]>
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> | |
c6d57f33 | 26 | #include <linux/ctype.h> |
ddbcc7e8 PM |
27 | #include <linux/errno.h> |
28 | #include <linux/fs.h> | |
29 | #include <linux/kernel.h> | |
30 | #include <linux/list.h> | |
31 | #include <linux/mm.h> | |
32 | #include <linux/mutex.h> | |
33 | #include <linux/mount.h> | |
34 | #include <linux/pagemap.h> | |
a424316c | 35 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
36 | #include <linux/rcupdate.h> |
37 | #include <linux/sched.h> | |
817929ec | 38 | #include <linux/backing-dev.h> |
ddbcc7e8 PM |
39 | #include <linux/seq_file.h> |
40 | #include <linux/slab.h> | |
41 | #include <linux/magic.h> | |
42 | #include <linux/spinlock.h> | |
43 | #include <linux/string.h> | |
bbcb81d0 | 44 | #include <linux/sort.h> |
81a6a5cd | 45 | #include <linux/kmod.h> |
846c7bb0 BS |
46 | #include <linux/delayacct.h> |
47 | #include <linux/cgroupstats.h> | |
472b1053 | 48 | #include <linux/hash.h> |
3f8206d4 | 49 | #include <linux/namei.h> |
337eb00a | 50 | #include <linux/smp_lock.h> |
096b7fe0 | 51 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 52 | #include <linux/idr.h> |
846c7bb0 | 53 | |
ddbcc7e8 PM |
54 | #include <asm/atomic.h> |
55 | ||
81a6a5cd PM |
56 | static DEFINE_MUTEX(cgroup_mutex); |
57 | ||
ddbcc7e8 PM |
58 | /* Generate an array of cgroup subsystem pointers */ |
59 | #define SUBSYS(_x) &_x ## _subsys, | |
60 | ||
61 | static struct cgroup_subsys *subsys[] = { | |
62 | #include <linux/cgroup_subsys.h> | |
63 | }; | |
64 | ||
c6d57f33 PM |
65 | #define MAX_CGROUP_ROOT_NAMELEN 64 |
66 | ||
ddbcc7e8 PM |
67 | /* |
68 | * A cgroupfs_root represents the root of a cgroup hierarchy, | |
69 | * and may be associated with a superblock to form an active | |
70 | * hierarchy | |
71 | */ | |
72 | struct cgroupfs_root { | |
73 | struct super_block *sb; | |
74 | ||
75 | /* | |
76 | * The bitmask of subsystems intended to be attached to this | |
77 | * hierarchy | |
78 | */ | |
79 | unsigned long subsys_bits; | |
80 | ||
2c6ab6d2 PM |
81 | /* Unique id for this hierarchy. */ |
82 | int hierarchy_id; | |
83 | ||
ddbcc7e8 PM |
84 | /* The bitmask of subsystems currently attached to this hierarchy */ |
85 | unsigned long actual_subsys_bits; | |
86 | ||
87 | /* A list running through the attached subsystems */ | |
88 | struct list_head subsys_list; | |
89 | ||
90 | /* The root cgroup for this hierarchy */ | |
91 | struct cgroup top_cgroup; | |
92 | ||
93 | /* Tracks how many cgroups are currently defined in hierarchy.*/ | |
94 | int number_of_cgroups; | |
95 | ||
e5f6a860 | 96 | /* A list running through the active hierarchies */ |
ddbcc7e8 PM |
97 | struct list_head root_list; |
98 | ||
99 | /* Hierarchy-specific flags */ | |
100 | unsigned long flags; | |
81a6a5cd | 101 | |
e788e066 | 102 | /* The path to use for release notifications. */ |
81a6a5cd | 103 | char release_agent_path[PATH_MAX]; |
c6d57f33 PM |
104 | |
105 | /* The name for this hierarchy - may be empty */ | |
106 | char name[MAX_CGROUP_ROOT_NAMELEN]; | |
ddbcc7e8 PM |
107 | }; |
108 | ||
ddbcc7e8 PM |
109 | /* |
110 | * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the | |
111 | * subsystems that are otherwise unattached - it never has more than a | |
112 | * single cgroup, and all tasks are part of that cgroup. | |
113 | */ | |
114 | static struct cgroupfs_root rootnode; | |
115 | ||
38460b48 KH |
116 | /* |
117 | * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when | |
118 | * cgroup_subsys->use_id != 0. | |
119 | */ | |
120 | #define CSS_ID_MAX (65535) | |
121 | struct css_id { | |
122 | /* | |
123 | * The css to which this ID points. This pointer is set to valid value | |
124 | * after cgroup is populated. If cgroup is removed, this will be NULL. | |
125 | * This pointer is expected to be RCU-safe because destroy() | |
126 | * is called after synchronize_rcu(). But for safe use, css_is_removed() | |
127 | * css_tryget() should be used for avoiding race. | |
128 | */ | |
129 | struct cgroup_subsys_state *css; | |
130 | /* | |
131 | * ID of this css. | |
132 | */ | |
133 | unsigned short id; | |
134 | /* | |
135 | * Depth in hierarchy which this ID belongs to. | |
136 | */ | |
137 | unsigned short depth; | |
138 | /* | |
139 | * ID is freed by RCU. (and lookup routine is RCU safe.) | |
140 | */ | |
141 | struct rcu_head rcu_head; | |
142 | /* | |
143 | * Hierarchy of CSS ID belongs to. | |
144 | */ | |
145 | unsigned short stack[0]; /* Array of Length (depth+1) */ | |
146 | }; | |
147 | ||
148 | ||
ddbcc7e8 PM |
149 | /* The list of hierarchy roots */ |
150 | ||
151 | static LIST_HEAD(roots); | |
817929ec | 152 | static int root_count; |
ddbcc7e8 | 153 | |
2c6ab6d2 PM |
154 | static DEFINE_IDA(hierarchy_ida); |
155 | static int next_hierarchy_id; | |
156 | static DEFINE_SPINLOCK(hierarchy_id_lock); | |
157 | ||
ddbcc7e8 PM |
158 | /* dummytop is a shorthand for the dummy hierarchy's top cgroup */ |
159 | #define dummytop (&rootnode.top_cgroup) | |
160 | ||
161 | /* This flag indicates whether tasks in the fork and exit paths should | |
a043e3b2 LZ |
162 | * check for fork/exit handlers to call. This avoids us having to do |
163 | * extra work in the fork/exit path if none of the subsystems need to | |
164 | * be called. | |
ddbcc7e8 | 165 | */ |
8947f9d5 | 166 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 167 | |
ddbcc7e8 | 168 | /* convenient tests for these bits */ |
bd89aabc | 169 | inline int cgroup_is_removed(const struct cgroup *cgrp) |
ddbcc7e8 | 170 | { |
bd89aabc | 171 | return test_bit(CGRP_REMOVED, &cgrp->flags); |
ddbcc7e8 PM |
172 | } |
173 | ||
174 | /* bits in struct cgroupfs_root flags field */ | |
175 | enum { | |
176 | ROOT_NOPREFIX, /* mounted subsystems have no named prefix */ | |
177 | }; | |
178 | ||
e9685a03 | 179 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
180 | { |
181 | const int bits = | |
bd89aabc PM |
182 | (1 << CGRP_RELEASABLE) | |
183 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
184 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
185 | } |
186 | ||
e9685a03 | 187 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 188 | { |
bd89aabc | 189 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
190 | } |
191 | ||
ddbcc7e8 PM |
192 | /* |
193 | * for_each_subsys() allows you to iterate on each subsystem attached to | |
194 | * an active hierarchy | |
195 | */ | |
196 | #define for_each_subsys(_root, _ss) \ | |
197 | list_for_each_entry(_ss, &_root->subsys_list, sibling) | |
198 | ||
e5f6a860 LZ |
199 | /* for_each_active_root() allows you to iterate across the active hierarchies */ |
200 | #define for_each_active_root(_root) \ | |
ddbcc7e8 PM |
201 | list_for_each_entry(_root, &roots, root_list) |
202 | ||
81a6a5cd PM |
203 | /* the list of cgroups eligible for automatic release. Protected by |
204 | * release_list_lock */ | |
205 | static LIST_HEAD(release_list); | |
206 | static DEFINE_SPINLOCK(release_list_lock); | |
207 | static void cgroup_release_agent(struct work_struct *work); | |
208 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 209 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 210 | |
817929ec PM |
211 | /* Link structure for associating css_set objects with cgroups */ |
212 | struct cg_cgroup_link { | |
213 | /* | |
214 | * List running through cg_cgroup_links associated with a | |
215 | * cgroup, anchored on cgroup->css_sets | |
216 | */ | |
bd89aabc | 217 | struct list_head cgrp_link_list; |
7717f7ba | 218 | struct cgroup *cgrp; |
817929ec PM |
219 | /* |
220 | * List running through cg_cgroup_links pointing at a | |
221 | * single css_set object, anchored on css_set->cg_links | |
222 | */ | |
223 | struct list_head cg_link_list; | |
224 | struct css_set *cg; | |
225 | }; | |
226 | ||
227 | /* The default css_set - used by init and its children prior to any | |
228 | * hierarchies being mounted. It contains a pointer to the root state | |
229 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
230 | * reference-counted, to improve performance when child cgroups | |
231 | * haven't been created. | |
232 | */ | |
233 | ||
234 | static struct css_set init_css_set; | |
235 | static struct cg_cgroup_link init_css_set_link; | |
236 | ||
38460b48 KH |
237 | static int cgroup_subsys_init_idr(struct cgroup_subsys *ss); |
238 | ||
817929ec PM |
239 | /* css_set_lock protects the list of css_set objects, and the |
240 | * chain of tasks off each css_set. Nests outside task->alloc_lock | |
241 | * due to cgroup_iter_start() */ | |
242 | static DEFINE_RWLOCK(css_set_lock); | |
243 | static int css_set_count; | |
244 | ||
7717f7ba PM |
245 | /* |
246 | * hash table for cgroup groups. This improves the performance to find | |
247 | * an existing css_set. This hash doesn't (currently) take into | |
248 | * account cgroups in empty hierarchies. | |
249 | */ | |
472b1053 LZ |
250 | #define CSS_SET_HASH_BITS 7 |
251 | #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS) | |
252 | static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE]; | |
253 | ||
254 | static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[]) | |
255 | { | |
256 | int i; | |
257 | int index; | |
258 | unsigned long tmp = 0UL; | |
259 | ||
260 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) | |
261 | tmp += (unsigned long)css[i]; | |
262 | tmp = (tmp >> 16) ^ tmp; | |
263 | ||
264 | index = hash_long(tmp, CSS_SET_HASH_BITS); | |
265 | ||
266 | return &css_set_table[index]; | |
267 | } | |
268 | ||
817929ec PM |
269 | /* We don't maintain the lists running through each css_set to its |
270 | * task until after the first call to cgroup_iter_start(). This | |
271 | * reduces the fork()/exit() overhead for people who have cgroups | |
272 | * compiled into their kernel but not actually in use */ | |
8947f9d5 | 273 | static int use_task_css_set_links __read_mostly; |
817929ec | 274 | |
2c6ab6d2 | 275 | static void __put_css_set(struct css_set *cg, int taskexit) |
b4f48b63 | 276 | { |
71cbb949 KM |
277 | struct cg_cgroup_link *link; |
278 | struct cg_cgroup_link *saved_link; | |
146aa1bd LJ |
279 | /* |
280 | * Ensure that the refcount doesn't hit zero while any readers | |
281 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
282 | * rwlock | |
283 | */ | |
284 | if (atomic_add_unless(&cg->refcount, -1, 1)) | |
285 | return; | |
286 | write_lock(&css_set_lock); | |
287 | if (!atomic_dec_and_test(&cg->refcount)) { | |
288 | write_unlock(&css_set_lock); | |
289 | return; | |
290 | } | |
81a6a5cd | 291 | |
2c6ab6d2 PM |
292 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
293 | hlist_del(&cg->hlist); | |
294 | css_set_count--; | |
295 | ||
296 | list_for_each_entry_safe(link, saved_link, &cg->cg_links, | |
297 | cg_link_list) { | |
298 | struct cgroup *cgrp = link->cgrp; | |
299 | list_del(&link->cg_link_list); | |
300 | list_del(&link->cgrp_link_list); | |
bd89aabc PM |
301 | if (atomic_dec_and_test(&cgrp->count) && |
302 | notify_on_release(cgrp)) { | |
81a6a5cd | 303 | if (taskexit) |
bd89aabc PM |
304 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
305 | check_for_release(cgrp); | |
81a6a5cd | 306 | } |
2c6ab6d2 PM |
307 | |
308 | kfree(link); | |
81a6a5cd | 309 | } |
2c6ab6d2 PM |
310 | |
311 | write_unlock(&css_set_lock); | |
817929ec | 312 | kfree(cg); |
b4f48b63 PM |
313 | } |
314 | ||
817929ec PM |
315 | /* |
316 | * refcounted get/put for css_set objects | |
317 | */ | |
318 | static inline void get_css_set(struct css_set *cg) | |
319 | { | |
146aa1bd | 320 | atomic_inc(&cg->refcount); |
817929ec PM |
321 | } |
322 | ||
323 | static inline void put_css_set(struct css_set *cg) | |
324 | { | |
146aa1bd | 325 | __put_css_set(cg, 0); |
817929ec PM |
326 | } |
327 | ||
81a6a5cd PM |
328 | static inline void put_css_set_taskexit(struct css_set *cg) |
329 | { | |
146aa1bd | 330 | __put_css_set(cg, 1); |
81a6a5cd PM |
331 | } |
332 | ||
7717f7ba PM |
333 | /* |
334 | * compare_css_sets - helper function for find_existing_css_set(). | |
335 | * @cg: candidate css_set being tested | |
336 | * @old_cg: existing css_set for a task | |
337 | * @new_cgrp: cgroup that's being entered by the task | |
338 | * @template: desired set of css pointers in css_set (pre-calculated) | |
339 | * | |
340 | * Returns true if "cg" matches "old_cg" except for the hierarchy | |
341 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". | |
342 | */ | |
343 | static bool compare_css_sets(struct css_set *cg, | |
344 | struct css_set *old_cg, | |
345 | struct cgroup *new_cgrp, | |
346 | struct cgroup_subsys_state *template[]) | |
347 | { | |
348 | struct list_head *l1, *l2; | |
349 | ||
350 | if (memcmp(template, cg->subsys, sizeof(cg->subsys))) { | |
351 | /* Not all subsystems matched */ | |
352 | return false; | |
353 | } | |
354 | ||
355 | /* | |
356 | * Compare cgroup pointers in order to distinguish between | |
357 | * different cgroups in heirarchies with no subsystems. We | |
358 | * could get by with just this check alone (and skip the | |
359 | * memcmp above) but on most setups the memcmp check will | |
360 | * avoid the need for this more expensive check on almost all | |
361 | * candidates. | |
362 | */ | |
363 | ||
364 | l1 = &cg->cg_links; | |
365 | l2 = &old_cg->cg_links; | |
366 | while (1) { | |
367 | struct cg_cgroup_link *cgl1, *cgl2; | |
368 | struct cgroup *cg1, *cg2; | |
369 | ||
370 | l1 = l1->next; | |
371 | l2 = l2->next; | |
372 | /* See if we reached the end - both lists are equal length. */ | |
373 | if (l1 == &cg->cg_links) { | |
374 | BUG_ON(l2 != &old_cg->cg_links); | |
375 | break; | |
376 | } else { | |
377 | BUG_ON(l2 == &old_cg->cg_links); | |
378 | } | |
379 | /* Locate the cgroups associated with these links. */ | |
380 | cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list); | |
381 | cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list); | |
382 | cg1 = cgl1->cgrp; | |
383 | cg2 = cgl2->cgrp; | |
384 | /* Hierarchies should be linked in the same order. */ | |
385 | BUG_ON(cg1->root != cg2->root); | |
386 | ||
387 | /* | |
388 | * If this hierarchy is the hierarchy of the cgroup | |
389 | * that's changing, then we need to check that this | |
390 | * css_set points to the new cgroup; if it's any other | |
391 | * hierarchy, then this css_set should point to the | |
392 | * same cgroup as the old css_set. | |
393 | */ | |
394 | if (cg1->root == new_cgrp->root) { | |
395 | if (cg1 != new_cgrp) | |
396 | return false; | |
397 | } else { | |
398 | if (cg1 != cg2) | |
399 | return false; | |
400 | } | |
401 | } | |
402 | return true; | |
403 | } | |
404 | ||
817929ec PM |
405 | /* |
406 | * find_existing_css_set() is a helper for | |
407 | * find_css_set(), and checks to see whether an existing | |
472b1053 | 408 | * css_set is suitable. |
817929ec PM |
409 | * |
410 | * oldcg: the cgroup group that we're using before the cgroup | |
411 | * transition | |
412 | * | |
bd89aabc | 413 | * cgrp: the cgroup that we're moving into |
817929ec PM |
414 | * |
415 | * template: location in which to build the desired set of subsystem | |
416 | * state objects for the new cgroup group | |
417 | */ | |
817929ec PM |
418 | static struct css_set *find_existing_css_set( |
419 | struct css_set *oldcg, | |
bd89aabc | 420 | struct cgroup *cgrp, |
817929ec | 421 | struct cgroup_subsys_state *template[]) |
b4f48b63 PM |
422 | { |
423 | int i; | |
bd89aabc | 424 | struct cgroupfs_root *root = cgrp->root; |
472b1053 LZ |
425 | struct hlist_head *hhead; |
426 | struct hlist_node *node; | |
427 | struct css_set *cg; | |
817929ec PM |
428 | |
429 | /* Built the set of subsystem state objects that we want to | |
430 | * see in the new css_set */ | |
431 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
8d53d55d | 432 | if (root->subsys_bits & (1UL << i)) { |
817929ec PM |
433 | /* Subsystem is in this hierarchy. So we want |
434 | * the subsystem state from the new | |
435 | * cgroup */ | |
bd89aabc | 436 | template[i] = cgrp->subsys[i]; |
817929ec PM |
437 | } else { |
438 | /* Subsystem is not in this hierarchy, so we | |
439 | * don't want to change the subsystem state */ | |
440 | template[i] = oldcg->subsys[i]; | |
441 | } | |
442 | } | |
443 | ||
472b1053 LZ |
444 | hhead = css_set_hash(template); |
445 | hlist_for_each_entry(cg, node, hhead, hlist) { | |
7717f7ba PM |
446 | if (!compare_css_sets(cg, oldcg, cgrp, template)) |
447 | continue; | |
448 | ||
449 | /* This css_set matches what we need */ | |
450 | return cg; | |
472b1053 | 451 | } |
817929ec PM |
452 | |
453 | /* No existing cgroup group matched */ | |
454 | return NULL; | |
455 | } | |
456 | ||
36553434 LZ |
457 | static void free_cg_links(struct list_head *tmp) |
458 | { | |
459 | struct cg_cgroup_link *link; | |
460 | struct cg_cgroup_link *saved_link; | |
461 | ||
462 | list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) { | |
463 | list_del(&link->cgrp_link_list); | |
464 | kfree(link); | |
465 | } | |
466 | } | |
467 | ||
817929ec PM |
468 | /* |
469 | * allocate_cg_links() allocates "count" cg_cgroup_link structures | |
bd89aabc | 470 | * and chains them on tmp through their cgrp_link_list fields. Returns 0 on |
817929ec PM |
471 | * success or a negative error |
472 | */ | |
817929ec PM |
473 | static int allocate_cg_links(int count, struct list_head *tmp) |
474 | { | |
475 | struct cg_cgroup_link *link; | |
476 | int i; | |
477 | INIT_LIST_HEAD(tmp); | |
478 | for (i = 0; i < count; i++) { | |
479 | link = kmalloc(sizeof(*link), GFP_KERNEL); | |
480 | if (!link) { | |
36553434 | 481 | free_cg_links(tmp); |
817929ec PM |
482 | return -ENOMEM; |
483 | } | |
bd89aabc | 484 | list_add(&link->cgrp_link_list, tmp); |
817929ec PM |
485 | } |
486 | return 0; | |
487 | } | |
488 | ||
c12f65d4 LZ |
489 | /** |
490 | * link_css_set - a helper function to link a css_set to a cgroup | |
491 | * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links() | |
492 | * @cg: the css_set to be linked | |
493 | * @cgrp: the destination cgroup | |
494 | */ | |
495 | static void link_css_set(struct list_head *tmp_cg_links, | |
496 | struct css_set *cg, struct cgroup *cgrp) | |
497 | { | |
498 | struct cg_cgroup_link *link; | |
499 | ||
500 | BUG_ON(list_empty(tmp_cg_links)); | |
501 | link = list_first_entry(tmp_cg_links, struct cg_cgroup_link, | |
502 | cgrp_link_list); | |
503 | link->cg = cg; | |
7717f7ba | 504 | link->cgrp = cgrp; |
2c6ab6d2 | 505 | atomic_inc(&cgrp->count); |
c12f65d4 | 506 | list_move(&link->cgrp_link_list, &cgrp->css_sets); |
7717f7ba PM |
507 | /* |
508 | * Always add links to the tail of the list so that the list | |
509 | * is sorted by order of hierarchy creation | |
510 | */ | |
511 | list_add_tail(&link->cg_link_list, &cg->cg_links); | |
c12f65d4 LZ |
512 | } |
513 | ||
817929ec PM |
514 | /* |
515 | * find_css_set() takes an existing cgroup group and a | |
516 | * cgroup object, and returns a css_set object that's | |
517 | * equivalent to the old group, but with the given cgroup | |
518 | * substituted into the appropriate hierarchy. Must be called with | |
519 | * cgroup_mutex held | |
520 | */ | |
817929ec | 521 | static struct css_set *find_css_set( |
bd89aabc | 522 | struct css_set *oldcg, struct cgroup *cgrp) |
817929ec PM |
523 | { |
524 | struct css_set *res; | |
525 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT]; | |
817929ec PM |
526 | |
527 | struct list_head tmp_cg_links; | |
817929ec | 528 | |
472b1053 | 529 | struct hlist_head *hhead; |
7717f7ba | 530 | struct cg_cgroup_link *link; |
472b1053 | 531 | |
817929ec PM |
532 | /* First see if we already have a cgroup group that matches |
533 | * the desired set */ | |
7e9abd89 | 534 | read_lock(&css_set_lock); |
bd89aabc | 535 | res = find_existing_css_set(oldcg, cgrp, template); |
817929ec PM |
536 | if (res) |
537 | get_css_set(res); | |
7e9abd89 | 538 | read_unlock(&css_set_lock); |
817929ec PM |
539 | |
540 | if (res) | |
541 | return res; | |
542 | ||
543 | res = kmalloc(sizeof(*res), GFP_KERNEL); | |
544 | if (!res) | |
545 | return NULL; | |
546 | ||
547 | /* Allocate all the cg_cgroup_link objects that we'll need */ | |
548 | if (allocate_cg_links(root_count, &tmp_cg_links) < 0) { | |
549 | kfree(res); | |
550 | return NULL; | |
551 | } | |
552 | ||
146aa1bd | 553 | atomic_set(&res->refcount, 1); |
817929ec PM |
554 | INIT_LIST_HEAD(&res->cg_links); |
555 | INIT_LIST_HEAD(&res->tasks); | |
472b1053 | 556 | INIT_HLIST_NODE(&res->hlist); |
817929ec PM |
557 | |
558 | /* Copy the set of subsystem state objects generated in | |
559 | * find_existing_css_set() */ | |
560 | memcpy(res->subsys, template, sizeof(res->subsys)); | |
561 | ||
562 | write_lock(&css_set_lock); | |
563 | /* Add reference counts and links from the new css_set. */ | |
7717f7ba PM |
564 | list_for_each_entry(link, &oldcg->cg_links, cg_link_list) { |
565 | struct cgroup *c = link->cgrp; | |
566 | if (c->root == cgrp->root) | |
567 | c = cgrp; | |
568 | link_css_set(&tmp_cg_links, res, c); | |
569 | } | |
817929ec PM |
570 | |
571 | BUG_ON(!list_empty(&tmp_cg_links)); | |
572 | ||
817929ec | 573 | css_set_count++; |
472b1053 LZ |
574 | |
575 | /* Add this cgroup group to the hash table */ | |
576 | hhead = css_set_hash(res->subsys); | |
577 | hlist_add_head(&res->hlist, hhead); | |
578 | ||
817929ec PM |
579 | write_unlock(&css_set_lock); |
580 | ||
581 | return res; | |
b4f48b63 PM |
582 | } |
583 | ||
7717f7ba PM |
584 | /* |
585 | * Return the cgroup for "task" from the given hierarchy. Must be | |
586 | * called with cgroup_mutex held. | |
587 | */ | |
588 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
589 | struct cgroupfs_root *root) | |
590 | { | |
591 | struct css_set *css; | |
592 | struct cgroup *res = NULL; | |
593 | ||
594 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
595 | read_lock(&css_set_lock); | |
596 | /* | |
597 | * No need to lock the task - since we hold cgroup_mutex the | |
598 | * task can't change groups, so the only thing that can happen | |
599 | * is that it exits and its css is set back to init_css_set. | |
600 | */ | |
601 | css = task->cgroups; | |
602 | if (css == &init_css_set) { | |
603 | res = &root->top_cgroup; | |
604 | } else { | |
605 | struct cg_cgroup_link *link; | |
606 | list_for_each_entry(link, &css->cg_links, cg_link_list) { | |
607 | struct cgroup *c = link->cgrp; | |
608 | if (c->root == root) { | |
609 | res = c; | |
610 | break; | |
611 | } | |
612 | } | |
613 | } | |
614 | read_unlock(&css_set_lock); | |
615 | BUG_ON(!res); | |
616 | return res; | |
617 | } | |
618 | ||
ddbcc7e8 PM |
619 | /* |
620 | * There is one global cgroup mutex. We also require taking | |
621 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
622 | * See "The task_lock() exception", at the end of this comment. | |
623 | * | |
624 | * A task must hold cgroup_mutex to modify cgroups. | |
625 | * | |
626 | * Any task can increment and decrement the count field without lock. | |
627 | * So in general, code holding cgroup_mutex can't rely on the count | |
628 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 629 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
630 | * means that no tasks are currently attached, therefore there is no |
631 | * way a task attached to that cgroup can fork (the other way to | |
632 | * increment the count). So code holding cgroup_mutex can safely | |
633 | * assume that if the count is zero, it will stay zero. Similarly, if | |
634 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
635 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
636 | * needs that mutex. | |
637 | * | |
ddbcc7e8 PM |
638 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
639 | * (usually) take cgroup_mutex. These are the two most performance | |
640 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
641 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
642 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
643 | * to the release agent with the name of the cgroup (path relative to |
644 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
645 | * |
646 | * A cgroup can only be deleted if both its 'count' of using tasks | |
647 | * is zero, and its list of 'children' cgroups is empty. Since all | |
648 | * tasks in the system use _some_ cgroup, and since there is always at | |
649 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
650 | * always has either children cgroups and/or using tasks. So we don't | |
651 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
652 | * | |
653 | * The task_lock() exception | |
654 | * | |
655 | * The need for this exception arises from the action of | |
956db3ca | 656 | * cgroup_attach_task(), which overwrites one tasks cgroup pointer with |
a043e3b2 | 657 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
658 | * several performance critical places that need to reference |
659 | * task->cgroup without the expense of grabbing a system global | |
660 | * mutex. Therefore except as noted below, when dereferencing or, as | |
956db3ca | 661 | * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use |
ddbcc7e8 PM |
662 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
663 | * the task_struct routinely used for such matters. | |
664 | * | |
665 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 666 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
667 | */ |
668 | ||
ddbcc7e8 PM |
669 | /** |
670 | * cgroup_lock - lock out any changes to cgroup structures | |
671 | * | |
672 | */ | |
ddbcc7e8 PM |
673 | void cgroup_lock(void) |
674 | { | |
675 | mutex_lock(&cgroup_mutex); | |
676 | } | |
677 | ||
678 | /** | |
679 | * cgroup_unlock - release lock on cgroup changes | |
680 | * | |
681 | * Undo the lock taken in a previous cgroup_lock() call. | |
682 | */ | |
ddbcc7e8 PM |
683 | void cgroup_unlock(void) |
684 | { | |
685 | mutex_unlock(&cgroup_mutex); | |
686 | } | |
687 | ||
688 | /* | |
689 | * A couple of forward declarations required, due to cyclic reference loop: | |
690 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> | |
691 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations | |
692 | * -> cgroup_mkdir. | |
693 | */ | |
694 | ||
695 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode); | |
696 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); | |
bd89aabc | 697 | static int cgroup_populate_dir(struct cgroup *cgrp); |
6e1d5dcc | 698 | static const struct inode_operations cgroup_dir_inode_operations; |
a424316c PM |
699 | static struct file_operations proc_cgroupstats_operations; |
700 | ||
701 | static struct backing_dev_info cgroup_backing_dev_info = { | |
d993831f | 702 | .name = "cgroup", |
e4ad08fe | 703 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
a424316c | 704 | }; |
ddbcc7e8 | 705 | |
38460b48 KH |
706 | static int alloc_css_id(struct cgroup_subsys *ss, |
707 | struct cgroup *parent, struct cgroup *child); | |
708 | ||
ddbcc7e8 PM |
709 | static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb) |
710 | { | |
711 | struct inode *inode = new_inode(sb); | |
ddbcc7e8 PM |
712 | |
713 | if (inode) { | |
714 | inode->i_mode = mode; | |
76aac0e9 DH |
715 | inode->i_uid = current_fsuid(); |
716 | inode->i_gid = current_fsgid(); | |
ddbcc7e8 PM |
717 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
718 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; | |
719 | } | |
720 | return inode; | |
721 | } | |
722 | ||
4fca88c8 KH |
723 | /* |
724 | * Call subsys's pre_destroy handler. | |
725 | * This is called before css refcnt check. | |
726 | */ | |
ec64f515 | 727 | static int cgroup_call_pre_destroy(struct cgroup *cgrp) |
4fca88c8 KH |
728 | { |
729 | struct cgroup_subsys *ss; | |
ec64f515 KH |
730 | int ret = 0; |
731 | ||
4fca88c8 | 732 | for_each_subsys(cgrp->root, ss) |
ec64f515 KH |
733 | if (ss->pre_destroy) { |
734 | ret = ss->pre_destroy(ss, cgrp); | |
735 | if (ret) | |
736 | break; | |
737 | } | |
738 | return ret; | |
4fca88c8 KH |
739 | } |
740 | ||
a47295e6 PM |
741 | static void free_cgroup_rcu(struct rcu_head *obj) |
742 | { | |
743 | struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head); | |
744 | ||
745 | kfree(cgrp); | |
746 | } | |
747 | ||
ddbcc7e8 PM |
748 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) |
749 | { | |
750 | /* is dentry a directory ? if so, kfree() associated cgroup */ | |
751 | if (S_ISDIR(inode->i_mode)) { | |
bd89aabc | 752 | struct cgroup *cgrp = dentry->d_fsdata; |
8dc4f3e1 | 753 | struct cgroup_subsys *ss; |
bd89aabc | 754 | BUG_ON(!(cgroup_is_removed(cgrp))); |
81a6a5cd PM |
755 | /* It's possible for external users to be holding css |
756 | * reference counts on a cgroup; css_put() needs to | |
757 | * be able to access the cgroup after decrementing | |
758 | * the reference count in order to know if it needs to | |
759 | * queue the cgroup to be handled by the release | |
760 | * agent */ | |
761 | synchronize_rcu(); | |
8dc4f3e1 PM |
762 | |
763 | mutex_lock(&cgroup_mutex); | |
764 | /* | |
765 | * Release the subsystem state objects. | |
766 | */ | |
75139b82 LZ |
767 | for_each_subsys(cgrp->root, ss) |
768 | ss->destroy(ss, cgrp); | |
8dc4f3e1 PM |
769 | |
770 | cgrp->root->number_of_cgroups--; | |
771 | mutex_unlock(&cgroup_mutex); | |
772 | ||
a47295e6 PM |
773 | /* |
774 | * Drop the active superblock reference that we took when we | |
775 | * created the cgroup | |
776 | */ | |
8dc4f3e1 PM |
777 | deactivate_super(cgrp->root->sb); |
778 | ||
a47295e6 | 779 | call_rcu(&cgrp->rcu_head, free_cgroup_rcu); |
ddbcc7e8 PM |
780 | } |
781 | iput(inode); | |
782 | } | |
783 | ||
784 | static void remove_dir(struct dentry *d) | |
785 | { | |
786 | struct dentry *parent = dget(d->d_parent); | |
787 | ||
788 | d_delete(d); | |
789 | simple_rmdir(parent->d_inode, d); | |
790 | dput(parent); | |
791 | } | |
792 | ||
793 | static void cgroup_clear_directory(struct dentry *dentry) | |
794 | { | |
795 | struct list_head *node; | |
796 | ||
797 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
798 | spin_lock(&dcache_lock); | |
799 | node = dentry->d_subdirs.next; | |
800 | while (node != &dentry->d_subdirs) { | |
801 | struct dentry *d = list_entry(node, struct dentry, d_u.d_child); | |
802 | list_del_init(node); | |
803 | if (d->d_inode) { | |
804 | /* This should never be called on a cgroup | |
805 | * directory with child cgroups */ | |
806 | BUG_ON(d->d_inode->i_mode & S_IFDIR); | |
807 | d = dget_locked(d); | |
808 | spin_unlock(&dcache_lock); | |
809 | d_delete(d); | |
810 | simple_unlink(dentry->d_inode, d); | |
811 | dput(d); | |
812 | spin_lock(&dcache_lock); | |
813 | } | |
814 | node = dentry->d_subdirs.next; | |
815 | } | |
816 | spin_unlock(&dcache_lock); | |
817 | } | |
818 | ||
819 | /* | |
820 | * NOTE : the dentry must have been dget()'ed | |
821 | */ | |
822 | static void cgroup_d_remove_dir(struct dentry *dentry) | |
823 | { | |
824 | cgroup_clear_directory(dentry); | |
825 | ||
826 | spin_lock(&dcache_lock); | |
827 | list_del_init(&dentry->d_u.d_child); | |
828 | spin_unlock(&dcache_lock); | |
829 | remove_dir(dentry); | |
830 | } | |
831 | ||
ec64f515 KH |
832 | /* |
833 | * A queue for waiters to do rmdir() cgroup. A tasks will sleep when | |
834 | * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some | |
835 | * reference to css->refcnt. In general, this refcnt is expected to goes down | |
836 | * to zero, soon. | |
837 | * | |
88703267 | 838 | * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex; |
ec64f515 KH |
839 | */ |
840 | DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq); | |
841 | ||
88703267 | 842 | static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp) |
ec64f515 | 843 | { |
88703267 | 844 | if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))) |
ec64f515 KH |
845 | wake_up_all(&cgroup_rmdir_waitq); |
846 | } | |
847 | ||
88703267 KH |
848 | void cgroup_exclude_rmdir(struct cgroup_subsys_state *css) |
849 | { | |
850 | css_get(css); | |
851 | } | |
852 | ||
853 | void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css) | |
854 | { | |
855 | cgroup_wakeup_rmdir_waiter(css->cgroup); | |
856 | css_put(css); | |
857 | } | |
858 | ||
859 | ||
ddbcc7e8 PM |
860 | static int rebind_subsystems(struct cgroupfs_root *root, |
861 | unsigned long final_bits) | |
862 | { | |
863 | unsigned long added_bits, removed_bits; | |
bd89aabc | 864 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
865 | int i; |
866 | ||
867 | removed_bits = root->actual_subsys_bits & ~final_bits; | |
868 | added_bits = final_bits & ~root->actual_subsys_bits; | |
869 | /* Check that any added subsystems are currently free */ | |
870 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
8d53d55d | 871 | unsigned long bit = 1UL << i; |
ddbcc7e8 PM |
872 | struct cgroup_subsys *ss = subsys[i]; |
873 | if (!(bit & added_bits)) | |
874 | continue; | |
875 | if (ss->root != &rootnode) { | |
876 | /* Subsystem isn't free */ | |
877 | return -EBUSY; | |
878 | } | |
879 | } | |
880 | ||
881 | /* Currently we don't handle adding/removing subsystems when | |
882 | * any child cgroups exist. This is theoretically supportable | |
883 | * but involves complex error handling, so it's being left until | |
884 | * later */ | |
307257cf | 885 | if (root->number_of_cgroups > 1) |
ddbcc7e8 PM |
886 | return -EBUSY; |
887 | ||
888 | /* Process each subsystem */ | |
889 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
890 | struct cgroup_subsys *ss = subsys[i]; | |
891 | unsigned long bit = 1UL << i; | |
892 | if (bit & added_bits) { | |
893 | /* We're binding this subsystem to this hierarchy */ | |
bd89aabc | 894 | BUG_ON(cgrp->subsys[i]); |
ddbcc7e8 PM |
895 | BUG_ON(!dummytop->subsys[i]); |
896 | BUG_ON(dummytop->subsys[i]->cgroup != dummytop); | |
999cd8a4 | 897 | mutex_lock(&ss->hierarchy_mutex); |
bd89aabc PM |
898 | cgrp->subsys[i] = dummytop->subsys[i]; |
899 | cgrp->subsys[i]->cgroup = cgrp; | |
33a68ac1 | 900 | list_move(&ss->sibling, &root->subsys_list); |
b2aa30f7 | 901 | ss->root = root; |
ddbcc7e8 | 902 | if (ss->bind) |
bd89aabc | 903 | ss->bind(ss, cgrp); |
999cd8a4 | 904 | mutex_unlock(&ss->hierarchy_mutex); |
ddbcc7e8 PM |
905 | } else if (bit & removed_bits) { |
906 | /* We're removing this subsystem */ | |
bd89aabc PM |
907 | BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]); |
908 | BUG_ON(cgrp->subsys[i]->cgroup != cgrp); | |
999cd8a4 | 909 | mutex_lock(&ss->hierarchy_mutex); |
ddbcc7e8 PM |
910 | if (ss->bind) |
911 | ss->bind(ss, dummytop); | |
912 | dummytop->subsys[i]->cgroup = dummytop; | |
bd89aabc | 913 | cgrp->subsys[i] = NULL; |
b2aa30f7 | 914 | subsys[i]->root = &rootnode; |
33a68ac1 | 915 | list_move(&ss->sibling, &rootnode.subsys_list); |
999cd8a4 | 916 | mutex_unlock(&ss->hierarchy_mutex); |
ddbcc7e8 PM |
917 | } else if (bit & final_bits) { |
918 | /* Subsystem state should already exist */ | |
bd89aabc | 919 | BUG_ON(!cgrp->subsys[i]); |
ddbcc7e8 PM |
920 | } else { |
921 | /* Subsystem state shouldn't exist */ | |
bd89aabc | 922 | BUG_ON(cgrp->subsys[i]); |
ddbcc7e8 PM |
923 | } |
924 | } | |
925 | root->subsys_bits = root->actual_subsys_bits = final_bits; | |
926 | synchronize_rcu(); | |
927 | ||
928 | return 0; | |
929 | } | |
930 | ||
931 | static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs) | |
932 | { | |
933 | struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info; | |
934 | struct cgroup_subsys *ss; | |
935 | ||
936 | mutex_lock(&cgroup_mutex); | |
937 | for_each_subsys(root, ss) | |
938 | seq_printf(seq, ",%s", ss->name); | |
939 | if (test_bit(ROOT_NOPREFIX, &root->flags)) | |
940 | seq_puts(seq, ",noprefix"); | |
81a6a5cd PM |
941 | if (strlen(root->release_agent_path)) |
942 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
c6d57f33 PM |
943 | if (strlen(root->name)) |
944 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
945 | mutex_unlock(&cgroup_mutex); |
946 | return 0; | |
947 | } | |
948 | ||
949 | struct cgroup_sb_opts { | |
950 | unsigned long subsys_bits; | |
951 | unsigned long flags; | |
81a6a5cd | 952 | char *release_agent; |
c6d57f33 | 953 | char *name; |
2c6ab6d2 PM |
954 | /* User explicitly requested empty subsystem */ |
955 | bool none; | |
c6d57f33 PM |
956 | |
957 | struct cgroupfs_root *new_root; | |
2c6ab6d2 | 958 | |
ddbcc7e8 PM |
959 | }; |
960 | ||
961 | /* Convert a hierarchy specifier into a bitmask of subsystems and | |
962 | * flags. */ | |
963 | static int parse_cgroupfs_options(char *data, | |
964 | struct cgroup_sb_opts *opts) | |
965 | { | |
966 | char *token, *o = data ?: "all"; | |
f9ab5b5b LZ |
967 | unsigned long mask = (unsigned long)-1; |
968 | ||
969 | #ifdef CONFIG_CPUSETS | |
970 | mask = ~(1UL << cpuset_subsys_id); | |
971 | #endif | |
ddbcc7e8 | 972 | |
c6d57f33 | 973 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
974 | |
975 | while ((token = strsep(&o, ",")) != NULL) { | |
976 | if (!*token) | |
977 | return -EINVAL; | |
978 | if (!strcmp(token, "all")) { | |
8bab8dde PM |
979 | /* Add all non-disabled subsystems */ |
980 | int i; | |
981 | opts->subsys_bits = 0; | |
982 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
983 | struct cgroup_subsys *ss = subsys[i]; | |
984 | if (!ss->disabled) | |
985 | opts->subsys_bits |= 1ul << i; | |
986 | } | |
2c6ab6d2 PM |
987 | } else if (!strcmp(token, "none")) { |
988 | /* Explicitly have no subsystems */ | |
989 | opts->none = true; | |
ddbcc7e8 PM |
990 | } else if (!strcmp(token, "noprefix")) { |
991 | set_bit(ROOT_NOPREFIX, &opts->flags); | |
81a6a5cd PM |
992 | } else if (!strncmp(token, "release_agent=", 14)) { |
993 | /* Specifying two release agents is forbidden */ | |
994 | if (opts->release_agent) | |
995 | return -EINVAL; | |
c6d57f33 PM |
996 | opts->release_agent = |
997 | kstrndup(token + 14, PATH_MAX, GFP_KERNEL); | |
81a6a5cd PM |
998 | if (!opts->release_agent) |
999 | return -ENOMEM; | |
c6d57f33 PM |
1000 | } else if (!strncmp(token, "name=", 5)) { |
1001 | int i; | |
1002 | const char *name = token + 5; | |
1003 | /* Can't specify an empty name */ | |
1004 | if (!strlen(name)) | |
1005 | return -EINVAL; | |
1006 | /* Must match [\w.-]+ */ | |
1007 | for (i = 0; i < strlen(name); i++) { | |
1008 | char c = name[i]; | |
1009 | if (isalnum(c)) | |
1010 | continue; | |
1011 | if ((c == '.') || (c == '-') || (c == '_')) | |
1012 | continue; | |
1013 | return -EINVAL; | |
1014 | } | |
1015 | /* Specifying two names is forbidden */ | |
1016 | if (opts->name) | |
1017 | return -EINVAL; | |
1018 | opts->name = kstrndup(name, | |
1019 | MAX_CGROUP_ROOT_NAMELEN, | |
1020 | GFP_KERNEL); | |
1021 | if (!opts->name) | |
1022 | return -ENOMEM; | |
ddbcc7e8 PM |
1023 | } else { |
1024 | struct cgroup_subsys *ss; | |
1025 | int i; | |
1026 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1027 | ss = subsys[i]; | |
1028 | if (!strcmp(token, ss->name)) { | |
8bab8dde PM |
1029 | if (!ss->disabled) |
1030 | set_bit(i, &opts->subsys_bits); | |
ddbcc7e8 PM |
1031 | break; |
1032 | } | |
1033 | } | |
1034 | if (i == CGROUP_SUBSYS_COUNT) | |
1035 | return -ENOENT; | |
1036 | } | |
1037 | } | |
1038 | ||
2c6ab6d2 PM |
1039 | /* Consistency checks */ |
1040 | ||
f9ab5b5b LZ |
1041 | /* |
1042 | * Option noprefix was introduced just for backward compatibility | |
1043 | * with the old cpuset, so we allow noprefix only if mounting just | |
1044 | * the cpuset subsystem. | |
1045 | */ | |
1046 | if (test_bit(ROOT_NOPREFIX, &opts->flags) && | |
1047 | (opts->subsys_bits & mask)) | |
1048 | return -EINVAL; | |
1049 | ||
2c6ab6d2 PM |
1050 | |
1051 | /* Can't specify "none" and some subsystems */ | |
1052 | if (opts->subsys_bits && opts->none) | |
1053 | return -EINVAL; | |
1054 | ||
1055 | /* | |
1056 | * We either have to specify by name or by subsystems. (So all | |
1057 | * empty hierarchies must have a name). | |
1058 | */ | |
c6d57f33 | 1059 | if (!opts->subsys_bits && !opts->name) |
ddbcc7e8 PM |
1060 | return -EINVAL; |
1061 | ||
1062 | return 0; | |
1063 | } | |
1064 | ||
1065 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) | |
1066 | { | |
1067 | int ret = 0; | |
1068 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1069 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
1070 | struct cgroup_sb_opts opts; |
1071 | ||
337eb00a | 1072 | lock_kernel(); |
bd89aabc | 1073 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
1074 | mutex_lock(&cgroup_mutex); |
1075 | ||
1076 | /* See what subsystems are wanted */ | |
1077 | ret = parse_cgroupfs_options(data, &opts); | |
1078 | if (ret) | |
1079 | goto out_unlock; | |
1080 | ||
1081 | /* Don't allow flags to change at remount */ | |
1082 | if (opts.flags != root->flags) { | |
1083 | ret = -EINVAL; | |
1084 | goto out_unlock; | |
1085 | } | |
1086 | ||
c6d57f33 PM |
1087 | /* Don't allow name to change at remount */ |
1088 | if (opts.name && strcmp(opts.name, root->name)) { | |
1089 | ret = -EINVAL; | |
1090 | goto out_unlock; | |
1091 | } | |
1092 | ||
ddbcc7e8 | 1093 | ret = rebind_subsystems(root, opts.subsys_bits); |
0670e08b LZ |
1094 | if (ret) |
1095 | goto out_unlock; | |
ddbcc7e8 PM |
1096 | |
1097 | /* (re)populate subsystem files */ | |
0670e08b | 1098 | cgroup_populate_dir(cgrp); |
ddbcc7e8 | 1099 | |
81a6a5cd PM |
1100 | if (opts.release_agent) |
1101 | strcpy(root->release_agent_path, opts.release_agent); | |
ddbcc7e8 | 1102 | out_unlock: |
66bdc9cf | 1103 | kfree(opts.release_agent); |
c6d57f33 | 1104 | kfree(opts.name); |
ddbcc7e8 | 1105 | mutex_unlock(&cgroup_mutex); |
bd89aabc | 1106 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
337eb00a | 1107 | unlock_kernel(); |
ddbcc7e8 PM |
1108 | return ret; |
1109 | } | |
1110 | ||
b87221de | 1111 | static const struct super_operations cgroup_ops = { |
ddbcc7e8 PM |
1112 | .statfs = simple_statfs, |
1113 | .drop_inode = generic_delete_inode, | |
1114 | .show_options = cgroup_show_options, | |
1115 | .remount_fs = cgroup_remount, | |
1116 | }; | |
1117 | ||
cc31edce PM |
1118 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1119 | { | |
1120 | INIT_LIST_HEAD(&cgrp->sibling); | |
1121 | INIT_LIST_HEAD(&cgrp->children); | |
1122 | INIT_LIST_HEAD(&cgrp->css_sets); | |
1123 | INIT_LIST_HEAD(&cgrp->release_list); | |
1124 | init_rwsem(&cgrp->pids_mutex); | |
1125 | } | |
c6d57f33 | 1126 | |
ddbcc7e8 PM |
1127 | static void init_cgroup_root(struct cgroupfs_root *root) |
1128 | { | |
bd89aabc | 1129 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
1130 | INIT_LIST_HEAD(&root->subsys_list); |
1131 | INIT_LIST_HEAD(&root->root_list); | |
1132 | root->number_of_cgroups = 1; | |
bd89aabc PM |
1133 | cgrp->root = root; |
1134 | cgrp->top_cgroup = cgrp; | |
cc31edce | 1135 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 PM |
1136 | } |
1137 | ||
2c6ab6d2 PM |
1138 | static bool init_root_id(struct cgroupfs_root *root) |
1139 | { | |
1140 | int ret = 0; | |
1141 | ||
1142 | do { | |
1143 | if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL)) | |
1144 | return false; | |
1145 | spin_lock(&hierarchy_id_lock); | |
1146 | /* Try to allocate the next unused ID */ | |
1147 | ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id, | |
1148 | &root->hierarchy_id); | |
1149 | if (ret == -ENOSPC) | |
1150 | /* Try again starting from 0 */ | |
1151 | ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id); | |
1152 | if (!ret) { | |
1153 | next_hierarchy_id = root->hierarchy_id + 1; | |
1154 | } else if (ret != -EAGAIN) { | |
1155 | /* Can only get here if the 31-bit IDR is full ... */ | |
1156 | BUG_ON(ret); | |
1157 | } | |
1158 | spin_unlock(&hierarchy_id_lock); | |
1159 | } while (ret); | |
1160 | return true; | |
1161 | } | |
1162 | ||
ddbcc7e8 PM |
1163 | static int cgroup_test_super(struct super_block *sb, void *data) |
1164 | { | |
c6d57f33 | 1165 | struct cgroup_sb_opts *opts = data; |
ddbcc7e8 PM |
1166 | struct cgroupfs_root *root = sb->s_fs_info; |
1167 | ||
c6d57f33 PM |
1168 | /* If we asked for a name then it must match */ |
1169 | if (opts->name && strcmp(opts->name, root->name)) | |
1170 | return 0; | |
ddbcc7e8 | 1171 | |
2c6ab6d2 PM |
1172 | /* |
1173 | * If we asked for subsystems (or explicitly for no | |
1174 | * subsystems) then they must match | |
1175 | */ | |
1176 | if ((opts->subsys_bits || opts->none) | |
1177 | && (opts->subsys_bits != root->subsys_bits)) | |
ddbcc7e8 PM |
1178 | return 0; |
1179 | ||
1180 | return 1; | |
1181 | } | |
1182 | ||
c6d57f33 PM |
1183 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1184 | { | |
1185 | struct cgroupfs_root *root; | |
1186 | ||
2c6ab6d2 | 1187 | if (!opts->subsys_bits && !opts->none) |
c6d57f33 PM |
1188 | return NULL; |
1189 | ||
1190 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1191 | if (!root) | |
1192 | return ERR_PTR(-ENOMEM); | |
1193 | ||
2c6ab6d2 PM |
1194 | if (!init_root_id(root)) { |
1195 | kfree(root); | |
1196 | return ERR_PTR(-ENOMEM); | |
1197 | } | |
c6d57f33 | 1198 | init_cgroup_root(root); |
2c6ab6d2 | 1199 | |
c6d57f33 PM |
1200 | root->subsys_bits = opts->subsys_bits; |
1201 | root->flags = opts->flags; | |
1202 | if (opts->release_agent) | |
1203 | strcpy(root->release_agent_path, opts->release_agent); | |
1204 | if (opts->name) | |
1205 | strcpy(root->name, opts->name); | |
1206 | return root; | |
1207 | } | |
1208 | ||
2c6ab6d2 PM |
1209 | static void cgroup_drop_root(struct cgroupfs_root *root) |
1210 | { | |
1211 | if (!root) | |
1212 | return; | |
1213 | ||
1214 | BUG_ON(!root->hierarchy_id); | |
1215 | spin_lock(&hierarchy_id_lock); | |
1216 | ida_remove(&hierarchy_ida, root->hierarchy_id); | |
1217 | spin_unlock(&hierarchy_id_lock); | |
1218 | kfree(root); | |
1219 | } | |
1220 | ||
ddbcc7e8 PM |
1221 | static int cgroup_set_super(struct super_block *sb, void *data) |
1222 | { | |
1223 | int ret; | |
c6d57f33 PM |
1224 | struct cgroup_sb_opts *opts = data; |
1225 | ||
1226 | /* If we don't have a new root, we can't set up a new sb */ | |
1227 | if (!opts->new_root) | |
1228 | return -EINVAL; | |
1229 | ||
2c6ab6d2 | 1230 | BUG_ON(!opts->subsys_bits && !opts->none); |
ddbcc7e8 PM |
1231 | |
1232 | ret = set_anon_super(sb, NULL); | |
1233 | if (ret) | |
1234 | return ret; | |
1235 | ||
c6d57f33 PM |
1236 | sb->s_fs_info = opts->new_root; |
1237 | opts->new_root->sb = sb; | |
ddbcc7e8 PM |
1238 | |
1239 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
1240 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
1241 | sb->s_magic = CGROUP_SUPER_MAGIC; | |
1242 | sb->s_op = &cgroup_ops; | |
1243 | ||
1244 | return 0; | |
1245 | } | |
1246 | ||
1247 | static int cgroup_get_rootdir(struct super_block *sb) | |
1248 | { | |
1249 | struct inode *inode = | |
1250 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); | |
1251 | struct dentry *dentry; | |
1252 | ||
1253 | if (!inode) | |
1254 | return -ENOMEM; | |
1255 | ||
ddbcc7e8 PM |
1256 | inode->i_fop = &simple_dir_operations; |
1257 | inode->i_op = &cgroup_dir_inode_operations; | |
1258 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
1259 | inc_nlink(inode); | |
1260 | dentry = d_alloc_root(inode); | |
1261 | if (!dentry) { | |
1262 | iput(inode); | |
1263 | return -ENOMEM; | |
1264 | } | |
1265 | sb->s_root = dentry; | |
1266 | return 0; | |
1267 | } | |
1268 | ||
1269 | static int cgroup_get_sb(struct file_system_type *fs_type, | |
1270 | int flags, const char *unused_dev_name, | |
1271 | void *data, struct vfsmount *mnt) | |
1272 | { | |
1273 | struct cgroup_sb_opts opts; | |
c6d57f33 | 1274 | struct cgroupfs_root *root; |
ddbcc7e8 PM |
1275 | int ret = 0; |
1276 | struct super_block *sb; | |
c6d57f33 | 1277 | struct cgroupfs_root *new_root; |
ddbcc7e8 PM |
1278 | |
1279 | /* First find the desired set of subsystems */ | |
1280 | ret = parse_cgroupfs_options(data, &opts); | |
c6d57f33 PM |
1281 | if (ret) |
1282 | goto out_err; | |
ddbcc7e8 | 1283 | |
c6d57f33 PM |
1284 | /* |
1285 | * Allocate a new cgroup root. We may not need it if we're | |
1286 | * reusing an existing hierarchy. | |
1287 | */ | |
1288 | new_root = cgroup_root_from_opts(&opts); | |
1289 | if (IS_ERR(new_root)) { | |
1290 | ret = PTR_ERR(new_root); | |
1291 | goto out_err; | |
81a6a5cd | 1292 | } |
c6d57f33 | 1293 | opts.new_root = new_root; |
ddbcc7e8 | 1294 | |
c6d57f33 PM |
1295 | /* Locate an existing or new sb for this hierarchy */ |
1296 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts); | |
ddbcc7e8 | 1297 | if (IS_ERR(sb)) { |
c6d57f33 | 1298 | ret = PTR_ERR(sb); |
2c6ab6d2 | 1299 | cgroup_drop_root(opts.new_root); |
c6d57f33 | 1300 | goto out_err; |
ddbcc7e8 PM |
1301 | } |
1302 | ||
c6d57f33 PM |
1303 | root = sb->s_fs_info; |
1304 | BUG_ON(!root); | |
1305 | if (root == opts.new_root) { | |
1306 | /* We used the new root structure, so this is a new hierarchy */ | |
1307 | struct list_head tmp_cg_links; | |
c12f65d4 | 1308 | struct cgroup *root_cgrp = &root->top_cgroup; |
817929ec | 1309 | struct inode *inode; |
c6d57f33 | 1310 | struct cgroupfs_root *existing_root; |
28fd5dfc | 1311 | int i; |
ddbcc7e8 PM |
1312 | |
1313 | BUG_ON(sb->s_root != NULL); | |
1314 | ||
1315 | ret = cgroup_get_rootdir(sb); | |
1316 | if (ret) | |
1317 | goto drop_new_super; | |
817929ec | 1318 | inode = sb->s_root->d_inode; |
ddbcc7e8 | 1319 | |
817929ec | 1320 | mutex_lock(&inode->i_mutex); |
ddbcc7e8 PM |
1321 | mutex_lock(&cgroup_mutex); |
1322 | ||
c6d57f33 PM |
1323 | if (strlen(root->name)) { |
1324 | /* Check for name clashes with existing mounts */ | |
1325 | for_each_active_root(existing_root) { | |
1326 | if (!strcmp(existing_root->name, root->name)) { | |
1327 | ret = -EBUSY; | |
1328 | mutex_unlock(&cgroup_mutex); | |
1329 | mutex_unlock(&inode->i_mutex); | |
1330 | goto drop_new_super; | |
1331 | } | |
1332 | } | |
1333 | } | |
1334 | ||
817929ec PM |
1335 | /* |
1336 | * We're accessing css_set_count without locking | |
1337 | * css_set_lock here, but that's OK - it can only be | |
1338 | * increased by someone holding cgroup_lock, and | |
1339 | * that's us. The worst that can happen is that we | |
1340 | * have some link structures left over | |
1341 | */ | |
1342 | ret = allocate_cg_links(css_set_count, &tmp_cg_links); | |
1343 | if (ret) { | |
1344 | mutex_unlock(&cgroup_mutex); | |
1345 | mutex_unlock(&inode->i_mutex); | |
1346 | goto drop_new_super; | |
1347 | } | |
1348 | ||
ddbcc7e8 PM |
1349 | ret = rebind_subsystems(root, root->subsys_bits); |
1350 | if (ret == -EBUSY) { | |
1351 | mutex_unlock(&cgroup_mutex); | |
817929ec | 1352 | mutex_unlock(&inode->i_mutex); |
c6d57f33 PM |
1353 | free_cg_links(&tmp_cg_links); |
1354 | goto drop_new_super; | |
ddbcc7e8 PM |
1355 | } |
1356 | ||
1357 | /* EBUSY should be the only error here */ | |
1358 | BUG_ON(ret); | |
1359 | ||
1360 | list_add(&root->root_list, &roots); | |
817929ec | 1361 | root_count++; |
ddbcc7e8 | 1362 | |
c12f65d4 | 1363 | sb->s_root->d_fsdata = root_cgrp; |
ddbcc7e8 PM |
1364 | root->top_cgroup.dentry = sb->s_root; |
1365 | ||
817929ec PM |
1366 | /* Link the top cgroup in this hierarchy into all |
1367 | * the css_set objects */ | |
1368 | write_lock(&css_set_lock); | |
28fd5dfc LZ |
1369 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { |
1370 | struct hlist_head *hhead = &css_set_table[i]; | |
1371 | struct hlist_node *node; | |
817929ec | 1372 | struct css_set *cg; |
28fd5dfc | 1373 | |
c12f65d4 LZ |
1374 | hlist_for_each_entry(cg, node, hhead, hlist) |
1375 | link_css_set(&tmp_cg_links, cg, root_cgrp); | |
28fd5dfc | 1376 | } |
817929ec PM |
1377 | write_unlock(&css_set_lock); |
1378 | ||
1379 | free_cg_links(&tmp_cg_links); | |
1380 | ||
c12f65d4 LZ |
1381 | BUG_ON(!list_empty(&root_cgrp->sibling)); |
1382 | BUG_ON(!list_empty(&root_cgrp->children)); | |
ddbcc7e8 PM |
1383 | BUG_ON(root->number_of_cgroups != 1); |
1384 | ||
c12f65d4 | 1385 | cgroup_populate_dir(root_cgrp); |
ddbcc7e8 | 1386 | mutex_unlock(&cgroup_mutex); |
34f77a90 | 1387 | mutex_unlock(&inode->i_mutex); |
c6d57f33 PM |
1388 | } else { |
1389 | /* | |
1390 | * We re-used an existing hierarchy - the new root (if | |
1391 | * any) is not needed | |
1392 | */ | |
2c6ab6d2 | 1393 | cgroup_drop_root(opts.new_root); |
ddbcc7e8 PM |
1394 | } |
1395 | ||
a3ec947c | 1396 | simple_set_mnt(mnt, sb); |
c6d57f33 PM |
1397 | kfree(opts.release_agent); |
1398 | kfree(opts.name); | |
a3ec947c | 1399 | return 0; |
ddbcc7e8 PM |
1400 | |
1401 | drop_new_super: | |
6f5bbff9 | 1402 | deactivate_locked_super(sb); |
c6d57f33 PM |
1403 | out_err: |
1404 | kfree(opts.release_agent); | |
1405 | kfree(opts.name); | |
1406 | ||
ddbcc7e8 PM |
1407 | return ret; |
1408 | } | |
1409 | ||
1410 | static void cgroup_kill_sb(struct super_block *sb) { | |
1411 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1412 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 | 1413 | int ret; |
71cbb949 KM |
1414 | struct cg_cgroup_link *link; |
1415 | struct cg_cgroup_link *saved_link; | |
ddbcc7e8 PM |
1416 | |
1417 | BUG_ON(!root); | |
1418 | ||
1419 | BUG_ON(root->number_of_cgroups != 1); | |
bd89aabc PM |
1420 | BUG_ON(!list_empty(&cgrp->children)); |
1421 | BUG_ON(!list_empty(&cgrp->sibling)); | |
ddbcc7e8 PM |
1422 | |
1423 | mutex_lock(&cgroup_mutex); | |
1424 | ||
1425 | /* Rebind all subsystems back to the default hierarchy */ | |
1426 | ret = rebind_subsystems(root, 0); | |
1427 | /* Shouldn't be able to fail ... */ | |
1428 | BUG_ON(ret); | |
1429 | ||
817929ec PM |
1430 | /* |
1431 | * Release all the links from css_sets to this hierarchy's | |
1432 | * root cgroup | |
1433 | */ | |
1434 | write_lock(&css_set_lock); | |
71cbb949 KM |
1435 | |
1436 | list_for_each_entry_safe(link, saved_link, &cgrp->css_sets, | |
1437 | cgrp_link_list) { | |
817929ec | 1438 | list_del(&link->cg_link_list); |
bd89aabc | 1439 | list_del(&link->cgrp_link_list); |
817929ec PM |
1440 | kfree(link); |
1441 | } | |
1442 | write_unlock(&css_set_lock); | |
1443 | ||
839ec545 PM |
1444 | if (!list_empty(&root->root_list)) { |
1445 | list_del(&root->root_list); | |
1446 | root_count--; | |
1447 | } | |
e5f6a860 | 1448 | |
ddbcc7e8 PM |
1449 | mutex_unlock(&cgroup_mutex); |
1450 | ||
ddbcc7e8 | 1451 | kill_litter_super(sb); |
2c6ab6d2 | 1452 | cgroup_drop_root(root); |
ddbcc7e8 PM |
1453 | } |
1454 | ||
1455 | static struct file_system_type cgroup_fs_type = { | |
1456 | .name = "cgroup", | |
1457 | .get_sb = cgroup_get_sb, | |
1458 | .kill_sb = cgroup_kill_sb, | |
1459 | }; | |
1460 | ||
bd89aabc | 1461 | static inline struct cgroup *__d_cgrp(struct dentry *dentry) |
ddbcc7e8 PM |
1462 | { |
1463 | return dentry->d_fsdata; | |
1464 | } | |
1465 | ||
1466 | static inline struct cftype *__d_cft(struct dentry *dentry) | |
1467 | { | |
1468 | return dentry->d_fsdata; | |
1469 | } | |
1470 | ||
a043e3b2 LZ |
1471 | /** |
1472 | * cgroup_path - generate the path of a cgroup | |
1473 | * @cgrp: the cgroup in question | |
1474 | * @buf: the buffer to write the path into | |
1475 | * @buflen: the length of the buffer | |
1476 | * | |
a47295e6 PM |
1477 | * Called with cgroup_mutex held or else with an RCU-protected cgroup |
1478 | * reference. Writes path of cgroup into buf. Returns 0 on success, | |
1479 | * -errno on error. | |
ddbcc7e8 | 1480 | */ |
bd89aabc | 1481 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
ddbcc7e8 PM |
1482 | { |
1483 | char *start; | |
a47295e6 | 1484 | struct dentry *dentry = rcu_dereference(cgrp->dentry); |
ddbcc7e8 | 1485 | |
a47295e6 | 1486 | if (!dentry || cgrp == dummytop) { |
ddbcc7e8 PM |
1487 | /* |
1488 | * Inactive subsystems have no dentry for their root | |
1489 | * cgroup | |
1490 | */ | |
1491 | strcpy(buf, "/"); | |
1492 | return 0; | |
1493 | } | |
1494 | ||
1495 | start = buf + buflen; | |
1496 | ||
1497 | *--start = '\0'; | |
1498 | for (;;) { | |
a47295e6 | 1499 | int len = dentry->d_name.len; |
ddbcc7e8 PM |
1500 | if ((start -= len) < buf) |
1501 | return -ENAMETOOLONG; | |
bd89aabc PM |
1502 | memcpy(start, cgrp->dentry->d_name.name, len); |
1503 | cgrp = cgrp->parent; | |
1504 | if (!cgrp) | |
ddbcc7e8 | 1505 | break; |
a47295e6 | 1506 | dentry = rcu_dereference(cgrp->dentry); |
bd89aabc | 1507 | if (!cgrp->parent) |
ddbcc7e8 PM |
1508 | continue; |
1509 | if (--start < buf) | |
1510 | return -ENAMETOOLONG; | |
1511 | *start = '/'; | |
1512 | } | |
1513 | memmove(buf, start, buf + buflen - start); | |
1514 | return 0; | |
1515 | } | |
1516 | ||
a043e3b2 LZ |
1517 | /** |
1518 | * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp' | |
1519 | * @cgrp: the cgroup the task is attaching to | |
1520 | * @tsk: the task to be attached | |
bbcb81d0 | 1521 | * |
a043e3b2 LZ |
1522 | * Call holding cgroup_mutex. May take task_lock of |
1523 | * the task 'tsk' during call. | |
bbcb81d0 | 1524 | */ |
956db3ca | 1525 | int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
bbcb81d0 PM |
1526 | { |
1527 | int retval = 0; | |
1528 | struct cgroup_subsys *ss; | |
bd89aabc | 1529 | struct cgroup *oldcgrp; |
77efecd9 | 1530 | struct css_set *cg; |
817929ec | 1531 | struct css_set *newcg; |
bd89aabc | 1532 | struct cgroupfs_root *root = cgrp->root; |
bbcb81d0 PM |
1533 | |
1534 | /* Nothing to do if the task is already in that cgroup */ | |
7717f7ba | 1535 | oldcgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 1536 | if (cgrp == oldcgrp) |
bbcb81d0 PM |
1537 | return 0; |
1538 | ||
1539 | for_each_subsys(root, ss) { | |
1540 | if (ss->can_attach) { | |
bd89aabc | 1541 | retval = ss->can_attach(ss, cgrp, tsk); |
e18f6318 | 1542 | if (retval) |
bbcb81d0 | 1543 | return retval; |
bbcb81d0 PM |
1544 | } |
1545 | } | |
1546 | ||
77efecd9 LJ |
1547 | task_lock(tsk); |
1548 | cg = tsk->cgroups; | |
1549 | get_css_set(cg); | |
1550 | task_unlock(tsk); | |
817929ec PM |
1551 | /* |
1552 | * Locate or allocate a new css_set for this task, | |
1553 | * based on its final set of cgroups | |
1554 | */ | |
bd89aabc | 1555 | newcg = find_css_set(cg, cgrp); |
77efecd9 | 1556 | put_css_set(cg); |
e18f6318 | 1557 | if (!newcg) |
817929ec | 1558 | return -ENOMEM; |
817929ec | 1559 | |
bbcb81d0 PM |
1560 | task_lock(tsk); |
1561 | if (tsk->flags & PF_EXITING) { | |
1562 | task_unlock(tsk); | |
817929ec | 1563 | put_css_set(newcg); |
bbcb81d0 PM |
1564 | return -ESRCH; |
1565 | } | |
817929ec | 1566 | rcu_assign_pointer(tsk->cgroups, newcg); |
bbcb81d0 PM |
1567 | task_unlock(tsk); |
1568 | ||
817929ec PM |
1569 | /* Update the css_set linked lists if we're using them */ |
1570 | write_lock(&css_set_lock); | |
1571 | if (!list_empty(&tsk->cg_list)) { | |
1572 | list_del(&tsk->cg_list); | |
1573 | list_add(&tsk->cg_list, &newcg->tasks); | |
1574 | } | |
1575 | write_unlock(&css_set_lock); | |
1576 | ||
bbcb81d0 | 1577 | for_each_subsys(root, ss) { |
e18f6318 | 1578 | if (ss->attach) |
bd89aabc | 1579 | ss->attach(ss, cgrp, oldcgrp, tsk); |
bbcb81d0 | 1580 | } |
bd89aabc | 1581 | set_bit(CGRP_RELEASABLE, &oldcgrp->flags); |
bbcb81d0 | 1582 | synchronize_rcu(); |
817929ec | 1583 | put_css_set(cg); |
ec64f515 KH |
1584 | |
1585 | /* | |
1586 | * wake up rmdir() waiter. the rmdir should fail since the cgroup | |
1587 | * is no longer empty. | |
1588 | */ | |
88703267 | 1589 | cgroup_wakeup_rmdir_waiter(cgrp); |
bbcb81d0 PM |
1590 | return 0; |
1591 | } | |
1592 | ||
1593 | /* | |
af351026 PM |
1594 | * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex |
1595 | * held. May take task_lock of task | |
bbcb81d0 | 1596 | */ |
af351026 | 1597 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid) |
bbcb81d0 | 1598 | { |
bbcb81d0 | 1599 | struct task_struct *tsk; |
c69e8d9c | 1600 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
1601 | int ret; |
1602 | ||
bbcb81d0 PM |
1603 | if (pid) { |
1604 | rcu_read_lock(); | |
73507f33 | 1605 | tsk = find_task_by_vpid(pid); |
bbcb81d0 PM |
1606 | if (!tsk || tsk->flags & PF_EXITING) { |
1607 | rcu_read_unlock(); | |
1608 | return -ESRCH; | |
1609 | } | |
bbcb81d0 | 1610 | |
c69e8d9c DH |
1611 | tcred = __task_cred(tsk); |
1612 | if (cred->euid && | |
1613 | cred->euid != tcred->uid && | |
1614 | cred->euid != tcred->suid) { | |
1615 | rcu_read_unlock(); | |
bbcb81d0 PM |
1616 | return -EACCES; |
1617 | } | |
c69e8d9c DH |
1618 | get_task_struct(tsk); |
1619 | rcu_read_unlock(); | |
bbcb81d0 PM |
1620 | } else { |
1621 | tsk = current; | |
1622 | get_task_struct(tsk); | |
1623 | } | |
1624 | ||
956db3ca | 1625 | ret = cgroup_attach_task(cgrp, tsk); |
bbcb81d0 PM |
1626 | put_task_struct(tsk); |
1627 | return ret; | |
1628 | } | |
1629 | ||
af351026 PM |
1630 | static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid) |
1631 | { | |
1632 | int ret; | |
1633 | if (!cgroup_lock_live_group(cgrp)) | |
1634 | return -ENODEV; | |
1635 | ret = attach_task_by_pid(cgrp, pid); | |
1636 | cgroup_unlock(); | |
1637 | return ret; | |
1638 | } | |
1639 | ||
ddbcc7e8 | 1640 | /* The various types of files and directories in a cgroup file system */ |
ddbcc7e8 PM |
1641 | enum cgroup_filetype { |
1642 | FILE_ROOT, | |
1643 | FILE_DIR, | |
1644 | FILE_TASKLIST, | |
81a6a5cd | 1645 | FILE_NOTIFY_ON_RELEASE, |
81a6a5cd | 1646 | FILE_RELEASE_AGENT, |
ddbcc7e8 PM |
1647 | }; |
1648 | ||
e788e066 PM |
1649 | /** |
1650 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
1651 | * @cgrp: the cgroup to be checked for liveness | |
1652 | * | |
84eea842 PM |
1653 | * On success, returns true; the lock should be later released with |
1654 | * cgroup_unlock(). On failure returns false with no lock held. | |
e788e066 | 1655 | */ |
84eea842 | 1656 | bool cgroup_lock_live_group(struct cgroup *cgrp) |
e788e066 PM |
1657 | { |
1658 | mutex_lock(&cgroup_mutex); | |
1659 | if (cgroup_is_removed(cgrp)) { | |
1660 | mutex_unlock(&cgroup_mutex); | |
1661 | return false; | |
1662 | } | |
1663 | return true; | |
1664 | } | |
1665 | ||
1666 | static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft, | |
1667 | const char *buffer) | |
1668 | { | |
1669 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
1670 | if (!cgroup_lock_live_group(cgrp)) | |
1671 | return -ENODEV; | |
1672 | strcpy(cgrp->root->release_agent_path, buffer); | |
84eea842 | 1673 | cgroup_unlock(); |
e788e066 PM |
1674 | return 0; |
1675 | } | |
1676 | ||
1677 | static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft, | |
1678 | struct seq_file *seq) | |
1679 | { | |
1680 | if (!cgroup_lock_live_group(cgrp)) | |
1681 | return -ENODEV; | |
1682 | seq_puts(seq, cgrp->root->release_agent_path); | |
1683 | seq_putc(seq, '\n'); | |
84eea842 | 1684 | cgroup_unlock(); |
e788e066 PM |
1685 | return 0; |
1686 | } | |
1687 | ||
84eea842 PM |
1688 | /* A buffer size big enough for numbers or short strings */ |
1689 | #define CGROUP_LOCAL_BUFFER_SIZE 64 | |
1690 | ||
e73d2c61 | 1691 | static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft, |
f4c753b7 PM |
1692 | struct file *file, |
1693 | const char __user *userbuf, | |
1694 | size_t nbytes, loff_t *unused_ppos) | |
355e0c48 | 1695 | { |
84eea842 | 1696 | char buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
355e0c48 | 1697 | int retval = 0; |
355e0c48 PM |
1698 | char *end; |
1699 | ||
1700 | if (!nbytes) | |
1701 | return -EINVAL; | |
1702 | if (nbytes >= sizeof(buffer)) | |
1703 | return -E2BIG; | |
1704 | if (copy_from_user(buffer, userbuf, nbytes)) | |
1705 | return -EFAULT; | |
1706 | ||
1707 | buffer[nbytes] = 0; /* nul-terminate */ | |
b7269dfc | 1708 | strstrip(buffer); |
e73d2c61 PM |
1709 | if (cft->write_u64) { |
1710 | u64 val = simple_strtoull(buffer, &end, 0); | |
1711 | if (*end) | |
1712 | return -EINVAL; | |
1713 | retval = cft->write_u64(cgrp, cft, val); | |
1714 | } else { | |
1715 | s64 val = simple_strtoll(buffer, &end, 0); | |
1716 | if (*end) | |
1717 | return -EINVAL; | |
1718 | retval = cft->write_s64(cgrp, cft, val); | |
1719 | } | |
355e0c48 PM |
1720 | if (!retval) |
1721 | retval = nbytes; | |
1722 | return retval; | |
1723 | } | |
1724 | ||
db3b1497 PM |
1725 | static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft, |
1726 | struct file *file, | |
1727 | const char __user *userbuf, | |
1728 | size_t nbytes, loff_t *unused_ppos) | |
1729 | { | |
84eea842 | 1730 | char local_buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
db3b1497 PM |
1731 | int retval = 0; |
1732 | size_t max_bytes = cft->max_write_len; | |
1733 | char *buffer = local_buffer; | |
1734 | ||
1735 | if (!max_bytes) | |
1736 | max_bytes = sizeof(local_buffer) - 1; | |
1737 | if (nbytes >= max_bytes) | |
1738 | return -E2BIG; | |
1739 | /* Allocate a dynamic buffer if we need one */ | |
1740 | if (nbytes >= sizeof(local_buffer)) { | |
1741 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); | |
1742 | if (buffer == NULL) | |
1743 | return -ENOMEM; | |
1744 | } | |
5a3eb9f6 LZ |
1745 | if (nbytes && copy_from_user(buffer, userbuf, nbytes)) { |
1746 | retval = -EFAULT; | |
1747 | goto out; | |
1748 | } | |
db3b1497 PM |
1749 | |
1750 | buffer[nbytes] = 0; /* nul-terminate */ | |
1751 | strstrip(buffer); | |
1752 | retval = cft->write_string(cgrp, cft, buffer); | |
1753 | if (!retval) | |
1754 | retval = nbytes; | |
5a3eb9f6 | 1755 | out: |
db3b1497 PM |
1756 | if (buffer != local_buffer) |
1757 | kfree(buffer); | |
1758 | return retval; | |
1759 | } | |
1760 | ||
ddbcc7e8 PM |
1761 | static ssize_t cgroup_file_write(struct file *file, const char __user *buf, |
1762 | size_t nbytes, loff_t *ppos) | |
1763 | { | |
1764 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 1765 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 1766 | |
75139b82 | 1767 | if (cgroup_is_removed(cgrp)) |
ddbcc7e8 | 1768 | return -ENODEV; |
355e0c48 | 1769 | if (cft->write) |
bd89aabc | 1770 | return cft->write(cgrp, cft, file, buf, nbytes, ppos); |
e73d2c61 PM |
1771 | if (cft->write_u64 || cft->write_s64) |
1772 | return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos); | |
db3b1497 PM |
1773 | if (cft->write_string) |
1774 | return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos); | |
d447ea2f PE |
1775 | if (cft->trigger) { |
1776 | int ret = cft->trigger(cgrp, (unsigned int)cft->private); | |
1777 | return ret ? ret : nbytes; | |
1778 | } | |
355e0c48 | 1779 | return -EINVAL; |
ddbcc7e8 PM |
1780 | } |
1781 | ||
f4c753b7 PM |
1782 | static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft, |
1783 | struct file *file, | |
1784 | char __user *buf, size_t nbytes, | |
1785 | loff_t *ppos) | |
ddbcc7e8 | 1786 | { |
84eea842 | 1787 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
f4c753b7 | 1788 | u64 val = cft->read_u64(cgrp, cft); |
ddbcc7e8 PM |
1789 | int len = sprintf(tmp, "%llu\n", (unsigned long long) val); |
1790 | ||
1791 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
1792 | } | |
1793 | ||
e73d2c61 PM |
1794 | static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft, |
1795 | struct file *file, | |
1796 | char __user *buf, size_t nbytes, | |
1797 | loff_t *ppos) | |
1798 | { | |
84eea842 | 1799 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
e73d2c61 PM |
1800 | s64 val = cft->read_s64(cgrp, cft); |
1801 | int len = sprintf(tmp, "%lld\n", (long long) val); | |
1802 | ||
1803 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
1804 | } | |
1805 | ||
ddbcc7e8 PM |
1806 | static ssize_t cgroup_file_read(struct file *file, char __user *buf, |
1807 | size_t nbytes, loff_t *ppos) | |
1808 | { | |
1809 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 1810 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 1811 | |
75139b82 | 1812 | if (cgroup_is_removed(cgrp)) |
ddbcc7e8 PM |
1813 | return -ENODEV; |
1814 | ||
1815 | if (cft->read) | |
bd89aabc | 1816 | return cft->read(cgrp, cft, file, buf, nbytes, ppos); |
f4c753b7 PM |
1817 | if (cft->read_u64) |
1818 | return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos); | |
e73d2c61 PM |
1819 | if (cft->read_s64) |
1820 | return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos); | |
ddbcc7e8 PM |
1821 | return -EINVAL; |
1822 | } | |
1823 | ||
91796569 PM |
1824 | /* |
1825 | * seqfile ops/methods for returning structured data. Currently just | |
1826 | * supports string->u64 maps, but can be extended in future. | |
1827 | */ | |
1828 | ||
1829 | struct cgroup_seqfile_state { | |
1830 | struct cftype *cft; | |
1831 | struct cgroup *cgroup; | |
1832 | }; | |
1833 | ||
1834 | static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value) | |
1835 | { | |
1836 | struct seq_file *sf = cb->state; | |
1837 | return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value); | |
1838 | } | |
1839 | ||
1840 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) | |
1841 | { | |
1842 | struct cgroup_seqfile_state *state = m->private; | |
1843 | struct cftype *cft = state->cft; | |
29486df3 SH |
1844 | if (cft->read_map) { |
1845 | struct cgroup_map_cb cb = { | |
1846 | .fill = cgroup_map_add, | |
1847 | .state = m, | |
1848 | }; | |
1849 | return cft->read_map(state->cgroup, cft, &cb); | |
1850 | } | |
1851 | return cft->read_seq_string(state->cgroup, cft, m); | |
91796569 PM |
1852 | } |
1853 | ||
96930a63 | 1854 | static int cgroup_seqfile_release(struct inode *inode, struct file *file) |
91796569 PM |
1855 | { |
1856 | struct seq_file *seq = file->private_data; | |
1857 | kfree(seq->private); | |
1858 | return single_release(inode, file); | |
1859 | } | |
1860 | ||
1861 | static struct file_operations cgroup_seqfile_operations = { | |
1862 | .read = seq_read, | |
e788e066 | 1863 | .write = cgroup_file_write, |
91796569 PM |
1864 | .llseek = seq_lseek, |
1865 | .release = cgroup_seqfile_release, | |
1866 | }; | |
1867 | ||
ddbcc7e8 PM |
1868 | static int cgroup_file_open(struct inode *inode, struct file *file) |
1869 | { | |
1870 | int err; | |
1871 | struct cftype *cft; | |
1872 | ||
1873 | err = generic_file_open(inode, file); | |
1874 | if (err) | |
1875 | return err; | |
ddbcc7e8 | 1876 | cft = __d_cft(file->f_dentry); |
75139b82 | 1877 | |
29486df3 | 1878 | if (cft->read_map || cft->read_seq_string) { |
91796569 PM |
1879 | struct cgroup_seqfile_state *state = |
1880 | kzalloc(sizeof(*state), GFP_USER); | |
1881 | if (!state) | |
1882 | return -ENOMEM; | |
1883 | state->cft = cft; | |
1884 | state->cgroup = __d_cgrp(file->f_dentry->d_parent); | |
1885 | file->f_op = &cgroup_seqfile_operations; | |
1886 | err = single_open(file, cgroup_seqfile_show, state); | |
1887 | if (err < 0) | |
1888 | kfree(state); | |
1889 | } else if (cft->open) | |
ddbcc7e8 PM |
1890 | err = cft->open(inode, file); |
1891 | else | |
1892 | err = 0; | |
1893 | ||
1894 | return err; | |
1895 | } | |
1896 | ||
1897 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
1898 | { | |
1899 | struct cftype *cft = __d_cft(file->f_dentry); | |
1900 | if (cft->release) | |
1901 | return cft->release(inode, file); | |
1902 | return 0; | |
1903 | } | |
1904 | ||
1905 | /* | |
1906 | * cgroup_rename - Only allow simple rename of directories in place. | |
1907 | */ | |
1908 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1909 | struct inode *new_dir, struct dentry *new_dentry) | |
1910 | { | |
1911 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) | |
1912 | return -ENOTDIR; | |
1913 | if (new_dentry->d_inode) | |
1914 | return -EEXIST; | |
1915 | if (old_dir != new_dir) | |
1916 | return -EIO; | |
1917 | return simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
1918 | } | |
1919 | ||
1920 | static struct file_operations cgroup_file_operations = { | |
1921 | .read = cgroup_file_read, | |
1922 | .write = cgroup_file_write, | |
1923 | .llseek = generic_file_llseek, | |
1924 | .open = cgroup_file_open, | |
1925 | .release = cgroup_file_release, | |
1926 | }; | |
1927 | ||
6e1d5dcc | 1928 | static const struct inode_operations cgroup_dir_inode_operations = { |
ddbcc7e8 PM |
1929 | .lookup = simple_lookup, |
1930 | .mkdir = cgroup_mkdir, | |
1931 | .rmdir = cgroup_rmdir, | |
1932 | .rename = cgroup_rename, | |
1933 | }; | |
1934 | ||
099fca32 | 1935 | static int cgroup_create_file(struct dentry *dentry, mode_t mode, |
ddbcc7e8 PM |
1936 | struct super_block *sb) |
1937 | { | |
3ba13d17 | 1938 | static const struct dentry_operations cgroup_dops = { |
ddbcc7e8 PM |
1939 | .d_iput = cgroup_diput, |
1940 | }; | |
1941 | ||
1942 | struct inode *inode; | |
1943 | ||
1944 | if (!dentry) | |
1945 | return -ENOENT; | |
1946 | if (dentry->d_inode) | |
1947 | return -EEXIST; | |
1948 | ||
1949 | inode = cgroup_new_inode(mode, sb); | |
1950 | if (!inode) | |
1951 | return -ENOMEM; | |
1952 | ||
1953 | if (S_ISDIR(mode)) { | |
1954 | inode->i_op = &cgroup_dir_inode_operations; | |
1955 | inode->i_fop = &simple_dir_operations; | |
1956 | ||
1957 | /* start off with i_nlink == 2 (for "." entry) */ | |
1958 | inc_nlink(inode); | |
1959 | ||
1960 | /* start with the directory inode held, so that we can | |
1961 | * populate it without racing with another mkdir */ | |
817929ec | 1962 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); |
ddbcc7e8 PM |
1963 | } else if (S_ISREG(mode)) { |
1964 | inode->i_size = 0; | |
1965 | inode->i_fop = &cgroup_file_operations; | |
1966 | } | |
1967 | dentry->d_op = &cgroup_dops; | |
1968 | d_instantiate(dentry, inode); | |
1969 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1970 | return 0; | |
1971 | } | |
1972 | ||
1973 | /* | |
a043e3b2 LZ |
1974 | * cgroup_create_dir - create a directory for an object. |
1975 | * @cgrp: the cgroup we create the directory for. It must have a valid | |
1976 | * ->parent field. And we are going to fill its ->dentry field. | |
1977 | * @dentry: dentry of the new cgroup | |
1978 | * @mode: mode to set on new directory. | |
ddbcc7e8 | 1979 | */ |
bd89aabc | 1980 | static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry, |
099fca32 | 1981 | mode_t mode) |
ddbcc7e8 PM |
1982 | { |
1983 | struct dentry *parent; | |
1984 | int error = 0; | |
1985 | ||
bd89aabc PM |
1986 | parent = cgrp->parent->dentry; |
1987 | error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb); | |
ddbcc7e8 | 1988 | if (!error) { |
bd89aabc | 1989 | dentry->d_fsdata = cgrp; |
ddbcc7e8 | 1990 | inc_nlink(parent->d_inode); |
a47295e6 | 1991 | rcu_assign_pointer(cgrp->dentry, dentry); |
ddbcc7e8 PM |
1992 | dget(dentry); |
1993 | } | |
1994 | dput(dentry); | |
1995 | ||
1996 | return error; | |
1997 | } | |
1998 | ||
099fca32 LZ |
1999 | /** |
2000 | * cgroup_file_mode - deduce file mode of a control file | |
2001 | * @cft: the control file in question | |
2002 | * | |
2003 | * returns cft->mode if ->mode is not 0 | |
2004 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
2005 | * returns S_IRUGO if it has only a read handler | |
2006 | * returns S_IWUSR if it has only a write hander | |
2007 | */ | |
2008 | static mode_t cgroup_file_mode(const struct cftype *cft) | |
2009 | { | |
2010 | mode_t mode = 0; | |
2011 | ||
2012 | if (cft->mode) | |
2013 | return cft->mode; | |
2014 | ||
2015 | if (cft->read || cft->read_u64 || cft->read_s64 || | |
2016 | cft->read_map || cft->read_seq_string) | |
2017 | mode |= S_IRUGO; | |
2018 | ||
2019 | if (cft->write || cft->write_u64 || cft->write_s64 || | |
2020 | cft->write_string || cft->trigger) | |
2021 | mode |= S_IWUSR; | |
2022 | ||
2023 | return mode; | |
2024 | } | |
2025 | ||
bd89aabc | 2026 | int cgroup_add_file(struct cgroup *cgrp, |
ddbcc7e8 PM |
2027 | struct cgroup_subsys *subsys, |
2028 | const struct cftype *cft) | |
2029 | { | |
bd89aabc | 2030 | struct dentry *dir = cgrp->dentry; |
ddbcc7e8 PM |
2031 | struct dentry *dentry; |
2032 | int error; | |
099fca32 | 2033 | mode_t mode; |
ddbcc7e8 PM |
2034 | |
2035 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; | |
bd89aabc | 2036 | if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) { |
ddbcc7e8 PM |
2037 | strcpy(name, subsys->name); |
2038 | strcat(name, "."); | |
2039 | } | |
2040 | strcat(name, cft->name); | |
2041 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); | |
2042 | dentry = lookup_one_len(name, dir, strlen(name)); | |
2043 | if (!IS_ERR(dentry)) { | |
099fca32 LZ |
2044 | mode = cgroup_file_mode(cft); |
2045 | error = cgroup_create_file(dentry, mode | S_IFREG, | |
bd89aabc | 2046 | cgrp->root->sb); |
ddbcc7e8 PM |
2047 | if (!error) |
2048 | dentry->d_fsdata = (void *)cft; | |
2049 | dput(dentry); | |
2050 | } else | |
2051 | error = PTR_ERR(dentry); | |
2052 | return error; | |
2053 | } | |
2054 | ||
bd89aabc | 2055 | int cgroup_add_files(struct cgroup *cgrp, |
ddbcc7e8 PM |
2056 | struct cgroup_subsys *subsys, |
2057 | const struct cftype cft[], | |
2058 | int count) | |
2059 | { | |
2060 | int i, err; | |
2061 | for (i = 0; i < count; i++) { | |
bd89aabc | 2062 | err = cgroup_add_file(cgrp, subsys, &cft[i]); |
ddbcc7e8 PM |
2063 | if (err) |
2064 | return err; | |
2065 | } | |
2066 | return 0; | |
2067 | } | |
2068 | ||
a043e3b2 LZ |
2069 | /** |
2070 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2071 | * @cgrp: the cgroup in question | |
2072 | * | |
2073 | * Return the number of tasks in the cgroup. | |
2074 | */ | |
bd89aabc | 2075 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2076 | { |
2077 | int count = 0; | |
71cbb949 | 2078 | struct cg_cgroup_link *link; |
817929ec PM |
2079 | |
2080 | read_lock(&css_set_lock); | |
71cbb949 | 2081 | list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) { |
146aa1bd | 2082 | count += atomic_read(&link->cg->refcount); |
817929ec PM |
2083 | } |
2084 | read_unlock(&css_set_lock); | |
bbcb81d0 PM |
2085 | return count; |
2086 | } | |
2087 | ||
817929ec PM |
2088 | /* |
2089 | * Advance a list_head iterator. The iterator should be positioned at | |
2090 | * the start of a css_set | |
2091 | */ | |
bd89aabc | 2092 | static void cgroup_advance_iter(struct cgroup *cgrp, |
7717f7ba | 2093 | struct cgroup_iter *it) |
817929ec PM |
2094 | { |
2095 | struct list_head *l = it->cg_link; | |
2096 | struct cg_cgroup_link *link; | |
2097 | struct css_set *cg; | |
2098 | ||
2099 | /* Advance to the next non-empty css_set */ | |
2100 | do { | |
2101 | l = l->next; | |
bd89aabc | 2102 | if (l == &cgrp->css_sets) { |
817929ec PM |
2103 | it->cg_link = NULL; |
2104 | return; | |
2105 | } | |
bd89aabc | 2106 | link = list_entry(l, struct cg_cgroup_link, cgrp_link_list); |
817929ec PM |
2107 | cg = link->cg; |
2108 | } while (list_empty(&cg->tasks)); | |
2109 | it->cg_link = l; | |
2110 | it->task = cg->tasks.next; | |
2111 | } | |
2112 | ||
31a7df01 CW |
2113 | /* |
2114 | * To reduce the fork() overhead for systems that are not actually | |
2115 | * using their cgroups capability, we don't maintain the lists running | |
2116 | * through each css_set to its tasks until we see the list actually | |
2117 | * used - in other words after the first call to cgroup_iter_start(). | |
2118 | * | |
2119 | * The tasklist_lock is not held here, as do_each_thread() and | |
2120 | * while_each_thread() are protected by RCU. | |
2121 | */ | |
3df91fe3 | 2122 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2123 | { |
2124 | struct task_struct *p, *g; | |
2125 | write_lock(&css_set_lock); | |
2126 | use_task_css_set_links = 1; | |
2127 | do_each_thread(g, p) { | |
2128 | task_lock(p); | |
0e04388f LZ |
2129 | /* |
2130 | * We should check if the process is exiting, otherwise | |
2131 | * it will race with cgroup_exit() in that the list | |
2132 | * entry won't be deleted though the process has exited. | |
2133 | */ | |
2134 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
31a7df01 CW |
2135 | list_add(&p->cg_list, &p->cgroups->tasks); |
2136 | task_unlock(p); | |
2137 | } while_each_thread(g, p); | |
2138 | write_unlock(&css_set_lock); | |
2139 | } | |
2140 | ||
bd89aabc | 2141 | void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it) |
817929ec PM |
2142 | { |
2143 | /* | |
2144 | * The first time anyone tries to iterate across a cgroup, | |
2145 | * we need to enable the list linking each css_set to its | |
2146 | * tasks, and fix up all existing tasks. | |
2147 | */ | |
31a7df01 CW |
2148 | if (!use_task_css_set_links) |
2149 | cgroup_enable_task_cg_lists(); | |
2150 | ||
817929ec | 2151 | read_lock(&css_set_lock); |
bd89aabc PM |
2152 | it->cg_link = &cgrp->css_sets; |
2153 | cgroup_advance_iter(cgrp, it); | |
817929ec PM |
2154 | } |
2155 | ||
bd89aabc | 2156 | struct task_struct *cgroup_iter_next(struct cgroup *cgrp, |
817929ec PM |
2157 | struct cgroup_iter *it) |
2158 | { | |
2159 | struct task_struct *res; | |
2160 | struct list_head *l = it->task; | |
2019f634 | 2161 | struct cg_cgroup_link *link; |
817929ec PM |
2162 | |
2163 | /* If the iterator cg is NULL, we have no tasks */ | |
2164 | if (!it->cg_link) | |
2165 | return NULL; | |
2166 | res = list_entry(l, struct task_struct, cg_list); | |
2167 | /* Advance iterator to find next entry */ | |
2168 | l = l->next; | |
2019f634 LJ |
2169 | link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list); |
2170 | if (l == &link->cg->tasks) { | |
817929ec PM |
2171 | /* We reached the end of this task list - move on to |
2172 | * the next cg_cgroup_link */ | |
bd89aabc | 2173 | cgroup_advance_iter(cgrp, it); |
817929ec PM |
2174 | } else { |
2175 | it->task = l; | |
2176 | } | |
2177 | return res; | |
2178 | } | |
2179 | ||
bd89aabc | 2180 | void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it) |
817929ec PM |
2181 | { |
2182 | read_unlock(&css_set_lock); | |
2183 | } | |
2184 | ||
31a7df01 CW |
2185 | static inline int started_after_time(struct task_struct *t1, |
2186 | struct timespec *time, | |
2187 | struct task_struct *t2) | |
2188 | { | |
2189 | int start_diff = timespec_compare(&t1->start_time, time); | |
2190 | if (start_diff > 0) { | |
2191 | return 1; | |
2192 | } else if (start_diff < 0) { | |
2193 | return 0; | |
2194 | } else { | |
2195 | /* | |
2196 | * Arbitrarily, if two processes started at the same | |
2197 | * time, we'll say that the lower pointer value | |
2198 | * started first. Note that t2 may have exited by now | |
2199 | * so this may not be a valid pointer any longer, but | |
2200 | * that's fine - it still serves to distinguish | |
2201 | * between two tasks started (effectively) simultaneously. | |
2202 | */ | |
2203 | return t1 > t2; | |
2204 | } | |
2205 | } | |
2206 | ||
2207 | /* | |
2208 | * This function is a callback from heap_insert() and is used to order | |
2209 | * the heap. | |
2210 | * In this case we order the heap in descending task start time. | |
2211 | */ | |
2212 | static inline int started_after(void *p1, void *p2) | |
2213 | { | |
2214 | struct task_struct *t1 = p1; | |
2215 | struct task_struct *t2 = p2; | |
2216 | return started_after_time(t1, &t2->start_time, t2); | |
2217 | } | |
2218 | ||
2219 | /** | |
2220 | * cgroup_scan_tasks - iterate though all the tasks in a cgroup | |
2221 | * @scan: struct cgroup_scanner containing arguments for the scan | |
2222 | * | |
2223 | * Arguments include pointers to callback functions test_task() and | |
2224 | * process_task(). | |
2225 | * Iterate through all the tasks in a cgroup, calling test_task() for each, | |
2226 | * and if it returns true, call process_task() for it also. | |
2227 | * The test_task pointer may be NULL, meaning always true (select all tasks). | |
2228 | * Effectively duplicates cgroup_iter_{start,next,end}() | |
2229 | * but does not lock css_set_lock for the call to process_task(). | |
2230 | * The struct cgroup_scanner may be embedded in any structure of the caller's | |
2231 | * creation. | |
2232 | * It is guaranteed that process_task() will act on every task that | |
2233 | * is a member of the cgroup for the duration of this call. This | |
2234 | * function may or may not call process_task() for tasks that exit | |
2235 | * or move to a different cgroup during the call, or are forked or | |
2236 | * move into the cgroup during the call. | |
2237 | * | |
2238 | * Note that test_task() may be called with locks held, and may in some | |
2239 | * situations be called multiple times for the same task, so it should | |
2240 | * be cheap. | |
2241 | * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been | |
2242 | * pre-allocated and will be used for heap operations (and its "gt" member will | |
2243 | * be overwritten), else a temporary heap will be used (allocation of which | |
2244 | * may cause this function to fail). | |
2245 | */ | |
2246 | int cgroup_scan_tasks(struct cgroup_scanner *scan) | |
2247 | { | |
2248 | int retval, i; | |
2249 | struct cgroup_iter it; | |
2250 | struct task_struct *p, *dropped; | |
2251 | /* Never dereference latest_task, since it's not refcounted */ | |
2252 | struct task_struct *latest_task = NULL; | |
2253 | struct ptr_heap tmp_heap; | |
2254 | struct ptr_heap *heap; | |
2255 | struct timespec latest_time = { 0, 0 }; | |
2256 | ||
2257 | if (scan->heap) { | |
2258 | /* The caller supplied our heap and pre-allocated its memory */ | |
2259 | heap = scan->heap; | |
2260 | heap->gt = &started_after; | |
2261 | } else { | |
2262 | /* We need to allocate our own heap memory */ | |
2263 | heap = &tmp_heap; | |
2264 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
2265 | if (retval) | |
2266 | /* cannot allocate the heap */ | |
2267 | return retval; | |
2268 | } | |
2269 | ||
2270 | again: | |
2271 | /* | |
2272 | * Scan tasks in the cgroup, using the scanner's "test_task" callback | |
2273 | * to determine which are of interest, and using the scanner's | |
2274 | * "process_task" callback to process any of them that need an update. | |
2275 | * Since we don't want to hold any locks during the task updates, | |
2276 | * gather tasks to be processed in a heap structure. | |
2277 | * The heap is sorted by descending task start time. | |
2278 | * If the statically-sized heap fills up, we overflow tasks that | |
2279 | * started later, and in future iterations only consider tasks that | |
2280 | * started after the latest task in the previous pass. This | |
2281 | * guarantees forward progress and that we don't miss any tasks. | |
2282 | */ | |
2283 | heap->size = 0; | |
2284 | cgroup_iter_start(scan->cg, &it); | |
2285 | while ((p = cgroup_iter_next(scan->cg, &it))) { | |
2286 | /* | |
2287 | * Only affect tasks that qualify per the caller's callback, | |
2288 | * if he provided one | |
2289 | */ | |
2290 | if (scan->test_task && !scan->test_task(p, scan)) | |
2291 | continue; | |
2292 | /* | |
2293 | * Only process tasks that started after the last task | |
2294 | * we processed | |
2295 | */ | |
2296 | if (!started_after_time(p, &latest_time, latest_task)) | |
2297 | continue; | |
2298 | dropped = heap_insert(heap, p); | |
2299 | if (dropped == NULL) { | |
2300 | /* | |
2301 | * The new task was inserted; the heap wasn't | |
2302 | * previously full | |
2303 | */ | |
2304 | get_task_struct(p); | |
2305 | } else if (dropped != p) { | |
2306 | /* | |
2307 | * The new task was inserted, and pushed out a | |
2308 | * different task | |
2309 | */ | |
2310 | get_task_struct(p); | |
2311 | put_task_struct(dropped); | |
2312 | } | |
2313 | /* | |
2314 | * Else the new task was newer than anything already in | |
2315 | * the heap and wasn't inserted | |
2316 | */ | |
2317 | } | |
2318 | cgroup_iter_end(scan->cg, &it); | |
2319 | ||
2320 | if (heap->size) { | |
2321 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 2322 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 2323 | if (i == 0) { |
4fe91d51 PJ |
2324 | latest_time = q->start_time; |
2325 | latest_task = q; | |
31a7df01 CW |
2326 | } |
2327 | /* Process the task per the caller's callback */ | |
4fe91d51 PJ |
2328 | scan->process_task(q, scan); |
2329 | put_task_struct(q); | |
31a7df01 CW |
2330 | } |
2331 | /* | |
2332 | * If we had to process any tasks at all, scan again | |
2333 | * in case some of them were in the middle of forking | |
2334 | * children that didn't get processed. | |
2335 | * Not the most efficient way to do it, but it avoids | |
2336 | * having to take callback_mutex in the fork path | |
2337 | */ | |
2338 | goto again; | |
2339 | } | |
2340 | if (heap == &tmp_heap) | |
2341 | heap_free(&tmp_heap); | |
2342 | return 0; | |
2343 | } | |
2344 | ||
bbcb81d0 PM |
2345 | /* |
2346 | * Stuff for reading the 'tasks' file. | |
2347 | * | |
2348 | * Reading this file can return large amounts of data if a cgroup has | |
2349 | * *lots* of attached tasks. So it may need several calls to read(), | |
2350 | * but we cannot guarantee that the information we produce is correct | |
2351 | * unless we produce it entirely atomically. | |
2352 | * | |
bbcb81d0 | 2353 | */ |
bbcb81d0 PM |
2354 | |
2355 | /* | |
2356 | * Load into 'pidarray' up to 'npids' of the tasks using cgroup | |
bd89aabc | 2357 | * 'cgrp'. Return actual number of pids loaded. No need to |
bbcb81d0 PM |
2358 | * task_lock(p) when reading out p->cgroup, since we're in an RCU |
2359 | * read section, so the css_set can't go away, and is | |
2360 | * immutable after creation. | |
2361 | */ | |
bd89aabc | 2362 | static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cgrp) |
bbcb81d0 | 2363 | { |
e7b80bb6 | 2364 | int n = 0, pid; |
817929ec PM |
2365 | struct cgroup_iter it; |
2366 | struct task_struct *tsk; | |
bd89aabc PM |
2367 | cgroup_iter_start(cgrp, &it); |
2368 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
817929ec PM |
2369 | if (unlikely(n == npids)) |
2370 | break; | |
e7b80bb6 G |
2371 | pid = task_pid_vnr(tsk); |
2372 | if (pid > 0) | |
2373 | pidarray[n++] = pid; | |
817929ec | 2374 | } |
bd89aabc | 2375 | cgroup_iter_end(cgrp, &it); |
bbcb81d0 PM |
2376 | return n; |
2377 | } | |
2378 | ||
846c7bb0 | 2379 | /** |
a043e3b2 | 2380 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
2381 | * @stats: cgroupstats to fill information into |
2382 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
2383 | * been requested. | |
a043e3b2 LZ |
2384 | * |
2385 | * Build and fill cgroupstats so that taskstats can export it to user | |
2386 | * space. | |
846c7bb0 BS |
2387 | */ |
2388 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
2389 | { | |
2390 | int ret = -EINVAL; | |
bd89aabc | 2391 | struct cgroup *cgrp; |
846c7bb0 BS |
2392 | struct cgroup_iter it; |
2393 | struct task_struct *tsk; | |
33d283be | 2394 | |
846c7bb0 | 2395 | /* |
33d283be LZ |
2396 | * Validate dentry by checking the superblock operations, |
2397 | * and make sure it's a directory. | |
846c7bb0 | 2398 | */ |
33d283be LZ |
2399 | if (dentry->d_sb->s_op != &cgroup_ops || |
2400 | !S_ISDIR(dentry->d_inode->i_mode)) | |
846c7bb0 BS |
2401 | goto err; |
2402 | ||
2403 | ret = 0; | |
bd89aabc | 2404 | cgrp = dentry->d_fsdata; |
846c7bb0 | 2405 | |
bd89aabc PM |
2406 | cgroup_iter_start(cgrp, &it); |
2407 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
846c7bb0 BS |
2408 | switch (tsk->state) { |
2409 | case TASK_RUNNING: | |
2410 | stats->nr_running++; | |
2411 | break; | |
2412 | case TASK_INTERRUPTIBLE: | |
2413 | stats->nr_sleeping++; | |
2414 | break; | |
2415 | case TASK_UNINTERRUPTIBLE: | |
2416 | stats->nr_uninterruptible++; | |
2417 | break; | |
2418 | case TASK_STOPPED: | |
2419 | stats->nr_stopped++; | |
2420 | break; | |
2421 | default: | |
2422 | if (delayacct_is_task_waiting_on_io(tsk)) | |
2423 | stats->nr_io_wait++; | |
2424 | break; | |
2425 | } | |
2426 | } | |
bd89aabc | 2427 | cgroup_iter_end(cgrp, &it); |
846c7bb0 | 2428 | |
846c7bb0 BS |
2429 | err: |
2430 | return ret; | |
2431 | } | |
2432 | ||
bbcb81d0 PM |
2433 | static int cmppid(const void *a, const void *b) |
2434 | { | |
2435 | return *(pid_t *)a - *(pid_t *)b; | |
2436 | } | |
2437 | ||
8f3ff208 | 2438 | |
bbcb81d0 | 2439 | /* |
cc31edce PM |
2440 | * seq_file methods for the "tasks" file. The seq_file position is the |
2441 | * next pid to display; the seq_file iterator is a pointer to the pid | |
2442 | * in the cgroup->tasks_pids array. | |
bbcb81d0 | 2443 | */ |
cc31edce PM |
2444 | |
2445 | static void *cgroup_tasks_start(struct seq_file *s, loff_t *pos) | |
bbcb81d0 | 2446 | { |
cc31edce PM |
2447 | /* |
2448 | * Initially we receive a position value that corresponds to | |
2449 | * one more than the last pid shown (or 0 on the first call or | |
2450 | * after a seek to the start). Use a binary-search to find the | |
2451 | * next pid to display, if any | |
2452 | */ | |
8f3ff208 | 2453 | struct cgroup *cgrp = s->private; |
cc31edce PM |
2454 | int index = 0, pid = *pos; |
2455 | int *iter; | |
2456 | ||
2457 | down_read(&cgrp->pids_mutex); | |
2458 | if (pid) { | |
8f3ff208 | 2459 | int end = cgrp->pids_length; |
20777766 | 2460 | |
cc31edce PM |
2461 | while (index < end) { |
2462 | int mid = (index + end) / 2; | |
8f3ff208 | 2463 | if (cgrp->tasks_pids[mid] == pid) { |
cc31edce PM |
2464 | index = mid; |
2465 | break; | |
8f3ff208 | 2466 | } else if (cgrp->tasks_pids[mid] <= pid) |
cc31edce PM |
2467 | index = mid + 1; |
2468 | else | |
2469 | end = mid; | |
2470 | } | |
2471 | } | |
2472 | /* If we're off the end of the array, we're done */ | |
8f3ff208 | 2473 | if (index >= cgrp->pids_length) |
cc31edce PM |
2474 | return NULL; |
2475 | /* Update the abstract position to be the actual pid that we found */ | |
8f3ff208 | 2476 | iter = cgrp->tasks_pids + index; |
cc31edce PM |
2477 | *pos = *iter; |
2478 | return iter; | |
2479 | } | |
2480 | ||
2481 | static void cgroup_tasks_stop(struct seq_file *s, void *v) | |
2482 | { | |
8f3ff208 | 2483 | struct cgroup *cgrp = s->private; |
cc31edce PM |
2484 | up_read(&cgrp->pids_mutex); |
2485 | } | |
2486 | ||
2487 | static void *cgroup_tasks_next(struct seq_file *s, void *v, loff_t *pos) | |
2488 | { | |
8f3ff208 | 2489 | struct cgroup *cgrp = s->private; |
cc31edce | 2490 | int *p = v; |
8f3ff208 | 2491 | int *end = cgrp->tasks_pids + cgrp->pids_length; |
cc31edce PM |
2492 | |
2493 | /* | |
2494 | * Advance to the next pid in the array. If this goes off the | |
2495 | * end, we're done | |
2496 | */ | |
2497 | p++; | |
2498 | if (p >= end) { | |
2499 | return NULL; | |
2500 | } else { | |
2501 | *pos = *p; | |
2502 | return p; | |
2503 | } | |
2504 | } | |
2505 | ||
2506 | static int cgroup_tasks_show(struct seq_file *s, void *v) | |
2507 | { | |
2508 | return seq_printf(s, "%d\n", *(int *)v); | |
2509 | } | |
bbcb81d0 | 2510 | |
88e9d34c | 2511 | static const struct seq_operations cgroup_tasks_seq_operations = { |
cc31edce PM |
2512 | .start = cgroup_tasks_start, |
2513 | .stop = cgroup_tasks_stop, | |
2514 | .next = cgroup_tasks_next, | |
2515 | .show = cgroup_tasks_show, | |
2516 | }; | |
2517 | ||
8f3ff208 | 2518 | static void release_cgroup_pid_array(struct cgroup *cgrp) |
cc31edce PM |
2519 | { |
2520 | down_write(&cgrp->pids_mutex); | |
8f3ff208 PM |
2521 | BUG_ON(!cgrp->pids_use_count); |
2522 | if (!--cgrp->pids_use_count) { | |
2523 | kfree(cgrp->tasks_pids); | |
2524 | cgrp->tasks_pids = NULL; | |
2525 | cgrp->pids_length = 0; | |
cc31edce PM |
2526 | } |
2527 | up_write(&cgrp->pids_mutex); | |
bbcb81d0 PM |
2528 | } |
2529 | ||
cc31edce PM |
2530 | static int cgroup_tasks_release(struct inode *inode, struct file *file) |
2531 | { | |
8f3ff208 | 2532 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
cc31edce PM |
2533 | |
2534 | if (!(file->f_mode & FMODE_READ)) | |
2535 | return 0; | |
2536 | ||
8f3ff208 | 2537 | release_cgroup_pid_array(cgrp); |
cc31edce PM |
2538 | return seq_release(inode, file); |
2539 | } | |
2540 | ||
2541 | static struct file_operations cgroup_tasks_operations = { | |
2542 | .read = seq_read, | |
2543 | .llseek = seq_lseek, | |
2544 | .write = cgroup_file_write, | |
2545 | .release = cgroup_tasks_release, | |
2546 | }; | |
2547 | ||
bbcb81d0 | 2548 | /* |
cc31edce | 2549 | * Handle an open on 'tasks' file. Prepare an array containing the |
bbcb81d0 | 2550 | * process id's of tasks currently attached to the cgroup being opened. |
bbcb81d0 | 2551 | */ |
cc31edce | 2552 | |
bbcb81d0 PM |
2553 | static int cgroup_tasks_open(struct inode *unused, struct file *file) |
2554 | { | |
bd89aabc | 2555 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
bbcb81d0 PM |
2556 | pid_t *pidarray; |
2557 | int npids; | |
cc31edce | 2558 | int retval; |
bbcb81d0 | 2559 | |
cc31edce | 2560 | /* Nothing to do for write-only files */ |
bbcb81d0 PM |
2561 | if (!(file->f_mode & FMODE_READ)) |
2562 | return 0; | |
2563 | ||
bbcb81d0 PM |
2564 | /* |
2565 | * If cgroup gets more users after we read count, we won't have | |
2566 | * enough space - tough. This race is indistinguishable to the | |
2567 | * caller from the case that the additional cgroup users didn't | |
2568 | * show up until sometime later on. | |
2569 | */ | |
bd89aabc | 2570 | npids = cgroup_task_count(cgrp); |
cc31edce PM |
2571 | pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL); |
2572 | if (!pidarray) | |
2573 | return -ENOMEM; | |
2574 | npids = pid_array_load(pidarray, npids, cgrp); | |
2575 | sort(pidarray, npids, sizeof(pid_t), cmppid, NULL); | |
bbcb81d0 | 2576 | |
cc31edce PM |
2577 | /* |
2578 | * Store the array in the cgroup, freeing the old | |
2579 | * array if necessary | |
2580 | */ | |
2581 | down_write(&cgrp->pids_mutex); | |
8f3ff208 PM |
2582 | kfree(cgrp->tasks_pids); |
2583 | cgrp->tasks_pids = pidarray; | |
2584 | cgrp->pids_length = npids; | |
2585 | cgrp->pids_use_count++; | |
cc31edce PM |
2586 | up_write(&cgrp->pids_mutex); |
2587 | ||
2588 | file->f_op = &cgroup_tasks_operations; | |
2589 | ||
2590 | retval = seq_open(file, &cgroup_tasks_seq_operations); | |
2591 | if (retval) { | |
8f3ff208 | 2592 | release_cgroup_pid_array(cgrp); |
cc31edce | 2593 | return retval; |
bbcb81d0 | 2594 | } |
8f3ff208 | 2595 | ((struct seq_file *)file->private_data)->private = cgrp; |
bbcb81d0 PM |
2596 | return 0; |
2597 | } | |
2598 | ||
bd89aabc | 2599 | static u64 cgroup_read_notify_on_release(struct cgroup *cgrp, |
81a6a5cd PM |
2600 | struct cftype *cft) |
2601 | { | |
bd89aabc | 2602 | return notify_on_release(cgrp); |
81a6a5cd PM |
2603 | } |
2604 | ||
6379c106 PM |
2605 | static int cgroup_write_notify_on_release(struct cgroup *cgrp, |
2606 | struct cftype *cft, | |
2607 | u64 val) | |
2608 | { | |
2609 | clear_bit(CGRP_RELEASABLE, &cgrp->flags); | |
2610 | if (val) | |
2611 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
2612 | else | |
2613 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
2614 | return 0; | |
2615 | } | |
2616 | ||
bbcb81d0 PM |
2617 | /* |
2618 | * for the common functions, 'private' gives the type of file | |
2619 | */ | |
81a6a5cd PM |
2620 | static struct cftype files[] = { |
2621 | { | |
2622 | .name = "tasks", | |
2623 | .open = cgroup_tasks_open, | |
af351026 | 2624 | .write_u64 = cgroup_tasks_write, |
81a6a5cd PM |
2625 | .release = cgroup_tasks_release, |
2626 | .private = FILE_TASKLIST, | |
099fca32 | 2627 | .mode = S_IRUGO | S_IWUSR, |
81a6a5cd PM |
2628 | }, |
2629 | ||
2630 | { | |
2631 | .name = "notify_on_release", | |
f4c753b7 | 2632 | .read_u64 = cgroup_read_notify_on_release, |
6379c106 | 2633 | .write_u64 = cgroup_write_notify_on_release, |
81a6a5cd PM |
2634 | .private = FILE_NOTIFY_ON_RELEASE, |
2635 | }, | |
81a6a5cd PM |
2636 | }; |
2637 | ||
2638 | static struct cftype cft_release_agent = { | |
2639 | .name = "release_agent", | |
e788e066 PM |
2640 | .read_seq_string = cgroup_release_agent_show, |
2641 | .write_string = cgroup_release_agent_write, | |
2642 | .max_write_len = PATH_MAX, | |
81a6a5cd | 2643 | .private = FILE_RELEASE_AGENT, |
bbcb81d0 PM |
2644 | }; |
2645 | ||
bd89aabc | 2646 | static int cgroup_populate_dir(struct cgroup *cgrp) |
ddbcc7e8 PM |
2647 | { |
2648 | int err; | |
2649 | struct cgroup_subsys *ss; | |
2650 | ||
2651 | /* First clear out any existing files */ | |
bd89aabc | 2652 | cgroup_clear_directory(cgrp->dentry); |
ddbcc7e8 | 2653 | |
bd89aabc | 2654 | err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files)); |
bbcb81d0 PM |
2655 | if (err < 0) |
2656 | return err; | |
2657 | ||
bd89aabc PM |
2658 | if (cgrp == cgrp->top_cgroup) { |
2659 | if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0) | |
81a6a5cd PM |
2660 | return err; |
2661 | } | |
2662 | ||
bd89aabc PM |
2663 | for_each_subsys(cgrp->root, ss) { |
2664 | if (ss->populate && (err = ss->populate(ss, cgrp)) < 0) | |
ddbcc7e8 PM |
2665 | return err; |
2666 | } | |
38460b48 KH |
2667 | /* This cgroup is ready now */ |
2668 | for_each_subsys(cgrp->root, ss) { | |
2669 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
2670 | /* | |
2671 | * Update id->css pointer and make this css visible from | |
2672 | * CSS ID functions. This pointer will be dereferened | |
2673 | * from RCU-read-side without locks. | |
2674 | */ | |
2675 | if (css->id) | |
2676 | rcu_assign_pointer(css->id->css, css); | |
2677 | } | |
ddbcc7e8 PM |
2678 | |
2679 | return 0; | |
2680 | } | |
2681 | ||
2682 | static void init_cgroup_css(struct cgroup_subsys_state *css, | |
2683 | struct cgroup_subsys *ss, | |
bd89aabc | 2684 | struct cgroup *cgrp) |
ddbcc7e8 | 2685 | { |
bd89aabc | 2686 | css->cgroup = cgrp; |
e7c5ec91 | 2687 | atomic_set(&css->refcnt, 1); |
ddbcc7e8 | 2688 | css->flags = 0; |
38460b48 | 2689 | css->id = NULL; |
bd89aabc | 2690 | if (cgrp == dummytop) |
ddbcc7e8 | 2691 | set_bit(CSS_ROOT, &css->flags); |
bd89aabc PM |
2692 | BUG_ON(cgrp->subsys[ss->subsys_id]); |
2693 | cgrp->subsys[ss->subsys_id] = css; | |
ddbcc7e8 PM |
2694 | } |
2695 | ||
999cd8a4 PM |
2696 | static void cgroup_lock_hierarchy(struct cgroupfs_root *root) |
2697 | { | |
2698 | /* We need to take each hierarchy_mutex in a consistent order */ | |
2699 | int i; | |
2700 | ||
2701 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2702 | struct cgroup_subsys *ss = subsys[i]; | |
2703 | if (ss->root == root) | |
cfebe563 | 2704 | mutex_lock(&ss->hierarchy_mutex); |
999cd8a4 PM |
2705 | } |
2706 | } | |
2707 | ||
2708 | static void cgroup_unlock_hierarchy(struct cgroupfs_root *root) | |
2709 | { | |
2710 | int i; | |
2711 | ||
2712 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2713 | struct cgroup_subsys *ss = subsys[i]; | |
2714 | if (ss->root == root) | |
2715 | mutex_unlock(&ss->hierarchy_mutex); | |
2716 | } | |
2717 | } | |
2718 | ||
ddbcc7e8 | 2719 | /* |
a043e3b2 LZ |
2720 | * cgroup_create - create a cgroup |
2721 | * @parent: cgroup that will be parent of the new cgroup | |
2722 | * @dentry: dentry of the new cgroup | |
2723 | * @mode: mode to set on new inode | |
ddbcc7e8 | 2724 | * |
a043e3b2 | 2725 | * Must be called with the mutex on the parent inode held |
ddbcc7e8 | 2726 | */ |
ddbcc7e8 | 2727 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, |
099fca32 | 2728 | mode_t mode) |
ddbcc7e8 | 2729 | { |
bd89aabc | 2730 | struct cgroup *cgrp; |
ddbcc7e8 PM |
2731 | struct cgroupfs_root *root = parent->root; |
2732 | int err = 0; | |
2733 | struct cgroup_subsys *ss; | |
2734 | struct super_block *sb = root->sb; | |
2735 | ||
bd89aabc PM |
2736 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
2737 | if (!cgrp) | |
ddbcc7e8 PM |
2738 | return -ENOMEM; |
2739 | ||
2740 | /* Grab a reference on the superblock so the hierarchy doesn't | |
2741 | * get deleted on unmount if there are child cgroups. This | |
2742 | * can be done outside cgroup_mutex, since the sb can't | |
2743 | * disappear while someone has an open control file on the | |
2744 | * fs */ | |
2745 | atomic_inc(&sb->s_active); | |
2746 | ||
2747 | mutex_lock(&cgroup_mutex); | |
2748 | ||
cc31edce | 2749 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 2750 | |
bd89aabc PM |
2751 | cgrp->parent = parent; |
2752 | cgrp->root = parent->root; | |
2753 | cgrp->top_cgroup = parent->top_cgroup; | |
ddbcc7e8 | 2754 | |
b6abdb0e LZ |
2755 | if (notify_on_release(parent)) |
2756 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
2757 | ||
ddbcc7e8 | 2758 | for_each_subsys(root, ss) { |
bd89aabc | 2759 | struct cgroup_subsys_state *css = ss->create(ss, cgrp); |
ddbcc7e8 PM |
2760 | if (IS_ERR(css)) { |
2761 | err = PTR_ERR(css); | |
2762 | goto err_destroy; | |
2763 | } | |
bd89aabc | 2764 | init_cgroup_css(css, ss, cgrp); |
38460b48 KH |
2765 | if (ss->use_id) |
2766 | if (alloc_css_id(ss, parent, cgrp)) | |
2767 | goto err_destroy; | |
2768 | /* At error, ->destroy() callback has to free assigned ID. */ | |
ddbcc7e8 PM |
2769 | } |
2770 | ||
999cd8a4 | 2771 | cgroup_lock_hierarchy(root); |
bd89aabc | 2772 | list_add(&cgrp->sibling, &cgrp->parent->children); |
999cd8a4 | 2773 | cgroup_unlock_hierarchy(root); |
ddbcc7e8 PM |
2774 | root->number_of_cgroups++; |
2775 | ||
bd89aabc | 2776 | err = cgroup_create_dir(cgrp, dentry, mode); |
ddbcc7e8 PM |
2777 | if (err < 0) |
2778 | goto err_remove; | |
2779 | ||
2780 | /* The cgroup directory was pre-locked for us */ | |
bd89aabc | 2781 | BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex)); |
ddbcc7e8 | 2782 | |
bd89aabc | 2783 | err = cgroup_populate_dir(cgrp); |
ddbcc7e8 PM |
2784 | /* If err < 0, we have a half-filled directory - oh well ;) */ |
2785 | ||
2786 | mutex_unlock(&cgroup_mutex); | |
bd89aabc | 2787 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
2788 | |
2789 | return 0; | |
2790 | ||
2791 | err_remove: | |
2792 | ||
baef99a0 | 2793 | cgroup_lock_hierarchy(root); |
bd89aabc | 2794 | list_del(&cgrp->sibling); |
baef99a0 | 2795 | cgroup_unlock_hierarchy(root); |
ddbcc7e8 PM |
2796 | root->number_of_cgroups--; |
2797 | ||
2798 | err_destroy: | |
2799 | ||
2800 | for_each_subsys(root, ss) { | |
bd89aabc PM |
2801 | if (cgrp->subsys[ss->subsys_id]) |
2802 | ss->destroy(ss, cgrp); | |
ddbcc7e8 PM |
2803 | } |
2804 | ||
2805 | mutex_unlock(&cgroup_mutex); | |
2806 | ||
2807 | /* Release the reference count that we took on the superblock */ | |
2808 | deactivate_super(sb); | |
2809 | ||
bd89aabc | 2810 | kfree(cgrp); |
ddbcc7e8 PM |
2811 | return err; |
2812 | } | |
2813 | ||
2814 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
2815 | { | |
2816 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
2817 | ||
2818 | /* the vfs holds inode->i_mutex already */ | |
2819 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
2820 | } | |
2821 | ||
55b6fd01 | 2822 | static int cgroup_has_css_refs(struct cgroup *cgrp) |
81a6a5cd PM |
2823 | { |
2824 | /* Check the reference count on each subsystem. Since we | |
2825 | * already established that there are no tasks in the | |
e7c5ec91 | 2826 | * cgroup, if the css refcount is also 1, then there should |
81a6a5cd PM |
2827 | * be no outstanding references, so the subsystem is safe to |
2828 | * destroy. We scan across all subsystems rather than using | |
2829 | * the per-hierarchy linked list of mounted subsystems since | |
2830 | * we can be called via check_for_release() with no | |
2831 | * synchronization other than RCU, and the subsystem linked | |
2832 | * list isn't RCU-safe */ | |
2833 | int i; | |
2834 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
2835 | struct cgroup_subsys *ss = subsys[i]; | |
2836 | struct cgroup_subsys_state *css; | |
2837 | /* Skip subsystems not in this hierarchy */ | |
bd89aabc | 2838 | if (ss->root != cgrp->root) |
81a6a5cd | 2839 | continue; |
bd89aabc | 2840 | css = cgrp->subsys[ss->subsys_id]; |
81a6a5cd PM |
2841 | /* When called from check_for_release() it's possible |
2842 | * that by this point the cgroup has been removed | |
2843 | * and the css deleted. But a false-positive doesn't | |
2844 | * matter, since it can only happen if the cgroup | |
2845 | * has been deleted and hence no longer needs the | |
2846 | * release agent to be called anyway. */ | |
e7c5ec91 | 2847 | if (css && (atomic_read(&css->refcnt) > 1)) |
81a6a5cd | 2848 | return 1; |
81a6a5cd PM |
2849 | } |
2850 | return 0; | |
2851 | } | |
2852 | ||
e7c5ec91 PM |
2853 | /* |
2854 | * Atomically mark all (or else none) of the cgroup's CSS objects as | |
2855 | * CSS_REMOVED. Return true on success, or false if the cgroup has | |
2856 | * busy subsystems. Call with cgroup_mutex held | |
2857 | */ | |
2858 | ||
2859 | static int cgroup_clear_css_refs(struct cgroup *cgrp) | |
2860 | { | |
2861 | struct cgroup_subsys *ss; | |
2862 | unsigned long flags; | |
2863 | bool failed = false; | |
2864 | local_irq_save(flags); | |
2865 | for_each_subsys(cgrp->root, ss) { | |
2866 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
2867 | int refcnt; | |
804b3c28 | 2868 | while (1) { |
e7c5ec91 PM |
2869 | /* We can only remove a CSS with a refcnt==1 */ |
2870 | refcnt = atomic_read(&css->refcnt); | |
2871 | if (refcnt > 1) { | |
2872 | failed = true; | |
2873 | goto done; | |
2874 | } | |
2875 | BUG_ON(!refcnt); | |
2876 | /* | |
2877 | * Drop the refcnt to 0 while we check other | |
2878 | * subsystems. This will cause any racing | |
2879 | * css_tryget() to spin until we set the | |
2880 | * CSS_REMOVED bits or abort | |
2881 | */ | |
804b3c28 PM |
2882 | if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt) |
2883 | break; | |
2884 | cpu_relax(); | |
2885 | } | |
e7c5ec91 PM |
2886 | } |
2887 | done: | |
2888 | for_each_subsys(cgrp->root, ss) { | |
2889 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
2890 | if (failed) { | |
2891 | /* | |
2892 | * Restore old refcnt if we previously managed | |
2893 | * to clear it from 1 to 0 | |
2894 | */ | |
2895 | if (!atomic_read(&css->refcnt)) | |
2896 | atomic_set(&css->refcnt, 1); | |
2897 | } else { | |
2898 | /* Commit the fact that the CSS is removed */ | |
2899 | set_bit(CSS_REMOVED, &css->flags); | |
2900 | } | |
2901 | } | |
2902 | local_irq_restore(flags); | |
2903 | return !failed; | |
2904 | } | |
2905 | ||
ddbcc7e8 PM |
2906 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) |
2907 | { | |
bd89aabc | 2908 | struct cgroup *cgrp = dentry->d_fsdata; |
ddbcc7e8 PM |
2909 | struct dentry *d; |
2910 | struct cgroup *parent; | |
ec64f515 KH |
2911 | DEFINE_WAIT(wait); |
2912 | int ret; | |
ddbcc7e8 PM |
2913 | |
2914 | /* the vfs holds both inode->i_mutex already */ | |
ec64f515 | 2915 | again: |
ddbcc7e8 | 2916 | mutex_lock(&cgroup_mutex); |
bd89aabc | 2917 | if (atomic_read(&cgrp->count) != 0) { |
ddbcc7e8 PM |
2918 | mutex_unlock(&cgroup_mutex); |
2919 | return -EBUSY; | |
2920 | } | |
bd89aabc | 2921 | if (!list_empty(&cgrp->children)) { |
ddbcc7e8 PM |
2922 | mutex_unlock(&cgroup_mutex); |
2923 | return -EBUSY; | |
2924 | } | |
3fa59dfb | 2925 | mutex_unlock(&cgroup_mutex); |
a043e3b2 | 2926 | |
88703267 KH |
2927 | /* |
2928 | * In general, subsystem has no css->refcnt after pre_destroy(). But | |
2929 | * in racy cases, subsystem may have to get css->refcnt after | |
2930 | * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes | |
2931 | * make rmdir return -EBUSY too often. To avoid that, we use waitqueue | |
2932 | * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir | |
2933 | * and subsystem's reference count handling. Please see css_get/put | |
2934 | * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation. | |
2935 | */ | |
2936 | set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
2937 | ||
4fca88c8 | 2938 | /* |
a043e3b2 LZ |
2939 | * Call pre_destroy handlers of subsys. Notify subsystems |
2940 | * that rmdir() request comes. | |
4fca88c8 | 2941 | */ |
ec64f515 | 2942 | ret = cgroup_call_pre_destroy(cgrp); |
88703267 KH |
2943 | if (ret) { |
2944 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
ec64f515 | 2945 | return ret; |
88703267 | 2946 | } |
ddbcc7e8 | 2947 | |
3fa59dfb KH |
2948 | mutex_lock(&cgroup_mutex); |
2949 | parent = cgrp->parent; | |
ec64f515 | 2950 | if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) { |
88703267 | 2951 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); |
ddbcc7e8 PM |
2952 | mutex_unlock(&cgroup_mutex); |
2953 | return -EBUSY; | |
2954 | } | |
ec64f515 | 2955 | prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE); |
ec64f515 KH |
2956 | if (!cgroup_clear_css_refs(cgrp)) { |
2957 | mutex_unlock(&cgroup_mutex); | |
88703267 KH |
2958 | /* |
2959 | * Because someone may call cgroup_wakeup_rmdir_waiter() before | |
2960 | * prepare_to_wait(), we need to check this flag. | |
2961 | */ | |
2962 | if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)) | |
2963 | schedule(); | |
ec64f515 KH |
2964 | finish_wait(&cgroup_rmdir_waitq, &wait); |
2965 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
2966 | if (signal_pending(current)) | |
2967 | return -EINTR; | |
2968 | goto again; | |
2969 | } | |
2970 | /* NO css_tryget() can success after here. */ | |
2971 | finish_wait(&cgroup_rmdir_waitq, &wait); | |
2972 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
ddbcc7e8 | 2973 | |
81a6a5cd | 2974 | spin_lock(&release_list_lock); |
bd89aabc PM |
2975 | set_bit(CGRP_REMOVED, &cgrp->flags); |
2976 | if (!list_empty(&cgrp->release_list)) | |
2977 | list_del(&cgrp->release_list); | |
81a6a5cd | 2978 | spin_unlock(&release_list_lock); |
999cd8a4 PM |
2979 | |
2980 | cgroup_lock_hierarchy(cgrp->root); | |
2981 | /* delete this cgroup from parent->children */ | |
bd89aabc | 2982 | list_del(&cgrp->sibling); |
999cd8a4 PM |
2983 | cgroup_unlock_hierarchy(cgrp->root); |
2984 | ||
bd89aabc PM |
2985 | spin_lock(&cgrp->dentry->d_lock); |
2986 | d = dget(cgrp->dentry); | |
ddbcc7e8 PM |
2987 | spin_unlock(&d->d_lock); |
2988 | ||
2989 | cgroup_d_remove_dir(d); | |
2990 | dput(d); | |
ddbcc7e8 | 2991 | |
bd89aabc | 2992 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd PM |
2993 | check_for_release(parent); |
2994 | ||
ddbcc7e8 | 2995 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
2996 | return 0; |
2997 | } | |
2998 | ||
06a11920 | 2999 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 3000 | { |
ddbcc7e8 | 3001 | struct cgroup_subsys_state *css; |
cfe36bde DC |
3002 | |
3003 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 PM |
3004 | |
3005 | /* Create the top cgroup state for this subsystem */ | |
33a68ac1 | 3006 | list_add(&ss->sibling, &rootnode.subsys_list); |
ddbcc7e8 PM |
3007 | ss->root = &rootnode; |
3008 | css = ss->create(ss, dummytop); | |
3009 | /* We don't handle early failures gracefully */ | |
3010 | BUG_ON(IS_ERR(css)); | |
3011 | init_cgroup_css(css, ss, dummytop); | |
3012 | ||
e8d55fde | 3013 | /* Update the init_css_set to contain a subsys |
817929ec | 3014 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
3015 | * newly registered, all tasks and hence the |
3016 | * init_css_set is in the subsystem's top cgroup. */ | |
3017 | init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id]; | |
ddbcc7e8 PM |
3018 | |
3019 | need_forkexit_callback |= ss->fork || ss->exit; | |
3020 | ||
e8d55fde LZ |
3021 | /* At system boot, before all subsystems have been |
3022 | * registered, no tasks have been forked, so we don't | |
3023 | * need to invoke fork callbacks here. */ | |
3024 | BUG_ON(!list_empty(&init_task.tasks)); | |
3025 | ||
999cd8a4 | 3026 | mutex_init(&ss->hierarchy_mutex); |
cfebe563 | 3027 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); |
ddbcc7e8 PM |
3028 | ss->active = 1; |
3029 | } | |
3030 | ||
3031 | /** | |
a043e3b2 LZ |
3032 | * cgroup_init_early - cgroup initialization at system boot |
3033 | * | |
3034 | * Initialize cgroups at system boot, and initialize any | |
3035 | * subsystems that request early init. | |
ddbcc7e8 PM |
3036 | */ |
3037 | int __init cgroup_init_early(void) | |
3038 | { | |
3039 | int i; | |
146aa1bd | 3040 | atomic_set(&init_css_set.refcount, 1); |
817929ec PM |
3041 | INIT_LIST_HEAD(&init_css_set.cg_links); |
3042 | INIT_LIST_HEAD(&init_css_set.tasks); | |
472b1053 | 3043 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 3044 | css_set_count = 1; |
ddbcc7e8 | 3045 | init_cgroup_root(&rootnode); |
817929ec PM |
3046 | root_count = 1; |
3047 | init_task.cgroups = &init_css_set; | |
3048 | ||
3049 | init_css_set_link.cg = &init_css_set; | |
7717f7ba | 3050 | init_css_set_link.cgrp = dummytop; |
bd89aabc | 3051 | list_add(&init_css_set_link.cgrp_link_list, |
817929ec PM |
3052 | &rootnode.top_cgroup.css_sets); |
3053 | list_add(&init_css_set_link.cg_link_list, | |
3054 | &init_css_set.cg_links); | |
ddbcc7e8 | 3055 | |
472b1053 LZ |
3056 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) |
3057 | INIT_HLIST_HEAD(&css_set_table[i]); | |
3058 | ||
ddbcc7e8 PM |
3059 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
3060 | struct cgroup_subsys *ss = subsys[i]; | |
3061 | ||
3062 | BUG_ON(!ss->name); | |
3063 | BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); | |
3064 | BUG_ON(!ss->create); | |
3065 | BUG_ON(!ss->destroy); | |
3066 | if (ss->subsys_id != i) { | |
cfe36bde | 3067 | printk(KERN_ERR "cgroup: Subsys %s id == %d\n", |
ddbcc7e8 PM |
3068 | ss->name, ss->subsys_id); |
3069 | BUG(); | |
3070 | } | |
3071 | ||
3072 | if (ss->early_init) | |
3073 | cgroup_init_subsys(ss); | |
3074 | } | |
3075 | return 0; | |
3076 | } | |
3077 | ||
3078 | /** | |
a043e3b2 LZ |
3079 | * cgroup_init - cgroup initialization |
3080 | * | |
3081 | * Register cgroup filesystem and /proc file, and initialize | |
3082 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
3083 | */ |
3084 | int __init cgroup_init(void) | |
3085 | { | |
3086 | int err; | |
3087 | int i; | |
472b1053 | 3088 | struct hlist_head *hhead; |
a424316c PM |
3089 | |
3090 | err = bdi_init(&cgroup_backing_dev_info); | |
3091 | if (err) | |
3092 | return err; | |
ddbcc7e8 PM |
3093 | |
3094 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
3095 | struct cgroup_subsys *ss = subsys[i]; | |
3096 | if (!ss->early_init) | |
3097 | cgroup_init_subsys(ss); | |
38460b48 KH |
3098 | if (ss->use_id) |
3099 | cgroup_subsys_init_idr(ss); | |
ddbcc7e8 PM |
3100 | } |
3101 | ||
472b1053 LZ |
3102 | /* Add init_css_set to the hash table */ |
3103 | hhead = css_set_hash(init_css_set.subsys); | |
3104 | hlist_add_head(&init_css_set.hlist, hhead); | |
2c6ab6d2 | 3105 | BUG_ON(!init_root_id(&rootnode)); |
ddbcc7e8 PM |
3106 | err = register_filesystem(&cgroup_fs_type); |
3107 | if (err < 0) | |
3108 | goto out; | |
3109 | ||
46ae220b | 3110 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
a424316c | 3111 | |
ddbcc7e8 | 3112 | out: |
a424316c PM |
3113 | if (err) |
3114 | bdi_destroy(&cgroup_backing_dev_info); | |
3115 | ||
ddbcc7e8 PM |
3116 | return err; |
3117 | } | |
b4f48b63 | 3118 | |
a424316c PM |
3119 | /* |
3120 | * proc_cgroup_show() | |
3121 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
3122 | * - Used for /proc/<pid>/cgroup. | |
3123 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
3124 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 3125 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
3126 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
3127 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
3128 | * cgroup to top_cgroup. | |
3129 | */ | |
3130 | ||
3131 | /* TODO: Use a proper seq_file iterator */ | |
3132 | static int proc_cgroup_show(struct seq_file *m, void *v) | |
3133 | { | |
3134 | struct pid *pid; | |
3135 | struct task_struct *tsk; | |
3136 | char *buf; | |
3137 | int retval; | |
3138 | struct cgroupfs_root *root; | |
3139 | ||
3140 | retval = -ENOMEM; | |
3141 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
3142 | if (!buf) | |
3143 | goto out; | |
3144 | ||
3145 | retval = -ESRCH; | |
3146 | pid = m->private; | |
3147 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
3148 | if (!tsk) | |
3149 | goto out_free; | |
3150 | ||
3151 | retval = 0; | |
3152 | ||
3153 | mutex_lock(&cgroup_mutex); | |
3154 | ||
e5f6a860 | 3155 | for_each_active_root(root) { |
a424316c | 3156 | struct cgroup_subsys *ss; |
bd89aabc | 3157 | struct cgroup *cgrp; |
a424316c PM |
3158 | int count = 0; |
3159 | ||
2c6ab6d2 | 3160 | seq_printf(m, "%d:", root->hierarchy_id); |
a424316c PM |
3161 | for_each_subsys(root, ss) |
3162 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
3163 | if (strlen(root->name)) |
3164 | seq_printf(m, "%sname=%s", count ? "," : "", | |
3165 | root->name); | |
a424316c | 3166 | seq_putc(m, ':'); |
7717f7ba | 3167 | cgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 3168 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); |
a424316c PM |
3169 | if (retval < 0) |
3170 | goto out_unlock; | |
3171 | seq_puts(m, buf); | |
3172 | seq_putc(m, '\n'); | |
3173 | } | |
3174 | ||
3175 | out_unlock: | |
3176 | mutex_unlock(&cgroup_mutex); | |
3177 | put_task_struct(tsk); | |
3178 | out_free: | |
3179 | kfree(buf); | |
3180 | out: | |
3181 | return retval; | |
3182 | } | |
3183 | ||
3184 | static int cgroup_open(struct inode *inode, struct file *file) | |
3185 | { | |
3186 | struct pid *pid = PROC_I(inode)->pid; | |
3187 | return single_open(file, proc_cgroup_show, pid); | |
3188 | } | |
3189 | ||
3190 | struct file_operations proc_cgroup_operations = { | |
3191 | .open = cgroup_open, | |
3192 | .read = seq_read, | |
3193 | .llseek = seq_lseek, | |
3194 | .release = single_release, | |
3195 | }; | |
3196 | ||
3197 | /* Display information about each subsystem and each hierarchy */ | |
3198 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
3199 | { | |
3200 | int i; | |
a424316c | 3201 | |
8bab8dde | 3202 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
a424316c | 3203 | mutex_lock(&cgroup_mutex); |
a424316c PM |
3204 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
3205 | struct cgroup_subsys *ss = subsys[i]; | |
2c6ab6d2 PM |
3206 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3207 | ss->name, ss->root->hierarchy_id, | |
8bab8dde | 3208 | ss->root->number_of_cgroups, !ss->disabled); |
a424316c PM |
3209 | } |
3210 | mutex_unlock(&cgroup_mutex); | |
3211 | return 0; | |
3212 | } | |
3213 | ||
3214 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
3215 | { | |
9dce07f1 | 3216 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
3217 | } |
3218 | ||
3219 | static struct file_operations proc_cgroupstats_operations = { | |
3220 | .open = cgroupstats_open, | |
3221 | .read = seq_read, | |
3222 | .llseek = seq_lseek, | |
3223 | .release = single_release, | |
3224 | }; | |
3225 | ||
b4f48b63 PM |
3226 | /** |
3227 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 3228 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
3229 | * |
3230 | * Description: A task inherits its parent's cgroup at fork(). | |
3231 | * | |
3232 | * A pointer to the shared css_set was automatically copied in | |
3233 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
3234 | * it was not made under the protection of RCU or cgroup_mutex, so | |
956db3ca | 3235 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might |
817929ec PM |
3236 | * have already changed current->cgroups, allowing the previously |
3237 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
3238 | * |
3239 | * At the point that cgroup_fork() is called, 'current' is the parent | |
3240 | * task, and the passed argument 'child' points to the child task. | |
3241 | */ | |
3242 | void cgroup_fork(struct task_struct *child) | |
3243 | { | |
817929ec PM |
3244 | task_lock(current); |
3245 | child->cgroups = current->cgroups; | |
3246 | get_css_set(child->cgroups); | |
3247 | task_unlock(current); | |
3248 | INIT_LIST_HEAD(&child->cg_list); | |
b4f48b63 PM |
3249 | } |
3250 | ||
3251 | /** | |
a043e3b2 LZ |
3252 | * cgroup_fork_callbacks - run fork callbacks |
3253 | * @child: the new task | |
3254 | * | |
3255 | * Called on a new task very soon before adding it to the | |
3256 | * tasklist. No need to take any locks since no-one can | |
3257 | * be operating on this task. | |
b4f48b63 PM |
3258 | */ |
3259 | void cgroup_fork_callbacks(struct task_struct *child) | |
3260 | { | |
3261 | if (need_forkexit_callback) { | |
3262 | int i; | |
3263 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
3264 | struct cgroup_subsys *ss = subsys[i]; | |
3265 | if (ss->fork) | |
3266 | ss->fork(ss, child); | |
3267 | } | |
3268 | } | |
3269 | } | |
3270 | ||
817929ec | 3271 | /** |
a043e3b2 LZ |
3272 | * cgroup_post_fork - called on a new task after adding it to the task list |
3273 | * @child: the task in question | |
3274 | * | |
3275 | * Adds the task to the list running through its css_set if necessary. | |
3276 | * Has to be after the task is visible on the task list in case we race | |
3277 | * with the first call to cgroup_iter_start() - to guarantee that the | |
3278 | * new task ends up on its list. | |
3279 | */ | |
817929ec PM |
3280 | void cgroup_post_fork(struct task_struct *child) |
3281 | { | |
3282 | if (use_task_css_set_links) { | |
3283 | write_lock(&css_set_lock); | |
b12b533f | 3284 | task_lock(child); |
817929ec PM |
3285 | if (list_empty(&child->cg_list)) |
3286 | list_add(&child->cg_list, &child->cgroups->tasks); | |
b12b533f | 3287 | task_unlock(child); |
817929ec PM |
3288 | write_unlock(&css_set_lock); |
3289 | } | |
3290 | } | |
b4f48b63 PM |
3291 | /** |
3292 | * cgroup_exit - detach cgroup from exiting task | |
3293 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 3294 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
3295 | * |
3296 | * Description: Detach cgroup from @tsk and release it. | |
3297 | * | |
3298 | * Note that cgroups marked notify_on_release force every task in | |
3299 | * them to take the global cgroup_mutex mutex when exiting. | |
3300 | * This could impact scaling on very large systems. Be reluctant to | |
3301 | * use notify_on_release cgroups where very high task exit scaling | |
3302 | * is required on large systems. | |
3303 | * | |
3304 | * the_top_cgroup_hack: | |
3305 | * | |
3306 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
3307 | * | |
3308 | * We call cgroup_exit() while the task is still competent to | |
3309 | * handle notify_on_release(), then leave the task attached to the | |
3310 | * root cgroup in each hierarchy for the remainder of its exit. | |
3311 | * | |
3312 | * To do this properly, we would increment the reference count on | |
3313 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
3314 | * code we would add a second cgroup function call, to drop that | |
3315 | * reference. This would just create an unnecessary hot spot on | |
3316 | * the top_cgroup reference count, to no avail. | |
3317 | * | |
3318 | * Normally, holding a reference to a cgroup without bumping its | |
3319 | * count is unsafe. The cgroup could go away, or someone could | |
3320 | * attach us to a different cgroup, decrementing the count on | |
3321 | * the first cgroup that we never incremented. But in this case, | |
3322 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
3323 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
3324 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
3325 | */ |
3326 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
3327 | { | |
3328 | int i; | |
817929ec | 3329 | struct css_set *cg; |
b4f48b63 PM |
3330 | |
3331 | if (run_callbacks && need_forkexit_callback) { | |
3332 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
3333 | struct cgroup_subsys *ss = subsys[i]; | |
3334 | if (ss->exit) | |
3335 | ss->exit(ss, tsk); | |
3336 | } | |
3337 | } | |
817929ec PM |
3338 | |
3339 | /* | |
3340 | * Unlink from the css_set task list if necessary. | |
3341 | * Optimistically check cg_list before taking | |
3342 | * css_set_lock | |
3343 | */ | |
3344 | if (!list_empty(&tsk->cg_list)) { | |
3345 | write_lock(&css_set_lock); | |
3346 | if (!list_empty(&tsk->cg_list)) | |
3347 | list_del(&tsk->cg_list); | |
3348 | write_unlock(&css_set_lock); | |
3349 | } | |
3350 | ||
b4f48b63 PM |
3351 | /* Reassign the task to the init_css_set. */ |
3352 | task_lock(tsk); | |
817929ec PM |
3353 | cg = tsk->cgroups; |
3354 | tsk->cgroups = &init_css_set; | |
b4f48b63 | 3355 | task_unlock(tsk); |
817929ec | 3356 | if (cg) |
81a6a5cd | 3357 | put_css_set_taskexit(cg); |
b4f48b63 | 3358 | } |
697f4161 PM |
3359 | |
3360 | /** | |
a043e3b2 LZ |
3361 | * cgroup_clone - clone the cgroup the given subsystem is attached to |
3362 | * @tsk: the task to be moved | |
3363 | * @subsys: the given subsystem | |
e885dcde | 3364 | * @nodename: the name for the new cgroup |
a043e3b2 LZ |
3365 | * |
3366 | * Duplicate the current cgroup in the hierarchy that the given | |
3367 | * subsystem is attached to, and move this task into the new | |
3368 | * child. | |
697f4161 | 3369 | */ |
e885dcde SH |
3370 | int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys, |
3371 | char *nodename) | |
697f4161 PM |
3372 | { |
3373 | struct dentry *dentry; | |
3374 | int ret = 0; | |
697f4161 PM |
3375 | struct cgroup *parent, *child; |
3376 | struct inode *inode; | |
3377 | struct css_set *cg; | |
3378 | struct cgroupfs_root *root; | |
3379 | struct cgroup_subsys *ss; | |
3380 | ||
3381 | /* We shouldn't be called by an unregistered subsystem */ | |
3382 | BUG_ON(!subsys->active); | |
3383 | ||
3384 | /* First figure out what hierarchy and cgroup we're dealing | |
3385 | * with, and pin them so we can drop cgroup_mutex */ | |
3386 | mutex_lock(&cgroup_mutex); | |
3387 | again: | |
3388 | root = subsys->root; | |
3389 | if (root == &rootnode) { | |
697f4161 PM |
3390 | mutex_unlock(&cgroup_mutex); |
3391 | return 0; | |
3392 | } | |
697f4161 | 3393 | |
697f4161 | 3394 | /* Pin the hierarchy */ |
1404f065 | 3395 | if (!atomic_inc_not_zero(&root->sb->s_active)) { |
7b574b7b LZ |
3396 | /* We race with the final deactivate_super() */ |
3397 | mutex_unlock(&cgroup_mutex); | |
3398 | return 0; | |
3399 | } | |
697f4161 | 3400 | |
817929ec | 3401 | /* Keep the cgroup alive */ |
1404f065 LZ |
3402 | task_lock(tsk); |
3403 | parent = task_cgroup(tsk, subsys->subsys_id); | |
3404 | cg = tsk->cgroups; | |
817929ec | 3405 | get_css_set(cg); |
104cbd55 | 3406 | task_unlock(tsk); |
1404f065 | 3407 | |
697f4161 PM |
3408 | mutex_unlock(&cgroup_mutex); |
3409 | ||
3410 | /* Now do the VFS work to create a cgroup */ | |
3411 | inode = parent->dentry->d_inode; | |
3412 | ||
3413 | /* Hold the parent directory mutex across this operation to | |
3414 | * stop anyone else deleting the new cgroup */ | |
3415 | mutex_lock(&inode->i_mutex); | |
3416 | dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename)); | |
3417 | if (IS_ERR(dentry)) { | |
3418 | printk(KERN_INFO | |
cfe36bde | 3419 | "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename, |
697f4161 PM |
3420 | PTR_ERR(dentry)); |
3421 | ret = PTR_ERR(dentry); | |
3422 | goto out_release; | |
3423 | } | |
3424 | ||
3425 | /* Create the cgroup directory, which also creates the cgroup */ | |
75139b82 | 3426 | ret = vfs_mkdir(inode, dentry, 0755); |
bd89aabc | 3427 | child = __d_cgrp(dentry); |
697f4161 PM |
3428 | dput(dentry); |
3429 | if (ret) { | |
3430 | printk(KERN_INFO | |
3431 | "Failed to create cgroup %s: %d\n", nodename, | |
3432 | ret); | |
3433 | goto out_release; | |
3434 | } | |
3435 | ||
697f4161 PM |
3436 | /* The cgroup now exists. Retake cgroup_mutex and check |
3437 | * that we're still in the same state that we thought we | |
3438 | * were. */ | |
3439 | mutex_lock(&cgroup_mutex); | |
3440 | if ((root != subsys->root) || | |
3441 | (parent != task_cgroup(tsk, subsys->subsys_id))) { | |
3442 | /* Aargh, we raced ... */ | |
3443 | mutex_unlock(&inode->i_mutex); | |
817929ec | 3444 | put_css_set(cg); |
697f4161 | 3445 | |
1404f065 | 3446 | deactivate_super(root->sb); |
697f4161 PM |
3447 | /* The cgroup is still accessible in the VFS, but |
3448 | * we're not going to try to rmdir() it at this | |
3449 | * point. */ | |
3450 | printk(KERN_INFO | |
3451 | "Race in cgroup_clone() - leaking cgroup %s\n", | |
3452 | nodename); | |
3453 | goto again; | |
3454 | } | |
3455 | ||
3456 | /* do any required auto-setup */ | |
3457 | for_each_subsys(root, ss) { | |
3458 | if (ss->post_clone) | |
3459 | ss->post_clone(ss, child); | |
3460 | } | |
3461 | ||
3462 | /* All seems fine. Finish by moving the task into the new cgroup */ | |
956db3ca | 3463 | ret = cgroup_attach_task(child, tsk); |
697f4161 PM |
3464 | mutex_unlock(&cgroup_mutex); |
3465 | ||
3466 | out_release: | |
3467 | mutex_unlock(&inode->i_mutex); | |
81a6a5cd PM |
3468 | |
3469 | mutex_lock(&cgroup_mutex); | |
817929ec | 3470 | put_css_set(cg); |
81a6a5cd | 3471 | mutex_unlock(&cgroup_mutex); |
1404f065 | 3472 | deactivate_super(root->sb); |
697f4161 PM |
3473 | return ret; |
3474 | } | |
3475 | ||
a043e3b2 | 3476 | /** |
313e924c | 3477 | * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp |
a043e3b2 | 3478 | * @cgrp: the cgroup in question |
313e924c | 3479 | * @task: the task in question |
a043e3b2 | 3480 | * |
313e924c GN |
3481 | * See if @cgrp is a descendant of @task's cgroup in the appropriate |
3482 | * hierarchy. | |
697f4161 PM |
3483 | * |
3484 | * If we are sending in dummytop, then presumably we are creating | |
3485 | * the top cgroup in the subsystem. | |
3486 | * | |
3487 | * Called only by the ns (nsproxy) cgroup. | |
3488 | */ | |
313e924c | 3489 | int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task) |
697f4161 PM |
3490 | { |
3491 | int ret; | |
3492 | struct cgroup *target; | |
697f4161 | 3493 | |
bd89aabc | 3494 | if (cgrp == dummytop) |
697f4161 PM |
3495 | return 1; |
3496 | ||
7717f7ba | 3497 | target = task_cgroup_from_root(task, cgrp->root); |
bd89aabc PM |
3498 | while (cgrp != target && cgrp!= cgrp->top_cgroup) |
3499 | cgrp = cgrp->parent; | |
3500 | ret = (cgrp == target); | |
697f4161 PM |
3501 | return ret; |
3502 | } | |
81a6a5cd | 3503 | |
bd89aabc | 3504 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd PM |
3505 | { |
3506 | /* All of these checks rely on RCU to keep the cgroup | |
3507 | * structure alive */ | |
bd89aabc PM |
3508 | if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count) |
3509 | && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) { | |
81a6a5cd PM |
3510 | /* Control Group is currently removeable. If it's not |
3511 | * already queued for a userspace notification, queue | |
3512 | * it now */ | |
3513 | int need_schedule_work = 0; | |
3514 | spin_lock(&release_list_lock); | |
bd89aabc PM |
3515 | if (!cgroup_is_removed(cgrp) && |
3516 | list_empty(&cgrp->release_list)) { | |
3517 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
3518 | need_schedule_work = 1; |
3519 | } | |
3520 | spin_unlock(&release_list_lock); | |
3521 | if (need_schedule_work) | |
3522 | schedule_work(&release_agent_work); | |
3523 | } | |
3524 | } | |
3525 | ||
3526 | void __css_put(struct cgroup_subsys_state *css) | |
3527 | { | |
bd89aabc | 3528 | struct cgroup *cgrp = css->cgroup; |
81a6a5cd | 3529 | rcu_read_lock(); |
ec64f515 KH |
3530 | if (atomic_dec_return(&css->refcnt) == 1) { |
3531 | if (notify_on_release(cgrp)) { | |
3532 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
3533 | check_for_release(cgrp); | |
3534 | } | |
88703267 | 3535 | cgroup_wakeup_rmdir_waiter(cgrp); |
81a6a5cd PM |
3536 | } |
3537 | rcu_read_unlock(); | |
3538 | } | |
3539 | ||
3540 | /* | |
3541 | * Notify userspace when a cgroup is released, by running the | |
3542 | * configured release agent with the name of the cgroup (path | |
3543 | * relative to the root of cgroup file system) as the argument. | |
3544 | * | |
3545 | * Most likely, this user command will try to rmdir this cgroup. | |
3546 | * | |
3547 | * This races with the possibility that some other task will be | |
3548 | * attached to this cgroup before it is removed, or that some other | |
3549 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
3550 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
3551 | * unused, and this cgroup will be reprieved from its death sentence, | |
3552 | * to continue to serve a useful existence. Next time it's released, | |
3553 | * we will get notified again, if it still has 'notify_on_release' set. | |
3554 | * | |
3555 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
3556 | * means only wait until the task is successfully execve()'d. The | |
3557 | * separate release agent task is forked by call_usermodehelper(), | |
3558 | * then control in this thread returns here, without waiting for the | |
3559 | * release agent task. We don't bother to wait because the caller of | |
3560 | * this routine has no use for the exit status of the release agent | |
3561 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 3562 | */ |
81a6a5cd PM |
3563 | static void cgroup_release_agent(struct work_struct *work) |
3564 | { | |
3565 | BUG_ON(work != &release_agent_work); | |
3566 | mutex_lock(&cgroup_mutex); | |
3567 | spin_lock(&release_list_lock); | |
3568 | while (!list_empty(&release_list)) { | |
3569 | char *argv[3], *envp[3]; | |
3570 | int i; | |
e788e066 | 3571 | char *pathbuf = NULL, *agentbuf = NULL; |
bd89aabc | 3572 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
3573 | struct cgroup, |
3574 | release_list); | |
bd89aabc | 3575 | list_del_init(&cgrp->release_list); |
81a6a5cd PM |
3576 | spin_unlock(&release_list_lock); |
3577 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
e788e066 PM |
3578 | if (!pathbuf) |
3579 | goto continue_free; | |
3580 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
3581 | goto continue_free; | |
3582 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
3583 | if (!agentbuf) | |
3584 | goto continue_free; | |
81a6a5cd PM |
3585 | |
3586 | i = 0; | |
e788e066 PM |
3587 | argv[i++] = agentbuf; |
3588 | argv[i++] = pathbuf; | |
81a6a5cd PM |
3589 | argv[i] = NULL; |
3590 | ||
3591 | i = 0; | |
3592 | /* minimal command environment */ | |
3593 | envp[i++] = "HOME=/"; | |
3594 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
3595 | envp[i] = NULL; | |
3596 | ||
3597 | /* Drop the lock while we invoke the usermode helper, | |
3598 | * since the exec could involve hitting disk and hence | |
3599 | * be a slow process */ | |
3600 | mutex_unlock(&cgroup_mutex); | |
3601 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 3602 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
3603 | continue_free: |
3604 | kfree(pathbuf); | |
3605 | kfree(agentbuf); | |
81a6a5cd PM |
3606 | spin_lock(&release_list_lock); |
3607 | } | |
3608 | spin_unlock(&release_list_lock); | |
3609 | mutex_unlock(&cgroup_mutex); | |
3610 | } | |
8bab8dde PM |
3611 | |
3612 | static int __init cgroup_disable(char *str) | |
3613 | { | |
3614 | int i; | |
3615 | char *token; | |
3616 | ||
3617 | while ((token = strsep(&str, ",")) != NULL) { | |
3618 | if (!*token) | |
3619 | continue; | |
3620 | ||
3621 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
3622 | struct cgroup_subsys *ss = subsys[i]; | |
3623 | ||
3624 | if (!strcmp(token, ss->name)) { | |
3625 | ss->disabled = 1; | |
3626 | printk(KERN_INFO "Disabling %s control group" | |
3627 | " subsystem\n", ss->name); | |
3628 | break; | |
3629 | } | |
3630 | } | |
3631 | } | |
3632 | return 1; | |
3633 | } | |
3634 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 KH |
3635 | |
3636 | /* | |
3637 | * Functons for CSS ID. | |
3638 | */ | |
3639 | ||
3640 | /* | |
3641 | *To get ID other than 0, this should be called when !cgroup_is_removed(). | |
3642 | */ | |
3643 | unsigned short css_id(struct cgroup_subsys_state *css) | |
3644 | { | |
3645 | struct css_id *cssid = rcu_dereference(css->id); | |
3646 | ||
3647 | if (cssid) | |
3648 | return cssid->id; | |
3649 | return 0; | |
3650 | } | |
3651 | ||
3652 | unsigned short css_depth(struct cgroup_subsys_state *css) | |
3653 | { | |
3654 | struct css_id *cssid = rcu_dereference(css->id); | |
3655 | ||
3656 | if (cssid) | |
3657 | return cssid->depth; | |
3658 | return 0; | |
3659 | } | |
3660 | ||
3661 | bool css_is_ancestor(struct cgroup_subsys_state *child, | |
0b7f569e | 3662 | const struct cgroup_subsys_state *root) |
38460b48 KH |
3663 | { |
3664 | struct css_id *child_id = rcu_dereference(child->id); | |
3665 | struct css_id *root_id = rcu_dereference(root->id); | |
3666 | ||
3667 | if (!child_id || !root_id || (child_id->depth < root_id->depth)) | |
3668 | return false; | |
3669 | return child_id->stack[root_id->depth] == root_id->id; | |
3670 | } | |
3671 | ||
3672 | static void __free_css_id_cb(struct rcu_head *head) | |
3673 | { | |
3674 | struct css_id *id; | |
3675 | ||
3676 | id = container_of(head, struct css_id, rcu_head); | |
3677 | kfree(id); | |
3678 | } | |
3679 | ||
3680 | void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css) | |
3681 | { | |
3682 | struct css_id *id = css->id; | |
3683 | /* When this is called before css_id initialization, id can be NULL */ | |
3684 | if (!id) | |
3685 | return; | |
3686 | ||
3687 | BUG_ON(!ss->use_id); | |
3688 | ||
3689 | rcu_assign_pointer(id->css, NULL); | |
3690 | rcu_assign_pointer(css->id, NULL); | |
3691 | spin_lock(&ss->id_lock); | |
3692 | idr_remove(&ss->idr, id->id); | |
3693 | spin_unlock(&ss->id_lock); | |
3694 | call_rcu(&id->rcu_head, __free_css_id_cb); | |
3695 | } | |
3696 | ||
3697 | /* | |
3698 | * This is called by init or create(). Then, calls to this function are | |
3699 | * always serialized (By cgroup_mutex() at create()). | |
3700 | */ | |
3701 | ||
3702 | static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth) | |
3703 | { | |
3704 | struct css_id *newid; | |
3705 | int myid, error, size; | |
3706 | ||
3707 | BUG_ON(!ss->use_id); | |
3708 | ||
3709 | size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1); | |
3710 | newid = kzalloc(size, GFP_KERNEL); | |
3711 | if (!newid) | |
3712 | return ERR_PTR(-ENOMEM); | |
3713 | /* get id */ | |
3714 | if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) { | |
3715 | error = -ENOMEM; | |
3716 | goto err_out; | |
3717 | } | |
3718 | spin_lock(&ss->id_lock); | |
3719 | /* Don't use 0. allocates an ID of 1-65535 */ | |
3720 | error = idr_get_new_above(&ss->idr, newid, 1, &myid); | |
3721 | spin_unlock(&ss->id_lock); | |
3722 | ||
3723 | /* Returns error when there are no free spaces for new ID.*/ | |
3724 | if (error) { | |
3725 | error = -ENOSPC; | |
3726 | goto err_out; | |
3727 | } | |
3728 | if (myid > CSS_ID_MAX) | |
3729 | goto remove_idr; | |
3730 | ||
3731 | newid->id = myid; | |
3732 | newid->depth = depth; | |
3733 | return newid; | |
3734 | remove_idr: | |
3735 | error = -ENOSPC; | |
3736 | spin_lock(&ss->id_lock); | |
3737 | idr_remove(&ss->idr, myid); | |
3738 | spin_unlock(&ss->id_lock); | |
3739 | err_out: | |
3740 | kfree(newid); | |
3741 | return ERR_PTR(error); | |
3742 | ||
3743 | } | |
3744 | ||
3745 | static int __init cgroup_subsys_init_idr(struct cgroup_subsys *ss) | |
3746 | { | |
3747 | struct css_id *newid; | |
3748 | struct cgroup_subsys_state *rootcss; | |
3749 | ||
3750 | spin_lock_init(&ss->id_lock); | |
3751 | idr_init(&ss->idr); | |
3752 | ||
3753 | rootcss = init_css_set.subsys[ss->subsys_id]; | |
3754 | newid = get_new_cssid(ss, 0); | |
3755 | if (IS_ERR(newid)) | |
3756 | return PTR_ERR(newid); | |
3757 | ||
3758 | newid->stack[0] = newid->id; | |
3759 | newid->css = rootcss; | |
3760 | rootcss->id = newid; | |
3761 | return 0; | |
3762 | } | |
3763 | ||
3764 | static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent, | |
3765 | struct cgroup *child) | |
3766 | { | |
3767 | int subsys_id, i, depth = 0; | |
3768 | struct cgroup_subsys_state *parent_css, *child_css; | |
3769 | struct css_id *child_id, *parent_id = NULL; | |
3770 | ||
3771 | subsys_id = ss->subsys_id; | |
3772 | parent_css = parent->subsys[subsys_id]; | |
3773 | child_css = child->subsys[subsys_id]; | |
3774 | depth = css_depth(parent_css) + 1; | |
3775 | parent_id = parent_css->id; | |
3776 | ||
3777 | child_id = get_new_cssid(ss, depth); | |
3778 | if (IS_ERR(child_id)) | |
3779 | return PTR_ERR(child_id); | |
3780 | ||
3781 | for (i = 0; i < depth; i++) | |
3782 | child_id->stack[i] = parent_id->stack[i]; | |
3783 | child_id->stack[depth] = child_id->id; | |
3784 | /* | |
3785 | * child_id->css pointer will be set after this cgroup is available | |
3786 | * see cgroup_populate_dir() | |
3787 | */ | |
3788 | rcu_assign_pointer(child_css->id, child_id); | |
3789 | ||
3790 | return 0; | |
3791 | } | |
3792 | ||
3793 | /** | |
3794 | * css_lookup - lookup css by id | |
3795 | * @ss: cgroup subsys to be looked into. | |
3796 | * @id: the id | |
3797 | * | |
3798 | * Returns pointer to cgroup_subsys_state if there is valid one with id. | |
3799 | * NULL if not. Should be called under rcu_read_lock() | |
3800 | */ | |
3801 | struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id) | |
3802 | { | |
3803 | struct css_id *cssid = NULL; | |
3804 | ||
3805 | BUG_ON(!ss->use_id); | |
3806 | cssid = idr_find(&ss->idr, id); | |
3807 | ||
3808 | if (unlikely(!cssid)) | |
3809 | return NULL; | |
3810 | ||
3811 | return rcu_dereference(cssid->css); | |
3812 | } | |
3813 | ||
3814 | /** | |
3815 | * css_get_next - lookup next cgroup under specified hierarchy. | |
3816 | * @ss: pointer to subsystem | |
3817 | * @id: current position of iteration. | |
3818 | * @root: pointer to css. search tree under this. | |
3819 | * @foundid: position of found object. | |
3820 | * | |
3821 | * Search next css under the specified hierarchy of rootid. Calling under | |
3822 | * rcu_read_lock() is necessary. Returns NULL if it reaches the end. | |
3823 | */ | |
3824 | struct cgroup_subsys_state * | |
3825 | css_get_next(struct cgroup_subsys *ss, int id, | |
3826 | struct cgroup_subsys_state *root, int *foundid) | |
3827 | { | |
3828 | struct cgroup_subsys_state *ret = NULL; | |
3829 | struct css_id *tmp; | |
3830 | int tmpid; | |
3831 | int rootid = css_id(root); | |
3832 | int depth = css_depth(root); | |
3833 | ||
3834 | if (!rootid) | |
3835 | return NULL; | |
3836 | ||
3837 | BUG_ON(!ss->use_id); | |
3838 | /* fill start point for scan */ | |
3839 | tmpid = id; | |
3840 | while (1) { | |
3841 | /* | |
3842 | * scan next entry from bitmap(tree), tmpid is updated after | |
3843 | * idr_get_next(). | |
3844 | */ | |
3845 | spin_lock(&ss->id_lock); | |
3846 | tmp = idr_get_next(&ss->idr, &tmpid); | |
3847 | spin_unlock(&ss->id_lock); | |
3848 | ||
3849 | if (!tmp) | |
3850 | break; | |
3851 | if (tmp->depth >= depth && tmp->stack[depth] == rootid) { | |
3852 | ret = rcu_dereference(tmp->css); | |
3853 | if (ret) { | |
3854 | *foundid = tmpid; | |
3855 | break; | |
3856 | } | |
3857 | } | |
3858 | /* continue to scan from next id */ | |
3859 | tmpid = tmpid + 1; | |
3860 | } | |
3861 | return ret; | |
3862 | } | |
3863 | ||
fe693435 PM |
3864 | #ifdef CONFIG_CGROUP_DEBUG |
3865 | static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss, | |
3866 | struct cgroup *cont) | |
3867 | { | |
3868 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
3869 | ||
3870 | if (!css) | |
3871 | return ERR_PTR(-ENOMEM); | |
3872 | ||
3873 | return css; | |
3874 | } | |
3875 | ||
3876 | static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
3877 | { | |
3878 | kfree(cont->subsys[debug_subsys_id]); | |
3879 | } | |
3880 | ||
3881 | static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft) | |
3882 | { | |
3883 | return atomic_read(&cont->count); | |
3884 | } | |
3885 | ||
3886 | static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft) | |
3887 | { | |
3888 | return cgroup_task_count(cont); | |
3889 | } | |
3890 | ||
3891 | static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft) | |
3892 | { | |
3893 | return (u64)(unsigned long)current->cgroups; | |
3894 | } | |
3895 | ||
3896 | static u64 current_css_set_refcount_read(struct cgroup *cont, | |
3897 | struct cftype *cft) | |
3898 | { | |
3899 | u64 count; | |
3900 | ||
3901 | rcu_read_lock(); | |
3902 | count = atomic_read(¤t->cgroups->refcount); | |
3903 | rcu_read_unlock(); | |
3904 | return count; | |
3905 | } | |
3906 | ||
7717f7ba PM |
3907 | static int current_css_set_cg_links_read(struct cgroup *cont, |
3908 | struct cftype *cft, | |
3909 | struct seq_file *seq) | |
3910 | { | |
3911 | struct cg_cgroup_link *link; | |
3912 | struct css_set *cg; | |
3913 | ||
3914 | read_lock(&css_set_lock); | |
3915 | rcu_read_lock(); | |
3916 | cg = rcu_dereference(current->cgroups); | |
3917 | list_for_each_entry(link, &cg->cg_links, cg_link_list) { | |
3918 | struct cgroup *c = link->cgrp; | |
3919 | const char *name; | |
3920 | ||
3921 | if (c->dentry) | |
3922 | name = c->dentry->d_name.name; | |
3923 | else | |
3924 | name = "?"; | |
2c6ab6d2 PM |
3925 | seq_printf(seq, "Root %d group %s\n", |
3926 | c->root->hierarchy_id, name); | |
7717f7ba PM |
3927 | } |
3928 | rcu_read_unlock(); | |
3929 | read_unlock(&css_set_lock); | |
3930 | return 0; | |
3931 | } | |
3932 | ||
3933 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
3934 | static int cgroup_css_links_read(struct cgroup *cont, | |
3935 | struct cftype *cft, | |
3936 | struct seq_file *seq) | |
3937 | { | |
3938 | struct cg_cgroup_link *link; | |
3939 | ||
3940 | read_lock(&css_set_lock); | |
3941 | list_for_each_entry(link, &cont->css_sets, cgrp_link_list) { | |
3942 | struct css_set *cg = link->cg; | |
3943 | struct task_struct *task; | |
3944 | int count = 0; | |
3945 | seq_printf(seq, "css_set %p\n", cg); | |
3946 | list_for_each_entry(task, &cg->tasks, cg_list) { | |
3947 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { | |
3948 | seq_puts(seq, " ...\n"); | |
3949 | break; | |
3950 | } else { | |
3951 | seq_printf(seq, " task %d\n", | |
3952 | task_pid_vnr(task)); | |
3953 | } | |
3954 | } | |
3955 | } | |
3956 | read_unlock(&css_set_lock); | |
3957 | return 0; | |
3958 | } | |
3959 | ||
fe693435 PM |
3960 | static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft) |
3961 | { | |
3962 | return test_bit(CGRP_RELEASABLE, &cgrp->flags); | |
3963 | } | |
3964 | ||
3965 | static struct cftype debug_files[] = { | |
3966 | { | |
3967 | .name = "cgroup_refcount", | |
3968 | .read_u64 = cgroup_refcount_read, | |
3969 | }, | |
3970 | { | |
3971 | .name = "taskcount", | |
3972 | .read_u64 = debug_taskcount_read, | |
3973 | }, | |
3974 | ||
3975 | { | |
3976 | .name = "current_css_set", | |
3977 | .read_u64 = current_css_set_read, | |
3978 | }, | |
3979 | ||
3980 | { | |
3981 | .name = "current_css_set_refcount", | |
3982 | .read_u64 = current_css_set_refcount_read, | |
3983 | }, | |
3984 | ||
7717f7ba PM |
3985 | { |
3986 | .name = "current_css_set_cg_links", | |
3987 | .read_seq_string = current_css_set_cg_links_read, | |
3988 | }, | |
3989 | ||
3990 | { | |
3991 | .name = "cgroup_css_links", | |
3992 | .read_seq_string = cgroup_css_links_read, | |
3993 | }, | |
3994 | ||
fe693435 PM |
3995 | { |
3996 | .name = "releasable", | |
3997 | .read_u64 = releasable_read, | |
3998 | }, | |
3999 | }; | |
4000 | ||
4001 | static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
4002 | { | |
4003 | return cgroup_add_files(cont, ss, debug_files, | |
4004 | ARRAY_SIZE(debug_files)); | |
4005 | } | |
4006 | ||
4007 | struct cgroup_subsys debug_subsys = { | |
4008 | .name = "debug", | |
4009 | .create = debug_create, | |
4010 | .destroy = debug_destroy, | |
4011 | .populate = debug_populate, | |
4012 | .subsys_id = debug_subsys_id, | |
4013 | }; | |
4014 | #endif /* CONFIG_CGROUP_DEBUG */ |