2269 | return 1; | |
2270 | } else if (start_diff < 0) { | |
2271 | return 0; | |
2272 | } else { | |
2273 | /* | |
2274 | * Arbitrarily, if two processes started at the same | |
2275 | * time, we'll say that the lower pointer value | |
2276 | * started first. Note that t2 may have exited by now | |
2277 | * so this may not be a valid pointer any longer, but | |
2278 | * that's fine - it still serves to distinguish | |
2279 | * between two tasks started (effectively) simultaneously. | |
2280 | */ | |
2281 | return t1 > t2; | |
2282 | } | |
2283 | } | |
2284 | ||
2285 | /* | |
2286 | * This function is a callback from heap_insert() and is used to order | |
2287 | * the heap. | |
2288 | * In this case we order the heap in descending task start time. | |
2289 | */ | |
2290 | static inline int started_after(void *p1, void *p2) | |
2291 | { | |
2292 | struct task_struct *t1 = p1; | |
2293 | struct task_struct *t2 = p2; | |
2294 | return started_after_time(t1, &t2->start_time, t2); | |
2295 | } | |
2296 | ||
2297 | /** | |
2298 | * cgroup_scan_tasks - iterate though all the tasks in a cgroup | |
2299 | * @scan: struct cgroup_scanner containing arguments for the scan | |
2300 | * | |
2301 | * Arguments include pointers to callback functions test_task() and | |
2302 | * process_task(). | |
2303 | * Iterate through all the tasks in a cgroup, calling test_task() for each, | |
2304 | * and if it returns true, call process_task() for it also. | |
2305 | * The test_task pointer may be NULL, meaning always true (select all tasks). | |
2306 | * Effectively duplicates cgroup_iter_{start,next,end}() | |
2307 | * but does not lock css_set_lock for the call to process_task(). | |
2308 | * The struct cgroup_scanner may be embedded in any structure of the caller's | |
2309 | * creation. | |
2310 | * It is guaranteed that process_task() will act on every task that | |
2311 | * is a member of the cgroup for the duration of this call. This | |
2312 | * function may or may not call process_task() for tasks that exit | |
2313 | * or move to a different cgroup during the call, or are forked or | |
2314 | * move into the cgroup during the call. | |
2315 | * | |
2316 | * Note that test_task() may be called with locks held, and may in some | |
2317 | * situations be called multiple times for the same task, so it should | |
2318 | * be cheap. | |
2319 | * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been | |
2320 | * pre-allocated and will be used for heap operations (and its "gt" member will | |
2321 | * be overwritten), else a temporary heap will be used (allocation of which | |
2322 | * may cause this function to fail). | |
2323 | */ | |
2324 | int cgroup_scan_tasks(struct cgroup_scanner *scan) | |
2325 | { | |
2326 | int retval, i; | |
2327 | struct cgroup_iter it; | |
2328 | struct task_struct *p, *dropped; | |
2329 | /* Never dereference latest_task, since it's not refcounted */ | |
2330 | struct task_struct *latest_task = NULL; | |
2331 | struct ptr_heap tmp_heap; | |
2332 | struct ptr_heap *heap; | |
2333 | struct timespec latest_time = { 0, 0 }; | |
2334 | ||
2335 | if (scan->heap) { | |
2336 | /* The caller supplied our heap and pre-allocated its memory */ | |
2337 | heap = scan->heap; | |
2338 | heap->gt = &started_after; | |
2339 | } else { | |
2340 | /* We need to allocate our own heap memory */ | |
2341 | heap = &tmp_heap; | |
2342 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
2343 | if (retval) | |
2344 | /* cannot allocate the heap */ | |
2345 | return retval; | |
2346 | } | |
2347 | ||
2348 | again: | |
2349 | /* | |
2350 | * Scan tasks in the cgroup, using the scanner's "test_task" callback | |
2351 | * to determine which are of interest, and using the scanner's | |
2352 | * "process_task" callback to process any of them that need an update. | |
2353 | * Since we don't want to hold any locks during the task updates, | |
2354 | * gather tasks to be processed in a heap structure. | |
2355 | * The heap is sorted by descending task start time. | |
2356 | * If the statically-sized heap fills up, we overflow tasks that | |
2357 | * started later, and in future iterations only consider tasks that | |
2358 | * started after the latest task in the previous pass. This | |
2359 | * guarantees forward progress and that we don't miss any tasks. | |
2360 | */ | |
2361 | heap->size = 0; | |
2362 | cgroup_iter_start(scan->cg, &it); | |
2363 | while ((p = cgroup_iter_next(scan->cg, &it))) { | |
2364 | /* | |
2365 | * Only affect tasks that qualify per the caller's callback, | |
2366 | * if he provided one | |
2367 | */ | |
2368 | if (scan->test_task && !scan->test_task(p, scan)) | |
2369 | continue; | |
2370 | /* | |
2371 | * Only process tasks that started after the last task | |
2372 | * we processed | |
2373 | */ | |
2374 | if (!started_after_time(p, &latest_time, latest_task)) | |
2375 | continue; | |
2376 | dropped = heap_insert(heap, p); | |
2377 | if (dropped == NULL) { | |
2378 | /* | |
2379 | * The new task was inserted; the heap wasn't | |
2380 | * previously full | |
2381 | */ | |
2382 | get_task_struct(p); | |
2383 | } else if (dropped != p) { | |
2384 | /* | |
2385 | * The new task was inserted, and pushed out a | |
2386 | * different task | |
2387 | */ | |
2388 | get_task_struct(p); | |
2389 | put_task_struct(dropped); | |
2390 | } | |
2391 | /* | |
2392 | * Else the new task was newer than anything already in | |
2393 | * the heap and wasn't inserted | |
2394 | */ | |
2395 | } | |
2396 | cgroup_iter_end(scan->cg, &it); | |
2397 | ||
2398 | if (heap->size) { | |
2399 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 2400 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 2401 | if (i == 0) { |
4fe91d51 PJ |
2402 | latest_time = q->start_time; |
2403 | latest_task = q; | |
31a7df01 CW |
2404 | } |
2405 | /* Process the task per the caller's callback */ | |
4fe91d51 PJ |
2406 | scan->process_task(q, scan); |
2407 | put_task_struct(q); | |
31a7df01 CW |
2408 | } |
2409 | /* | |
2410 | * If we had to process any tasks at all, scan again | |
2411 | * in case some of them were in the middle of forking | |
2412 | * children that didn't get processed. | |
2413 | * Not the most efficient way to do it, but it avoids | |
2414 | * having to take callback_mutex in the fork path | |
2415 | */ | |
2416 | goto again; | |
2417 | } | |
2418 | if (heap == &tmp_heap) | |
2419 | heap_free(&tmp_heap); | |
2420 | return 0; | |
2421 | } | |
2422 | ||
bbcb81d0 | 2423 | /* |
102a775e | 2424 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
2425 | * |
2426 | * Reading this file can return large amounts of data if a cgroup has | |
2427 | * *lots* of attached tasks. So it may need several calls to read(), | |
2428 | * but we cannot guarantee that the information we produce is correct | |
2429 | * unless we produce it entirely atomically. | |
2430 | * | |
bbcb81d0 | 2431 | */ |
bbcb81d0 | 2432 | |
d1d9fd33 BB |
2433 | /* |
2434 | * The following two functions "fix" the issue where there are more pids | |
2435 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
2436 | * TODO: replace with a kernel-wide solution to this problem | |
2437 | */ | |
2438 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
2439 | static void *pidlist_allocate(int count) | |
2440 | { | |
2441 | if (PIDLIST_TOO_LARGE(count)) | |
2442 | return vmalloc(count * sizeof(pid_t)); | |
2443 | else | |
2444 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
2445 | } | |
2446 | static void pidlist_free(void *p) | |
2447 | { | |
2448 | if (is_vmalloc_addr(p)) | |
2449 | vfree(p); | |
2450 | else | |
2451 | kfree(p); | |
2452 | } | |
2453 | static void *pidlist_resize(void *p, int newcount) | |
2454 | { | |
2455 | void *newlist; | |
2456 | /* note: if new alloc fails, old p will still be valid either way */ | |
2457 | if (is_vmalloc_addr(p)) { | |
2458 | newlist = vmalloc(newcount * sizeof(pid_t)); | |
2459 | if (!newlist) | |
2460 | return NULL; | |
2461 | memcpy(newlist, p, newcount * sizeof(pid_t)); | |
2462 | vfree(p); | |
2463 | } else { | |
2464 | newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL); | |
2465 | } | |
2466 | return newlist; | |
2467 | } | |
2468 | ||
bbcb81d0 | 2469 | /* |
102a775e BB |
2470 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
2471 | * If the new stripped list is sufficiently smaller and there's enough memory | |
2472 | * to allocate a new buffer, will let go of the unneeded memory. Returns the | |
2473 | * number of unique elements. | |
bbcb81d0 | 2474 | */ |
102a775e BB |
2475 | /* is the size difference enough that we should re-allocate the array? */ |
2476 | #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new)) | |
2477 | static int pidlist_uniq(pid_t **p, int length) | |
bbcb81d0 | 2478 | { |
102a775e BB |
2479 | int src, dest = 1; |
2480 | pid_t *list = *p; | |
2481 | pid_t *newlist; | |
2482 | ||
2483 | /* | |
2484 | * we presume the 0th element is unique, so i starts at 1. trivial | |
2485 | * edge cases first; no work needs to be done for either | |
2486 | */ | |
2487 | if (length == 0 || length == 1) | |
2488 | return length; | |
2489 | /* src and dest walk down the list; dest counts unique elements */ | |
2490 | for (src = 1; src < length; src++) { | |
2491 | /* find next unique element */ | |
2492 | while (list[src] == list[src-1]) { | |
2493 | src++; | |
2494 | if (src == length) | |
2495 | goto after; | |
2496 | } | |
2497 | /* dest always points to where the next unique element goes */ | |
2498 | list[dest] = list[src]; | |
2499 | dest++; | |
2500 | } | |
2501 | after: | |
2502 | /* | |
2503 | * if the length difference is large enough, we want to allocate a | |
2504 | * smaller buffer to save memory. if this fails due to out of memory, | |
2505 | * we'll just stay with what we've got. | |
2506 | */ | |
2507 | if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) { | |
d1d9fd33 | 2508 | newlist = pidlist_resize(list, dest); |
102a775e BB |
2509 | if (newlist) |
2510 | *p = newlist; | |
2511 | } | |
2512 | return dest; | |
2513 | } | |
2514 | ||
2515 | static int cmppid(const void *a, const void *b) | |
2516 | { | |
2517 | return *(pid_t *)a - *(pid_t *)b; | |
2518 | } | |
2519 | ||
72a8cb30 BB |
2520 | /* |
2521 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
2522 | * returns with the lock on that pidlist already held, and takes care | |
2523 | * of the use count, or returns NULL with no locks held if we're out of | |
2524 | * memory. | |
2525 | */ | |
2526 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
2527 | enum cgroup_filetype type) | |
2528 | { | |
2529 | struct cgroup_pidlist *l; | |
2530 | /* don't need task_nsproxy() if we're looking at ourself */ | |
2531 | struct pid_namespace *ns = get_pid_ns(current->nsproxy->pid_ns); | |
2532 | /* | |
2533 | * We can't drop the pidlist_mutex before taking the l->mutex in case | |
2534 | * the last ref-holder is trying to remove l from the list at the same | |
2535 | * time. Holding the pidlist_mutex precludes somebody taking whichever | |
2536 | * list we find out from under us - compare release_pid_array(). | |
2537 | */ | |
2538 | mutex_lock(&cgrp->pidlist_mutex); | |
2539 | list_for_each_entry(l, &cgrp->pidlists, links) { | |
2540 | if (l->key.type == type && l->key.ns == ns) { | |
2541 | /* found a matching list - drop the extra refcount */ | |
2542 | put_pid_ns(ns); | |
2543 | /* make sure l doesn't vanish out from under us */ | |
2544 | down_write(&l->mutex); | |
2545 | mutex_unlock(&cgrp->pidlist_mutex); | |
72a8cb30 BB |
2546 | return l; |
2547 | } | |
2548 | } | |
2549 | /* entry not found; create a new one */ | |
2550 | l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); | |
2551 | if (!l) { | |
2552 | mutex_unlock(&cgrp->pidlist_mutex); | |
2553 | put_pid_ns(ns); | |
2554 | return l; | |
2555 | } | |
2556 | init_rwsem(&l->mutex); | |
2557 | down_write(&l->mutex); | |
2558 | l->key.type = type; | |
2559 | l->key.ns = ns; | |
2560 | l->use_count = 0; /* don't increment here */ | |
2561 | l->list = NULL; | |
2562 | l->owner = cgrp; | |
2563 | list_add(&l->links, &cgrp->pidlists); | |
2564 | mutex_unlock(&cgrp->pidlist_mutex); | |
2565 | return l; | |
2566 | } | |
2567 | ||
102a775e BB |
2568 | /* |
2569 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
2570 | */ | |
72a8cb30 BB |
2571 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
2572 | struct cgroup_pidlist **lp) | |
102a775e BB |
2573 | { |
2574 | pid_t *array; | |
2575 | int length; | |
2576 | int pid, n = 0; /* used for populating the array */ | |
817929ec PM |
2577 | struct cgroup_iter it; |
2578 | struct task_struct *tsk; | |
102a775e BB |
2579 | struct cgroup_pidlist *l; |
2580 | ||
2581 | /* | |
2582 | * If cgroup gets more users after we read count, we won't have | |
2583 | * enough space - tough. This race is indistinguishable to the | |
2584 | * caller from the case that the additional cgroup users didn't | |
2585 | * show up until sometime later on. | |
2586 | */ | |
2587 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 2588 | array = pidlist_allocate(length); |
102a775e BB |
2589 | if (!array) |
2590 | return -ENOMEM; | |
2591 | /* now, populate the array */ | |
bd89aabc PM |
2592 | cgroup_iter_start(cgrp, &it); |
2593 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
102a775e | 2594 | if (unlikely(n == length)) |
817929ec | 2595 | break; |
102a775e | 2596 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
2597 | if (type == CGROUP_FILE_PROCS) |
2598 | pid = task_tgid_vnr(tsk); | |
2599 | else | |
2600 | pid = task_pid_vnr(tsk); | |
102a775e BB |
2601 | if (pid > 0) /* make sure to only use valid results */ |
2602 | array[n++] = pid; | |
817929ec | 2603 | } |
bd89aabc | 2604 | cgroup_iter_end(cgrp, &it); |
102a775e BB |
2605 | length = n; |
2606 | /* now sort & (if procs) strip out duplicates */ | |
2607 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 2608 | if (type == CGROUP_FILE_PROCS) |
102a775e | 2609 | length = pidlist_uniq(&array, length); |
72a8cb30 BB |
2610 | l = cgroup_pidlist_find(cgrp, type); |
2611 | if (!l) { | |
d1d9fd33 | 2612 | pidlist_free(array); |
72a8cb30 | 2613 | return -ENOMEM; |
102a775e | 2614 | } |
72a8cb30 | 2615 | /* store array, freeing old if necessary - lock already held */ |
d1d9fd33 | 2616 | pidlist_free(l->list); |
102a775e BB |
2617 | l->list = array; |
2618 | l->length = length; | |
2619 | l->use_count++; | |
2620 | up_write(&l->mutex); | |
72a8cb30 | 2621 | *lp = l; |
102a775e | 2622 | return 0; |
bbcb81d0 PM |
2623 | } |
2624 | ||
846c7bb0 | 2625 | /** |
a043e3b2 | 2626 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
2627 | * @stats: cgroupstats to fill information into |
2628 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
2629 | * been requested. | |
a043e3b2 LZ |
2630 | * |
2631 | * Build and fill cgroupstats so that taskstats can export it to user | |
2632 | * space. | |
846c7bb0 BS |
2633 | */ |
2634 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
2635 | { | |
2636 | int ret = -EINVAL; | |
bd89aabc | 2637 | struct cgroup *cgrp; |
846c7bb0 BS |
2638 | struct cgroup_iter it; |
2639 | struct task_struct *tsk; | |
33d283be | 2640 | |
846c7bb0 | 2641 | /* |
33d283be LZ |
2642 | * Validate dentry by checking the superblock operations, |
2643 | * and make sure it's a directory. | |
846c7bb0 | 2644 | */ |
33d283be LZ |
2645 | if (dentry->d_sb->s_op != &cgroup_ops || |
2646 | !S_ISDIR(dentry->d_inode->i_mode)) | |
846c7bb0 BS |
2647 | goto err; |
2648 | ||
2649 | ret = 0; | |
bd89aabc | 2650 | cgrp = dentry->d_fsdata; |
846c7bb0 | 2651 | |
bd89aabc PM |
2652 | cgroup_iter_start(cgrp, &it); |
2653 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
846c7bb0 BS |
2654 | switch (tsk->state) { |
2655 | case TASK_RUNNING: | |
2656 | stats->nr_running++; | |
2657 | break; | |
2658 | case TASK_INTERRUPTIBLE: | |
2659 | stats->nr_sleeping++; | |
2660 | break; | |
2661 | case TASK_UNINTERRUPTIBLE: | |
2662 | stats->nr_uninterruptible++; | |
2663 | break; | |
2664 | case TASK_STOPPED: | |
2665 | stats->nr_stopped++; | |
2666 | break; | |
2667 | default: | |
2668 | if (delayacct_is_task_waiting_on_io(tsk)) | |
2669 | stats->nr_io_wait++; | |
2670 | break; | |
2671 | } | |
2672 | } | |
bd89aabc | 2673 | cgroup_iter_end(cgrp, &it); |
846c7bb0 | 2674 | |
846c7bb0 BS |
2675 | err: |
2676 | return ret; | |
2677 | } | |
2678 | ||
8f3ff208 | 2679 | |
bbcb81d0 | 2680 | /* |
102a775e | 2681 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 2682 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 2683 | * in the cgroup->l->list array. |
bbcb81d0 | 2684 | */ |
cc31edce | 2685 | |
102a775e | 2686 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 2687 | { |
cc31edce PM |
2688 | /* |
2689 | * Initially we receive a position value that corresponds to | |
2690 | * one more than the last pid shown (or 0 on the first call or | |
2691 | * after a seek to the start). Use a binary-search to find the | |
2692 | * next pid to display, if any | |
2693 | */ | |
102a775e | 2694 | struct cgroup_pidlist *l = s->private; |
cc31edce PM |
2695 | int index = 0, pid = *pos; |
2696 | int *iter; | |
2697 | ||
102a775e | 2698 | down_read(&l->mutex); |
cc31edce | 2699 | if (pid) { |
102a775e | 2700 | int end = l->length; |
20777766 | 2701 | |
cc31edce PM |
2702 | while (index < end) { |
2703 | int mid = (index + end) / 2; | |
102a775e | 2704 | if (l->list[mid] == pid) { |
cc31edce PM |
2705 | index = mid; |
2706 | break; | |
102a775e | 2707 | } else if (l->list[mid] <= pid) |
cc31edce PM |
2708 | index = mid + 1; |
2709 | else | |
2710 | end = mid; | |
2711 | } | |
2712 | } | |
2713 | /* If we're off the end of the array, we're done */ | |
102a775e | 2714 | if (index >= l->length) |
cc31edce PM |
2715 | return NULL; |
2716 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 2717 | iter = l->list + index; |
cc31edce PM |
2718 | *pos = *iter; |
2719 | return iter; | |
2720 | } | |
2721 | ||
102a775e | 2722 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 2723 | { |
102a775e BB |
2724 | struct cgroup_pidlist *l = s->private; |
2725 | up_read(&l->mutex); | |
cc31edce PM |
2726 | } |
2727 | ||
102a775e | 2728 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 2729 | { |
102a775e BB |
2730 | struct cgroup_pidlist *l = s->private; |
2731 | pid_t *p = v; | |
2732 | pid_t *end = l->list + l->length; | |
cc31edce PM |
2733 | /* |
2734 | * Advance to the next pid in the array. If this goes off the | |
2735 | * end, we're done | |
2736 | */ | |
2737 | p++; | |
2738 | if (p >= end) { | |
2739 | return NULL; | |
2740 | } else { | |
2741 | *pos = *p; | |
2742 | return p; | |
2743 | } | |
2744 | } | |
2745 | ||
102a775e | 2746 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
2747 | { |
2748 | return seq_printf(s, "%d\n", *(int *)v); | |
2749 | } | |
bbcb81d0 | 2750 | |
102a775e BB |
2751 | /* |
2752 | * seq_operations functions for iterating on pidlists through seq_file - | |
2753 | * independent of whether it's tasks or procs | |
2754 | */ | |
2755 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
2756 | .start = cgroup_pidlist_start, | |
2757 | .stop = cgroup_pidlist_stop, | |
2758 | .next = cgroup_pidlist_next, | |
2759 | .show = cgroup_pidlist_show, | |
cc31edce PM |
2760 | }; |
2761 | ||
102a775e | 2762 | static void cgroup_release_pid_array(struct cgroup_pidlist *l) |
cc31edce | 2763 | { |
72a8cb30 BB |
2764 | /* |
2765 | * the case where we're the last user of this particular pidlist will | |
2766 | * have us remove it from the cgroup's list, which entails taking the | |
2767 | * mutex. since in pidlist_find the pidlist->lock depends on cgroup-> | |
2768 | * pidlist_mutex, we have to take pidlist_mutex first. | |
2769 | */ | |
2770 | mutex_lock(&l->owner->pidlist_mutex); | |
102a775e BB |
2771 | down_write(&l->mutex); |
2772 | BUG_ON(!l->use_count); | |
2773 | if (!--l->use_count) { | |
72a8cb30 BB |
2774 | /* we're the last user if refcount is 0; remove and free */ |
2775 | list_del(&l->links); | |
2776 | mutex_unlock(&l->owner->pidlist_mutex); | |
d1d9fd33 | 2777 | pidlist_free(l->list); |
72a8cb30 BB |
2778 | put_pid_ns(l->key.ns); |
2779 | up_write(&l->mutex); | |
2780 | kfree(l); | |
2781 | return; | |
cc31edce | 2782 | } |
72a8cb30 | 2783 | mutex_unlock(&l->owner->pidlist_mutex); |
102a775e | 2784 | up_write(&l->mutex); |
bbcb81d0 PM |
2785 | } |
2786 | ||
102a775e | 2787 | static int cgroup_pidlist_release(struct inode *inode, struct file *file) |
cc31edce | 2788 | { |
102a775e | 2789 | struct cgroup_pidlist *l; |
cc31edce PM |
2790 | if (!(file->f_mode & FMODE_READ)) |
2791 | return 0; | |
102a775e BB |
2792 | /* |
2793 | * the seq_file will only be initialized if the file was opened for | |
2794 | * reading; hence we check if it's not null only in that case. | |
2795 | */ | |
2796 | l = ((struct seq_file *)file->private_data)->private; | |
2797 | cgroup_release_pid_array(l); | |
cc31edce PM |
2798 | return seq_release(inode, file); |
2799 | } | |
2800 | ||
102a775e | 2801 | static const struct file_operations cgroup_pidlist_operations = { |
cc31edce PM |
2802 | .read = seq_read, |
2803 | .llseek = seq_lseek, | |
2804 | .write = cgroup_file_write, | |
102a775e | 2805 | .release = cgroup_pidlist_release, |
cc31edce PM |
2806 | }; |
2807 | ||
bbcb81d0 | 2808 | /* |
102a775e BB |
2809 | * The following functions handle opens on a file that displays a pidlist |
2810 | * (tasks or procs). Prepare an array of the process/thread IDs of whoever's | |
2811 | * in the cgroup. | |
bbcb81d0 | 2812 | */ |
102a775e | 2813 | /* helper function for the two below it */ |
72a8cb30 | 2814 | static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type) |
bbcb81d0 | 2815 | { |
bd89aabc | 2816 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
72a8cb30 | 2817 | struct cgroup_pidlist *l; |
cc31edce | 2818 | int retval; |
bbcb81d0 | 2819 | |
cc31edce | 2820 | /* Nothing to do for write-only files */ |
bbcb81d0 PM |
2821 | if (!(file->f_mode & FMODE_READ)) |
2822 | return 0; | |
2823 | ||
102a775e | 2824 | /* have the array populated */ |
72a8cb30 | 2825 | retval = pidlist_array_load(cgrp, type, &l); |
102a775e BB |
2826 | if (retval) |
2827 | return retval; | |
2828 | /* configure file information */ | |
2829 | file->f_op = &cgroup_pidlist_operations; | |
cc31edce | 2830 | |
102a775e | 2831 | retval = seq_open(file, &cgroup_pidlist_seq_operations); |
cc31edce | 2832 | if (retval) { |
102a775e | 2833 | cgroup_release_pid_array(l); |
cc31edce | 2834 | return retval; |
bbcb81d0 | 2835 | } |
102a775e | 2836 | ((struct seq_file *)file->private_data)->private = l; |
bbcb81d0 PM |
2837 | return 0; |
2838 | } | |
102a775e BB |
2839 | static int cgroup_tasks_open(struct inode *unused, struct file *file) |
2840 | { | |
72a8cb30 | 2841 | return cgroup_pidlist_open(file, CGROUP_FILE_TASKS); |
102a775e BB |
2842 | } |
2843 | static int cgroup_procs_open(struct inode *unused, struct file *file) | |
2844 | { | |
72a8cb30 | 2845 | return cgroup_pidlist_open(file, CGROUP_FILE_PROCS); |
102a775e | 2846 | } |
bbcb81d0 | 2847 | |
bd89aabc | 2848 | static u64 cgroup_read_notify_on_release(struct cgroup *cgrp, |
81a6a5cd PM |
2849 | struct cftype *cft) |
2850 | { | |
bd89aabc | 2851 | return notify_on_release(cgrp); |
81a6a5cd PM |
2852 | } |
2853 | ||
6379c106 PM |
2854 | static int cgroup_write_notify_on_release(struct cgroup *cgrp, |
2855 | struct cftype *cft, | |
2856 | u64 val) | |
2857 | { | |
2858 | clear_bit(CGRP_RELEASABLE, &cgrp->flags); | |
2859 | if (val) | |
2860 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
2861 | else | |
2862 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
2863 | return 0; | |
2864 | } | |
2865 | ||
bbcb81d0 PM |
2866 | /* |
2867 | * for the common functions, 'private' gives the type of file | |
2868 | */ | |
102a775e BB |
2869 | /* for hysterical raisins, we can't put this on the older files */ |
2870 | #define CGROUP_FILE_GENERIC_PREFIX "cgroup." | |
81a6a5cd PM |
2871 | static struct cftype files[] = { |
2872 | { | |
2873 | .name = "tasks", | |
2874 | .open = cgroup_tasks_open, | |
af351026 | 2875 | .write_u64 = cgroup_tasks_write, |
102a775e | 2876 | .release = cgroup_pidlist_release, |
099fca32 | 2877 | .mode = S_IRUGO | S_IWUSR, |
81a6a5cd | 2878 | }, |
102a775e BB |
2879 | { |
2880 | .name = CGROUP_FILE_GENERIC_PREFIX "procs", | |
2881 | .open = cgroup_procs_open, | |
2882 | /* .write_u64 = cgroup_procs_write, TODO */ | |
2883 | .release = cgroup_pidlist_release, | |
2884 | .mode = S_IRUGO, | |
2885 | }, | |
81a6a5cd PM |
2886 | { |
2887 | .name = "notify_on_release", | |
f4c753b7 | 2888 | .read_u64 = cgroup_read_notify_on_release, |
6379c106 | 2889 | .write_u64 = cgroup_write_notify_on_release, |
81a6a5cd | 2890 | }, |
81a6a5cd PM |
2891 | }; |
2892 | ||
2893 | static struct cftype cft_release_agent = { | |
2894 | .name = "release_agent", | |
e788e066 PM |
2895 | .read_seq_string = cgroup_release_agent_show, |
2896 | .write_string = cgroup_release_agent_write, | |
2897 | .max_write_len = PATH_MAX, | |
bbcb81d0 PM |
2898 | }; |
2899 | ||
bd89aabc | 2900 | static int cgroup_populate_dir(struct cgroup *cgrp) |
ddbcc7e8 PM |
2901 | { |
2902 | int err; | |
2903 | struct cgroup_subsys *ss; | |
2904 | ||
2905 | /* First clear out any existing files */ | |
bd89aabc | 2906 | cgroup_clear_directory(cgrp->dentry); |
ddbcc7e8 | 2907 | |
bd89aabc | 2908 | err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files)); |
bbcb81d0 PM |
2909 | if (err < 0) |
2910 | return err; | |
2911 | ||
bd89aabc PM |
2912 | if (cgrp == cgrp->top_cgroup) { |
2913 | if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0) | |
81a6a5cd PM |
2914 | return err; |
2915 | } | |
2916 | ||
bd89aabc PM |
2917 | for_each_subsys(cgrp->root, ss) { |
2918 | if (ss->populate && (err = ss->populate(ss, cgrp)) < 0) | |
ddbcc7e8 PM |
2919 | return err; |
2920 | } | |
38460b48 KH |
2921 | /* This cgroup is ready now */ |
2922 | for_each_subsys(cgrp->root, ss) { | |
2923 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
2924 | /* | |
2925 | * Update id->css pointer and make this css visible from | |
2926 | * CSS ID functions. This pointer will be dereferened | |
2927 | * from RCU-read-side without locks. | |
2928 | */ | |
2929 | if (css->id) | |
2930 | rcu_assign_pointer(css->id->css, css); | |
2931 | } | |
ddbcc7e8 PM |
2932 | |
2933 | return 0; | |
2934 | } | |
2935 | ||
2936 | static void init_cgroup_css(struct cgroup_subsys_state *css, | |
2937 | struct cgroup_subsys *ss, | |
bd89aabc | 2938 | struct cgroup *cgrp) |
ddbcc7e8 | 2939 | { |
bd89aabc | 2940 | css->cgroup = cgrp; |
e7c5ec91 | 2941 | atomic_set(&css->refcnt, 1); |
ddbcc7e8 | 2942 | css->flags = 0; |
38460b48 | 2943 | css->id = NULL; |
bd89aabc | 2944 | if (cgrp == dummytop) |
ddbcc7e8 | 2945 | set_bit(CSS_ROOT, &css->flags); |
bd89aabc PM |
2946 | BUG_ON(cgrp->subsys[ss->subsys_id]); |
2947 | cgrp->subsys[ss->subsys_id] = css; | |
ddbcc7e8 PM |
2948 | } |
2949 | ||
999cd8a4 PM |
2950 | static void cgroup_lock_hierarchy(struct cgroupfs_root *root) |
2951 | { | |
2952 | /* We need to take each hierarchy_mutex in a consistent order */ | |
2953 | int i; | |
2954 | ||
aae8aab4 BB |
2955 | /* |
2956 | * No worry about a race with rebind_subsystems that might mess up the | |
2957 | * locking order, since both parties are under cgroup_mutex. | |
2958 | */ | |
999cd8a4 PM |
2959 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
2960 | struct cgroup_subsys *ss = subsys[i]; | |
aae8aab4 BB |
2961 | if (ss == NULL) |
2962 | continue; | |
999cd8a4 | 2963 | if (ss->root == root) |
cfebe563 | 2964 | mutex_lock(&ss->hierarchy_mutex); |
999cd8a4 PM |
2965 | } |
2966 | } | |
2967 | ||
2968 | static void cgroup_unlock_hierarchy(struct cgroupfs_root *root) | |
2969 | { | |
2970 | int i; | |
2971 | ||
2972 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2973 | struct cgroup_subsys *ss = subsys[i]; | |
aae8aab4 BB |
2974 | if (ss == NULL) |
2975 | continue; | |
999cd8a4 PM |
2976 | if (ss->root == root) |
2977 | mutex_unlock(&ss->hierarchy_mutex); | |
2978 | } | |
2979 | } | |
2980 | ||
ddbcc7e8 | 2981 | /* |
a043e3b2 LZ |
2982 | * cgroup_create - create a cgroup |
2983 | * @parent: cgroup that will be parent of the new cgroup | |
2984 | * @dentry: dentry of the new cgroup | |
2985 | * @mode: mode to set on new inode | |
ddbcc7e8 | 2986 | * |
a043e3b2 | 2987 | * Must be called with the mutex on the parent inode held |
ddbcc7e8 | 2988 | */ |
ddbcc7e8 | 2989 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, |
099fca32 | 2990 | mode_t mode) |
ddbcc7e8 | 2991 | { |
bd89aabc | 2992 | struct cgroup *cgrp; |
ddbcc7e8 PM |
2993 | struct cgroupfs_root *root = parent->root; |
2994 | int err = 0; | |
2995 | struct cgroup_subsys *ss; | |
2996 | struct super_block *sb = root->sb; | |
2997 | ||
bd89aabc PM |
2998 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
2999 | if (!cgrp) | |
ddbcc7e8 PM |
3000 | return -ENOMEM; |
3001 | ||
3002 | /* Grab a reference on the superblock so the hierarchy doesn't | |
3003 | * get deleted on unmount if there are child cgroups. This | |
3004 | * can be done outside cgroup_mutex, since the sb can't | |
3005 | * disappear while someone has an open control file on the | |
3006 | * fs */ | |
3007 | atomic_inc(&sb->s_active); | |
3008 | ||
3009 | mutex_lock(&cgroup_mutex); | |
3010 | ||
cc31edce | 3011 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3012 | |
bd89aabc PM |
3013 | cgrp->parent = parent; |
3014 | cgrp->root = parent->root; | |
3015 | cgrp->top_cgroup = parent->top_cgroup; | |
ddbcc7e8 | 3016 | |
b6abdb0e LZ |
3017 | if (notify_on_release(parent)) |
3018 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3019 | ||
ddbcc7e8 | 3020 | for_each_subsys(root, ss) { |
bd89aabc | 3021 | struct cgroup_subsys_state *css = ss->create(ss, cgrp); |
4528fd05 | 3022 | |
ddbcc7e8 PM |
3023 | if (IS_ERR(css)) { |
3024 | err = PTR_ERR(css); | |
3025 | goto err_destroy; | |
3026 | } | |
bd89aabc | 3027 | init_cgroup_css(css, ss, cgrp); |
4528fd05 LZ |
3028 | if (ss->use_id) { |
3029 | err = alloc_css_id(ss, parent, cgrp); | |
3030 | if (err) | |
38460b48 | 3031 | goto err_destroy; |
4528fd05 | 3032 | } |
38460b48 | 3033 | /* At error, ->destroy() callback has to free assigned ID. */ |
ddbcc7e8 PM |
3034 | } |
3035 | ||
999cd8a4 | 3036 | cgroup_lock_hierarchy(root); |
bd89aabc | 3037 | list_add(&cgrp->sibling, &cgrp->parent->children); |
999cd8a4 | 3038 | cgroup_unlock_hierarchy(root); |
ddbcc7e8 PM |
3039 | root->number_of_cgroups++; |
3040 | ||
bd89aabc | 3041 | err = cgroup_create_dir(cgrp, dentry, mode); |
ddbcc7e8 PM |
3042 | if (err < 0) |
3043 | goto err_remove; | |
3044 | ||
3045 | /* The cgroup directory was pre-locked for us */ | |
bd89aabc | 3046 | BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex)); |
ddbcc7e8 | 3047 | |
bd89aabc | 3048 | err = cgroup_populate_dir(cgrp); |
ddbcc7e8 PM |
3049 | /* If err < 0, we have a half-filled directory - oh well ;) */ |
3050 | ||
3051 | mutex_unlock(&cgroup_mutex); | |
bd89aabc | 3052 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
3053 | |
3054 | return 0; | |
3055 | ||
3056 | err_remove: | |
3057 | ||
baef99a0 | 3058 | cgroup_lock_hierarchy(root); |
bd89aabc | 3059 | list_del(&cgrp->sibling); |
baef99a0 | 3060 | cgroup_unlock_hierarchy(root); |
ddbcc7e8 PM |
3061 | root->number_of_cgroups--; |
3062 | ||
3063 | err_destroy: | |
3064 | ||
3065 | for_each_subsys(root, ss) { | |
bd89aabc PM |
3066 | if (cgrp->subsys[ss->subsys_id]) |
3067 | ss->destroy(ss, cgrp); | |
ddbcc7e8 PM |
3068 | } |
3069 | ||
3070 | mutex_unlock(&cgroup_mutex); | |
3071 | ||
3072 | /* Release the reference count that we took on the superblock */ | |
3073 | deactivate_super(sb); | |
3074 | ||
bd89aabc | 3075 | kfree(cgrp); |
ddbcc7e8 PM |
3076 | return err; |
3077 | } | |
3078 | ||
3079 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
3080 | { | |
3081 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
3082 | ||
3083 | /* the vfs holds inode->i_mutex already */ | |
3084 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
3085 | } | |
3086 | ||
55b6fd01 | 3087 | static int cgroup_has_css_refs(struct cgroup *cgrp) |
81a6a5cd PM |
3088 | { |
3089 | /* Check the reference count on each subsystem. Since we | |
3090 | * already established that there are no tasks in the | |
e7c5ec91 | 3091 | * cgroup, if the css refcount is also 1, then there should |
81a6a5cd PM |
3092 | * be no outstanding references, so the subsystem is safe to |
3093 | * destroy. We scan across all subsystems rather than using | |
3094 | * the per-hierarchy linked list of mounted subsystems since | |
3095 | * we can be called via check_for_release() with no | |
3096 | * synchronization other than RCU, and the subsystem linked | |
3097 | * list isn't RCU-safe */ | |
3098 | int i; | |
aae8aab4 BB |
3099 | /* |
3100 | * We won't need to lock the subsys array, because the subsystems | |
3101 | * we're concerned about aren't going anywhere since our cgroup root | |
3102 | * has a reference on them. | |
3103 | */ | |
81a6a5cd PM |
3104 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
3105 | struct cgroup_subsys *ss = subsys[i]; | |
3106 | struct cgroup_subsys_state *css; | |
aae8aab4 BB |
3107 | /* Skip subsystems not present or not in this hierarchy */ |
3108 | if (ss == NULL || ss->root != cgrp->root) | |
81a6a5cd | 3109 | continue; |
bd89aabc | 3110 | css = cgrp->subsys[ss->subsys_id]; |
81a6a5cd PM |
3111 | /* When called from check_for_release() it's possible |
3112 | * that by this point the cgroup has been removed | |
3113 | * and the css deleted. But a false-positive doesn't | |
3114 | * matter, since it can only happen if the cgroup | |
3115 | * has been deleted and hence no longer needs the | |
3116 | * release agent to be called anyway. */ | |
e7c5ec91 | 3117 | if (css && (atomic_read(&css->refcnt) > 1)) |
81a6a5cd | 3118 | return 1; |
81a6a5cd PM |
3119 | } |
3120 | return 0; | |
3121 | } | |
3122 | ||
e7c5ec91 PM |
3123 | /* |
3124 | * Atomically mark all (or else none) of the cgroup's CSS objects as | |
3125 | * CSS_REMOVED. Return true on success, or false if the cgroup has | |
3126 | * busy subsystems. Call with cgroup_mutex held | |
3127 | */ | |
3128 | ||
3129 | static int cgroup_clear_css_refs(struct cgroup *cgrp) | |
3130 | { | |
3131 | struct cgroup_subsys *ss; | |
3132 | unsigned long flags; | |
3133 | bool failed = false; | |
3134 | local_irq_save(flags); | |
3135 | for_each_subsys(cgrp->root, ss) { | |
3136 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3137 | int refcnt; | |
804b3c28 | 3138 | while (1) { |
e7c5ec91 PM |
3139 | /* We can only remove a CSS with a refcnt==1 */ |
3140 | refcnt = atomic_read(&css->refcnt); | |
3141 | if (refcnt > 1) { | |
3142 | failed = true; | |
3143 | goto done; | |
3144 | } | |
3145 | BUG_ON(!refcnt); | |
3146 | /* | |
3147 | * Drop the refcnt to 0 while we check other | |
3148 | * subsystems. This will cause any racing | |
3149 | * css_tryget() to spin until we set the | |
3150 | * CSS_REMOVED bits or abort | |
3151 | */ | |
804b3c28 PM |
3152 | if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt) |
3153 | break; | |
3154 | cpu_relax(); | |
3155 | } | |
e7c5ec91 PM |
3156 | } |
3157 | done: | |
3158 | for_each_subsys(cgrp->root, ss) { | |
3159 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3160 | if (failed) { | |
3161 | /* | |
3162 | * Restore old refcnt if we previously managed | |
3163 | * to clear it from 1 to 0 | |
3164 | */ | |
3165 | if (!atomic_read(&css->refcnt)) | |
3166 | atomic_set(&css->refcnt, 1); | |
3167 | } else { | |
3168 | /* Commit the fact that the CSS is removed */ | |
3169 | set_bit(CSS_REMOVED, &css->flags); | |
3170 | } | |
3171 | } | |
3172 | local_irq_restore(flags); | |
3173 | return !failed; | |
3174 | } | |
3175 | ||
ddbcc7e8 PM |
3176 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) |
3177 | { | |
bd89aabc | 3178 | struct cgroup *cgrp = dentry->d_fsdata; |
ddbcc7e8 PM |
3179 | struct dentry *d; |
3180 | struct cgroup *parent; | |
ec64f515 KH |
3181 | DEFINE_WAIT(wait); |
3182 | int ret; | |
ddbcc7e8 PM |
3183 | |
3184 | /* the vfs holds both inode->i_mutex already */ | |
ec64f515 | 3185 | again: |
ddbcc7e8 | 3186 | mutex_lock(&cgroup_mutex); |
bd89aabc | 3187 | if (atomic_read(&cgrp->count) != 0) { |
ddbcc7e8 PM |
3188 | mutex_unlock(&cgroup_mutex); |
3189 | return -EBUSY; | |
3190 | } | |
bd89aabc | 3191 | if (!list_empty(&cgrp->children)) { |
ddbcc7e8 PM |
3192 | mutex_unlock(&cgroup_mutex); |
3193 | return -EBUSY; | |
3194 | } | |
3fa59dfb | 3195 | mutex_unlock(&cgroup_mutex); |
a043e3b2 | 3196 | |
88703267 KH |
3197 | /* |
3198 | * In general, subsystem has no css->refcnt after pre_destroy(). But | |
3199 | * in racy cases, subsystem may have to get css->refcnt after | |
3200 | * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes | |
3201 | * make rmdir return -EBUSY too often. To avoid that, we use waitqueue | |
3202 | * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir | |
3203 | * and subsystem's reference count handling. Please see css_get/put | |
3204 | * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation. | |
3205 | */ | |
3206 | set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
3207 | ||
4fca88c8 | 3208 | /* |
a043e3b2 LZ |
3209 | * Call pre_destroy handlers of subsys. Notify subsystems |
3210 | * that rmdir() request comes. | |
4fca88c8 | 3211 | */ |
ec64f515 | 3212 | ret = cgroup_call_pre_destroy(cgrp); |
88703267 KH |
3213 | if (ret) { |
3214 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
ec64f515 | 3215 | return ret; |
88703267 | 3216 | } |
ddbcc7e8 | 3217 | |
3fa59dfb KH |
3218 | mutex_lock(&cgroup_mutex); |
3219 | parent = cgrp->parent; | |
ec64f515 | 3220 | if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) { |
88703267 | 3221 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); |
ddbcc7e8 PM |
3222 | mutex_unlock(&cgroup_mutex); |
3223 | return -EBUSY; | |
3224 | } | |
ec64f515 | 3225 | prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE); |
ec64f515 KH |
3226 | if (!cgroup_clear_css_refs(cgrp)) { |
3227 | mutex_unlock(&cgroup_mutex); | |
88703267 KH |
3228 | /* |
3229 | * Because someone may call cgroup_wakeup_rmdir_waiter() before | |
3230 | * prepare_to_wait(), we need to check this flag. | |
3231 | */ | |
3232 | if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)) | |
3233 | schedule(); | |
ec64f515 KH |
3234 | finish_wait(&cgroup_rmdir_waitq, &wait); |
3235 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
3236 | if (signal_pending(current)) | |
3237 | return -EINTR; | |
3238 | goto again; | |
3239 | } | |
3240 | /* NO css_tryget() can success after here. */ | |
3241 | finish_wait(&cgroup_rmdir_waitq, &wait); | |
3242 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
ddbcc7e8 | 3243 | |
81a6a5cd | 3244 | spin_lock(&release_list_lock); |
bd89aabc PM |
3245 | set_bit(CGRP_REMOVED, &cgrp->flags); |
3246 | if (!list_empty(&cgrp->release_list)) | |
3247 | list_del(&cgrp->release_list); | |
81a6a5cd | 3248 | spin_unlock(&release_list_lock); |
999cd8a4 PM |
3249 | |
3250 | cgroup_lock_hierarchy(cgrp->root); | |
3251 | /* delete this cgroup from parent->children */ | |
bd89aabc | 3252 | list_del(&cgrp->sibling); |
999cd8a4 PM |
3253 | cgroup_unlock_hierarchy(cgrp->root); |
3254 | ||
bd89aabc PM |
3255 | spin_lock(&cgrp->dentry->d_lock); |
3256 | d = dget(cgrp->dentry); | |
ddbcc7e8 PM |
3257 | spin_unlock(&d->d_lock); |
3258 | ||
3259 | cgroup_d_remove_dir(d); | |
3260 | dput(d); | |
ddbcc7e8 | 3261 | |
bd89aabc | 3262 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd PM |
3263 | check_for_release(parent); |
3264 | ||
ddbcc7e8 | 3265 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
3266 | return 0; |
3267 | } | |
3268 | ||
06a11920 | 3269 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 3270 | { |
ddbcc7e8 | 3271 | struct cgroup_subsys_state *css; |
cfe36bde DC |
3272 | |
3273 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 PM |
3274 | |
3275 | /* Create the top cgroup state for this subsystem */ | |
33a68ac1 | 3276 | list_add(&ss->sibling, &rootnode.subsys_list); |
ddbcc7e8 PM |
3277 | ss->root = &rootnode; |
3278 | css = ss->create(ss, dummytop); | |
3279 | /* We don't handle early failures gracefully */ | |
3280 | BUG_ON(IS_ERR(css)); | |
3281 | init_cgroup_css(css, ss, dummytop); | |
3282 | ||
e8d55fde | 3283 | /* Update the init_css_set to contain a subsys |
817929ec | 3284 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
3285 | * newly registered, all tasks and hence the |
3286 | * init_css_set is in the subsystem's top cgroup. */ | |
3287 | init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id]; | |
ddbcc7e8 PM |
3288 | |
3289 | need_forkexit_callback |= ss->fork || ss->exit; | |
3290 | ||
e8d55fde LZ |
3291 | /* At system boot, before all subsystems have been |
3292 | * registered, no tasks have been forked, so we don't | |
3293 | * need to invoke fork callbacks here. */ | |
3294 | BUG_ON(!list_empty(&init_task.tasks)); | |
3295 | ||
999cd8a4 | 3296 | mutex_init(&ss->hierarchy_mutex); |
cfebe563 | 3297 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); |
ddbcc7e8 | 3298 | ss->active = 1; |
e6a1105b BB |
3299 | |
3300 | /* this function shouldn't be used with modular subsystems, since they | |
3301 | * need to register a subsys_id, among other things */ | |
3302 | BUG_ON(ss->module); | |
3303 | } | |
3304 | ||
3305 | /** | |
3306 | * cgroup_load_subsys: load and register a modular subsystem at runtime | |
3307 | * @ss: the subsystem to load | |
3308 | * | |
3309 | * This function should be called in a modular subsystem's initcall. If the | |
3310 | * subsytem is built as a module, it will be assigned a new subsys_id and set | |
3311 | * up for use. If the subsystem is built-in anyway, work is delegated to the | |
3312 | * simpler cgroup_init_subsys. | |
3313 | */ | |
3314 | int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss) | |
3315 | { | |
3316 | int i; | |
3317 | struct cgroup_subsys_state *css; | |
3318 | ||
3319 | /* check name and function validity */ | |
3320 | if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN || | |
3321 | ss->create == NULL || ss->destroy == NULL) | |
3322 | return -EINVAL; | |
3323 | ||
3324 | /* | |
3325 | * we don't support callbacks in modular subsystems. this check is | |
3326 | * before the ss->module check for consistency; a subsystem that could | |
3327 | * be a module should still have no callbacks even if the user isn't | |
3328 | * compiling it as one. | |
3329 | */ | |
3330 | if (ss->fork || ss->exit) | |
3331 | return -EINVAL; | |
3332 | ||
3333 | /* | |
3334 | * an optionally modular subsystem is built-in: we want to do nothing, | |
3335 | * since cgroup_init_subsys will have already taken care of it. | |
3336 | */ | |
3337 | if (ss->module == NULL) { | |
3338 | /* a few sanity checks */ | |
3339 | BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT); | |
3340 | BUG_ON(subsys[ss->subsys_id] != ss); | |
3341 | return 0; | |
3342 | } | |
3343 | ||
3344 | /* | |
3345 | * need to register a subsys id before anything else - for example, | |
3346 | * init_cgroup_css needs it. | |
3347 | */ | |
3348 | mutex_lock(&cgroup_mutex); | |
3349 | /* find the first empty slot in the array */ | |
3350 | for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) { | |
3351 | if (subsys[i] == NULL) | |
3352 | break; | |
3353 | } | |
3354 | if (i == CGROUP_SUBSYS_COUNT) { | |
3355 | /* maximum number of subsystems already registered! */ | |
3356 | mutex_unlock(&cgroup_mutex); | |
3357 | return -EBUSY; | |
3358 | } | |
3359 | /* assign ourselves the subsys_id */ | |
3360 | ss->subsys_id = i; | |
3361 | subsys[i] = ss; | |
3362 | ||
3363 | /* | |
3364 | * no ss->create seems to need anything important in the ss struct, so | |
3365 | * this can happen first (i.e. before the rootnode attachment). | |
3366 | */ | |
3367 | css = ss->create(ss, dummytop); | |
3368 | if (IS_ERR(css)) { | |
3369 | /* failure case - need to deassign the subsys[] slot. */ | |
3370 | subsys[i] = NULL; | |
3371 | mutex_unlock(&cgroup_mutex); | |
3372 | return PTR_ERR(css); | |
3373 | } | |
3374 | ||
3375 | list_add(&ss->sibling, &rootnode.subsys_list); | |
3376 | ss->root = &rootnode; | |
3377 | ||
3378 | /* our new subsystem will be attached to the dummy hierarchy. */ | |
3379 | init_cgroup_css(css, ss, dummytop); | |
3380 | /* init_idr must be after init_cgroup_css because it sets css->id. */ | |
3381 | if (ss->use_id) { | |
3382 | int ret = cgroup_init_idr(ss, css); | |
3383 | if (ret) { | |
3384 | dummytop->subsys[ss->subsys_id] = NULL; | |
3385 | ss->destroy(ss, dummytop); | |
3386 | subsys[i] = NULL; | |
3387 | mutex_unlock(&cgroup_mutex); | |
3388 | return ret; | |
3389 | } | |
3390 | } | |
3391 | ||
3392 | /* | |
3393 | * Now we need to entangle the css into the existing css_sets. unlike | |
3394 | * in cgroup_init_subsys, there are now multiple css_sets, so each one | |
3395 | * will need a new pointer to it; done by iterating the css_set_table. | |
3396 | * furthermore, modifying the existing css_sets will corrupt the hash | |
3397 | * table state, so each changed css_set will need its hash recomputed. | |
3398 | * this is all done under the css_set_lock. | |
3399 | */ | |
3400 | write_lock(&css_set_lock); | |
3401 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { | |
3402 | struct css_set *cg; | |
3403 | struct hlist_node *node, *tmp; | |
3404 | struct hlist_head *bucket = &css_set_table[i], *new_bucket; | |
3405 | ||
3406 | hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) { | |
3407 | /* skip entries that we already rehashed */ | |
3408 | if (cg->subsys[ss->subsys_id]) | |
3409 | continue; | |
3410 | /* remove existing entry */ | |
3411 | hlist_del(&cg->hlist); | |
3412 | /* set new value */ | |
3413 | cg->subsys[ss->subsys_id] = css; | |
3414 | /* recompute hash and restore entry */ | |
3415 | new_bucket = css_set_hash(cg->subsys); | |
3416 | hlist_add_head(&cg->hlist, new_bucket); | |
3417 | } | |
3418 | } | |
3419 | write_unlock(&css_set_lock); | |
3420 | ||
3421 | mutex_init(&ss->hierarchy_mutex); | |
3422 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); | |
3423 | ss->active = 1; | |
3424 | ||
3425 | /* | |
3426 | * pin the subsystem's module so it doesn't go away. this shouldn't | |
3427 | * fail, since the module's initcall calls us. | |
3428 | * TODO: with module unloading, move this elsewhere | |
3429 | */ | |
3430 | BUG_ON(!try_module_get(ss->module)); | |
3431 | ||
3432 | /* success! */ | |
3433 | mutex_unlock(&cgroup_mutex); | |
3434 | return 0; | |
ddbcc7e8 | 3435 | } |
e6a1105b | 3436 | EXPORT_SYMBOL_GPL(cgroup_load_subsys); |
ddbcc7e8 PM |
3437 | |
3438 | /** | |
a043e3b2 LZ |
3439 | * cgroup_init_early - cgroup initialization at system boot |
3440 | * | |
3441 | * Initialize cgroups at system boot, and initialize any | |
3442 | * subsystems that request early init. | |
ddbcc7e8 PM |
3443 | */ |
3444 | int __init cgroup_init_early(void) | |
3445 | { | |
3446 | int i; | |
146aa1bd | 3447 | atomic_set(&init_css_set.refcount, 1); |
817929ec PM |
3448 | INIT_LIST_HEAD(&init_css_set.cg_links); |
3449 | INIT_LIST_HEAD(&init_css_set.tasks); | |
472b1053 | 3450 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 3451 | css_set_count = 1; |
ddbcc7e8 | 3452 | init_cgroup_root(&rootnode); |
817929ec PM |
3453 | root_count = 1; |
3454 | init_task.cgroups = &init_css_set; | |
3455 | ||
3456 | init_css_set_link.cg = &init_css_set; | |
7717f7ba | 3457 | init_css_set_link.cgrp = dummytop; |
bd89aabc | 3458 | list_add(&init_css_set_link.cgrp_link_list, |
817929ec PM |
3459 | &rootnode.top_cgroup.css_sets); |
3460 | list_add(&init_css_set_link.cg_link_list, | |
3461 | &init_css_set.cg_links); | |
ddbcc7e8 | 3462 | |
472b1053 LZ |
3463 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) |
3464 | INIT_HLIST_HEAD(&css_set_table[i]); | |
3465 | ||
aae8aab4 BB |
3466 | /* at bootup time, we don't worry about modular subsystems */ |
3467 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
ddbcc7e8 PM |
3468 | struct cgroup_subsys *ss = subsys[i]; |
3469 | ||
3470 | BUG_ON(!ss->name); | |
3471 | BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); | |
3472 | BUG_ON(!ss->create); | |
3473 | BUG_ON(!ss->destroy); | |
3474 | if (ss->subsys_id != i) { | |
cfe36bde | 3475 | printk(KERN_ERR "cgroup: Subsys %s id == %d\n", |
ddbcc7e8 PM |
3476 | ss->name, ss->subsys_id); |
3477 | BUG(); | |
3478 | } | |
3479 | ||
3480 | if (ss->early_init) | |
3481 | cgroup_init_subsys(ss); | |
3482 | } | |
3483 | return 0; | |
3484 | } | |
3485 | ||
3486 | /** | |
a043e3b2 LZ |
3487 | * cgroup_init - cgroup initialization |
3488 | * | |
3489 | * Register cgroup filesystem and /proc file, and initialize | |
3490 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
3491 | */ |
3492 | int __init cgroup_init(void) | |
3493 | { | |
3494 | int err; | |
3495 | int i; | |
472b1053 | 3496 | struct hlist_head *hhead; |
a424316c PM |
3497 | |
3498 | err = bdi_init(&cgroup_backing_dev_info); | |
3499 | if (err) | |
3500 | return err; | |
ddbcc7e8 | 3501 | |
aae8aab4 BB |
3502 | /* at bootup time, we don't worry about modular subsystems */ |
3503 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
ddbcc7e8 PM |
3504 | struct cgroup_subsys *ss = subsys[i]; |
3505 | if (!ss->early_init) | |
3506 | cgroup_init_subsys(ss); | |
38460b48 | 3507 | if (ss->use_id) |
e6a1105b | 3508 | cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]); |
ddbcc7e8 PM |
3509 | } |
3510 | ||
472b1053 LZ |
3511 | /* Add init_css_set to the hash table */ |
3512 | hhead = css_set_hash(init_css_set.subsys); | |
3513 | hlist_add_head(&init_css_set.hlist, hhead); | |
2c6ab6d2 | 3514 | BUG_ON(!init_root_id(&rootnode)); |
ddbcc7e8 PM |
3515 | err = register_filesystem(&cgroup_fs_type); |
3516 | if (err < 0) | |
3517 | goto out; | |
3518 | ||
46ae220b | 3519 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
a424316c | 3520 | |
ddbcc7e8 | 3521 | out: |
a424316c PM |
3522 | if (err) |
3523 | bdi_destroy(&cgroup_backing_dev_info); | |
3524 | ||
ddbcc7e8 PM |
3525 | return err; |
3526 | } | |
b4f48b63 | 3527 | |
a424316c PM |
3528 | /* |
3529 | * proc_cgroup_show() | |
3530 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
3531 | * - Used for /proc/<pid>/cgroup. | |
3532 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
3533 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 3534 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
3535 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
3536 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
3537 | * cgroup to top_cgroup. | |
3538 | */ | |
3539 | ||
3540 | /* TODO: Use a proper seq_file iterator */ | |
3541 | static int proc_cgroup_show(struct seq_file *m, void *v) | |
3542 | { | |
3543 | struct pid *pid; | |
3544 | struct task_struct *tsk; | |
3545 | char *buf; | |
3546 | int retval; | |
3547 | struct cgroupfs_root *root; | |
3548 | ||
3549 | retval = -ENOMEM; | |
3550 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
3551 | if (!buf) | |
3552 | goto out; | |
3553 | ||
3554 | retval = -ESRCH; | |
3555 | pid = m->private; | |
3556 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
3557 | if (!tsk) | |
3558 | goto out_free; | |
3559 | ||
3560 | retval = 0; | |
3561 | ||
3562 | mutex_lock(&cgroup_mutex); | |
3563 | ||
e5f6a860 | 3564 | for_each_active_root(root) { |
a424316c | 3565 | struct cgroup_subsys *ss; |
bd89aabc | 3566 | struct cgroup *cgrp; |
a424316c PM |
3567 | int count = 0; |
3568 | ||
2c6ab6d2 | 3569 | seq_printf(m, "%d:", root->hierarchy_id); |
a424316c PM |
3570 | for_each_subsys(root, ss) |
3571 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
3572 | if (strlen(root->name)) |
3573 | seq_printf(m, "%sname=%s", count ? "," : "", | |
3574 | root->name); | |
a424316c | 3575 | seq_putc(m, ':'); |
7717f7ba | 3576 | cgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 3577 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); |
a424316c PM |
3578 | if (retval < 0) |
3579 | goto out_unlock; | |
3580 | seq_puts(m, buf); | |
3581 | seq_putc(m, '\n'); | |
3582 | } | |
3583 | ||
3584 | out_unlock: | |
3585 | mutex_unlock(&cgroup_mutex); | |
3586 | put_task_struct(tsk); | |
3587 | out_free: | |
3588 | kfree(buf); | |
3589 | out: | |
3590 | return retval; | |
3591 | } | |
3592 | ||
3593 | static int cgroup_open(struct inode *inode, struct file *file) | |
3594 | { | |
3595 | struct pid *pid = PROC_I(inode)->pid; | |
3596 | return single_open(file, proc_cgroup_show, pid); | |
3597 | } | |
3598 | ||
828c0950 | 3599 | const struct file_operations proc_cgroup_operations = { |
a424316c PM |
3600 | .open = cgroup_open, |
3601 | .read = seq_read, | |
3602 | .llseek = seq_lseek, | |
3603 | .release = single_release, | |
3604 | }; | |
3605 | ||
3606 | /* Display information about each subsystem and each hierarchy */ | |
3607 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
3608 | { | |
3609 | int i; | |
a424316c | 3610 | |
8bab8dde | 3611 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
3612 | /* |
3613 | * ideally we don't want subsystems moving around while we do this. | |
3614 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
3615 | * subsys/hierarchy state. | |
3616 | */ | |
a424316c | 3617 | mutex_lock(&cgroup_mutex); |
a424316c PM |
3618 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
3619 | struct cgroup_subsys *ss = subsys[i]; | |
aae8aab4 BB |
3620 | if (ss == NULL) |
3621 | continue; | |
2c6ab6d2 PM |
3622 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3623 | ss->name, ss->root->hierarchy_id, | |
8bab8dde | 3624 | ss->root->number_of_cgroups, !ss->disabled); |
a424316c PM |
3625 | } |
3626 | mutex_unlock(&cgroup_mutex); | |
3627 | return 0; | |
3628 | } | |
3629 | ||
3630 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
3631 | { | |
9dce07f1 | 3632 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
3633 | } |
3634 | ||
828c0950 | 3635 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
3636 | .open = cgroupstats_open, |
3637 | .read = seq_read, | |
3638 | .llseek = seq_lseek, | |
3639 | .release = single_release, | |
3640 | }; | |
3641 | ||
b4f48b63 PM |
3642 | /** |
3643 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 3644 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
3645 | * |
3646 | * Description: A task inherits its parent's cgroup at fork(). | |
3647 | * | |
3648 | * A pointer to the shared css_set was automatically copied in | |
3649 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
3650 | * it was not made under the protection of RCU or cgroup_mutex, so | |
956db3ca | 3651 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might |
817929ec PM |
3652 | * have already changed current->cgroups, allowing the previously |
3653 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
3654 | * |
3655 | * At the point that cgroup_fork() is called, 'current' is the parent | |
3656 | * task, and the passed argument 'child' points to the child task. | |
3657 | */ | |
3658 | void cgroup_fork(struct task_struct *child) | |
3659 | { | |
817929ec PM |
3660 | task_lock(current); |
3661 | child->cgroups = current->cgroups; | |
3662 | get_css_set(child->cgroups); | |
3663 | task_unlock(current); | |
3664 | INIT_LIST_HEAD(&child->cg_list); | |
b4f48b63 PM |
3665 | } |
3666 | ||
3667 | /** | |
a043e3b2 LZ |
3668 | * cgroup_fork_callbacks - run fork callbacks |
3669 | * @child: the new task | |
3670 | * | |
3671 | * Called on a new task very soon before adding it to the | |
3672 | * tasklist. No need to take any locks since no-one can | |
3673 | * be operating on this task. | |
b4f48b63 PM |
3674 | */ |
3675 | void cgroup_fork_callbacks(struct task_struct *child) | |
3676 | { | |
3677 | if (need_forkexit_callback) { | |
3678 | int i; | |
aae8aab4 BB |
3679 | /* |
3680 | * forkexit callbacks are only supported for builtin | |
3681 | * subsystems, and the builtin section of the subsys array is | |
3682 | * immutable, so we don't need to lock the subsys array here. | |
3683 | */ | |
3684 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
b4f48b63 PM |
3685 | struct cgroup_subsys *ss = subsys[i]; |
3686 | if (ss->fork) | |
3687 | ss->fork(ss, child); | |
3688 | } | |
3689 | } | |
3690 | } | |
3691 | ||
817929ec | 3692 | /** |
a043e3b2 LZ |
3693 | * cgroup_post_fork - called on a new task after adding it to the task list |
3694 | * @child: the task in question | |
3695 | * | |
3696 | * Adds the task to the list running through its css_set if necessary. | |
3697 | * Has to be after the task is visible on the task list in case we race | |
3698 | * with the first call to cgroup_iter_start() - to guarantee that the | |
3699 | * new task ends up on its list. | |
3700 | */ | |
817929ec PM |
3701 | void cgroup_post_fork(struct task_struct *child) |
3702 | { | |
3703 | if (use_task_css_set_links) { | |
3704 | write_lock(&css_set_lock); | |
b12b533f | 3705 | task_lock(child); |
817929ec PM |
3706 | if (list_empty(&child->cg_list)) |
3707 | list_add(&child->cg_list, &child->cgroups->tasks); | |
b12b533f | 3708 | task_unlock(child); |
817929ec PM |
3709 | write_unlock(&css_set_lock); |
3710 | } | |
3711 | } | |
b4f48b63 PM |
3712 | /** |
3713 | * cgroup_exit - detach cgroup from exiting task | |
3714 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 3715 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
3716 | * |
3717 | * Description: Detach cgroup from @tsk and release it. | |
3718 | * | |
3719 | * Note that cgroups marked notify_on_release force every task in | |
3720 | * them to take the global cgroup_mutex mutex when exiting. | |
3721 | * This could impact scaling on very large systems. Be reluctant to | |
3722 | * use notify_on_release cgroups where very high task exit scaling | |
3723 | * is required on large systems. | |
3724 | * | |
3725 | * the_top_cgroup_hack: | |
3726 | * | |
3727 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
3728 | * | |
3729 | * We call cgroup_exit() while the task is still competent to | |
3730 | * handle notify_on_release(), then leave the task attached to the | |
3731 | * root cgroup in each hierarchy for the remainder of its exit. | |
3732 | * | |
3733 | * To do this properly, we would increment the reference count on | |
3734 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
3735 | * code we would add a second cgroup function call, to drop that | |
3736 | * reference. This would just create an unnecessary hot spot on | |
3737 | * the top_cgroup reference count, to no avail. | |
3738 | * | |
3739 | * Normally, holding a reference to a cgroup without bumping its | |
3740 | * count is unsafe. The cgroup could go away, or someone could | |
3741 | * attach us to a different cgroup, decrementing the count on | |
3742 | * the first cgroup that we never incremented. But in this case, | |
3743 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
3744 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
3745 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
3746 | */ |
3747 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
3748 | { | |
3749 | int i; | |
817929ec | 3750 | struct css_set *cg; |
b4f48b63 PM |
3751 | |
3752 | if (run_callbacks && need_forkexit_callback) { | |
aae8aab4 BB |
3753 | /* |
3754 | * modular subsystems can't use callbacks, so no need to lock | |
3755 | * the subsys array | |
3756 | */ | |
3757 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
b4f48b63 PM |
3758 | struct cgroup_subsys *ss = subsys[i]; |
3759 | if (ss->exit) | |
3760 | ss->exit(ss, tsk); | |
3761 | } | |
3762 | } | |
817929ec PM |
3763 | |
3764 | /* | |
3765 | * Unlink from the css_set task list if necessary. | |
3766 | * Optimistically check cg_list before taking | |
3767 | * css_set_lock | |
3768 | */ | |
3769 | if (!list_empty(&tsk->cg_list)) { | |
3770 | write_lock(&css_set_lock); | |
3771 | if (!list_empty(&tsk->cg_list)) | |
3772 | list_del(&tsk->cg_list); | |
3773 | write_unlock(&css_set_lock); | |
3774 | } | |
3775 | ||
b4f48b63 PM |
3776 | /* Reassign the task to the init_css_set. */ |
3777 | task_lock(tsk); | |
817929ec PM |
3778 | cg = tsk->cgroups; |
3779 | tsk->cgroups = &init_css_set; | |
b4f48b63 | 3780 | task_unlock(tsk); |
817929ec | 3781 | if (cg) |
81a6a5cd | 3782 | put_css_set_taskexit(cg); |
b4f48b63 | 3783 | } |
697f4161 PM |
3784 | |
3785 | /** | |
a043e3b2 LZ |
3786 | * cgroup_clone - clone the cgroup the given subsystem is attached to |
3787 | * @tsk: the task to be moved | |
3788 | * @subsys: the given subsystem | |
e885dcde | 3789 | * @nodename: the name for the new cgroup |
a043e3b2 LZ |
3790 | * |
3791 | * Duplicate the current cgroup in the hierarchy that the given | |
3792 | * subsystem is attached to, and move this task into the new | |
3793 | * child. | |
697f4161 | 3794 | */ |
e885dcde SH |
3795 | int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys, |
3796 | char *nodename) | |
697f4161 PM |
3797 | { |
3798 | struct dentry *dentry; | |
3799 | int ret = 0; | |
697f4161 PM |
3800 | struct cgroup *parent, *child; |
3801 | struct inode *inode; | |
3802 | struct css_set *cg; | |
3803 | struct cgroupfs_root *root; | |
3804 | struct cgroup_subsys *ss; | |
3805 | ||
3806 | /* We shouldn't be called by an unregistered subsystem */ | |
3807 | BUG_ON(!subsys->active); | |
3808 | ||
3809 | /* First figure out what hierarchy and cgroup we're dealing | |
3810 | * with, and pin them so we can drop cgroup_mutex */ | |
3811 | mutex_lock(&cgroup_mutex); | |
3812 | again: | |
3813 | root = subsys->root; | |
3814 | if (root == &rootnode) { | |
697f4161 PM |
3815 | mutex_unlock(&cgroup_mutex); |
3816 | return 0; | |
3817 | } | |
697f4161 | 3818 | |
697f4161 | 3819 | /* Pin the hierarchy */ |
1404f065 | 3820 | if (!atomic_inc_not_zero(&root->sb->s_active)) { |
7b574b7b LZ |
3821 | /* We race with the final deactivate_super() */ |
3822 | mutex_unlock(&cgroup_mutex); | |
3823 | return 0; | |
3824 | } | |
697f4161 | 3825 | |
817929ec | 3826 | /* Keep the cgroup alive */ |
1404f065 LZ |
3827 | task_lock(tsk); |
3828 | parent = task_cgroup(tsk, subsys->subsys_id); | |
3829 | cg = tsk->cgroups; | |
817929ec | 3830 | get_css_set(cg); |
104cbd55 | 3831 | task_unlock(tsk); |
1404f065 | 3832 | |
697f4161 PM |
3833 | mutex_unlock(&cgroup_mutex); |
3834 | ||
3835 | /* Now do the VFS work to create a cgroup */ | |
3836 | inode = parent->dentry->d_inode; | |
3837 | ||
3838 | /* Hold the parent directory mutex across this operation to | |
3839 | * stop anyone else deleting the new cgroup */ | |
3840 | mutex_lock(&inode->i_mutex); | |
3841 | dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename)); | |
3842 | if (IS_ERR(dentry)) { | |
3843 | printk(KERN_INFO | |
cfe36bde | 3844 | "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename, |
697f4161 PM |
3845 | PTR_ERR(dentry)); |
3846 | ret = PTR_ERR(dentry); | |
3847 | goto out_release; | |
3848 | } | |
3849 | ||
3850 | /* Create the cgroup directory, which also creates the cgroup */ | |
75139b82 | 3851 | ret = vfs_mkdir(inode, dentry, 0755); |
bd89aabc | 3852 | child = __d_cgrp(dentry); |
697f4161 PM |
3853 | dput(dentry); |
3854 | if (ret) { | |
3855 | printk(KERN_INFO | |
3856 | "Failed to create cgroup %s: %d\n", nodename, | |
3857 | ret); | |
3858 | goto out_release; | |
3859 | } | |
3860 | ||
697f4161 PM |
3861 | /* The cgroup now exists. Retake cgroup_mutex and check |
3862 | * that we're still in the same state that we thought we | |
3863 | * were. */ | |
3864 | mutex_lock(&cgroup_mutex); | |
3865 | if ((root != subsys->root) || | |
3866 | (parent != task_cgroup(tsk, subsys->subsys_id))) { | |
3867 | /* Aargh, we raced ... */ | |
3868 | mutex_unlock(&inode->i_mutex); | |
817929ec | 3869 | put_css_set(cg); |
697f4161 | 3870 | |
1404f065 | 3871 | deactivate_super(root->sb); |
697f4161 PM |
3872 | /* The cgroup is still accessible in the VFS, but |
3873 | * we're not going to try to rmdir() it at this | |
3874 | * point. */ | |
3875 | printk(KERN_INFO | |
3876 | "Race in cgroup_clone() - leaking cgroup %s\n", | |
3877 | nodename); | |
3878 | goto again; | |
3879 | } | |
3880 | ||
3881 | /* do any required auto-setup */ | |
3882 | for_each_subsys(root, ss) { | |
3883 | if (ss->post_clone) | |
3884 | ss->post_clone(ss, child); | |
3885 | } | |
3886 | ||
3887 | /* All seems fine. Finish by moving the task into the new cgroup */ | |
956db3ca | 3888 | ret = cgroup_attach_task(child, tsk); |
697f4161 PM |
3889 | mutex_unlock(&cgroup_mutex); |
3890 | ||
3891 | out_release: | |
3892 | mutex_unlock(&inode->i_mutex); | |
81a6a5cd PM |
3893 | |
3894 | mutex_lock(&cgroup_mutex); | |
817929ec | 3895 | put_css_set(cg); |
81a6a5cd | 3896 | mutex_unlock(&cgroup_mutex); |
1404f065 | 3897 | deactivate_super(root->sb); |
697f4161 PM |
3898 | return ret; |
3899 | } | |
3900 | ||
a043e3b2 | 3901 | /** |
313e924c | 3902 | * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp |
a043e3b2 | 3903 | * @cgrp: the cgroup in question |
313e924c | 3904 | * @task: the task in question |
a043e3b2 | 3905 | * |
313e924c GN |
3906 | * See if @cgrp is a descendant of @task's cgroup in the appropriate |
3907 | * hierarchy. | |
697f4161 PM |
3908 | * |
3909 | * If we are sending in dummytop, then presumably we are creating | |
3910 | * the top cgroup in the subsystem. | |
3911 | * | |
3912 | * Called only by the ns (nsproxy) cgroup. | |
3913 | */ | |
313e924c | 3914 | int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task) |
697f4161 PM |
3915 | { |
3916 | int ret; | |
3917 | struct cgroup *target; | |
697f4161 | 3918 | |
bd89aabc | 3919 | if (cgrp == dummytop) |
697f4161 PM |
3920 | return 1; |
3921 | ||
7717f7ba | 3922 | target = task_cgroup_from_root(task, cgrp->root); |
bd89aabc PM |
3923 | while (cgrp != target && cgrp!= cgrp->top_cgroup) |
3924 | cgrp = cgrp->parent; | |
3925 | ret = (cgrp == target); | |
697f4161 PM |
3926 | return ret; |
3927 | } | |
81a6a5cd | 3928 | |
bd89aabc | 3929 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd PM |
3930 | { |
3931 | /* All of these checks rely on RCU to keep the cgroup | |
3932 | * structure alive */ | |
bd89aabc PM |
3933 | if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count) |
3934 | && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) { | |
81a6a5cd PM |
3935 | /* Control Group is currently removeable. If it's not |
3936 | * already queued for a userspace notification, queue | |
3937 | * it now */ | |
3938 | int need_schedule_work = 0; | |
3939 | spin_lock(&release_list_lock); | |
bd89aabc PM |
3940 | if (!cgroup_is_removed(cgrp) && |
3941 | list_empty(&cgrp->release_list)) { | |
3942 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
3943 | need_schedule_work = 1; |
3944 | } | |
3945 | spin_unlock(&release_list_lock); | |
3946 | if (need_schedule_work) | |
3947 | schedule_work(&release_agent_work); | |
3948 | } | |
3949 | } | |
3950 | ||
d7b9fff7 DN |
3951 | /* Caller must verify that the css is not for root cgroup */ |
3952 | void __css_put(struct cgroup_subsys_state *css, int count) | |
81a6a5cd | 3953 | { |
bd89aabc | 3954 | struct cgroup *cgrp = css->cgroup; |
3dece834 | 3955 | int val; |
81a6a5cd | 3956 | rcu_read_lock(); |
d7b9fff7 | 3957 | val = atomic_sub_return(count, &css->refcnt); |
3dece834 | 3958 | if (val == 1) { |
ec64f515 KH |
3959 | if (notify_on_release(cgrp)) { |
3960 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
3961 | check_for_release(cgrp); | |
3962 | } | |
88703267 | 3963 | cgroup_wakeup_rmdir_waiter(cgrp); |
81a6a5cd PM |
3964 | } |
3965 | rcu_read_unlock(); | |
3dece834 | 3966 | WARN_ON_ONCE(val < 1); |
81a6a5cd PM |
3967 | } |
3968 | ||
3969 | /* | |
3970 | * Notify userspace when a cgroup is released, by running the | |
3971 | * configured release agent with the name of the cgroup (path | |
3972 | * relative to the root of cgroup file system) as the argument. | |
3973 | * | |
3974 | * Most likely, this user command will try to rmdir this cgroup. | |
3975 | * | |
3976 | * This races with the possibility that some other task will be | |
3977 | * attached to this cgroup before it is removed, or that some other | |
3978 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
3979 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
3980 | * unused, and this cgroup will be reprieved from its death sentence, | |
3981 | * to continue to serve a useful existence. Next time it's released, | |
3982 | * we will get notified again, if it still has 'notify_on_release' set. | |
3983 | * | |
3984 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
3985 | * means only wait until the task is successfully execve()'d. The | |
3986 | * separate release agent task is forked by call_usermodehelper(), | |
3987 | * then control in this thread returns here, without waiting for the | |
3988 | * release agent task. We don't bother to wait because the caller of | |
3989 | * this routine has no use for the exit status of the release agent | |
3990 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 3991 | */ |
81a6a5cd PM |
3992 | static void cgroup_release_agent(struct work_struct *work) |
3993 | { | |
3994 | BUG_ON(work != &release_agent_work); | |
3995 | mutex_lock(&cgroup_mutex); | |
3996 | spin_lock(&release_list_lock); | |
3997 | while (!list_empty(&release_list)) { | |
3998 | char *argv[3], *envp[3]; | |
3999 | int i; | |
e788e066 | 4000 | char *pathbuf = NULL, *agentbuf = NULL; |
bd89aabc | 4001 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4002 | struct cgroup, |
4003 | release_list); | |
bd89aabc | 4004 | list_del_init(&cgrp->release_list); |
81a6a5cd PM |
4005 | spin_unlock(&release_list_lock); |
4006 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
e788e066 PM |
4007 | if (!pathbuf) |
4008 | goto continue_free; | |
4009 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
4010 | goto continue_free; | |
4011 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4012 | if (!agentbuf) | |
4013 | goto continue_free; | |
81a6a5cd PM |
4014 | |
4015 | i = 0; | |
e788e066 PM |
4016 | argv[i++] = agentbuf; |
4017 | argv[i++] = pathbuf; | |
81a6a5cd PM |
4018 | argv[i] = NULL; |
4019 | ||
4020 | i = 0; | |
4021 | /* minimal command environment */ | |
4022 | envp[i++] = "HOME=/"; | |
4023 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4024 | envp[i] = NULL; | |
4025 | ||
4026 | /* Drop the lock while we invoke the usermode helper, | |
4027 | * since the exec could involve hitting disk and hence | |
4028 | * be a slow process */ | |
4029 | mutex_unlock(&cgroup_mutex); | |
4030 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4031 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4032 | continue_free: |
4033 | kfree(pathbuf); | |
4034 | kfree(agentbuf); | |
81a6a5cd PM |
4035 | spin_lock(&release_list_lock); |
4036 | } | |
4037 | spin_unlock(&release_list_lock); | |
4038 | mutex_unlock(&cgroup_mutex); | |
4039 | } | |
8bab8dde PM |
4040 | |
4041 | static int __init cgroup_disable(char *str) | |
4042 | { | |
4043 | int i; | |
4044 | char *token; | |
4045 | ||
4046 | while ((token = strsep(&str, ",")) != NULL) { | |
4047 | if (!*token) | |
4048 | continue; | |
aae8aab4 BB |
4049 | /* |
4050 | * cgroup_disable, being at boot time, can't know about module | |
4051 | * subsystems, so we don't worry about them. | |
4052 | */ | |
4053 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
8bab8dde PM |
4054 | struct cgroup_subsys *ss = subsys[i]; |
4055 | ||
4056 | if (!strcmp(token, ss->name)) { | |
4057 | ss->disabled = 1; | |
4058 | printk(KERN_INFO "Disabling %s control group" | |
4059 | " subsystem\n", ss->name); | |
4060 | break; | |
4061 | } | |
4062 | } | |
4063 | } | |
4064 | return 1; | |
4065 | } | |
4066 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 KH |
4067 | |
4068 | /* | |
4069 | * Functons for CSS ID. | |
4070 | */ | |
4071 | ||
4072 | /* | |
4073 | *To get ID other than 0, this should be called when !cgroup_is_removed(). | |
4074 | */ | |
4075 | unsigned short css_id(struct cgroup_subsys_state *css) | |
4076 | { | |
4077 | struct css_id *cssid = rcu_dereference(css->id); | |
4078 | ||
4079 | if (cssid) | |
4080 | return cssid->id; | |
4081 | return 0; | |
4082 | } | |
4083 | ||
4084 | unsigned short css_depth(struct cgroup_subsys_state *css) | |
4085 | { | |
4086 | struct css_id *cssid = rcu_dereference(css->id); | |
4087 | ||
4088 | if (cssid) | |
4089 | return cssid->depth; | |
4090 | return 0; | |
4091 | } | |
4092 | ||
4093 | bool css_is_ancestor(struct cgroup_subsys_state *child, | |
0b7f569e | 4094 | const struct cgroup_subsys_state *root) |
38460b48 KH |
4095 | { |
4096 | struct css_id *child_id = rcu_dereference(child->id); | |
4097 | struct css_id *root_id = rcu_dereference(root->id); | |
4098 | ||
4099 | if (!child_id || !root_id || (child_id->depth < root_id->depth)) | |
4100 | return false; | |
4101 | return child_id->stack[root_id->depth] == root_id->id; | |
4102 | } | |
4103 | ||
4104 | static void __free_css_id_cb(struct rcu_head *head) | |
4105 | { | |
4106 | struct css_id *id; | |
4107 | ||
4108 | id = container_of(head, struct css_id, rcu_head); | |
4109 | kfree(id); | |
4110 | } | |
4111 | ||
4112 | void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css) | |
4113 | { | |
4114 | struct css_id *id = css->id; | |
4115 | /* When this is called before css_id initialization, id can be NULL */ | |
4116 | if (!id) | |
4117 | return; | |
4118 | ||
4119 | BUG_ON(!ss->use_id); | |
4120 | ||
4121 | rcu_assign_pointer(id->css, NULL); | |
4122 | rcu_assign_pointer(css->id, NULL); | |
4123 | spin_lock(&ss->id_lock); | |
4124 | idr_remove(&ss->idr, id->id); | |
4125 | spin_unlock(&ss->id_lock); | |
4126 | call_rcu(&id->rcu_head, __free_css_id_cb); | |
4127 | } | |
4128 | ||
4129 | /* | |
4130 | * This is called by init or create(). Then, calls to this function are | |
4131 | * always serialized (By cgroup_mutex() at create()). | |
4132 | */ | |
4133 | ||
4134 | static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth) | |
4135 | { | |
4136 | struct css_id *newid; | |
4137 | int myid, error, size; | |
4138 | ||
4139 | BUG_ON(!ss->use_id); | |
4140 | ||
4141 | size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1); | |
4142 | newid = kzalloc(size, GFP_KERNEL); | |
4143 | if (!newid) | |
4144 | return ERR_PTR(-ENOMEM); | |
4145 | /* get id */ | |
4146 | if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) { | |
4147 | error = -ENOMEM; | |
4148 | goto err_out; | |
4149 | } | |
4150 | spin_lock(&ss->id_lock); | |
4151 | /* Don't use 0. allocates an ID of 1-65535 */ | |
4152 | error = idr_get_new_above(&ss->idr, newid, 1, &myid); | |
4153 | spin_unlock(&ss->id_lock); | |
4154 | ||
4155 | /* Returns error when there are no free spaces for new ID.*/ | |
4156 | if (error) { | |
4157 | error = -ENOSPC; | |
4158 | goto err_out; | |
4159 | } | |
4160 | if (myid > CSS_ID_MAX) | |
4161 | goto remove_idr; | |
4162 | ||
4163 | newid->id = myid; | |
4164 | newid->depth = depth; | |
4165 | return newid; | |
4166 | remove_idr: | |
4167 | error = -ENOSPC; | |
4168 | spin_lock(&ss->id_lock); | |
4169 | idr_remove(&ss->idr, myid); | |
4170 | spin_unlock(&ss->id_lock); | |
4171 | err_out: | |
4172 | kfree(newid); | |
4173 | return ERR_PTR(error); | |
4174 | ||
4175 | } | |
4176 | ||
e6a1105b BB |
4177 | static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss, |
4178 | struct cgroup_subsys_state *rootcss) | |
38460b48 KH |
4179 | { |
4180 | struct css_id *newid; | |
38460b48 KH |
4181 | |
4182 | spin_lock_init(&ss->id_lock); | |
4183 | idr_init(&ss->idr); | |
4184 | ||
38460b48 KH |
4185 | newid = get_new_cssid(ss, 0); |
4186 | if (IS_ERR(newid)) | |
4187 | return PTR_ERR(newid); | |
4188 | ||
4189 | newid->stack[0] = newid->id; | |
4190 | newid->css = rootcss; | |
4191 | rootcss->id = newid; | |
4192 | return 0; | |
4193 | } | |
4194 | ||
4195 | static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent, | |
4196 | struct cgroup *child) | |
4197 | { | |
4198 | int subsys_id, i, depth = 0; | |
4199 | struct cgroup_subsys_state *parent_css, *child_css; | |
4200 | struct css_id *child_id, *parent_id = NULL; | |
4201 | ||
4202 | subsys_id = ss->subsys_id; | |
4203 | parent_css = parent->subsys[subsys_id]; | |
4204 | child_css = child->subsys[subsys_id]; | |
4205 | depth = css_depth(parent_css) + 1; | |
4206 | parent_id = parent_css->id; | |
4207 | ||
4208 | child_id = get_new_cssid(ss, depth); | |
4209 | if (IS_ERR(child_id)) | |
4210 | return PTR_ERR(child_id); | |
4211 | ||
4212 | for (i = 0; i < depth; i++) | |
4213 | child_id->stack[i] = parent_id->stack[i]; | |
4214 | child_id->stack[depth] = child_id->id; | |
4215 | /* | |
4216 | * child_id->css pointer will be set after this cgroup is available | |
4217 | * see cgroup_populate_dir() | |
4218 | */ | |
4219 | rcu_assign_pointer(child_css->id, child_id); | |
4220 | ||
4221 | return 0; | |
4222 | } | |
4223 | ||
4224 | /** | |
4225 | * css_lookup - lookup css by id | |
4226 | * @ss: cgroup subsys to be looked into. | |
4227 | * @id: the id | |
4228 | * | |
4229 | * Returns pointer to cgroup_subsys_state if there is valid one with id. | |
4230 | * NULL if not. Should be called under rcu_read_lock() | |
4231 | */ | |
4232 | struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id) | |
4233 | { | |
4234 | struct css_id *cssid = NULL; | |
4235 | ||
4236 | BUG_ON(!ss->use_id); | |
4237 | cssid = idr_find(&ss->idr, id); | |
4238 | ||
4239 | if (unlikely(!cssid)) | |
4240 | return NULL; | |
4241 | ||
4242 | return rcu_dereference(cssid->css); | |
4243 | } | |
4244 | ||
4245 | /** | |
4246 | * css_get_next - lookup next cgroup under specified hierarchy. | |
4247 | * @ss: pointer to subsystem | |
4248 | * @id: current position of iteration. | |
4249 | * @root: pointer to css. search tree under this. | |
4250 | * @foundid: position of found object. | |
4251 | * | |
4252 | * Search next css under the specified hierarchy of rootid. Calling under | |
4253 | * rcu_read_lock() is necessary. Returns NULL if it reaches the end. | |
4254 | */ | |
4255 | struct cgroup_subsys_state * | |
4256 | css_get_next(struct cgroup_subsys *ss, int id, | |
4257 | struct cgroup_subsys_state *root, int *foundid) | |
4258 | { | |
4259 | struct cgroup_subsys_state *ret = NULL; | |
4260 | struct css_id *tmp; | |
4261 | int tmpid; | |
4262 | int rootid = css_id(root); | |
4263 | int depth = css_depth(root); | |
4264 | ||
4265 | if (!rootid) | |
4266 | return NULL; | |
4267 | ||
4268 | BUG_ON(!ss->use_id); | |
4269 | /* fill start point for scan */ | |
4270 | tmpid = id; | |
4271 | while (1) { | |
4272 | /* | |
4273 | * scan next entry from bitmap(tree), tmpid is updated after | |
4274 | * idr_get_next(). | |
4275 | */ | |
4276 | spin_lock(&ss->id_lock); | |
4277 | tmp = idr_get_next(&ss->idr, &tmpid); | |
4278 | spin_unlock(&ss->id_lock); | |
4279 | ||
4280 | if (!tmp) | |
4281 | break; | |
4282 | if (tmp->depth >= depth && tmp->stack[depth] == rootid) { | |
4283 | ret = rcu_dereference(tmp->css); | |
4284 | if (ret) { | |
4285 | *foundid = tmpid; | |
4286 | break; | |
4287 | } | |
4288 | } | |
4289 | /* continue to scan from next id */ | |
4290 | tmpid = tmpid + 1; | |
4291 | } | |
4292 | return ret; | |
4293 | } | |
4294 | ||
fe693435 PM |
4295 | #ifdef CONFIG_CGROUP_DEBUG |
4296 | static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss, | |
4297 | struct cgroup *cont) | |
4298 | { | |
4299 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
4300 | ||
4301 | if (!css) | |
4302 | return ERR_PTR(-ENOMEM); | |
4303 | ||
4304 | return css; | |
4305 | } | |
4306 | ||
4307 | static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
4308 | { | |
4309 | kfree(cont->subsys[debug_subsys_id]); | |
4310 | } | |
4311 | ||
4312 | static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft) | |
4313 | { | |
4314 | return atomic_read(&cont->count); | |
4315 | } | |
4316 | ||
4317 | static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft) | |
4318 | { | |
4319 | return cgroup_task_count(cont); | |
4320 | } | |
4321 | ||
4322 | static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft) | |
4323 | { | |
4324 | return (u64)(unsigned long)current->cgroups; | |
4325 | } | |
4326 | ||
4327 | static u64 current_css_set_refcount_read(struct cgroup *cont, | |
4328 | struct cftype *cft) | |
4329 | { | |
4330 | u64 count; | |
4331 | ||
4332 | rcu_read_lock(); | |
4333 | count = atomic_read(¤t->cgroups->refcount); | |
4334 | rcu_read_unlock(); | |
4335 | return count; | |
4336 | } | |
4337 | ||
7717f7ba PM |
4338 | static int current_css_set_cg_links_read(struct cgroup *cont, |
4339 | struct cftype *cft, | |
4340 | struct seq_file *seq) | |
4341 | { | |
4342 | struct cg_cgroup_link *link; | |
4343 | struct css_set *cg; | |
4344 | ||
4345 | read_lock(&css_set_lock); | |
4346 | rcu_read_lock(); | |
4347 | cg = rcu_dereference(current->cgroups); | |
4348 | list_for_each_entry(link, &cg->cg_links, cg_link_list) { | |
4349 | struct cgroup *c = link->cgrp; | |
4350 | const char *name; | |
4351 | ||
4352 | if (c->dentry) | |
4353 | name = c->dentry->d_name.name; | |
4354 | else | |
4355 | name = "?"; | |
2c6ab6d2 PM |
4356 | seq_printf(seq, "Root %d group %s\n", |
4357 | c->root->hierarchy_id, name); | |
7717f7ba PM |
4358 | } |
4359 | rcu_read_unlock(); | |
4360 | read_unlock(&css_set_lock); | |
4361 | return 0; | |
4362 | } | |
4363 | ||
4364 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
4365 | static int cgroup_css_links_read(struct cgroup *cont, | |
4366 | struct cftype *cft, | |
4367 | struct seq_file *seq) | |
4368 | { | |
4369 | struct cg_cgroup_link *link; | |
4370 | ||
4371 | read_lock(&css_set_lock); | |
4372 | list_for_each_entry(link, &cont->css_sets, cgrp_link_list) { | |
4373 | struct css_set *cg = link->cg; | |
4374 | struct task_struct *task; | |
4375 | int count = 0; | |
4376 | seq_printf(seq, "css_set %p\n", cg); | |
4377 | list_for_each_entry(task, &cg->tasks, cg_list) { | |
4378 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { | |
4379 | seq_puts(seq, " ...\n"); | |
4380 | break; | |
4381 | } else { | |
4382 | seq_printf(seq, " task %d\n", | |
4383 | task_pid_vnr(task)); | |
4384 | } | |
4385 | } | |
4386 | } | |
4387 | read_unlock(&css_set_lock); | |
4388 | return 0; | |
4389 | } | |
4390 | ||
fe693435 PM |
4391 | static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft) |
4392 | { | |
4393 | return test_bit(CGRP_RELEASABLE, &cgrp->flags); | |
4394 | } | |
4395 | ||
4396 | static struct cftype debug_files[] = { | |
4397 | { | |
4398 | .name = "cgroup_refcount", | |
4399 | .read_u64 = cgroup_refcount_read, | |
4400 | }, | |
4401 | { | |
4402 | .name = "taskcount", | |
4403 | .read_u64 = debug_taskcount_read, | |
4404 | }, | |
4405 | ||
4406 | { | |
4407 | .name = "current_css_set", | |
4408 | .read_u64 = current_css_set_read, | |
4409 | }, | |
4410 | ||
4411 | { | |
4412 | .name = "current_css_set_refcount", | |
4413 | .read_u64 = current_css_set_refcount_read, | |
4414 | }, | |
4415 | ||
7717f7ba PM |
4416 | { |
4417 | .name = "current_css_set_cg_links", | |
4418 | .read_seq_string = current_css_set_cg_links_read, | |
4419 | }, | |
4420 | ||
4421 | { | |
4422 | .name = "cgroup_css_links", | |
4423 | .read_seq_string = cgroup_css_links_read, | |
4424 | }, | |
4425 | ||
fe693435 PM |
4426 | { |
4427 | .name = "releasable", | |
4428 | .read_u64 = releasable_read, | |
4429 | }, | |
4430 | }; | |
4431 | ||
4432 | static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
4433 | { | |
4434 | return cgroup_add_files(cont, ss, debug_files, | |
4435 | ARRAY_SIZE(debug_files)); | |
4436 | } | |
4437 | ||
4438 | struct cgroup_subsys debug_subsys = { | |
4439 | .name = "debug", | |
4440 | .create = debug_create, | |
4441 | .destroy = debug_destroy, | |
4442 | .populate = debug_populate, | |
4443 | .subsys_id = debug_subsys_id, | |
4444 | }; | |
4445 | #endif /* CONFIG_CGROUP_DEBUG */ |