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