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 | * | |
0dea1168 KS |
7 | * Notifications support |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
ddbcc7e8 PM |
11 | * Copyright notices from the original cpuset code: |
12 | * -------------------------------------------------- | |
13 | * Copyright (C) 2003 BULL SA. | |
14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
15 | * | |
16 | * Portions derived from Patrick Mochel's sysfs code. | |
17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
18 | * | |
19 | * 2003-10-10 Written by Simon Derr. | |
20 | * 2003-10-22 Updates by Stephen Hemminger. | |
21 | * 2004 May-July Rework by Paul Jackson. | |
22 | * --------------------------------------------------- | |
23 | * | |
24 | * This file is subject to the terms and conditions of the GNU General Public | |
25 | * License. See the file COPYING in the main directory of the Linux | |
26 | * distribution for more details. | |
27 | */ | |
28 | ||
29 | #include <linux/cgroup.h> | |
2ce9738b | 30 | #include <linux/cred.h> |
c6d57f33 | 31 | #include <linux/ctype.h> |
ddbcc7e8 PM |
32 | #include <linux/errno.h> |
33 | #include <linux/fs.h> | |
2ce9738b | 34 | #include <linux/init_task.h> |
ddbcc7e8 PM |
35 | #include <linux/kernel.h> |
36 | #include <linux/list.h> | |
37 | #include <linux/mm.h> | |
38 | #include <linux/mutex.h> | |
39 | #include <linux/mount.h> | |
40 | #include <linux/pagemap.h> | |
a424316c | 41 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
42 | #include <linux/rcupdate.h> |
43 | #include <linux/sched.h> | |
817929ec | 44 | #include <linux/backing-dev.h> |
ddbcc7e8 PM |
45 | #include <linux/seq_file.h> |
46 | #include <linux/slab.h> | |
47 | #include <linux/magic.h> | |
48 | #include <linux/spinlock.h> | |
49 | #include <linux/string.h> | |
bbcb81d0 | 50 | #include <linux/sort.h> |
81a6a5cd | 51 | #include <linux/kmod.h> |
e6a1105b | 52 | #include <linux/module.h> |
846c7bb0 BS |
53 | #include <linux/delayacct.h> |
54 | #include <linux/cgroupstats.h> | |
472b1053 | 55 | #include <linux/hash.h> |
3f8206d4 | 56 | #include <linux/namei.h> |
096b7fe0 | 57 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 58 | #include <linux/idr.h> |
d1d9fd33 | 59 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
0dea1168 KS |
60 | #include <linux/eventfd.h> |
61 | #include <linux/poll.h> | |
d846687d | 62 | #include <linux/flex_array.h> /* used in cgroup_attach_proc */ |
846c7bb0 | 63 | |
60063497 | 64 | #include <linux/atomic.h> |
ddbcc7e8 | 65 | |
e25e2cbb TH |
66 | /* |
67 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
68 | * hierarchy must be performed while holding it. | |
69 | * | |
70 | * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify | |
71 | * cgroupfs_root of any cgroup hierarchy - subsys list, flags, | |
72 | * release_agent_path and so on. Modifying requires both cgroup_mutex and | |
73 | * cgroup_root_mutex. Readers can acquire either of the two. This is to | |
74 | * break the following locking order cycle. | |
75 | * | |
76 | * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem | |
77 | * B. namespace_sem -> cgroup_mutex | |
78 | * | |
79 | * B happens only through cgroup_show_options() and using cgroup_root_mutex | |
80 | * breaks it. | |
81 | */ | |
81a6a5cd | 82 | static DEFINE_MUTEX(cgroup_mutex); |
e25e2cbb | 83 | static DEFINE_MUTEX(cgroup_root_mutex); |
81a6a5cd | 84 | |
aae8aab4 BB |
85 | /* |
86 | * Generate an array of cgroup subsystem pointers. At boot time, this is | |
87 | * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are | |
88 | * registered after that. The mutable section of this array is protected by | |
89 | * cgroup_mutex. | |
90 | */ | |
ddbcc7e8 | 91 | #define SUBSYS(_x) &_x ## _subsys, |
aae8aab4 | 92 | static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = { |
ddbcc7e8 PM |
93 | #include <linux/cgroup_subsys.h> |
94 | }; | |
95 | ||
c6d57f33 PM |
96 | #define MAX_CGROUP_ROOT_NAMELEN 64 |
97 | ||
ddbcc7e8 PM |
98 | /* |
99 | * A cgroupfs_root represents the root of a cgroup hierarchy, | |
100 | * and may be associated with a superblock to form an active | |
101 | * hierarchy | |
102 | */ | |
103 | struct cgroupfs_root { | |
104 | struct super_block *sb; | |
105 | ||
106 | /* | |
107 | * The bitmask of subsystems intended to be attached to this | |
108 | * hierarchy | |
109 | */ | |
110 | unsigned long subsys_bits; | |
111 | ||
2c6ab6d2 PM |
112 | /* Unique id for this hierarchy. */ |
113 | int hierarchy_id; | |
114 | ||
ddbcc7e8 PM |
115 | /* The bitmask of subsystems currently attached to this hierarchy */ |
116 | unsigned long actual_subsys_bits; | |
117 | ||
118 | /* A list running through the attached subsystems */ | |
119 | struct list_head subsys_list; | |
120 | ||
121 | /* The root cgroup for this hierarchy */ | |
122 | struct cgroup top_cgroup; | |
123 | ||
124 | /* Tracks how many cgroups are currently defined in hierarchy.*/ | |
125 | int number_of_cgroups; | |
126 | ||
e5f6a860 | 127 | /* A list running through the active hierarchies */ |
ddbcc7e8 PM |
128 | struct list_head root_list; |
129 | ||
130 | /* Hierarchy-specific flags */ | |
131 | unsigned long flags; | |
81a6a5cd | 132 | |
e788e066 | 133 | /* The path to use for release notifications. */ |
81a6a5cd | 134 | char release_agent_path[PATH_MAX]; |
c6d57f33 PM |
135 | |
136 | /* The name for this hierarchy - may be empty */ | |
137 | char name[MAX_CGROUP_ROOT_NAMELEN]; | |
ddbcc7e8 PM |
138 | }; |
139 | ||
ddbcc7e8 PM |
140 | /* |
141 | * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the | |
142 | * subsystems that are otherwise unattached - it never has more than a | |
143 | * single cgroup, and all tasks are part of that cgroup. | |
144 | */ | |
145 | static struct cgroupfs_root rootnode; | |
146 | ||
38460b48 KH |
147 | /* |
148 | * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when | |
149 | * cgroup_subsys->use_id != 0. | |
150 | */ | |
151 | #define CSS_ID_MAX (65535) | |
152 | struct css_id { | |
153 | /* | |
154 | * The css to which this ID points. This pointer is set to valid value | |
155 | * after cgroup is populated. If cgroup is removed, this will be NULL. | |
156 | * This pointer is expected to be RCU-safe because destroy() | |
157 | * is called after synchronize_rcu(). But for safe use, css_is_removed() | |
158 | * css_tryget() should be used for avoiding race. | |
159 | */ | |
2c392b8c | 160 | struct cgroup_subsys_state __rcu *css; |
38460b48 KH |
161 | /* |
162 | * ID of this css. | |
163 | */ | |
164 | unsigned short id; | |
165 | /* | |
166 | * Depth in hierarchy which this ID belongs to. | |
167 | */ | |
168 | unsigned short depth; | |
169 | /* | |
170 | * ID is freed by RCU. (and lookup routine is RCU safe.) | |
171 | */ | |
172 | struct rcu_head rcu_head; | |
173 | /* | |
174 | * Hierarchy of CSS ID belongs to. | |
175 | */ | |
176 | unsigned short stack[0]; /* Array of Length (depth+1) */ | |
177 | }; | |
178 | ||
0dea1168 | 179 | /* |
25985edc | 180 | * cgroup_event represents events which userspace want to receive. |
0dea1168 KS |
181 | */ |
182 | struct cgroup_event { | |
183 | /* | |
184 | * Cgroup which the event belongs to. | |
185 | */ | |
186 | struct cgroup *cgrp; | |
187 | /* | |
188 | * Control file which the event associated. | |
189 | */ | |
190 | struct cftype *cft; | |
191 | /* | |
192 | * eventfd to signal userspace about the event. | |
193 | */ | |
194 | struct eventfd_ctx *eventfd; | |
195 | /* | |
196 | * Each of these stored in a list by the cgroup. | |
197 | */ | |
198 | struct list_head list; | |
199 | /* | |
200 | * All fields below needed to unregister event when | |
201 | * userspace closes eventfd. | |
202 | */ | |
203 | poll_table pt; | |
204 | wait_queue_head_t *wqh; | |
205 | wait_queue_t wait; | |
206 | struct work_struct remove; | |
207 | }; | |
38460b48 | 208 | |
ddbcc7e8 PM |
209 | /* The list of hierarchy roots */ |
210 | ||
211 | static LIST_HEAD(roots); | |
817929ec | 212 | static int root_count; |
ddbcc7e8 | 213 | |
2c6ab6d2 PM |
214 | static DEFINE_IDA(hierarchy_ida); |
215 | static int next_hierarchy_id; | |
216 | static DEFINE_SPINLOCK(hierarchy_id_lock); | |
217 | ||
ddbcc7e8 PM |
218 | /* dummytop is a shorthand for the dummy hierarchy's top cgroup */ |
219 | #define dummytop (&rootnode.top_cgroup) | |
220 | ||
221 | /* This flag indicates whether tasks in the fork and exit paths should | |
a043e3b2 LZ |
222 | * check for fork/exit handlers to call. This avoids us having to do |
223 | * extra work in the fork/exit path if none of the subsystems need to | |
224 | * be called. | |
ddbcc7e8 | 225 | */ |
8947f9d5 | 226 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 227 | |
d11c563d PM |
228 | #ifdef CONFIG_PROVE_LOCKING |
229 | int cgroup_lock_is_held(void) | |
230 | { | |
231 | return lockdep_is_held(&cgroup_mutex); | |
232 | } | |
233 | #else /* #ifdef CONFIG_PROVE_LOCKING */ | |
234 | int cgroup_lock_is_held(void) | |
235 | { | |
236 | return mutex_is_locked(&cgroup_mutex); | |
237 | } | |
238 | #endif /* #else #ifdef CONFIG_PROVE_LOCKING */ | |
239 | ||
240 | EXPORT_SYMBOL_GPL(cgroup_lock_is_held); | |
241 | ||
ddbcc7e8 | 242 | /* convenient tests for these bits */ |
bd89aabc | 243 | inline int cgroup_is_removed(const struct cgroup *cgrp) |
ddbcc7e8 | 244 | { |
bd89aabc | 245 | return test_bit(CGRP_REMOVED, &cgrp->flags); |
ddbcc7e8 PM |
246 | } |
247 | ||
248 | /* bits in struct cgroupfs_root flags field */ | |
249 | enum { | |
250 | ROOT_NOPREFIX, /* mounted subsystems have no named prefix */ | |
251 | }; | |
252 | ||
e9685a03 | 253 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
254 | { |
255 | const int bits = | |
bd89aabc PM |
256 | (1 << CGRP_RELEASABLE) | |
257 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
258 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
259 | } |
260 | ||
e9685a03 | 261 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 262 | { |
bd89aabc | 263 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
264 | } |
265 | ||
97978e6d DL |
266 | static int clone_children(const struct cgroup *cgrp) |
267 | { | |
268 | return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); | |
269 | } | |
270 | ||
ddbcc7e8 PM |
271 | /* |
272 | * for_each_subsys() allows you to iterate on each subsystem attached to | |
273 | * an active hierarchy | |
274 | */ | |
275 | #define for_each_subsys(_root, _ss) \ | |
276 | list_for_each_entry(_ss, &_root->subsys_list, sibling) | |
277 | ||
e5f6a860 LZ |
278 | /* for_each_active_root() allows you to iterate across the active hierarchies */ |
279 | #define for_each_active_root(_root) \ | |
ddbcc7e8 PM |
280 | list_for_each_entry(_root, &roots, root_list) |
281 | ||
81a6a5cd PM |
282 | /* the list of cgroups eligible for automatic release. Protected by |
283 | * release_list_lock */ | |
284 | static LIST_HEAD(release_list); | |
cdcc136f | 285 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
286 | static void cgroup_release_agent(struct work_struct *work); |
287 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 288 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 289 | |
817929ec PM |
290 | /* Link structure for associating css_set objects with cgroups */ |
291 | struct cg_cgroup_link { | |
292 | /* | |
293 | * List running through cg_cgroup_links associated with a | |
294 | * cgroup, anchored on cgroup->css_sets | |
295 | */ | |
bd89aabc | 296 | struct list_head cgrp_link_list; |
7717f7ba | 297 | struct cgroup *cgrp; |
817929ec PM |
298 | /* |
299 | * List running through cg_cgroup_links pointing at a | |
300 | * single css_set object, anchored on css_set->cg_links | |
301 | */ | |
302 | struct list_head cg_link_list; | |
303 | struct css_set *cg; | |
304 | }; | |
305 | ||
306 | /* The default css_set - used by init and its children prior to any | |
307 | * hierarchies being mounted. It contains a pointer to the root state | |
308 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
309 | * reference-counted, to improve performance when child cgroups | |
310 | * haven't been created. | |
311 | */ | |
312 | ||
313 | static struct css_set init_css_set; | |
314 | static struct cg_cgroup_link init_css_set_link; | |
315 | ||
e6a1105b BB |
316 | static int cgroup_init_idr(struct cgroup_subsys *ss, |
317 | struct cgroup_subsys_state *css); | |
38460b48 | 318 | |
817929ec PM |
319 | /* css_set_lock protects the list of css_set objects, and the |
320 | * chain of tasks off each css_set. Nests outside task->alloc_lock | |
321 | * due to cgroup_iter_start() */ | |
322 | static DEFINE_RWLOCK(css_set_lock); | |
323 | static int css_set_count; | |
324 | ||
7717f7ba PM |
325 | /* |
326 | * hash table for cgroup groups. This improves the performance to find | |
327 | * an existing css_set. This hash doesn't (currently) take into | |
328 | * account cgroups in empty hierarchies. | |
329 | */ | |
472b1053 LZ |
330 | #define CSS_SET_HASH_BITS 7 |
331 | #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS) | |
332 | static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE]; | |
333 | ||
334 | static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[]) | |
335 | { | |
336 | int i; | |
337 | int index; | |
338 | unsigned long tmp = 0UL; | |
339 | ||
340 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) | |
341 | tmp += (unsigned long)css[i]; | |
342 | tmp = (tmp >> 16) ^ tmp; | |
343 | ||
344 | index = hash_long(tmp, CSS_SET_HASH_BITS); | |
345 | ||
346 | return &css_set_table[index]; | |
347 | } | |
348 | ||
817929ec PM |
349 | /* We don't maintain the lists running through each css_set to its |
350 | * task until after the first call to cgroup_iter_start(). This | |
351 | * reduces the fork()/exit() overhead for people who have cgroups | |
352 | * compiled into their kernel but not actually in use */ | |
8947f9d5 | 353 | static int use_task_css_set_links __read_mostly; |
817929ec | 354 | |
2c6ab6d2 | 355 | static void __put_css_set(struct css_set *cg, int taskexit) |
b4f48b63 | 356 | { |
71cbb949 KM |
357 | struct cg_cgroup_link *link; |
358 | struct cg_cgroup_link *saved_link; | |
146aa1bd LJ |
359 | /* |
360 | * Ensure that the refcount doesn't hit zero while any readers | |
361 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
362 | * rwlock | |
363 | */ | |
364 | if (atomic_add_unless(&cg->refcount, -1, 1)) | |
365 | return; | |
366 | write_lock(&css_set_lock); | |
367 | if (!atomic_dec_and_test(&cg->refcount)) { | |
368 | write_unlock(&css_set_lock); | |
369 | return; | |
370 | } | |
81a6a5cd | 371 | |
2c6ab6d2 PM |
372 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
373 | hlist_del(&cg->hlist); | |
374 | css_set_count--; | |
375 | ||
376 | list_for_each_entry_safe(link, saved_link, &cg->cg_links, | |
377 | cg_link_list) { | |
378 | struct cgroup *cgrp = link->cgrp; | |
379 | list_del(&link->cg_link_list); | |
380 | list_del(&link->cgrp_link_list); | |
bd89aabc PM |
381 | if (atomic_dec_and_test(&cgrp->count) && |
382 | notify_on_release(cgrp)) { | |
81a6a5cd | 383 | if (taskexit) |
bd89aabc PM |
384 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
385 | check_for_release(cgrp); | |
81a6a5cd | 386 | } |
2c6ab6d2 PM |
387 | |
388 | kfree(link); | |
81a6a5cd | 389 | } |
2c6ab6d2 PM |
390 | |
391 | write_unlock(&css_set_lock); | |
30088ad8 | 392 | kfree_rcu(cg, rcu_head); |
b4f48b63 PM |
393 | } |
394 | ||
817929ec PM |
395 | /* |
396 | * refcounted get/put for css_set objects | |
397 | */ | |
398 | static inline void get_css_set(struct css_set *cg) | |
399 | { | |
146aa1bd | 400 | atomic_inc(&cg->refcount); |
817929ec PM |
401 | } |
402 | ||
403 | static inline void put_css_set(struct css_set *cg) | |
404 | { | |
146aa1bd | 405 | __put_css_set(cg, 0); |
817929ec PM |
406 | } |
407 | ||
81a6a5cd PM |
408 | static inline void put_css_set_taskexit(struct css_set *cg) |
409 | { | |
146aa1bd | 410 | __put_css_set(cg, 1); |
81a6a5cd PM |
411 | } |
412 | ||
7717f7ba PM |
413 | /* |
414 | * compare_css_sets - helper function for find_existing_css_set(). | |
415 | * @cg: candidate css_set being tested | |
416 | * @old_cg: existing css_set for a task | |
417 | * @new_cgrp: cgroup that's being entered by the task | |
418 | * @template: desired set of css pointers in css_set (pre-calculated) | |
419 | * | |
420 | * Returns true if "cg" matches "old_cg" except for the hierarchy | |
421 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". | |
422 | */ | |
423 | static bool compare_css_sets(struct css_set *cg, | |
424 | struct css_set *old_cg, | |
425 | struct cgroup *new_cgrp, | |
426 | struct cgroup_subsys_state *template[]) | |
427 | { | |
428 | struct list_head *l1, *l2; | |
429 | ||
430 | if (memcmp(template, cg->subsys, sizeof(cg->subsys))) { | |
431 | /* Not all subsystems matched */ | |
432 | return false; | |
433 | } | |
434 | ||
435 | /* | |
436 | * Compare cgroup pointers in order to distinguish between | |
437 | * different cgroups in heirarchies with no subsystems. We | |
438 | * could get by with just this check alone (and skip the | |
439 | * memcmp above) but on most setups the memcmp check will | |
440 | * avoid the need for this more expensive check on almost all | |
441 | * candidates. | |
442 | */ | |
443 | ||
444 | l1 = &cg->cg_links; | |
445 | l2 = &old_cg->cg_links; | |
446 | while (1) { | |
447 | struct cg_cgroup_link *cgl1, *cgl2; | |
448 | struct cgroup *cg1, *cg2; | |
449 | ||
450 | l1 = l1->next; | |
451 | l2 = l2->next; | |
452 | /* See if we reached the end - both lists are equal length. */ | |
453 | if (l1 == &cg->cg_links) { | |
454 | BUG_ON(l2 != &old_cg->cg_links); | |
455 | break; | |
456 | } else { | |
457 | BUG_ON(l2 == &old_cg->cg_links); | |
458 | } | |
459 | /* Locate the cgroups associated with these links. */ | |
460 | cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list); | |
461 | cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list); | |
462 | cg1 = cgl1->cgrp; | |
463 | cg2 = cgl2->cgrp; | |
464 | /* Hierarchies should be linked in the same order. */ | |
465 | BUG_ON(cg1->root != cg2->root); | |
466 | ||
467 | /* | |
468 | * If this hierarchy is the hierarchy of the cgroup | |
469 | * that's changing, then we need to check that this | |
470 | * css_set points to the new cgroup; if it's any other | |
471 | * hierarchy, then this css_set should point to the | |
472 | * same cgroup as the old css_set. | |
473 | */ | |
474 | if (cg1->root == new_cgrp->root) { | |
475 | if (cg1 != new_cgrp) | |
476 | return false; | |
477 | } else { | |
478 | if (cg1 != cg2) | |
479 | return false; | |
480 | } | |
481 | } | |
482 | return true; | |
483 | } | |
484 | ||
817929ec PM |
485 | /* |
486 | * find_existing_css_set() is a helper for | |
487 | * find_css_set(), and checks to see whether an existing | |
472b1053 | 488 | * css_set is suitable. |
817929ec PM |
489 | * |
490 | * oldcg: the cgroup group that we're using before the cgroup | |
491 | * transition | |
492 | * | |
bd89aabc | 493 | * cgrp: the cgroup that we're moving into |
817929ec PM |
494 | * |
495 | * template: location in which to build the desired set of subsystem | |
496 | * state objects for the new cgroup group | |
497 | */ | |
817929ec PM |
498 | static struct css_set *find_existing_css_set( |
499 | struct css_set *oldcg, | |
bd89aabc | 500 | struct cgroup *cgrp, |
817929ec | 501 | struct cgroup_subsys_state *template[]) |
b4f48b63 PM |
502 | { |
503 | int i; | |
bd89aabc | 504 | struct cgroupfs_root *root = cgrp->root; |
472b1053 LZ |
505 | struct hlist_head *hhead; |
506 | struct hlist_node *node; | |
507 | struct css_set *cg; | |
817929ec | 508 | |
aae8aab4 BB |
509 | /* |
510 | * Build the set of subsystem state objects that we want to see in the | |
511 | * new css_set. while subsystems can change globally, the entries here | |
512 | * won't change, so no need for locking. | |
513 | */ | |
817929ec | 514 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
8d53d55d | 515 | if (root->subsys_bits & (1UL << i)) { |
817929ec PM |
516 | /* Subsystem is in this hierarchy. So we want |
517 | * the subsystem state from the new | |
518 | * cgroup */ | |
bd89aabc | 519 | template[i] = cgrp->subsys[i]; |
817929ec PM |
520 | } else { |
521 | /* Subsystem is not in this hierarchy, so we | |
522 | * don't want to change the subsystem state */ | |
523 | template[i] = oldcg->subsys[i]; | |
524 | } | |
525 | } | |
526 | ||
472b1053 LZ |
527 | hhead = css_set_hash(template); |
528 | hlist_for_each_entry(cg, node, hhead, hlist) { | |
7717f7ba PM |
529 | if (!compare_css_sets(cg, oldcg, cgrp, template)) |
530 | continue; | |
531 | ||
532 | /* This css_set matches what we need */ | |
533 | return cg; | |
472b1053 | 534 | } |
817929ec PM |
535 | |
536 | /* No existing cgroup group matched */ | |
537 | return NULL; | |
538 | } | |
539 | ||
36553434 LZ |
540 | static void free_cg_links(struct list_head *tmp) |
541 | { | |
542 | struct cg_cgroup_link *link; | |
543 | struct cg_cgroup_link *saved_link; | |
544 | ||
545 | list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) { | |
546 | list_del(&link->cgrp_link_list); | |
547 | kfree(link); | |
548 | } | |
549 | } | |
550 | ||
817929ec PM |
551 | /* |
552 | * allocate_cg_links() allocates "count" cg_cgroup_link structures | |
bd89aabc | 553 | * and chains them on tmp through their cgrp_link_list fields. Returns 0 on |
817929ec PM |
554 | * success or a negative error |
555 | */ | |
817929ec PM |
556 | static int allocate_cg_links(int count, struct list_head *tmp) |
557 | { | |
558 | struct cg_cgroup_link *link; | |
559 | int i; | |
560 | INIT_LIST_HEAD(tmp); | |
561 | for (i = 0; i < count; i++) { | |
562 | link = kmalloc(sizeof(*link), GFP_KERNEL); | |
563 | if (!link) { | |
36553434 | 564 | free_cg_links(tmp); |
817929ec PM |
565 | return -ENOMEM; |
566 | } | |
bd89aabc | 567 | list_add(&link->cgrp_link_list, tmp); |
817929ec PM |
568 | } |
569 | return 0; | |
570 | } | |
571 | ||
c12f65d4 LZ |
572 | /** |
573 | * link_css_set - a helper function to link a css_set to a cgroup | |
574 | * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links() | |
575 | * @cg: the css_set to be linked | |
576 | * @cgrp: the destination cgroup | |
577 | */ | |
578 | static void link_css_set(struct list_head *tmp_cg_links, | |
579 | struct css_set *cg, struct cgroup *cgrp) | |
580 | { | |
581 | struct cg_cgroup_link *link; | |
582 | ||
583 | BUG_ON(list_empty(tmp_cg_links)); | |
584 | link = list_first_entry(tmp_cg_links, struct cg_cgroup_link, | |
585 | cgrp_link_list); | |
586 | link->cg = cg; | |
7717f7ba | 587 | link->cgrp = cgrp; |
2c6ab6d2 | 588 | atomic_inc(&cgrp->count); |
c12f65d4 | 589 | list_move(&link->cgrp_link_list, &cgrp->css_sets); |
7717f7ba PM |
590 | /* |
591 | * Always add links to the tail of the list so that the list | |
592 | * is sorted by order of hierarchy creation | |
593 | */ | |
594 | list_add_tail(&link->cg_link_list, &cg->cg_links); | |
c12f65d4 LZ |
595 | } |
596 | ||
817929ec PM |
597 | /* |
598 | * find_css_set() takes an existing cgroup group and a | |
599 | * cgroup object, and returns a css_set object that's | |
600 | * equivalent to the old group, but with the given cgroup | |
601 | * substituted into the appropriate hierarchy. Must be called with | |
602 | * cgroup_mutex held | |
603 | */ | |
817929ec | 604 | static struct css_set *find_css_set( |
bd89aabc | 605 | struct css_set *oldcg, struct cgroup *cgrp) |
817929ec PM |
606 | { |
607 | struct css_set *res; | |
608 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT]; | |
817929ec PM |
609 | |
610 | struct list_head tmp_cg_links; | |
817929ec | 611 | |
472b1053 | 612 | struct hlist_head *hhead; |
7717f7ba | 613 | struct cg_cgroup_link *link; |
472b1053 | 614 | |
817929ec PM |
615 | /* First see if we already have a cgroup group that matches |
616 | * the desired set */ | |
7e9abd89 | 617 | read_lock(&css_set_lock); |
bd89aabc | 618 | res = find_existing_css_set(oldcg, cgrp, template); |
817929ec PM |
619 | if (res) |
620 | get_css_set(res); | |
7e9abd89 | 621 | read_unlock(&css_set_lock); |
817929ec PM |
622 | |
623 | if (res) | |
624 | return res; | |
625 | ||
626 | res = kmalloc(sizeof(*res), GFP_KERNEL); | |
627 | if (!res) | |
628 | return NULL; | |
629 | ||
630 | /* Allocate all the cg_cgroup_link objects that we'll need */ | |
631 | if (allocate_cg_links(root_count, &tmp_cg_links) < 0) { | |
632 | kfree(res); | |
633 | return NULL; | |
634 | } | |
635 | ||
146aa1bd | 636 | atomic_set(&res->refcount, 1); |
817929ec PM |
637 | INIT_LIST_HEAD(&res->cg_links); |
638 | INIT_LIST_HEAD(&res->tasks); | |
472b1053 | 639 | INIT_HLIST_NODE(&res->hlist); |
817929ec PM |
640 | |
641 | /* Copy the set of subsystem state objects generated in | |
642 | * find_existing_css_set() */ | |
643 | memcpy(res->subsys, template, sizeof(res->subsys)); | |
644 | ||
645 | write_lock(&css_set_lock); | |
646 | /* Add reference counts and links from the new css_set. */ | |
7717f7ba PM |
647 | list_for_each_entry(link, &oldcg->cg_links, cg_link_list) { |
648 | struct cgroup *c = link->cgrp; | |
649 | if (c->root == cgrp->root) | |
650 | c = cgrp; | |
651 | link_css_set(&tmp_cg_links, res, c); | |
652 | } | |
817929ec PM |
653 | |
654 | BUG_ON(!list_empty(&tmp_cg_links)); | |
655 | ||
817929ec | 656 | css_set_count++; |
472b1053 LZ |
657 | |
658 | /* Add this cgroup group to the hash table */ | |
659 | hhead = css_set_hash(res->subsys); | |
660 | hlist_add_head(&res->hlist, hhead); | |
661 | ||
817929ec PM |
662 | write_unlock(&css_set_lock); |
663 | ||
664 | return res; | |
b4f48b63 PM |
665 | } |
666 | ||
7717f7ba PM |
667 | /* |
668 | * Return the cgroup for "task" from the given hierarchy. Must be | |
669 | * called with cgroup_mutex held. | |
670 | */ | |
671 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
672 | struct cgroupfs_root *root) | |
673 | { | |
674 | struct css_set *css; | |
675 | struct cgroup *res = NULL; | |
676 | ||
677 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
678 | read_lock(&css_set_lock); | |
679 | /* | |
680 | * No need to lock the task - since we hold cgroup_mutex the | |
681 | * task can't change groups, so the only thing that can happen | |
682 | * is that it exits and its css is set back to init_css_set. | |
683 | */ | |
684 | css = task->cgroups; | |
685 | if (css == &init_css_set) { | |
686 | res = &root->top_cgroup; | |
687 | } else { | |
688 | struct cg_cgroup_link *link; | |
689 | list_for_each_entry(link, &css->cg_links, cg_link_list) { | |
690 | struct cgroup *c = link->cgrp; | |
691 | if (c->root == root) { | |
692 | res = c; | |
693 | break; | |
694 | } | |
695 | } | |
696 | } | |
697 | read_unlock(&css_set_lock); | |
698 | BUG_ON(!res); | |
699 | return res; | |
700 | } | |
701 | ||
ddbcc7e8 PM |
702 | /* |
703 | * There is one global cgroup mutex. We also require taking | |
704 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
705 | * See "The task_lock() exception", at the end of this comment. | |
706 | * | |
707 | * A task must hold cgroup_mutex to modify cgroups. | |
708 | * | |
709 | * Any task can increment and decrement the count field without lock. | |
710 | * So in general, code holding cgroup_mutex can't rely on the count | |
711 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 712 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
713 | * means that no tasks are currently attached, therefore there is no |
714 | * way a task attached to that cgroup can fork (the other way to | |
715 | * increment the count). So code holding cgroup_mutex can safely | |
716 | * assume that if the count is zero, it will stay zero. Similarly, if | |
717 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
718 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
719 | * needs that mutex. | |
720 | * | |
ddbcc7e8 PM |
721 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
722 | * (usually) take cgroup_mutex. These are the two most performance | |
723 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
724 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
725 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
726 | * to the release agent with the name of the cgroup (path relative to |
727 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
728 | * |
729 | * A cgroup can only be deleted if both its 'count' of using tasks | |
730 | * is zero, and its list of 'children' cgroups is empty. Since all | |
731 | * tasks in the system use _some_ cgroup, and since there is always at | |
732 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
733 | * always has either children cgroups and/or using tasks. So we don't | |
734 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
735 | * | |
736 | * The task_lock() exception | |
737 | * | |
738 | * The need for this exception arises from the action of | |
956db3ca | 739 | * cgroup_attach_task(), which overwrites one tasks cgroup pointer with |
a043e3b2 | 740 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
741 | * several performance critical places that need to reference |
742 | * task->cgroup without the expense of grabbing a system global | |
743 | * mutex. Therefore except as noted below, when dereferencing or, as | |
956db3ca | 744 | * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use |
ddbcc7e8 PM |
745 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
746 | * the task_struct routinely used for such matters. | |
747 | * | |
748 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 749 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
750 | */ |
751 | ||
ddbcc7e8 PM |
752 | /** |
753 | * cgroup_lock - lock out any changes to cgroup structures | |
754 | * | |
755 | */ | |
ddbcc7e8 PM |
756 | void cgroup_lock(void) |
757 | { | |
758 | mutex_lock(&cgroup_mutex); | |
759 | } | |
67523c48 | 760 | EXPORT_SYMBOL_GPL(cgroup_lock); |
ddbcc7e8 PM |
761 | |
762 | /** | |
763 | * cgroup_unlock - release lock on cgroup changes | |
764 | * | |
765 | * Undo the lock taken in a previous cgroup_lock() call. | |
766 | */ | |
ddbcc7e8 PM |
767 | void cgroup_unlock(void) |
768 | { | |
769 | mutex_unlock(&cgroup_mutex); | |
770 | } | |
67523c48 | 771 | EXPORT_SYMBOL_GPL(cgroup_unlock); |
ddbcc7e8 PM |
772 | |
773 | /* | |
774 | * A couple of forward declarations required, due to cyclic reference loop: | |
775 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> | |
776 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations | |
777 | * -> cgroup_mkdir. | |
778 | */ | |
779 | ||
18bb1db3 | 780 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode); |
c72a04e3 | 781 | static struct dentry *cgroup_lookup(struct inode *, struct dentry *, struct nameidata *); |
ddbcc7e8 | 782 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); |
bd89aabc | 783 | static int cgroup_populate_dir(struct cgroup *cgrp); |
6e1d5dcc | 784 | static const struct inode_operations cgroup_dir_inode_operations; |
828c0950 | 785 | static const struct file_operations proc_cgroupstats_operations; |
a424316c PM |
786 | |
787 | static struct backing_dev_info cgroup_backing_dev_info = { | |
d993831f | 788 | .name = "cgroup", |
e4ad08fe | 789 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
a424316c | 790 | }; |
ddbcc7e8 | 791 | |
38460b48 KH |
792 | static int alloc_css_id(struct cgroup_subsys *ss, |
793 | struct cgroup *parent, struct cgroup *child); | |
794 | ||
a5e7ed32 | 795 | static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb) |
ddbcc7e8 PM |
796 | { |
797 | struct inode *inode = new_inode(sb); | |
ddbcc7e8 PM |
798 | |
799 | if (inode) { | |
85fe4025 | 800 | inode->i_ino = get_next_ino(); |
ddbcc7e8 | 801 | inode->i_mode = mode; |
76aac0e9 DH |
802 | inode->i_uid = current_fsuid(); |
803 | inode->i_gid = current_fsgid(); | |
ddbcc7e8 PM |
804 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
805 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; | |
806 | } | |
807 | return inode; | |
808 | } | |
809 | ||
4fca88c8 KH |
810 | /* |
811 | * Call subsys's pre_destroy handler. | |
812 | * This is called before css refcnt check. | |
813 | */ | |
ec64f515 | 814 | static int cgroup_call_pre_destroy(struct cgroup *cgrp) |
4fca88c8 KH |
815 | { |
816 | struct cgroup_subsys *ss; | |
ec64f515 KH |
817 | int ret = 0; |
818 | ||
4fca88c8 | 819 | for_each_subsys(cgrp->root, ss) |
ec64f515 | 820 | if (ss->pre_destroy) { |
761b3ef5 | 821 | ret = ss->pre_destroy(cgrp); |
ec64f515 | 822 | if (ret) |
4ab78683 | 823 | break; |
ec64f515 | 824 | } |
0dea1168 | 825 | |
ec64f515 | 826 | return ret; |
4fca88c8 KH |
827 | } |
828 | ||
ddbcc7e8 PM |
829 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) |
830 | { | |
831 | /* is dentry a directory ? if so, kfree() associated cgroup */ | |
832 | if (S_ISDIR(inode->i_mode)) { | |
bd89aabc | 833 | struct cgroup *cgrp = dentry->d_fsdata; |
8dc4f3e1 | 834 | struct cgroup_subsys *ss; |
bd89aabc | 835 | BUG_ON(!(cgroup_is_removed(cgrp))); |
81a6a5cd PM |
836 | /* It's possible for external users to be holding css |
837 | * reference counts on a cgroup; css_put() needs to | |
838 | * be able to access the cgroup after decrementing | |
839 | * the reference count in order to know if it needs to | |
840 | * queue the cgroup to be handled by the release | |
841 | * agent */ | |
842 | synchronize_rcu(); | |
8dc4f3e1 PM |
843 | |
844 | mutex_lock(&cgroup_mutex); | |
845 | /* | |
846 | * Release the subsystem state objects. | |
847 | */ | |
75139b82 | 848 | for_each_subsys(cgrp->root, ss) |
761b3ef5 | 849 | ss->destroy(cgrp); |
8dc4f3e1 PM |
850 | |
851 | cgrp->root->number_of_cgroups--; | |
852 | mutex_unlock(&cgroup_mutex); | |
853 | ||
a47295e6 PM |
854 | /* |
855 | * Drop the active superblock reference that we took when we | |
856 | * created the cgroup | |
857 | */ | |
8dc4f3e1 PM |
858 | deactivate_super(cgrp->root->sb); |
859 | ||
72a8cb30 BB |
860 | /* |
861 | * if we're getting rid of the cgroup, refcount should ensure | |
862 | * that there are no pidlists left. | |
863 | */ | |
864 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
865 | ||
f2da1c40 | 866 | kfree_rcu(cgrp, rcu_head); |
ddbcc7e8 PM |
867 | } |
868 | iput(inode); | |
869 | } | |
870 | ||
c72a04e3 AV |
871 | static int cgroup_delete(const struct dentry *d) |
872 | { | |
873 | return 1; | |
874 | } | |
875 | ||
ddbcc7e8 PM |
876 | static void remove_dir(struct dentry *d) |
877 | { | |
878 | struct dentry *parent = dget(d->d_parent); | |
879 | ||
880 | d_delete(d); | |
881 | simple_rmdir(parent->d_inode, d); | |
882 | dput(parent); | |
883 | } | |
884 | ||
885 | static void cgroup_clear_directory(struct dentry *dentry) | |
886 | { | |
887 | struct list_head *node; | |
888 | ||
889 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
2fd6b7f5 | 890 | spin_lock(&dentry->d_lock); |
ddbcc7e8 PM |
891 | node = dentry->d_subdirs.next; |
892 | while (node != &dentry->d_subdirs) { | |
893 | struct dentry *d = list_entry(node, struct dentry, d_u.d_child); | |
2fd6b7f5 NP |
894 | |
895 | spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED); | |
ddbcc7e8 PM |
896 | list_del_init(node); |
897 | if (d->d_inode) { | |
898 | /* This should never be called on a cgroup | |
899 | * directory with child cgroups */ | |
900 | BUG_ON(d->d_inode->i_mode & S_IFDIR); | |
dc0474be | 901 | dget_dlock(d); |
2fd6b7f5 NP |
902 | spin_unlock(&d->d_lock); |
903 | spin_unlock(&dentry->d_lock); | |
ddbcc7e8 PM |
904 | d_delete(d); |
905 | simple_unlink(dentry->d_inode, d); | |
906 | dput(d); | |
2fd6b7f5 NP |
907 | spin_lock(&dentry->d_lock); |
908 | } else | |
909 | spin_unlock(&d->d_lock); | |
ddbcc7e8 PM |
910 | node = dentry->d_subdirs.next; |
911 | } | |
2fd6b7f5 | 912 | spin_unlock(&dentry->d_lock); |
ddbcc7e8 PM |
913 | } |
914 | ||
915 | /* | |
916 | * NOTE : the dentry must have been dget()'ed | |
917 | */ | |
918 | static void cgroup_d_remove_dir(struct dentry *dentry) | |
919 | { | |
2fd6b7f5 NP |
920 | struct dentry *parent; |
921 | ||
ddbcc7e8 PM |
922 | cgroup_clear_directory(dentry); |
923 | ||
2fd6b7f5 NP |
924 | parent = dentry->d_parent; |
925 | spin_lock(&parent->d_lock); | |
3ec762ad | 926 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
ddbcc7e8 | 927 | list_del_init(&dentry->d_u.d_child); |
2fd6b7f5 NP |
928 | spin_unlock(&dentry->d_lock); |
929 | spin_unlock(&parent->d_lock); | |
ddbcc7e8 PM |
930 | remove_dir(dentry); |
931 | } | |
932 | ||
ec64f515 KH |
933 | /* |
934 | * A queue for waiters to do rmdir() cgroup. A tasks will sleep when | |
935 | * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some | |
936 | * reference to css->refcnt. In general, this refcnt is expected to goes down | |
937 | * to zero, soon. | |
938 | * | |
88703267 | 939 | * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex; |
ec64f515 | 940 | */ |
1c6c3fad | 941 | static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq); |
ec64f515 | 942 | |
88703267 | 943 | static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp) |
ec64f515 | 944 | { |
88703267 | 945 | if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))) |
ec64f515 KH |
946 | wake_up_all(&cgroup_rmdir_waitq); |
947 | } | |
948 | ||
88703267 KH |
949 | void cgroup_exclude_rmdir(struct cgroup_subsys_state *css) |
950 | { | |
951 | css_get(css); | |
952 | } | |
953 | ||
954 | void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css) | |
955 | { | |
956 | cgroup_wakeup_rmdir_waiter(css->cgroup); | |
957 | css_put(css); | |
958 | } | |
959 | ||
aae8aab4 | 960 | /* |
cf5d5941 BB |
961 | * Call with cgroup_mutex held. Drops reference counts on modules, including |
962 | * any duplicate ones that parse_cgroupfs_options took. If this function | |
963 | * returns an error, no reference counts are touched. | |
aae8aab4 | 964 | */ |
ddbcc7e8 PM |
965 | static int rebind_subsystems(struct cgroupfs_root *root, |
966 | unsigned long final_bits) | |
967 | { | |
968 | unsigned long added_bits, removed_bits; | |
bd89aabc | 969 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
970 | int i; |
971 | ||
aae8aab4 | 972 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
e25e2cbb | 973 | BUG_ON(!mutex_is_locked(&cgroup_root_mutex)); |
aae8aab4 | 974 | |
ddbcc7e8 PM |
975 | removed_bits = root->actual_subsys_bits & ~final_bits; |
976 | added_bits = final_bits & ~root->actual_subsys_bits; | |
977 | /* Check that any added subsystems are currently free */ | |
978 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
8d53d55d | 979 | unsigned long bit = 1UL << i; |
ddbcc7e8 PM |
980 | struct cgroup_subsys *ss = subsys[i]; |
981 | if (!(bit & added_bits)) | |
982 | continue; | |
aae8aab4 BB |
983 | /* |
984 | * Nobody should tell us to do a subsys that doesn't exist: | |
985 | * parse_cgroupfs_options should catch that case and refcounts | |
986 | * ensure that subsystems won't disappear once selected. | |
987 | */ | |
988 | BUG_ON(ss == NULL); | |
ddbcc7e8 PM |
989 | if (ss->root != &rootnode) { |
990 | /* Subsystem isn't free */ | |
991 | return -EBUSY; | |
992 | } | |
993 | } | |
994 | ||
995 | /* Currently we don't handle adding/removing subsystems when | |
996 | * any child cgroups exist. This is theoretically supportable | |
997 | * but involves complex error handling, so it's being left until | |
998 | * later */ | |
307257cf | 999 | if (root->number_of_cgroups > 1) |
ddbcc7e8 PM |
1000 | return -EBUSY; |
1001 | ||
1002 | /* Process each subsystem */ | |
1003 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1004 | struct cgroup_subsys *ss = subsys[i]; | |
1005 | unsigned long bit = 1UL << i; | |
1006 | if (bit & added_bits) { | |
1007 | /* We're binding this subsystem to this hierarchy */ | |
aae8aab4 | 1008 | BUG_ON(ss == NULL); |
bd89aabc | 1009 | BUG_ON(cgrp->subsys[i]); |
ddbcc7e8 PM |
1010 | BUG_ON(!dummytop->subsys[i]); |
1011 | BUG_ON(dummytop->subsys[i]->cgroup != dummytop); | |
999cd8a4 | 1012 | mutex_lock(&ss->hierarchy_mutex); |
bd89aabc PM |
1013 | cgrp->subsys[i] = dummytop->subsys[i]; |
1014 | cgrp->subsys[i]->cgroup = cgrp; | |
33a68ac1 | 1015 | list_move(&ss->sibling, &root->subsys_list); |
b2aa30f7 | 1016 | ss->root = root; |
ddbcc7e8 | 1017 | if (ss->bind) |
761b3ef5 | 1018 | ss->bind(cgrp); |
999cd8a4 | 1019 | mutex_unlock(&ss->hierarchy_mutex); |
cf5d5941 | 1020 | /* refcount was already taken, and we're keeping it */ |
ddbcc7e8 PM |
1021 | } else if (bit & removed_bits) { |
1022 | /* We're removing this subsystem */ | |
aae8aab4 | 1023 | BUG_ON(ss == NULL); |
bd89aabc PM |
1024 | BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]); |
1025 | BUG_ON(cgrp->subsys[i]->cgroup != cgrp); | |
999cd8a4 | 1026 | mutex_lock(&ss->hierarchy_mutex); |
ddbcc7e8 | 1027 | if (ss->bind) |
761b3ef5 | 1028 | ss->bind(dummytop); |
ddbcc7e8 | 1029 | dummytop->subsys[i]->cgroup = dummytop; |
bd89aabc | 1030 | cgrp->subsys[i] = NULL; |
b2aa30f7 | 1031 | subsys[i]->root = &rootnode; |
33a68ac1 | 1032 | list_move(&ss->sibling, &rootnode.subsys_list); |
999cd8a4 | 1033 | mutex_unlock(&ss->hierarchy_mutex); |
cf5d5941 BB |
1034 | /* subsystem is now free - drop reference on module */ |
1035 | module_put(ss->module); | |
ddbcc7e8 PM |
1036 | } else if (bit & final_bits) { |
1037 | /* Subsystem state should already exist */ | |
aae8aab4 | 1038 | BUG_ON(ss == NULL); |
bd89aabc | 1039 | BUG_ON(!cgrp->subsys[i]); |
cf5d5941 BB |
1040 | /* |
1041 | * a refcount was taken, but we already had one, so | |
1042 | * drop the extra reference. | |
1043 | */ | |
1044 | module_put(ss->module); | |
1045 | #ifdef CONFIG_MODULE_UNLOAD | |
1046 | BUG_ON(ss->module && !module_refcount(ss->module)); | |
1047 | #endif | |
ddbcc7e8 PM |
1048 | } else { |
1049 | /* Subsystem state shouldn't exist */ | |
bd89aabc | 1050 | BUG_ON(cgrp->subsys[i]); |
ddbcc7e8 PM |
1051 | } |
1052 | } | |
1053 | root->subsys_bits = root->actual_subsys_bits = final_bits; | |
1054 | synchronize_rcu(); | |
1055 | ||
1056 | return 0; | |
1057 | } | |
1058 | ||
34c80b1d | 1059 | static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry) |
ddbcc7e8 | 1060 | { |
34c80b1d | 1061 | struct cgroupfs_root *root = dentry->d_sb->s_fs_info; |
ddbcc7e8 PM |
1062 | struct cgroup_subsys *ss; |
1063 | ||
e25e2cbb | 1064 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1065 | for_each_subsys(root, ss) |
1066 | seq_printf(seq, ",%s", ss->name); | |
1067 | if (test_bit(ROOT_NOPREFIX, &root->flags)) | |
1068 | seq_puts(seq, ",noprefix"); | |
81a6a5cd PM |
1069 | if (strlen(root->release_agent_path)) |
1070 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
97978e6d DL |
1071 | if (clone_children(&root->top_cgroup)) |
1072 | seq_puts(seq, ",clone_children"); | |
c6d57f33 PM |
1073 | if (strlen(root->name)) |
1074 | seq_printf(seq, ",name=%s", root->name); | |
e25e2cbb | 1075 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1076 | return 0; |
1077 | } | |
1078 | ||
1079 | struct cgroup_sb_opts { | |
1080 | unsigned long subsys_bits; | |
1081 | unsigned long flags; | |
81a6a5cd | 1082 | char *release_agent; |
97978e6d | 1083 | bool clone_children; |
c6d57f33 | 1084 | char *name; |
2c6ab6d2 PM |
1085 | /* User explicitly requested empty subsystem */ |
1086 | bool none; | |
c6d57f33 PM |
1087 | |
1088 | struct cgroupfs_root *new_root; | |
2c6ab6d2 | 1089 | |
ddbcc7e8 PM |
1090 | }; |
1091 | ||
aae8aab4 BB |
1092 | /* |
1093 | * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call | |
cf5d5941 BB |
1094 | * with cgroup_mutex held to protect the subsys[] array. This function takes |
1095 | * refcounts on subsystems to be used, unless it returns error, in which case | |
1096 | * no refcounts are taken. | |
aae8aab4 | 1097 | */ |
cf5d5941 | 1098 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1099 | { |
32a8cf23 DL |
1100 | char *token, *o = data; |
1101 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1102 | unsigned long mask = (unsigned long)-1; |
cf5d5941 BB |
1103 | int i; |
1104 | bool module_pin_failed = false; | |
f9ab5b5b | 1105 | |
aae8aab4 BB |
1106 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1107 | ||
f9ab5b5b LZ |
1108 | #ifdef CONFIG_CPUSETS |
1109 | mask = ~(1UL << cpuset_subsys_id); | |
1110 | #endif | |
ddbcc7e8 | 1111 | |
c6d57f33 | 1112 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1113 | |
1114 | while ((token = strsep(&o, ",")) != NULL) { | |
1115 | if (!*token) | |
1116 | return -EINVAL; | |
32a8cf23 | 1117 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1118 | /* Explicitly have no subsystems */ |
1119 | opts->none = true; | |
32a8cf23 DL |
1120 | continue; |
1121 | } | |
1122 | if (!strcmp(token, "all")) { | |
1123 | /* Mutually exclusive option 'all' + subsystem name */ | |
1124 | if (one_ss) | |
1125 | return -EINVAL; | |
1126 | all_ss = true; | |
1127 | continue; | |
1128 | } | |
1129 | if (!strcmp(token, "noprefix")) { | |
ddbcc7e8 | 1130 | set_bit(ROOT_NOPREFIX, &opts->flags); |
32a8cf23 DL |
1131 | continue; |
1132 | } | |
1133 | if (!strcmp(token, "clone_children")) { | |
97978e6d | 1134 | opts->clone_children = true; |
32a8cf23 DL |
1135 | continue; |
1136 | } | |
1137 | if (!strncmp(token, "release_agent=", 14)) { | |
81a6a5cd PM |
1138 | /* Specifying two release agents is forbidden */ |
1139 | if (opts->release_agent) | |
1140 | return -EINVAL; | |
c6d57f33 | 1141 | opts->release_agent = |
e400c285 | 1142 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1143 | if (!opts->release_agent) |
1144 | return -ENOMEM; | |
32a8cf23 DL |
1145 | continue; |
1146 | } | |
1147 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1148 | const char *name = token + 5; |
1149 | /* Can't specify an empty name */ | |
1150 | if (!strlen(name)) | |
1151 | return -EINVAL; | |
1152 | /* Must match [\w.-]+ */ | |
1153 | for (i = 0; i < strlen(name); i++) { | |
1154 | char c = name[i]; | |
1155 | if (isalnum(c)) | |
1156 | continue; | |
1157 | if ((c == '.') || (c == '-') || (c == '_')) | |
1158 | continue; | |
1159 | return -EINVAL; | |
1160 | } | |
1161 | /* Specifying two names is forbidden */ | |
1162 | if (opts->name) | |
1163 | return -EINVAL; | |
1164 | opts->name = kstrndup(name, | |
e400c285 | 1165 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1166 | GFP_KERNEL); |
1167 | if (!opts->name) | |
1168 | return -ENOMEM; | |
32a8cf23 DL |
1169 | |
1170 | continue; | |
1171 | } | |
1172 | ||
1173 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1174 | struct cgroup_subsys *ss = subsys[i]; | |
1175 | if (ss == NULL) | |
1176 | continue; | |
1177 | if (strcmp(token, ss->name)) | |
1178 | continue; | |
1179 | if (ss->disabled) | |
1180 | continue; | |
1181 | ||
1182 | /* Mutually exclusive option 'all' + subsystem name */ | |
1183 | if (all_ss) | |
1184 | return -EINVAL; | |
1185 | set_bit(i, &opts->subsys_bits); | |
1186 | one_ss = true; | |
1187 | ||
1188 | break; | |
1189 | } | |
1190 | if (i == CGROUP_SUBSYS_COUNT) | |
1191 | return -ENOENT; | |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1196 | * otherwise if 'none', 'name=' and a subsystem name options |
1197 | * were not specified, let's default to 'all' | |
32a8cf23 | 1198 | */ |
0d19ea86 | 1199 | if (all_ss || (!one_ss && !opts->none && !opts->name)) { |
32a8cf23 DL |
1200 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
1201 | struct cgroup_subsys *ss = subsys[i]; | |
1202 | if (ss == NULL) | |
1203 | continue; | |
1204 | if (ss->disabled) | |
1205 | continue; | |
1206 | set_bit(i, &opts->subsys_bits); | |
ddbcc7e8 PM |
1207 | } |
1208 | } | |
1209 | ||
2c6ab6d2 PM |
1210 | /* Consistency checks */ |
1211 | ||
f9ab5b5b LZ |
1212 | /* |
1213 | * Option noprefix was introduced just for backward compatibility | |
1214 | * with the old cpuset, so we allow noprefix only if mounting just | |
1215 | * the cpuset subsystem. | |
1216 | */ | |
1217 | if (test_bit(ROOT_NOPREFIX, &opts->flags) && | |
1218 | (opts->subsys_bits & mask)) | |
1219 | return -EINVAL; | |
1220 | ||
2c6ab6d2 PM |
1221 | |
1222 | /* Can't specify "none" and some subsystems */ | |
1223 | if (opts->subsys_bits && opts->none) | |
1224 | return -EINVAL; | |
1225 | ||
1226 | /* | |
1227 | * We either have to specify by name or by subsystems. (So all | |
1228 | * empty hierarchies must have a name). | |
1229 | */ | |
c6d57f33 | 1230 | if (!opts->subsys_bits && !opts->name) |
ddbcc7e8 PM |
1231 | return -EINVAL; |
1232 | ||
cf5d5941 BB |
1233 | /* |
1234 | * Grab references on all the modules we'll need, so the subsystems | |
1235 | * don't dance around before rebind_subsystems attaches them. This may | |
1236 | * take duplicate reference counts on a subsystem that's already used, | |
1237 | * but rebind_subsystems handles this case. | |
1238 | */ | |
1239 | for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) { | |
1240 | unsigned long bit = 1UL << i; | |
1241 | ||
1242 | if (!(bit & opts->subsys_bits)) | |
1243 | continue; | |
1244 | if (!try_module_get(subsys[i]->module)) { | |
1245 | module_pin_failed = true; | |
1246 | break; | |
1247 | } | |
1248 | } | |
1249 | if (module_pin_failed) { | |
1250 | /* | |
1251 | * oops, one of the modules was going away. this means that we | |
1252 | * raced with a module_delete call, and to the user this is | |
1253 | * essentially a "subsystem doesn't exist" case. | |
1254 | */ | |
1255 | for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) { | |
1256 | /* drop refcounts only on the ones we took */ | |
1257 | unsigned long bit = 1UL << i; | |
1258 | ||
1259 | if (!(bit & opts->subsys_bits)) | |
1260 | continue; | |
1261 | module_put(subsys[i]->module); | |
1262 | } | |
1263 | return -ENOENT; | |
1264 | } | |
1265 | ||
ddbcc7e8 PM |
1266 | return 0; |
1267 | } | |
1268 | ||
cf5d5941 BB |
1269 | static void drop_parsed_module_refcounts(unsigned long subsys_bits) |
1270 | { | |
1271 | int i; | |
1272 | for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) { | |
1273 | unsigned long bit = 1UL << i; | |
1274 | ||
1275 | if (!(bit & subsys_bits)) | |
1276 | continue; | |
1277 | module_put(subsys[i]->module); | |
1278 | } | |
1279 | } | |
1280 | ||
ddbcc7e8 PM |
1281 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) |
1282 | { | |
1283 | int ret = 0; | |
1284 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1285 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
1286 | struct cgroup_sb_opts opts; |
1287 | ||
bd89aabc | 1288 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 | 1289 | mutex_lock(&cgroup_mutex); |
e25e2cbb | 1290 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1291 | |
1292 | /* See what subsystems are wanted */ | |
1293 | ret = parse_cgroupfs_options(data, &opts); | |
1294 | if (ret) | |
1295 | goto out_unlock; | |
1296 | ||
cf5d5941 BB |
1297 | /* Don't allow flags or name to change at remount */ |
1298 | if (opts.flags != root->flags || | |
1299 | (opts.name && strcmp(opts.name, root->name))) { | |
c6d57f33 | 1300 | ret = -EINVAL; |
cf5d5941 | 1301 | drop_parsed_module_refcounts(opts.subsys_bits); |
c6d57f33 PM |
1302 | goto out_unlock; |
1303 | } | |
1304 | ||
ddbcc7e8 | 1305 | ret = rebind_subsystems(root, opts.subsys_bits); |
cf5d5941 BB |
1306 | if (ret) { |
1307 | drop_parsed_module_refcounts(opts.subsys_bits); | |
0670e08b | 1308 | goto out_unlock; |
cf5d5941 | 1309 | } |
ddbcc7e8 PM |
1310 | |
1311 | /* (re)populate subsystem files */ | |
0670e08b | 1312 | cgroup_populate_dir(cgrp); |
ddbcc7e8 | 1313 | |
81a6a5cd PM |
1314 | if (opts.release_agent) |
1315 | strcpy(root->release_agent_path, opts.release_agent); | |
ddbcc7e8 | 1316 | out_unlock: |
66bdc9cf | 1317 | kfree(opts.release_agent); |
c6d57f33 | 1318 | kfree(opts.name); |
e25e2cbb | 1319 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 | 1320 | mutex_unlock(&cgroup_mutex); |
bd89aabc | 1321 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
1322 | return ret; |
1323 | } | |
1324 | ||
b87221de | 1325 | static const struct super_operations cgroup_ops = { |
ddbcc7e8 PM |
1326 | .statfs = simple_statfs, |
1327 | .drop_inode = generic_delete_inode, | |
1328 | .show_options = cgroup_show_options, | |
1329 | .remount_fs = cgroup_remount, | |
1330 | }; | |
1331 | ||
cc31edce PM |
1332 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1333 | { | |
1334 | INIT_LIST_HEAD(&cgrp->sibling); | |
1335 | INIT_LIST_HEAD(&cgrp->children); | |
1336 | INIT_LIST_HEAD(&cgrp->css_sets); | |
1337 | INIT_LIST_HEAD(&cgrp->release_list); | |
72a8cb30 BB |
1338 | INIT_LIST_HEAD(&cgrp->pidlists); |
1339 | mutex_init(&cgrp->pidlist_mutex); | |
0dea1168 KS |
1340 | INIT_LIST_HEAD(&cgrp->event_list); |
1341 | spin_lock_init(&cgrp->event_list_lock); | |
cc31edce | 1342 | } |
c6d57f33 | 1343 | |
ddbcc7e8 PM |
1344 | static void init_cgroup_root(struct cgroupfs_root *root) |
1345 | { | |
bd89aabc | 1346 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
1347 | INIT_LIST_HEAD(&root->subsys_list); |
1348 | INIT_LIST_HEAD(&root->root_list); | |
1349 | root->number_of_cgroups = 1; | |
bd89aabc PM |
1350 | cgrp->root = root; |
1351 | cgrp->top_cgroup = cgrp; | |
cc31edce | 1352 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 PM |
1353 | } |
1354 | ||
2c6ab6d2 PM |
1355 | static bool init_root_id(struct cgroupfs_root *root) |
1356 | { | |
1357 | int ret = 0; | |
1358 | ||
1359 | do { | |
1360 | if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL)) | |
1361 | return false; | |
1362 | spin_lock(&hierarchy_id_lock); | |
1363 | /* Try to allocate the next unused ID */ | |
1364 | ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id, | |
1365 | &root->hierarchy_id); | |
1366 | if (ret == -ENOSPC) | |
1367 | /* Try again starting from 0 */ | |
1368 | ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id); | |
1369 | if (!ret) { | |
1370 | next_hierarchy_id = root->hierarchy_id + 1; | |
1371 | } else if (ret != -EAGAIN) { | |
1372 | /* Can only get here if the 31-bit IDR is full ... */ | |
1373 | BUG_ON(ret); | |
1374 | } | |
1375 | spin_unlock(&hierarchy_id_lock); | |
1376 | } while (ret); | |
1377 | return true; | |
1378 | } | |
1379 | ||
ddbcc7e8 PM |
1380 | static int cgroup_test_super(struct super_block *sb, void *data) |
1381 | { | |
c6d57f33 | 1382 | struct cgroup_sb_opts *opts = data; |
ddbcc7e8 PM |
1383 | struct cgroupfs_root *root = sb->s_fs_info; |
1384 | ||
c6d57f33 PM |
1385 | /* If we asked for a name then it must match */ |
1386 | if (opts->name && strcmp(opts->name, root->name)) | |
1387 | return 0; | |
ddbcc7e8 | 1388 | |
2c6ab6d2 PM |
1389 | /* |
1390 | * If we asked for subsystems (or explicitly for no | |
1391 | * subsystems) then they must match | |
1392 | */ | |
1393 | if ((opts->subsys_bits || opts->none) | |
1394 | && (opts->subsys_bits != root->subsys_bits)) | |
ddbcc7e8 PM |
1395 | return 0; |
1396 | ||
1397 | return 1; | |
1398 | } | |
1399 | ||
c6d57f33 PM |
1400 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1401 | { | |
1402 | struct cgroupfs_root *root; | |
1403 | ||
2c6ab6d2 | 1404 | if (!opts->subsys_bits && !opts->none) |
c6d57f33 PM |
1405 | return NULL; |
1406 | ||
1407 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1408 | if (!root) | |
1409 | return ERR_PTR(-ENOMEM); | |
1410 | ||
2c6ab6d2 PM |
1411 | if (!init_root_id(root)) { |
1412 | kfree(root); | |
1413 | return ERR_PTR(-ENOMEM); | |
1414 | } | |
c6d57f33 | 1415 | init_cgroup_root(root); |
2c6ab6d2 | 1416 | |
c6d57f33 PM |
1417 | root->subsys_bits = opts->subsys_bits; |
1418 | root->flags = opts->flags; | |
1419 | if (opts->release_agent) | |
1420 | strcpy(root->release_agent_path, opts->release_agent); | |
1421 | if (opts->name) | |
1422 | strcpy(root->name, opts->name); | |
97978e6d DL |
1423 | if (opts->clone_children) |
1424 | set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1425 | return root; |
1426 | } | |
1427 | ||
2c6ab6d2 PM |
1428 | static void cgroup_drop_root(struct cgroupfs_root *root) |
1429 | { | |
1430 | if (!root) | |
1431 | return; | |
1432 | ||
1433 | BUG_ON(!root->hierarchy_id); | |
1434 | spin_lock(&hierarchy_id_lock); | |
1435 | ida_remove(&hierarchy_ida, root->hierarchy_id); | |
1436 | spin_unlock(&hierarchy_id_lock); | |
1437 | kfree(root); | |
1438 | } | |
1439 | ||
ddbcc7e8 PM |
1440 | static int cgroup_set_super(struct super_block *sb, void *data) |
1441 | { | |
1442 | int ret; | |
c6d57f33 PM |
1443 | struct cgroup_sb_opts *opts = data; |
1444 | ||
1445 | /* If we don't have a new root, we can't set up a new sb */ | |
1446 | if (!opts->new_root) | |
1447 | return -EINVAL; | |
1448 | ||
2c6ab6d2 | 1449 | BUG_ON(!opts->subsys_bits && !opts->none); |
ddbcc7e8 PM |
1450 | |
1451 | ret = set_anon_super(sb, NULL); | |
1452 | if (ret) | |
1453 | return ret; | |
1454 | ||
c6d57f33 PM |
1455 | sb->s_fs_info = opts->new_root; |
1456 | opts->new_root->sb = sb; | |
ddbcc7e8 PM |
1457 | |
1458 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
1459 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
1460 | sb->s_magic = CGROUP_SUPER_MAGIC; | |
1461 | sb->s_op = &cgroup_ops; | |
1462 | ||
1463 | return 0; | |
1464 | } | |
1465 | ||
1466 | static int cgroup_get_rootdir(struct super_block *sb) | |
1467 | { | |
0df6a63f AV |
1468 | static const struct dentry_operations cgroup_dops = { |
1469 | .d_iput = cgroup_diput, | |
c72a04e3 | 1470 | .d_delete = cgroup_delete, |
0df6a63f AV |
1471 | }; |
1472 | ||
ddbcc7e8 PM |
1473 | struct inode *inode = |
1474 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); | |
ddbcc7e8 PM |
1475 | |
1476 | if (!inode) | |
1477 | return -ENOMEM; | |
1478 | ||
ddbcc7e8 PM |
1479 | inode->i_fop = &simple_dir_operations; |
1480 | inode->i_op = &cgroup_dir_inode_operations; | |
1481 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
1482 | inc_nlink(inode); | |
48fde701 AV |
1483 | sb->s_root = d_make_root(inode); |
1484 | if (!sb->s_root) | |
ddbcc7e8 | 1485 | return -ENOMEM; |
0df6a63f AV |
1486 | /* for everything else we want ->d_op set */ |
1487 | sb->s_d_op = &cgroup_dops; | |
ddbcc7e8 PM |
1488 | return 0; |
1489 | } | |
1490 | ||
f7e83571 | 1491 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1492 | int flags, const char *unused_dev_name, |
f7e83571 | 1493 | void *data) |
ddbcc7e8 PM |
1494 | { |
1495 | struct cgroup_sb_opts opts; | |
c6d57f33 | 1496 | struct cgroupfs_root *root; |
ddbcc7e8 PM |
1497 | int ret = 0; |
1498 | struct super_block *sb; | |
c6d57f33 | 1499 | struct cgroupfs_root *new_root; |
e25e2cbb | 1500 | struct inode *inode; |
ddbcc7e8 PM |
1501 | |
1502 | /* First find the desired set of subsystems */ | |
aae8aab4 | 1503 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1504 | ret = parse_cgroupfs_options(data, &opts); |
aae8aab4 | 1505 | mutex_unlock(&cgroup_mutex); |
c6d57f33 PM |
1506 | if (ret) |
1507 | goto out_err; | |
ddbcc7e8 | 1508 | |
c6d57f33 PM |
1509 | /* |
1510 | * Allocate a new cgroup root. We may not need it if we're | |
1511 | * reusing an existing hierarchy. | |
1512 | */ | |
1513 | new_root = cgroup_root_from_opts(&opts); | |
1514 | if (IS_ERR(new_root)) { | |
1515 | ret = PTR_ERR(new_root); | |
cf5d5941 | 1516 | goto drop_modules; |
81a6a5cd | 1517 | } |
c6d57f33 | 1518 | opts.new_root = new_root; |
ddbcc7e8 | 1519 | |
c6d57f33 PM |
1520 | /* Locate an existing or new sb for this hierarchy */ |
1521 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts); | |
ddbcc7e8 | 1522 | if (IS_ERR(sb)) { |
c6d57f33 | 1523 | ret = PTR_ERR(sb); |
2c6ab6d2 | 1524 | cgroup_drop_root(opts.new_root); |
cf5d5941 | 1525 | goto drop_modules; |
ddbcc7e8 PM |
1526 | } |
1527 | ||
c6d57f33 PM |
1528 | root = sb->s_fs_info; |
1529 | BUG_ON(!root); | |
1530 | if (root == opts.new_root) { | |
1531 | /* We used the new root structure, so this is a new hierarchy */ | |
1532 | struct list_head tmp_cg_links; | |
c12f65d4 | 1533 | struct cgroup *root_cgrp = &root->top_cgroup; |
c6d57f33 | 1534 | struct cgroupfs_root *existing_root; |
2ce9738b | 1535 | const struct cred *cred; |
28fd5dfc | 1536 | int i; |
ddbcc7e8 PM |
1537 | |
1538 | BUG_ON(sb->s_root != NULL); | |
1539 | ||
1540 | ret = cgroup_get_rootdir(sb); | |
1541 | if (ret) | |
1542 | goto drop_new_super; | |
817929ec | 1543 | inode = sb->s_root->d_inode; |
ddbcc7e8 | 1544 | |
817929ec | 1545 | mutex_lock(&inode->i_mutex); |
ddbcc7e8 | 1546 | mutex_lock(&cgroup_mutex); |
e25e2cbb | 1547 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 | 1548 | |
e25e2cbb TH |
1549 | /* Check for name clashes with existing mounts */ |
1550 | ret = -EBUSY; | |
1551 | if (strlen(root->name)) | |
1552 | for_each_active_root(existing_root) | |
1553 | if (!strcmp(existing_root->name, root->name)) | |
1554 | goto unlock_drop; | |
c6d57f33 | 1555 | |
817929ec PM |
1556 | /* |
1557 | * We're accessing css_set_count without locking | |
1558 | * css_set_lock here, but that's OK - it can only be | |
1559 | * increased by someone holding cgroup_lock, and | |
1560 | * that's us. The worst that can happen is that we | |
1561 | * have some link structures left over | |
1562 | */ | |
1563 | ret = allocate_cg_links(css_set_count, &tmp_cg_links); | |
e25e2cbb TH |
1564 | if (ret) |
1565 | goto unlock_drop; | |
817929ec | 1566 | |
ddbcc7e8 PM |
1567 | ret = rebind_subsystems(root, root->subsys_bits); |
1568 | if (ret == -EBUSY) { | |
c6d57f33 | 1569 | free_cg_links(&tmp_cg_links); |
e25e2cbb | 1570 | goto unlock_drop; |
ddbcc7e8 | 1571 | } |
cf5d5941 BB |
1572 | /* |
1573 | * There must be no failure case after here, since rebinding | |
1574 | * takes care of subsystems' refcounts, which are explicitly | |
1575 | * dropped in the failure exit path. | |
1576 | */ | |
ddbcc7e8 PM |
1577 | |
1578 | /* EBUSY should be the only error here */ | |
1579 | BUG_ON(ret); | |
1580 | ||
1581 | list_add(&root->root_list, &roots); | |
817929ec | 1582 | root_count++; |
ddbcc7e8 | 1583 | |
c12f65d4 | 1584 | sb->s_root->d_fsdata = root_cgrp; |
ddbcc7e8 PM |
1585 | root->top_cgroup.dentry = sb->s_root; |
1586 | ||
817929ec PM |
1587 | /* Link the top cgroup in this hierarchy into all |
1588 | * the css_set objects */ | |
1589 | write_lock(&css_set_lock); | |
28fd5dfc LZ |
1590 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { |
1591 | struct hlist_head *hhead = &css_set_table[i]; | |
1592 | struct hlist_node *node; | |
817929ec | 1593 | struct css_set *cg; |
28fd5dfc | 1594 | |
c12f65d4 LZ |
1595 | hlist_for_each_entry(cg, node, hhead, hlist) |
1596 | link_css_set(&tmp_cg_links, cg, root_cgrp); | |
28fd5dfc | 1597 | } |
817929ec PM |
1598 | write_unlock(&css_set_lock); |
1599 | ||
1600 | free_cg_links(&tmp_cg_links); | |
1601 | ||
c12f65d4 LZ |
1602 | BUG_ON(!list_empty(&root_cgrp->sibling)); |
1603 | BUG_ON(!list_empty(&root_cgrp->children)); | |
ddbcc7e8 PM |
1604 | BUG_ON(root->number_of_cgroups != 1); |
1605 | ||
2ce9738b | 1606 | cred = override_creds(&init_cred); |
c12f65d4 | 1607 | cgroup_populate_dir(root_cgrp); |
2ce9738b | 1608 | revert_creds(cred); |
e25e2cbb | 1609 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 | 1610 | mutex_unlock(&cgroup_mutex); |
34f77a90 | 1611 | mutex_unlock(&inode->i_mutex); |
c6d57f33 PM |
1612 | } else { |
1613 | /* | |
1614 | * We re-used an existing hierarchy - the new root (if | |
1615 | * any) is not needed | |
1616 | */ | |
2c6ab6d2 | 1617 | cgroup_drop_root(opts.new_root); |
cf5d5941 BB |
1618 | /* no subsys rebinding, so refcounts don't change */ |
1619 | drop_parsed_module_refcounts(opts.subsys_bits); | |
ddbcc7e8 PM |
1620 | } |
1621 | ||
c6d57f33 PM |
1622 | kfree(opts.release_agent); |
1623 | kfree(opts.name); | |
f7e83571 | 1624 | return dget(sb->s_root); |
ddbcc7e8 | 1625 | |
e25e2cbb TH |
1626 | unlock_drop: |
1627 | mutex_unlock(&cgroup_root_mutex); | |
1628 | mutex_unlock(&cgroup_mutex); | |
1629 | mutex_unlock(&inode->i_mutex); | |
ddbcc7e8 | 1630 | drop_new_super: |
6f5bbff9 | 1631 | deactivate_locked_super(sb); |
cf5d5941 BB |
1632 | drop_modules: |
1633 | drop_parsed_module_refcounts(opts.subsys_bits); | |
c6d57f33 PM |
1634 | out_err: |
1635 | kfree(opts.release_agent); | |
1636 | kfree(opts.name); | |
f7e83571 | 1637 | return ERR_PTR(ret); |
ddbcc7e8 PM |
1638 | } |
1639 | ||
1640 | static void cgroup_kill_sb(struct super_block *sb) { | |
1641 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1642 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 | 1643 | int ret; |
71cbb949 KM |
1644 | struct cg_cgroup_link *link; |
1645 | struct cg_cgroup_link *saved_link; | |
ddbcc7e8 PM |
1646 | |
1647 | BUG_ON(!root); | |
1648 | ||
1649 | BUG_ON(root->number_of_cgroups != 1); | |
bd89aabc PM |
1650 | BUG_ON(!list_empty(&cgrp->children)); |
1651 | BUG_ON(!list_empty(&cgrp->sibling)); | |
ddbcc7e8 PM |
1652 | |
1653 | mutex_lock(&cgroup_mutex); | |
e25e2cbb | 1654 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1655 | |
1656 | /* Rebind all subsystems back to the default hierarchy */ | |
1657 | ret = rebind_subsystems(root, 0); | |
1658 | /* Shouldn't be able to fail ... */ | |
1659 | BUG_ON(ret); | |
1660 | ||
817929ec PM |
1661 | /* |
1662 | * Release all the links from css_sets to this hierarchy's | |
1663 | * root cgroup | |
1664 | */ | |
1665 | write_lock(&css_set_lock); | |
71cbb949 KM |
1666 | |
1667 | list_for_each_entry_safe(link, saved_link, &cgrp->css_sets, | |
1668 | cgrp_link_list) { | |
817929ec | 1669 | list_del(&link->cg_link_list); |
bd89aabc | 1670 | list_del(&link->cgrp_link_list); |
817929ec PM |
1671 | kfree(link); |
1672 | } | |
1673 | write_unlock(&css_set_lock); | |
1674 | ||
839ec545 PM |
1675 | if (!list_empty(&root->root_list)) { |
1676 | list_del(&root->root_list); | |
1677 | root_count--; | |
1678 | } | |
e5f6a860 | 1679 | |
e25e2cbb | 1680 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1681 | mutex_unlock(&cgroup_mutex); |
1682 | ||
ddbcc7e8 | 1683 | kill_litter_super(sb); |
2c6ab6d2 | 1684 | cgroup_drop_root(root); |
ddbcc7e8 PM |
1685 | } |
1686 | ||
1687 | static struct file_system_type cgroup_fs_type = { | |
1688 | .name = "cgroup", | |
f7e83571 | 1689 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1690 | .kill_sb = cgroup_kill_sb, |
1691 | }; | |
1692 | ||
676db4af GK |
1693 | static struct kobject *cgroup_kobj; |
1694 | ||
bd89aabc | 1695 | static inline struct cgroup *__d_cgrp(struct dentry *dentry) |
ddbcc7e8 PM |
1696 | { |
1697 | return dentry->d_fsdata; | |
1698 | } | |
1699 | ||
1700 | static inline struct cftype *__d_cft(struct dentry *dentry) | |
1701 | { | |
1702 | return dentry->d_fsdata; | |
1703 | } | |
1704 | ||
a043e3b2 LZ |
1705 | /** |
1706 | * cgroup_path - generate the path of a cgroup | |
1707 | * @cgrp: the cgroup in question | |
1708 | * @buf: the buffer to write the path into | |
1709 | * @buflen: the length of the buffer | |
1710 | * | |
a47295e6 PM |
1711 | * Called with cgroup_mutex held or else with an RCU-protected cgroup |
1712 | * reference. Writes path of cgroup into buf. Returns 0 on success, | |
1713 | * -errno on error. | |
ddbcc7e8 | 1714 | */ |
bd89aabc | 1715 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
ddbcc7e8 PM |
1716 | { |
1717 | char *start; | |
9a9686b6 | 1718 | struct dentry *dentry = rcu_dereference_check(cgrp->dentry, |
9a9686b6 | 1719 | cgroup_lock_is_held()); |
ddbcc7e8 | 1720 | |
a47295e6 | 1721 | if (!dentry || cgrp == dummytop) { |
ddbcc7e8 PM |
1722 | /* |
1723 | * Inactive subsystems have no dentry for their root | |
1724 | * cgroup | |
1725 | */ | |
1726 | strcpy(buf, "/"); | |
1727 | return 0; | |
1728 | } | |
1729 | ||
1730 | start = buf + buflen; | |
1731 | ||
1732 | *--start = '\0'; | |
1733 | for (;;) { | |
a47295e6 | 1734 | int len = dentry->d_name.len; |
9a9686b6 | 1735 | |
ddbcc7e8 PM |
1736 | if ((start -= len) < buf) |
1737 | return -ENAMETOOLONG; | |
9a9686b6 | 1738 | memcpy(start, dentry->d_name.name, len); |
bd89aabc PM |
1739 | cgrp = cgrp->parent; |
1740 | if (!cgrp) | |
ddbcc7e8 | 1741 | break; |
9a9686b6 LZ |
1742 | |
1743 | dentry = rcu_dereference_check(cgrp->dentry, | |
9a9686b6 | 1744 | cgroup_lock_is_held()); |
bd89aabc | 1745 | if (!cgrp->parent) |
ddbcc7e8 PM |
1746 | continue; |
1747 | if (--start < buf) | |
1748 | return -ENAMETOOLONG; | |
1749 | *start = '/'; | |
1750 | } | |
1751 | memmove(buf, start, buf + buflen - start); | |
1752 | return 0; | |
1753 | } | |
67523c48 | 1754 | EXPORT_SYMBOL_GPL(cgroup_path); |
ddbcc7e8 | 1755 | |
2f7ee569 TH |
1756 | /* |
1757 | * Control Group taskset | |
1758 | */ | |
134d3373 TH |
1759 | struct task_and_cgroup { |
1760 | struct task_struct *task; | |
1761 | struct cgroup *cgrp; | |
61d1d219 | 1762 | struct css_set *cg; |
134d3373 TH |
1763 | }; |
1764 | ||
2f7ee569 TH |
1765 | struct cgroup_taskset { |
1766 | struct task_and_cgroup single; | |
1767 | struct flex_array *tc_array; | |
1768 | int tc_array_len; | |
1769 | int idx; | |
1770 | struct cgroup *cur_cgrp; | |
1771 | }; | |
1772 | ||
1773 | /** | |
1774 | * cgroup_taskset_first - reset taskset and return the first task | |
1775 | * @tset: taskset of interest | |
1776 | * | |
1777 | * @tset iteration is initialized and the first task is returned. | |
1778 | */ | |
1779 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1780 | { | |
1781 | if (tset->tc_array) { | |
1782 | tset->idx = 0; | |
1783 | return cgroup_taskset_next(tset); | |
1784 | } else { | |
1785 | tset->cur_cgrp = tset->single.cgrp; | |
1786 | return tset->single.task; | |
1787 | } | |
1788 | } | |
1789 | EXPORT_SYMBOL_GPL(cgroup_taskset_first); | |
1790 | ||
1791 | /** | |
1792 | * cgroup_taskset_next - iterate to the next task in taskset | |
1793 | * @tset: taskset of interest | |
1794 | * | |
1795 | * Return the next task in @tset. Iteration must have been initialized | |
1796 | * with cgroup_taskset_first(). | |
1797 | */ | |
1798 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1799 | { | |
1800 | struct task_and_cgroup *tc; | |
1801 | ||
1802 | if (!tset->tc_array || tset->idx >= tset->tc_array_len) | |
1803 | return NULL; | |
1804 | ||
1805 | tc = flex_array_get(tset->tc_array, tset->idx++); | |
1806 | tset->cur_cgrp = tc->cgrp; | |
1807 | return tc->task; | |
1808 | } | |
1809 | EXPORT_SYMBOL_GPL(cgroup_taskset_next); | |
1810 | ||
1811 | /** | |
1812 | * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task | |
1813 | * @tset: taskset of interest | |
1814 | * | |
1815 | * Return the cgroup for the current (last returned) task of @tset. This | |
1816 | * function must be preceded by either cgroup_taskset_first() or | |
1817 | * cgroup_taskset_next(). | |
1818 | */ | |
1819 | struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset) | |
1820 | { | |
1821 | return tset->cur_cgrp; | |
1822 | } | |
1823 | EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup); | |
1824 | ||
1825 | /** | |
1826 | * cgroup_taskset_size - return the number of tasks in taskset | |
1827 | * @tset: taskset of interest | |
1828 | */ | |
1829 | int cgroup_taskset_size(struct cgroup_taskset *tset) | |
1830 | { | |
1831 | return tset->tc_array ? tset->tc_array_len : 1; | |
1832 | } | |
1833 | EXPORT_SYMBOL_GPL(cgroup_taskset_size); | |
1834 | ||
1835 | ||
74a1166d BB |
1836 | /* |
1837 | * cgroup_task_migrate - move a task from one cgroup to another. | |
1838 | * | |
1839 | * 'guarantee' is set if the caller promises that a new css_set for the task | |
1840 | * will already exist. If not set, this function might sleep, and can fail with | |
cd3d0952 | 1841 | * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked. |
74a1166d | 1842 | */ |
61d1d219 MSB |
1843 | static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp, |
1844 | struct task_struct *tsk, struct css_set *newcg) | |
74a1166d BB |
1845 | { |
1846 | struct css_set *oldcg; | |
74a1166d BB |
1847 | |
1848 | /* | |
026085ef MSB |
1849 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1850 | * setting such that we can't race against cgroup_exit() changing the | |
1851 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1852 | */ |
c84cdf75 | 1853 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
74a1166d | 1854 | oldcg = tsk->cgroups; |
74a1166d | 1855 | |
74a1166d | 1856 | task_lock(tsk); |
74a1166d BB |
1857 | rcu_assign_pointer(tsk->cgroups, newcg); |
1858 | task_unlock(tsk); | |
1859 | ||
1860 | /* Update the css_set linked lists if we're using them */ | |
1861 | write_lock(&css_set_lock); | |
1862 | if (!list_empty(&tsk->cg_list)) | |
1863 | list_move(&tsk->cg_list, &newcg->tasks); | |
1864 | write_unlock(&css_set_lock); | |
1865 | ||
1866 | /* | |
1867 | * We just gained a reference on oldcg by taking it from the task. As | |
1868 | * trading it for newcg is protected by cgroup_mutex, we're safe to drop | |
1869 | * it here; it will be freed under RCU. | |
1870 | */ | |
1871 | put_css_set(oldcg); | |
1872 | ||
1873 | set_bit(CGRP_RELEASABLE, &oldcgrp->flags); | |
74a1166d BB |
1874 | } |
1875 | ||
a043e3b2 LZ |
1876 | /** |
1877 | * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp' | |
1878 | * @cgrp: the cgroup the task is attaching to | |
1879 | * @tsk: the task to be attached | |
bbcb81d0 | 1880 | * |
cd3d0952 TH |
1881 | * Call with cgroup_mutex and threadgroup locked. May take task_lock of |
1882 | * @tsk during call. | |
bbcb81d0 | 1883 | */ |
956db3ca | 1884 | int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
bbcb81d0 | 1885 | { |
74a1166d | 1886 | int retval; |
2468c723 | 1887 | struct cgroup_subsys *ss, *failed_ss = NULL; |
bd89aabc | 1888 | struct cgroup *oldcgrp; |
bd89aabc | 1889 | struct cgroupfs_root *root = cgrp->root; |
2f7ee569 | 1890 | struct cgroup_taskset tset = { }; |
61d1d219 | 1891 | struct css_set *newcg; |
bbcb81d0 | 1892 | |
cd3d0952 TH |
1893 | /* @tsk either already exited or can't exit until the end */ |
1894 | if (tsk->flags & PF_EXITING) | |
1895 | return -ESRCH; | |
bbcb81d0 PM |
1896 | |
1897 | /* Nothing to do if the task is already in that cgroup */ | |
7717f7ba | 1898 | oldcgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 1899 | if (cgrp == oldcgrp) |
bbcb81d0 PM |
1900 | return 0; |
1901 | ||
2f7ee569 TH |
1902 | tset.single.task = tsk; |
1903 | tset.single.cgrp = oldcgrp; | |
1904 | ||
bbcb81d0 PM |
1905 | for_each_subsys(root, ss) { |
1906 | if (ss->can_attach) { | |
761b3ef5 | 1907 | retval = ss->can_attach(cgrp, &tset); |
2468c723 DN |
1908 | if (retval) { |
1909 | /* | |
1910 | * Remember on which subsystem the can_attach() | |
1911 | * failed, so that we only call cancel_attach() | |
1912 | * against the subsystems whose can_attach() | |
1913 | * succeeded. (See below) | |
1914 | */ | |
1915 | failed_ss = ss; | |
1916 | goto out; | |
1917 | } | |
bbcb81d0 PM |
1918 | } |
1919 | } | |
1920 | ||
61d1d219 MSB |
1921 | newcg = find_css_set(tsk->cgroups, cgrp); |
1922 | if (!newcg) { | |
1923 | retval = -ENOMEM; | |
2468c723 | 1924 | goto out; |
61d1d219 MSB |
1925 | } |
1926 | ||
1927 | cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg); | |
817929ec | 1928 | |
bbcb81d0 | 1929 | for_each_subsys(root, ss) { |
e18f6318 | 1930 | if (ss->attach) |
761b3ef5 | 1931 | ss->attach(cgrp, &tset); |
bbcb81d0 | 1932 | } |
74a1166d | 1933 | |
bbcb81d0 | 1934 | synchronize_rcu(); |
ec64f515 KH |
1935 | |
1936 | /* | |
1937 | * wake up rmdir() waiter. the rmdir should fail since the cgroup | |
1938 | * is no longer empty. | |
1939 | */ | |
88703267 | 1940 | cgroup_wakeup_rmdir_waiter(cgrp); |
2468c723 DN |
1941 | out: |
1942 | if (retval) { | |
1943 | for_each_subsys(root, ss) { | |
1944 | if (ss == failed_ss) | |
1945 | /* | |
1946 | * This subsystem was the one that failed the | |
1947 | * can_attach() check earlier, so we don't need | |
1948 | * to call cancel_attach() against it or any | |
1949 | * remaining subsystems. | |
1950 | */ | |
1951 | break; | |
1952 | if (ss->cancel_attach) | |
761b3ef5 | 1953 | ss->cancel_attach(cgrp, &tset); |
2468c723 DN |
1954 | } |
1955 | } | |
1956 | return retval; | |
bbcb81d0 PM |
1957 | } |
1958 | ||
d7926ee3 | 1959 | /** |
31583bb0 MT |
1960 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' |
1961 | * @from: attach to all cgroups of a given task | |
d7926ee3 SS |
1962 | * @tsk: the task to be attached |
1963 | */ | |
31583bb0 | 1964 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) |
d7926ee3 SS |
1965 | { |
1966 | struct cgroupfs_root *root; | |
d7926ee3 SS |
1967 | int retval = 0; |
1968 | ||
1969 | cgroup_lock(); | |
1970 | for_each_active_root(root) { | |
31583bb0 MT |
1971 | struct cgroup *from_cg = task_cgroup_from_root(from, root); |
1972 | ||
1973 | retval = cgroup_attach_task(from_cg, tsk); | |
d7926ee3 SS |
1974 | if (retval) |
1975 | break; | |
1976 | } | |
1977 | cgroup_unlock(); | |
1978 | ||
1979 | return retval; | |
1980 | } | |
31583bb0 | 1981 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); |
d7926ee3 | 1982 | |
74a1166d BB |
1983 | /** |
1984 | * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup | |
1985 | * @cgrp: the cgroup to attach to | |
1986 | * @leader: the threadgroup leader task_struct of the group to be attached | |
1987 | * | |
257058ae TH |
1988 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
1989 | * task_lock of each thread in leader's threadgroup individually in turn. | |
74a1166d | 1990 | */ |
1c6c3fad | 1991 | static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader) |
74a1166d BB |
1992 | { |
1993 | int retval, i, group_size; | |
1994 | struct cgroup_subsys *ss, *failed_ss = NULL; | |
74a1166d | 1995 | /* guaranteed to be initialized later, but the compiler needs this */ |
74a1166d BB |
1996 | struct cgroupfs_root *root = cgrp->root; |
1997 | /* threadgroup list cursor and array */ | |
1998 | struct task_struct *tsk; | |
134d3373 | 1999 | struct task_and_cgroup *tc; |
d846687d | 2000 | struct flex_array *group; |
2f7ee569 | 2001 | struct cgroup_taskset tset = { }; |
74a1166d BB |
2002 | |
2003 | /* | |
2004 | * step 0: in order to do expensive, possibly blocking operations for | |
2005 | * every thread, we cannot iterate the thread group list, since it needs | |
2006 | * rcu or tasklist locked. instead, build an array of all threads in the | |
257058ae TH |
2007 | * group - group_rwsem prevents new threads from appearing, and if |
2008 | * threads exit, this will just be an over-estimate. | |
74a1166d BB |
2009 | */ |
2010 | group_size = get_nr_threads(leader); | |
d846687d | 2011 | /* flex_array supports very large thread-groups better than kmalloc. */ |
134d3373 | 2012 | group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL); |
74a1166d BB |
2013 | if (!group) |
2014 | return -ENOMEM; | |
d846687d BB |
2015 | /* pre-allocate to guarantee space while iterating in rcu read-side. */ |
2016 | retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL); | |
2017 | if (retval) | |
2018 | goto out_free_group_list; | |
74a1166d | 2019 | |
74a1166d BB |
2020 | tsk = leader; |
2021 | i = 0; | |
fb5d2b4c MSB |
2022 | /* |
2023 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2024 | * already PF_EXITING could be freed from underneath us unless we | |
2025 | * take an rcu_read_lock. | |
2026 | */ | |
2027 | rcu_read_lock(); | |
74a1166d | 2028 | do { |
134d3373 TH |
2029 | struct task_and_cgroup ent; |
2030 | ||
cd3d0952 TH |
2031 | /* @tsk either already exited or can't exit until the end */ |
2032 | if (tsk->flags & PF_EXITING) | |
2033 | continue; | |
2034 | ||
74a1166d BB |
2035 | /* as per above, nr_threads may decrease, but not increase. */ |
2036 | BUG_ON(i >= group_size); | |
134d3373 TH |
2037 | ent.task = tsk; |
2038 | ent.cgrp = task_cgroup_from_root(tsk, root); | |
892a2b90 MSB |
2039 | /* nothing to do if this task is already in the cgroup */ |
2040 | if (ent.cgrp == cgrp) | |
2041 | continue; | |
61d1d219 MSB |
2042 | /* |
2043 | * saying GFP_ATOMIC has no effect here because we did prealloc | |
2044 | * earlier, but it's good form to communicate our expectations. | |
2045 | */ | |
134d3373 | 2046 | retval = flex_array_put(group, i, &ent, GFP_ATOMIC); |
d846687d | 2047 | BUG_ON(retval != 0); |
74a1166d BB |
2048 | i++; |
2049 | } while_each_thread(leader, tsk); | |
fb5d2b4c | 2050 | rcu_read_unlock(); |
74a1166d BB |
2051 | /* remember the number of threads in the array for later. */ |
2052 | group_size = i; | |
2f7ee569 TH |
2053 | tset.tc_array = group; |
2054 | tset.tc_array_len = group_size; | |
74a1166d | 2055 | |
134d3373 TH |
2056 | /* methods shouldn't be called if no task is actually migrating */ |
2057 | retval = 0; | |
892a2b90 | 2058 | if (!group_size) |
b07ef774 | 2059 | goto out_free_group_list; |
134d3373 | 2060 | |
74a1166d BB |
2061 | /* |
2062 | * step 1: check that we can legitimately attach to the cgroup. | |
2063 | */ | |
2064 | for_each_subsys(root, ss) { | |
2065 | if (ss->can_attach) { | |
761b3ef5 | 2066 | retval = ss->can_attach(cgrp, &tset); |
74a1166d BB |
2067 | if (retval) { |
2068 | failed_ss = ss; | |
2069 | goto out_cancel_attach; | |
2070 | } | |
2071 | } | |
74a1166d BB |
2072 | } |
2073 | ||
2074 | /* | |
2075 | * step 2: make sure css_sets exist for all threads to be migrated. | |
2076 | * we use find_css_set, which allocates a new one if necessary. | |
2077 | */ | |
74a1166d | 2078 | for (i = 0; i < group_size; i++) { |
134d3373 | 2079 | tc = flex_array_get(group, i); |
61d1d219 MSB |
2080 | tc->cg = find_css_set(tc->task->cgroups, cgrp); |
2081 | if (!tc->cg) { | |
2082 | retval = -ENOMEM; | |
2083 | goto out_put_css_set_refs; | |
74a1166d BB |
2084 | } |
2085 | } | |
2086 | ||
2087 | /* | |
494c167c TH |
2088 | * step 3: now that we're guaranteed success wrt the css_sets, |
2089 | * proceed to move all tasks to the new cgroup. There are no | |
2090 | * failure cases after here, so this is the commit point. | |
74a1166d | 2091 | */ |
74a1166d | 2092 | for (i = 0; i < group_size; i++) { |
134d3373 | 2093 | tc = flex_array_get(group, i); |
61d1d219 | 2094 | cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg); |
74a1166d BB |
2095 | } |
2096 | /* nothing is sensitive to fork() after this point. */ | |
2097 | ||
2098 | /* | |
494c167c | 2099 | * step 4: do subsystem attach callbacks. |
74a1166d BB |
2100 | */ |
2101 | for_each_subsys(root, ss) { | |
2102 | if (ss->attach) | |
761b3ef5 | 2103 | ss->attach(cgrp, &tset); |
74a1166d BB |
2104 | } |
2105 | ||
2106 | /* | |
2107 | * step 5: success! and cleanup | |
2108 | */ | |
2109 | synchronize_rcu(); | |
2110 | cgroup_wakeup_rmdir_waiter(cgrp); | |
2111 | retval = 0; | |
61d1d219 MSB |
2112 | out_put_css_set_refs: |
2113 | if (retval) { | |
2114 | for (i = 0; i < group_size; i++) { | |
2115 | tc = flex_array_get(group, i); | |
2116 | if (!tc->cg) | |
2117 | break; | |
2118 | put_css_set(tc->cg); | |
2119 | } | |
74a1166d BB |
2120 | } |
2121 | out_cancel_attach: | |
74a1166d BB |
2122 | if (retval) { |
2123 | for_each_subsys(root, ss) { | |
494c167c | 2124 | if (ss == failed_ss) |
74a1166d | 2125 | break; |
74a1166d | 2126 | if (ss->cancel_attach) |
761b3ef5 | 2127 | ss->cancel_attach(cgrp, &tset); |
74a1166d BB |
2128 | } |
2129 | } | |
74a1166d | 2130 | out_free_group_list: |
d846687d | 2131 | flex_array_free(group); |
74a1166d BB |
2132 | return retval; |
2133 | } | |
2134 | ||
2135 | /* | |
2136 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
2137 | * function to attach either it or all tasks in its threadgroup. Will lock |
2138 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 2139 | */ |
74a1166d | 2140 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2141 | { |
bbcb81d0 | 2142 | struct task_struct *tsk; |
c69e8d9c | 2143 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2144 | int ret; |
2145 | ||
74a1166d BB |
2146 | if (!cgroup_lock_live_group(cgrp)) |
2147 | return -ENODEV; | |
2148 | ||
b78949eb MSB |
2149 | retry_find_task: |
2150 | rcu_read_lock(); | |
bbcb81d0 | 2151 | if (pid) { |
73507f33 | 2152 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2153 | if (!tsk) { |
2154 | rcu_read_unlock(); | |
b78949eb MSB |
2155 | ret= -ESRCH; |
2156 | goto out_unlock_cgroup; | |
bbcb81d0 | 2157 | } |
74a1166d BB |
2158 | /* |
2159 | * even if we're attaching all tasks in the thread group, we | |
2160 | * only need to check permissions on one of them. | |
2161 | */ | |
c69e8d9c DH |
2162 | tcred = __task_cred(tsk); |
2163 | if (cred->euid && | |
2164 | cred->euid != tcred->uid && | |
2165 | cred->euid != tcred->suid) { | |
2166 | rcu_read_unlock(); | |
b78949eb MSB |
2167 | ret = -EACCES; |
2168 | goto out_unlock_cgroup; | |
bbcb81d0 | 2169 | } |
b78949eb MSB |
2170 | } else |
2171 | tsk = current; | |
cd3d0952 TH |
2172 | |
2173 | if (threadgroup) | |
b78949eb MSB |
2174 | tsk = tsk->group_leader; |
2175 | get_task_struct(tsk); | |
2176 | rcu_read_unlock(); | |
2177 | ||
2178 | threadgroup_lock(tsk); | |
2179 | if (threadgroup) { | |
2180 | if (!thread_group_leader(tsk)) { | |
2181 | /* | |
2182 | * a race with de_thread from another thread's exec() | |
2183 | * may strip us of our leadership, if this happens, | |
2184 | * there is no choice but to throw this task away and | |
2185 | * try again; this is | |
2186 | * "double-double-toil-and-trouble-check locking". | |
2187 | */ | |
2188 | threadgroup_unlock(tsk); | |
2189 | put_task_struct(tsk); | |
2190 | goto retry_find_task; | |
2191 | } | |
74a1166d | 2192 | ret = cgroup_attach_proc(cgrp, tsk); |
b78949eb | 2193 | } else |
74a1166d | 2194 | ret = cgroup_attach_task(cgrp, tsk); |
cd3d0952 TH |
2195 | threadgroup_unlock(tsk); |
2196 | ||
bbcb81d0 | 2197 | put_task_struct(tsk); |
b78949eb | 2198 | out_unlock_cgroup: |
74a1166d | 2199 | cgroup_unlock(); |
bbcb81d0 PM |
2200 | return ret; |
2201 | } | |
2202 | ||
af351026 | 2203 | static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid) |
74a1166d BB |
2204 | { |
2205 | return attach_task_by_pid(cgrp, pid, false); | |
2206 | } | |
2207 | ||
2208 | static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid) | |
af351026 | 2209 | { |
b78949eb | 2210 | return attach_task_by_pid(cgrp, tgid, true); |
af351026 PM |
2211 | } |
2212 | ||
e788e066 PM |
2213 | /** |
2214 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
2215 | * @cgrp: the cgroup to be checked for liveness | |
2216 | * | |
84eea842 PM |
2217 | * On success, returns true; the lock should be later released with |
2218 | * cgroup_unlock(). On failure returns false with no lock held. | |
e788e066 | 2219 | */ |
84eea842 | 2220 | bool cgroup_lock_live_group(struct cgroup *cgrp) |
e788e066 PM |
2221 | { |
2222 | mutex_lock(&cgroup_mutex); | |
2223 | if (cgroup_is_removed(cgrp)) { | |
2224 | mutex_unlock(&cgroup_mutex); | |
2225 | return false; | |
2226 | } | |
2227 | return true; | |
2228 | } | |
67523c48 | 2229 | EXPORT_SYMBOL_GPL(cgroup_lock_live_group); |
e788e066 PM |
2230 | |
2231 | static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft, | |
2232 | const char *buffer) | |
2233 | { | |
2234 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
f4a2589f EK |
2235 | if (strlen(buffer) >= PATH_MAX) |
2236 | return -EINVAL; | |
e788e066 PM |
2237 | if (!cgroup_lock_live_group(cgrp)) |
2238 | return -ENODEV; | |
e25e2cbb | 2239 | mutex_lock(&cgroup_root_mutex); |
e788e066 | 2240 | strcpy(cgrp->root->release_agent_path, buffer); |
e25e2cbb | 2241 | mutex_unlock(&cgroup_root_mutex); |
84eea842 | 2242 | cgroup_unlock(); |
e788e066 PM |
2243 | return 0; |
2244 | } | |
2245 | ||
2246 | static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft, | |
2247 | struct seq_file *seq) | |
2248 | { | |
2249 | if (!cgroup_lock_live_group(cgrp)) | |
2250 | return -ENODEV; | |
2251 | seq_puts(seq, cgrp->root->release_agent_path); | |
2252 | seq_putc(seq, '\n'); | |
84eea842 | 2253 | cgroup_unlock(); |
e788e066 PM |
2254 | return 0; |
2255 | } | |
2256 | ||
84eea842 PM |
2257 | /* A buffer size big enough for numbers or short strings */ |
2258 | #define CGROUP_LOCAL_BUFFER_SIZE 64 | |
2259 | ||
e73d2c61 | 2260 | static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft, |
f4c753b7 PM |
2261 | struct file *file, |
2262 | const char __user *userbuf, | |
2263 | size_t nbytes, loff_t *unused_ppos) | |
355e0c48 | 2264 | { |
84eea842 | 2265 | char buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
355e0c48 | 2266 | int retval = 0; |
355e0c48 PM |
2267 | char *end; |
2268 | ||
2269 | if (!nbytes) | |
2270 | return -EINVAL; | |
2271 | if (nbytes >= sizeof(buffer)) | |
2272 | return -E2BIG; | |
2273 | if (copy_from_user(buffer, userbuf, nbytes)) | |
2274 | return -EFAULT; | |
2275 | ||
2276 | buffer[nbytes] = 0; /* nul-terminate */ | |
e73d2c61 | 2277 | if (cft->write_u64) { |
478988d3 | 2278 | u64 val = simple_strtoull(strstrip(buffer), &end, 0); |
e73d2c61 PM |
2279 | if (*end) |
2280 | return -EINVAL; | |
2281 | retval = cft->write_u64(cgrp, cft, val); | |
2282 | } else { | |
478988d3 | 2283 | s64 val = simple_strtoll(strstrip(buffer), &end, 0); |
e73d2c61 PM |
2284 | if (*end) |
2285 | return -EINVAL; | |
2286 | retval = cft->write_s64(cgrp, cft, val); | |
2287 | } | |
355e0c48 PM |
2288 | if (!retval) |
2289 | retval = nbytes; | |
2290 | return retval; | |
2291 | } | |
2292 | ||
db3b1497 PM |
2293 | static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft, |
2294 | struct file *file, | |
2295 | const char __user *userbuf, | |
2296 | size_t nbytes, loff_t *unused_ppos) | |
2297 | { | |
84eea842 | 2298 | char local_buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
db3b1497 PM |
2299 | int retval = 0; |
2300 | size_t max_bytes = cft->max_write_len; | |
2301 | char *buffer = local_buffer; | |
2302 | ||
2303 | if (!max_bytes) | |
2304 | max_bytes = sizeof(local_buffer) - 1; | |
2305 | if (nbytes >= max_bytes) | |
2306 | return -E2BIG; | |
2307 | /* Allocate a dynamic buffer if we need one */ | |
2308 | if (nbytes >= sizeof(local_buffer)) { | |
2309 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); | |
2310 | if (buffer == NULL) | |
2311 | return -ENOMEM; | |
2312 | } | |
5a3eb9f6 LZ |
2313 | if (nbytes && copy_from_user(buffer, userbuf, nbytes)) { |
2314 | retval = -EFAULT; | |
2315 | goto out; | |
2316 | } | |
db3b1497 PM |
2317 | |
2318 | buffer[nbytes] = 0; /* nul-terminate */ | |
478988d3 | 2319 | retval = cft->write_string(cgrp, cft, strstrip(buffer)); |
db3b1497 PM |
2320 | if (!retval) |
2321 | retval = nbytes; | |
5a3eb9f6 | 2322 | out: |
db3b1497 PM |
2323 | if (buffer != local_buffer) |
2324 | kfree(buffer); | |
2325 | return retval; | |
2326 | } | |
2327 | ||
ddbcc7e8 PM |
2328 | static ssize_t cgroup_file_write(struct file *file, const char __user *buf, |
2329 | size_t nbytes, loff_t *ppos) | |
2330 | { | |
2331 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 2332 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 2333 | |
75139b82 | 2334 | if (cgroup_is_removed(cgrp)) |
ddbcc7e8 | 2335 | return -ENODEV; |
355e0c48 | 2336 | if (cft->write) |
bd89aabc | 2337 | return cft->write(cgrp, cft, file, buf, nbytes, ppos); |
e73d2c61 PM |
2338 | if (cft->write_u64 || cft->write_s64) |
2339 | return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos); | |
db3b1497 PM |
2340 | if (cft->write_string) |
2341 | return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos); | |
d447ea2f PE |
2342 | if (cft->trigger) { |
2343 | int ret = cft->trigger(cgrp, (unsigned int)cft->private); | |
2344 | return ret ? ret : nbytes; | |
2345 | } | |
355e0c48 | 2346 | return -EINVAL; |
ddbcc7e8 PM |
2347 | } |
2348 | ||
f4c753b7 PM |
2349 | static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft, |
2350 | struct file *file, | |
2351 | char __user *buf, size_t nbytes, | |
2352 | loff_t *ppos) | |
ddbcc7e8 | 2353 | { |
84eea842 | 2354 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
f4c753b7 | 2355 | u64 val = cft->read_u64(cgrp, cft); |
ddbcc7e8 PM |
2356 | int len = sprintf(tmp, "%llu\n", (unsigned long long) val); |
2357 | ||
2358 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
2359 | } | |
2360 | ||
e73d2c61 PM |
2361 | static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft, |
2362 | struct file *file, | |
2363 | char __user *buf, size_t nbytes, | |
2364 | loff_t *ppos) | |
2365 | { | |
84eea842 | 2366 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
e73d2c61 PM |
2367 | s64 val = cft->read_s64(cgrp, cft); |
2368 | int len = sprintf(tmp, "%lld\n", (long long) val); | |
2369 | ||
2370 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
2371 | } | |
2372 | ||
ddbcc7e8 PM |
2373 | static ssize_t cgroup_file_read(struct file *file, char __user *buf, |
2374 | size_t nbytes, loff_t *ppos) | |
2375 | { | |
2376 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 2377 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 2378 | |
75139b82 | 2379 | if (cgroup_is_removed(cgrp)) |
ddbcc7e8 PM |
2380 | return -ENODEV; |
2381 | ||
2382 | if (cft->read) | |
bd89aabc | 2383 | return cft->read(cgrp, cft, file, buf, nbytes, ppos); |
f4c753b7 PM |
2384 | if (cft->read_u64) |
2385 | return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos); | |
e73d2c61 PM |
2386 | if (cft->read_s64) |
2387 | return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos); | |
ddbcc7e8 PM |
2388 | return -EINVAL; |
2389 | } | |
2390 | ||
91796569 PM |
2391 | /* |
2392 | * seqfile ops/methods for returning structured data. Currently just | |
2393 | * supports string->u64 maps, but can be extended in future. | |
2394 | */ | |
2395 | ||
2396 | struct cgroup_seqfile_state { | |
2397 | struct cftype *cft; | |
2398 | struct cgroup *cgroup; | |
2399 | }; | |
2400 | ||
2401 | static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value) | |
2402 | { | |
2403 | struct seq_file *sf = cb->state; | |
2404 | return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value); | |
2405 | } | |
2406 | ||
2407 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) | |
2408 | { | |
2409 | struct cgroup_seqfile_state *state = m->private; | |
2410 | struct cftype *cft = state->cft; | |
29486df3 SH |
2411 | if (cft->read_map) { |
2412 | struct cgroup_map_cb cb = { | |
2413 | .fill = cgroup_map_add, | |
2414 | .state = m, | |
2415 | }; | |
2416 | return cft->read_map(state->cgroup, cft, &cb); | |
2417 | } | |
2418 | return cft->read_seq_string(state->cgroup, cft, m); | |
91796569 PM |
2419 | } |
2420 | ||
96930a63 | 2421 | static int cgroup_seqfile_release(struct inode *inode, struct file *file) |
91796569 PM |
2422 | { |
2423 | struct seq_file *seq = file->private_data; | |
2424 | kfree(seq->private); | |
2425 | return single_release(inode, file); | |
2426 | } | |
2427 | ||
828c0950 | 2428 | static const struct file_operations cgroup_seqfile_operations = { |
91796569 | 2429 | .read = seq_read, |
e788e066 | 2430 | .write = cgroup_file_write, |
91796569 PM |
2431 | .llseek = seq_lseek, |
2432 | .release = cgroup_seqfile_release, | |
2433 | }; | |
2434 | ||
ddbcc7e8 PM |
2435 | static int cgroup_file_open(struct inode *inode, struct file *file) |
2436 | { | |
2437 | int err; | |
2438 | struct cftype *cft; | |
2439 | ||
2440 | err = generic_file_open(inode, file); | |
2441 | if (err) | |
2442 | return err; | |
ddbcc7e8 | 2443 | cft = __d_cft(file->f_dentry); |
75139b82 | 2444 | |
29486df3 | 2445 | if (cft->read_map || cft->read_seq_string) { |
91796569 PM |
2446 | struct cgroup_seqfile_state *state = |
2447 | kzalloc(sizeof(*state), GFP_USER); | |
2448 | if (!state) | |
2449 | return -ENOMEM; | |
2450 | state->cft = cft; | |
2451 | state->cgroup = __d_cgrp(file->f_dentry->d_parent); | |
2452 | file->f_op = &cgroup_seqfile_operations; | |
2453 | err = single_open(file, cgroup_seqfile_show, state); | |
2454 | if (err < 0) | |
2455 | kfree(state); | |
2456 | } else if (cft->open) | |
ddbcc7e8 PM |
2457 | err = cft->open(inode, file); |
2458 | else | |
2459 | err = 0; | |
2460 | ||
2461 | return err; | |
2462 | } | |
2463 | ||
2464 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
2465 | { | |
2466 | struct cftype *cft = __d_cft(file->f_dentry); | |
2467 | if (cft->release) | |
2468 | return cft->release(inode, file); | |
2469 | return 0; | |
2470 | } | |
2471 | ||
2472 | /* | |
2473 | * cgroup_rename - Only allow simple rename of directories in place. | |
2474 | */ | |
2475 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
2476 | struct inode *new_dir, struct dentry *new_dentry) | |
2477 | { | |
2478 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) | |
2479 | return -ENOTDIR; | |
2480 | if (new_dentry->d_inode) | |
2481 | return -EEXIST; | |
2482 | if (old_dir != new_dir) | |
2483 | return -EIO; | |
2484 | return simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
2485 | } | |
2486 | ||
828c0950 | 2487 | static const struct file_operations cgroup_file_operations = { |
ddbcc7e8 PM |
2488 | .read = cgroup_file_read, |
2489 | .write = cgroup_file_write, | |
2490 | .llseek = generic_file_llseek, | |
2491 | .open = cgroup_file_open, | |
2492 | .release = cgroup_file_release, | |
2493 | }; | |
2494 | ||
6e1d5dcc | 2495 | static const struct inode_operations cgroup_dir_inode_operations = { |
c72a04e3 | 2496 | .lookup = cgroup_lookup, |
ddbcc7e8 PM |
2497 | .mkdir = cgroup_mkdir, |
2498 | .rmdir = cgroup_rmdir, | |
2499 | .rename = cgroup_rename, | |
2500 | }; | |
2501 | ||
c72a04e3 AV |
2502 | static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) |
2503 | { | |
2504 | if (dentry->d_name.len > NAME_MAX) | |
2505 | return ERR_PTR(-ENAMETOOLONG); | |
2506 | d_add(dentry, NULL); | |
2507 | return NULL; | |
2508 | } | |
2509 | ||
0dea1168 KS |
2510 | /* |
2511 | * Check if a file is a control file | |
2512 | */ | |
2513 | static inline struct cftype *__file_cft(struct file *file) | |
2514 | { | |
2515 | if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations) | |
2516 | return ERR_PTR(-EINVAL); | |
2517 | return __d_cft(file->f_dentry); | |
2518 | } | |
2519 | ||
a5e7ed32 | 2520 | static int cgroup_create_file(struct dentry *dentry, umode_t mode, |
5adcee1d NP |
2521 | struct super_block *sb) |
2522 | { | |
ddbcc7e8 PM |
2523 | struct inode *inode; |
2524 | ||
2525 | if (!dentry) | |
2526 | return -ENOENT; | |
2527 | if (dentry->d_inode) | |
2528 | return -EEXIST; | |
2529 | ||
2530 | inode = cgroup_new_inode(mode, sb); | |
2531 | if (!inode) | |
2532 | return -ENOMEM; | |
2533 | ||
2534 | if (S_ISDIR(mode)) { | |
2535 | inode->i_op = &cgroup_dir_inode_operations; | |
2536 | inode->i_fop = &simple_dir_operations; | |
2537 | ||
2538 | /* start off with i_nlink == 2 (for "." entry) */ | |
2539 | inc_nlink(inode); | |
2540 | ||
2541 | /* start with the directory inode held, so that we can | |
2542 | * populate it without racing with another mkdir */ | |
817929ec | 2543 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); |
ddbcc7e8 PM |
2544 | } else if (S_ISREG(mode)) { |
2545 | inode->i_size = 0; | |
2546 | inode->i_fop = &cgroup_file_operations; | |
2547 | } | |
ddbcc7e8 PM |
2548 | d_instantiate(dentry, inode); |
2549 | dget(dentry); /* Extra count - pin the dentry in core */ | |
2550 | return 0; | |
2551 | } | |
2552 | ||
2553 | /* | |
a043e3b2 LZ |
2554 | * cgroup_create_dir - create a directory for an object. |
2555 | * @cgrp: the cgroup we create the directory for. It must have a valid | |
2556 | * ->parent field. And we are going to fill its ->dentry field. | |
2557 | * @dentry: dentry of the new cgroup | |
2558 | * @mode: mode to set on new directory. | |
ddbcc7e8 | 2559 | */ |
bd89aabc | 2560 | static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry, |
a5e7ed32 | 2561 | umode_t mode) |
ddbcc7e8 PM |
2562 | { |
2563 | struct dentry *parent; | |
2564 | int error = 0; | |
2565 | ||
bd89aabc PM |
2566 | parent = cgrp->parent->dentry; |
2567 | error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb); | |
ddbcc7e8 | 2568 | if (!error) { |
bd89aabc | 2569 | dentry->d_fsdata = cgrp; |
ddbcc7e8 | 2570 | inc_nlink(parent->d_inode); |
a47295e6 | 2571 | rcu_assign_pointer(cgrp->dentry, dentry); |
ddbcc7e8 PM |
2572 | dget(dentry); |
2573 | } | |
2574 | dput(dentry); | |
2575 | ||
2576 | return error; | |
2577 | } | |
2578 | ||
099fca32 LZ |
2579 | /** |
2580 | * cgroup_file_mode - deduce file mode of a control file | |
2581 | * @cft: the control file in question | |
2582 | * | |
2583 | * returns cft->mode if ->mode is not 0 | |
2584 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
2585 | * returns S_IRUGO if it has only a read handler | |
2586 | * returns S_IWUSR if it has only a write hander | |
2587 | */ | |
a5e7ed32 | 2588 | static umode_t cgroup_file_mode(const struct cftype *cft) |
099fca32 | 2589 | { |
a5e7ed32 | 2590 | umode_t mode = 0; |
099fca32 LZ |
2591 | |
2592 | if (cft->mode) | |
2593 | return cft->mode; | |
2594 | ||
2595 | if (cft->read || cft->read_u64 || cft->read_s64 || | |
2596 | cft->read_map || cft->read_seq_string) | |
2597 | mode |= S_IRUGO; | |
2598 | ||
2599 | if (cft->write || cft->write_u64 || cft->write_s64 || | |
2600 | cft->write_string || cft->trigger) | |
2601 | mode |= S_IWUSR; | |
2602 | ||
2603 | return mode; | |
2604 | } | |
2605 | ||
bd89aabc | 2606 | int cgroup_add_file(struct cgroup *cgrp, |
ddbcc7e8 PM |
2607 | struct cgroup_subsys *subsys, |
2608 | const struct cftype *cft) | |
2609 | { | |
bd89aabc | 2610 | struct dentry *dir = cgrp->dentry; |
ddbcc7e8 PM |
2611 | struct dentry *dentry; |
2612 | int error; | |
a5e7ed32 | 2613 | umode_t mode; |
ddbcc7e8 PM |
2614 | |
2615 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; | |
bd89aabc | 2616 | if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) { |
ddbcc7e8 PM |
2617 | strcpy(name, subsys->name); |
2618 | strcat(name, "."); | |
2619 | } | |
2620 | strcat(name, cft->name); | |
2621 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); | |
2622 | dentry = lookup_one_len(name, dir, strlen(name)); | |
2623 | if (!IS_ERR(dentry)) { | |
099fca32 LZ |
2624 | mode = cgroup_file_mode(cft); |
2625 | error = cgroup_create_file(dentry, mode | S_IFREG, | |
bd89aabc | 2626 | cgrp->root->sb); |
ddbcc7e8 PM |
2627 | if (!error) |
2628 | dentry->d_fsdata = (void *)cft; | |
2629 | dput(dentry); | |
2630 | } else | |
2631 | error = PTR_ERR(dentry); | |
2632 | return error; | |
2633 | } | |
e6a1105b | 2634 | EXPORT_SYMBOL_GPL(cgroup_add_file); |
ddbcc7e8 | 2635 | |
bd89aabc | 2636 | int cgroup_add_files(struct cgroup *cgrp, |
ddbcc7e8 PM |
2637 | struct cgroup_subsys *subsys, |
2638 | const struct cftype cft[], | |
2639 | int count) | |
2640 | { | |
2641 | int i, err; | |
2642 | for (i = 0; i < count; i++) { | |
bd89aabc | 2643 | err = cgroup_add_file(cgrp, subsys, &cft[i]); |
ddbcc7e8 PM |
2644 | if (err) |
2645 | return err; | |
2646 | } | |
2647 | return 0; | |
2648 | } | |
e6a1105b | 2649 | EXPORT_SYMBOL_GPL(cgroup_add_files); |
ddbcc7e8 | 2650 | |
a043e3b2 LZ |
2651 | /** |
2652 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2653 | * @cgrp: the cgroup in question | |
2654 | * | |
2655 | * Return the number of tasks in the cgroup. | |
2656 | */ | |
bd89aabc | 2657 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2658 | { |
2659 | int count = 0; | |
71cbb949 | 2660 | struct cg_cgroup_link *link; |
817929ec PM |
2661 | |
2662 | read_lock(&css_set_lock); | |
71cbb949 | 2663 | list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) { |
146aa1bd | 2664 | count += atomic_read(&link->cg->refcount); |
817929ec PM |
2665 | } |
2666 | read_unlock(&css_set_lock); | |
bbcb81d0 PM |
2667 | return count; |
2668 | } | |
2669 | ||
817929ec PM |
2670 | /* |
2671 | * Advance a list_head iterator. The iterator should be positioned at | |
2672 | * the start of a css_set | |
2673 | */ | |
bd89aabc | 2674 | static void cgroup_advance_iter(struct cgroup *cgrp, |
7717f7ba | 2675 | struct cgroup_iter *it) |
817929ec PM |
2676 | { |
2677 | struct list_head *l = it->cg_link; | |
2678 | struct cg_cgroup_link *link; | |
2679 | struct css_set *cg; | |
2680 | ||
2681 | /* Advance to the next non-empty css_set */ | |
2682 | do { | |
2683 | l = l->next; | |
bd89aabc | 2684 | if (l == &cgrp->css_sets) { |
817929ec PM |
2685 | it->cg_link = NULL; |
2686 | return; | |
2687 | } | |
bd89aabc | 2688 | link = list_entry(l, struct cg_cgroup_link, cgrp_link_list); |
817929ec PM |
2689 | cg = link->cg; |
2690 | } while (list_empty(&cg->tasks)); | |
2691 | it->cg_link = l; | |
2692 | it->task = cg->tasks.next; | |
2693 | } | |
2694 | ||
31a7df01 CW |
2695 | /* |
2696 | * To reduce the fork() overhead for systems that are not actually | |
2697 | * using their cgroups capability, we don't maintain the lists running | |
2698 | * through each css_set to its tasks until we see the list actually | |
2699 | * used - in other words after the first call to cgroup_iter_start(). | |
31a7df01 | 2700 | */ |
3df91fe3 | 2701 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2702 | { |
2703 | struct task_struct *p, *g; | |
2704 | write_lock(&css_set_lock); | |
2705 | use_task_css_set_links = 1; | |
3ce3230a FW |
2706 | /* |
2707 | * We need tasklist_lock because RCU is not safe against | |
2708 | * while_each_thread(). Besides, a forking task that has passed | |
2709 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
2710 | * is not guaranteed to have its child immediately visible in the | |
2711 | * tasklist if we walk through it with RCU. | |
2712 | */ | |
2713 | read_lock(&tasklist_lock); | |
31a7df01 CW |
2714 | do_each_thread(g, p) { |
2715 | task_lock(p); | |
0e04388f LZ |
2716 | /* |
2717 | * We should check if the process is exiting, otherwise | |
2718 | * it will race with cgroup_exit() in that the list | |
2719 | * entry won't be deleted though the process has exited. | |
2720 | */ | |
2721 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
31a7df01 CW |
2722 | list_add(&p->cg_list, &p->cgroups->tasks); |
2723 | task_unlock(p); | |
2724 | } while_each_thread(g, p); | |
3ce3230a | 2725 | read_unlock(&tasklist_lock); |
31a7df01 CW |
2726 | write_unlock(&css_set_lock); |
2727 | } | |
2728 | ||
bd89aabc | 2729 | void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it) |
c6ca5750 | 2730 | __acquires(css_set_lock) |
817929ec PM |
2731 | { |
2732 | /* | |
2733 | * The first time anyone tries to iterate across a cgroup, | |
2734 | * we need to enable the list linking each css_set to its | |
2735 | * tasks, and fix up all existing tasks. | |
2736 | */ | |
31a7df01 CW |
2737 | if (!use_task_css_set_links) |
2738 | cgroup_enable_task_cg_lists(); | |
2739 | ||
817929ec | 2740 | read_lock(&css_set_lock); |
bd89aabc PM |
2741 | it->cg_link = &cgrp->css_sets; |
2742 | cgroup_advance_iter(cgrp, it); | |
817929ec PM |
2743 | } |
2744 | ||
bd89aabc | 2745 | struct task_struct *cgroup_iter_next(struct cgroup *cgrp, |
817929ec PM |
2746 | struct cgroup_iter *it) |
2747 | { | |
2748 | struct task_struct *res; | |
2749 | struct list_head *l = it->task; | |
2019f634 | 2750 | struct cg_cgroup_link *link; |
817929ec PM |
2751 | |
2752 | /* If the iterator cg is NULL, we have no tasks */ | |
2753 | if (!it->cg_link) | |
2754 | return NULL; | |
2755 | res = list_entry(l, struct task_struct, cg_list); | |
2756 | /* Advance iterator to find next entry */ | |
2757 | l = l->next; | |
2019f634 LJ |
2758 | link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list); |
2759 | if (l == &link->cg->tasks) { | |
817929ec PM |
2760 | /* We reached the end of this task list - move on to |
2761 | * the next cg_cgroup_link */ | |
bd89aabc | 2762 | cgroup_advance_iter(cgrp, it); |
817929ec PM |
2763 | } else { |
2764 | it->task = l; | |
2765 | } | |
2766 | return res; | |
2767 | } | |
2768 | ||
bd89aabc | 2769 | void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it) |
c6ca5750 | 2770 | __releases(css_set_lock) |
817929ec PM |
2771 | { |
2772 | read_unlock(&css_set_lock); | |
2773 | } | |
2774 | ||
31a7df01 CW |
2775 | static inline int started_after_time(struct task_struct *t1, |
2776 | struct timespec *time, | |
2777 | struct task_struct *t2) | |
2778 | { | |
2779 | int start_diff = timespec_compare(&t1->start_time, time); | |
2780 | if (start_diff > 0) { | |
2781 | return 1; | |
2782 | } else if (start_diff < 0) { | |
2783 | return 0; | |
2784 | } else { | |
2785 | /* | |
2786 | * Arbitrarily, if two processes started at the same | |
2787 | * time, we'll say that the lower pointer value | |
2788 | * started first. Note that t2 may have exited by now | |
2789 | * so this may not be a valid pointer any longer, but | |
2790 | * that's fine - it still serves to distinguish | |
2791 | * between two tasks started (effectively) simultaneously. | |
2792 | */ | |
2793 | return t1 > t2; | |
2794 | } | |
2795 | } | |
2796 | ||
2797 | /* | |
2798 | * This function is a callback from heap_insert() and is used to order | |
2799 | * the heap. | |
2800 | * In this case we order the heap in descending task start time. | |
2801 | */ | |
2802 | static inline int started_after(void *p1, void *p2) | |
2803 | { | |
2804 | struct task_struct *t1 = p1; | |
2805 | struct task_struct *t2 = p2; | |
2806 | return started_after_time(t1, &t2->start_time, t2); | |
2807 | } | |
2808 | ||
2809 | /** | |
2810 | * cgroup_scan_tasks - iterate though all the tasks in a cgroup | |
2811 | * @scan: struct cgroup_scanner containing arguments for the scan | |
2812 | * | |
2813 | * Arguments include pointers to callback functions test_task() and | |
2814 | * process_task(). | |
2815 | * Iterate through all the tasks in a cgroup, calling test_task() for each, | |
2816 | * and if it returns true, call process_task() for it also. | |
2817 | * The test_task pointer may be NULL, meaning always true (select all tasks). | |
2818 | * Effectively duplicates cgroup_iter_{start,next,end}() | |
2819 | * but does not lock css_set_lock for the call to process_task(). | |
2820 | * The struct cgroup_scanner may be embedded in any structure of the caller's | |
2821 | * creation. | |
2822 | * It is guaranteed that process_task() will act on every task that | |
2823 | * is a member of the cgroup for the duration of this call. This | |
2824 | * function may or may not call process_task() for tasks that exit | |
2825 | * or move to a different cgroup during the call, or are forked or | |
2826 | * move into the cgroup during the call. | |
2827 | * | |
2828 | * Note that test_task() may be called with locks held, and may in some | |
2829 | * situations be called multiple times for the same task, so it should | |
2830 | * be cheap. | |
2831 | * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been | |
2832 | * pre-allocated and will be used for heap operations (and its "gt" member will | |
2833 | * be overwritten), else a temporary heap will be used (allocation of which | |
2834 | * may cause this function to fail). | |
2835 | */ | |
2836 | int cgroup_scan_tasks(struct cgroup_scanner *scan) | |
2837 | { | |
2838 | int retval, i; | |
2839 | struct cgroup_iter it; | |
2840 | struct task_struct *p, *dropped; | |
2841 | /* Never dereference latest_task, since it's not refcounted */ | |
2842 | struct task_struct *latest_task = NULL; | |
2843 | struct ptr_heap tmp_heap; | |
2844 | struct ptr_heap *heap; | |
2845 | struct timespec latest_time = { 0, 0 }; | |
2846 | ||
2847 | if (scan->heap) { | |
2848 | /* The caller supplied our heap and pre-allocated its memory */ | |
2849 | heap = scan->heap; | |
2850 | heap->gt = &started_after; | |
2851 | } else { | |
2852 | /* We need to allocate our own heap memory */ | |
2853 | heap = &tmp_heap; | |
2854 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
2855 | if (retval) | |
2856 | /* cannot allocate the heap */ | |
2857 | return retval; | |
2858 | } | |
2859 | ||
2860 | again: | |
2861 | /* | |
2862 | * Scan tasks in the cgroup, using the scanner's "test_task" callback | |
2863 | * to determine which are of interest, and using the scanner's | |
2864 | * "process_task" callback to process any of them that need an update. | |
2865 | * Since we don't want to hold any locks during the task updates, | |
2866 | * gather tasks to be processed in a heap structure. | |
2867 | * The heap is sorted by descending task start time. | |
2868 | * If the statically-sized heap fills up, we overflow tasks that | |
2869 | * started later, and in future iterations only consider tasks that | |
2870 | * started after the latest task in the previous pass. This | |
2871 | * guarantees forward progress and that we don't miss any tasks. | |
2872 | */ | |
2873 | heap->size = 0; | |
2874 | cgroup_iter_start(scan->cg, &it); | |
2875 | while ((p = cgroup_iter_next(scan->cg, &it))) { | |
2876 | /* | |
2877 | * Only affect tasks that qualify per the caller's callback, | |
2878 | * if he provided one | |
2879 | */ | |
2880 | if (scan->test_task && !scan->test_task(p, scan)) | |
2881 | continue; | |
2882 | /* | |
2883 | * Only process tasks that started after the last task | |
2884 | * we processed | |
2885 | */ | |
2886 | if (!started_after_time(p, &latest_time, latest_task)) | |
2887 | continue; | |
2888 | dropped = heap_insert(heap, p); | |
2889 | if (dropped == NULL) { | |
2890 | /* | |
2891 | * The new task was inserted; the heap wasn't | |
2892 | * previously full | |
2893 | */ | |
2894 | get_task_struct(p); | |
2895 | } else if (dropped != p) { | |
2896 | /* | |
2897 | * The new task was inserted, and pushed out a | |
2898 | * different task | |
2899 | */ | |
2900 | get_task_struct(p); | |
2901 | put_task_struct(dropped); | |
2902 | } | |
2903 | /* | |
2904 | * Else the new task was newer than anything already in | |
2905 | * the heap and wasn't inserted | |
2906 | */ | |
2907 | } | |
2908 | cgroup_iter_end(scan->cg, &it); | |
2909 | ||
2910 | if (heap->size) { | |
2911 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 2912 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 2913 | if (i == 0) { |
4fe91d51 PJ |
2914 | latest_time = q->start_time; |
2915 | latest_task = q; | |
31a7df01 CW |
2916 | } |
2917 | /* Process the task per the caller's callback */ | |
4fe91d51 PJ |
2918 | scan->process_task(q, scan); |
2919 | put_task_struct(q); | |
31a7df01 CW |
2920 | } |
2921 | /* | |
2922 | * If we had to process any tasks at all, scan again | |
2923 | * in case some of them were in the middle of forking | |
2924 | * children that didn't get processed. | |
2925 | * Not the most efficient way to do it, but it avoids | |
2926 | * having to take callback_mutex in the fork path | |
2927 | */ | |
2928 | goto again; | |
2929 | } | |
2930 | if (heap == &tmp_heap) | |
2931 | heap_free(&tmp_heap); | |
2932 | return 0; | |
2933 | } | |
2934 | ||
bbcb81d0 | 2935 | /* |
102a775e | 2936 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
2937 | * |
2938 | * Reading this file can return large amounts of data if a cgroup has | |
2939 | * *lots* of attached tasks. So it may need several calls to read(), | |
2940 | * but we cannot guarantee that the information we produce is correct | |
2941 | * unless we produce it entirely atomically. | |
2942 | * | |
bbcb81d0 | 2943 | */ |
bbcb81d0 | 2944 | |
24528255 LZ |
2945 | /* which pidlist file are we talking about? */ |
2946 | enum cgroup_filetype { | |
2947 | CGROUP_FILE_PROCS, | |
2948 | CGROUP_FILE_TASKS, | |
2949 | }; | |
2950 | ||
2951 | /* | |
2952 | * A pidlist is a list of pids that virtually represents the contents of one | |
2953 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
2954 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
2955 | * to the cgroup. | |
2956 | */ | |
2957 | struct cgroup_pidlist { | |
2958 | /* | |
2959 | * used to find which pidlist is wanted. doesn't change as long as | |
2960 | * this particular list stays in the list. | |
2961 | */ | |
2962 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
2963 | /* array of xids */ | |
2964 | pid_t *list; | |
2965 | /* how many elements the above list has */ | |
2966 | int length; | |
2967 | /* how many files are using the current array */ | |
2968 | int use_count; | |
2969 | /* each of these stored in a list by its cgroup */ | |
2970 | struct list_head links; | |
2971 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
2972 | struct cgroup *owner; | |
2973 | /* protects the other fields */ | |
2974 | struct rw_semaphore mutex; | |
2975 | }; | |
2976 | ||
d1d9fd33 BB |
2977 | /* |
2978 | * The following two functions "fix" the issue where there are more pids | |
2979 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
2980 | * TODO: replace with a kernel-wide solution to this problem | |
2981 | */ | |
2982 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
2983 | static void *pidlist_allocate(int count) | |
2984 | { | |
2985 | if (PIDLIST_TOO_LARGE(count)) | |
2986 | return vmalloc(count * sizeof(pid_t)); | |
2987 | else | |
2988 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
2989 | } | |
2990 | static void pidlist_free(void *p) | |
2991 | { | |
2992 | if (is_vmalloc_addr(p)) | |
2993 | vfree(p); | |
2994 | else | |
2995 | kfree(p); | |
2996 | } | |
2997 | static void *pidlist_resize(void *p, int newcount) | |
2998 | { | |
2999 | void *newlist; | |
3000 | /* note: if new alloc fails, old p will still be valid either way */ | |
3001 | if (is_vmalloc_addr(p)) { | |
3002 | newlist = vmalloc(newcount * sizeof(pid_t)); | |
3003 | if (!newlist) | |
3004 | return NULL; | |
3005 | memcpy(newlist, p, newcount * sizeof(pid_t)); | |
3006 | vfree(p); | |
3007 | } else { | |
3008 | newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL); | |
3009 | } | |
3010 | return newlist; | |
3011 | } | |
3012 | ||
bbcb81d0 | 3013 | /* |
102a775e BB |
3014 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
3015 | * If the new stripped list is sufficiently smaller and there's enough memory | |
3016 | * to allocate a new buffer, will let go of the unneeded memory. Returns the | |
3017 | * number of unique elements. | |
bbcb81d0 | 3018 | */ |
102a775e BB |
3019 | /* is the size difference enough that we should re-allocate the array? */ |
3020 | #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new)) | |
3021 | static int pidlist_uniq(pid_t **p, int length) | |
bbcb81d0 | 3022 | { |
102a775e BB |
3023 | int src, dest = 1; |
3024 | pid_t *list = *p; | |
3025 | pid_t *newlist; | |
3026 | ||
3027 | /* | |
3028 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3029 | * edge cases first; no work needs to be done for either | |
3030 | */ | |
3031 | if (length == 0 || length == 1) | |
3032 | return length; | |
3033 | /* src and dest walk down the list; dest counts unique elements */ | |
3034 | for (src = 1; src < length; src++) { | |
3035 | /* find next unique element */ | |
3036 | while (list[src] == list[src-1]) { | |
3037 | src++; | |
3038 | if (src == length) | |
3039 | goto after; | |
3040 | } | |
3041 | /* dest always points to where the next unique element goes */ | |
3042 | list[dest] = list[src]; | |
3043 | dest++; | |
3044 | } | |
3045 | after: | |
3046 | /* | |
3047 | * if the length difference is large enough, we want to allocate a | |
3048 | * smaller buffer to save memory. if this fails due to out of memory, | |
3049 | * we'll just stay with what we've got. | |
3050 | */ | |
3051 | if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) { | |
d1d9fd33 | 3052 | newlist = pidlist_resize(list, dest); |
102a775e BB |
3053 | if (newlist) |
3054 | *p = newlist; | |
3055 | } | |
3056 | return dest; | |
3057 | } | |
3058 | ||
3059 | static int cmppid(const void *a, const void *b) | |
3060 | { | |
3061 | return *(pid_t *)a - *(pid_t *)b; | |
3062 | } | |
3063 | ||
72a8cb30 BB |
3064 | /* |
3065 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3066 | * returns with the lock on that pidlist already held, and takes care | |
3067 | * of the use count, or returns NULL with no locks held if we're out of | |
3068 | * memory. | |
3069 | */ | |
3070 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
3071 | enum cgroup_filetype type) | |
3072 | { | |
3073 | struct cgroup_pidlist *l; | |
3074 | /* don't need task_nsproxy() if we're looking at ourself */ | |
b70cc5fd LZ |
3075 | struct pid_namespace *ns = current->nsproxy->pid_ns; |
3076 | ||
72a8cb30 BB |
3077 | /* |
3078 | * We can't drop the pidlist_mutex before taking the l->mutex in case | |
3079 | * the last ref-holder is trying to remove l from the list at the same | |
3080 | * time. Holding the pidlist_mutex precludes somebody taking whichever | |
3081 | * list we find out from under us - compare release_pid_array(). | |
3082 | */ | |
3083 | mutex_lock(&cgrp->pidlist_mutex); | |
3084 | list_for_each_entry(l, &cgrp->pidlists, links) { | |
3085 | if (l->key.type == type && l->key.ns == ns) { | |
72a8cb30 BB |
3086 | /* make sure l doesn't vanish out from under us */ |
3087 | down_write(&l->mutex); | |
3088 | mutex_unlock(&cgrp->pidlist_mutex); | |
72a8cb30 BB |
3089 | return l; |
3090 | } | |
3091 | } | |
3092 | /* entry not found; create a new one */ | |
3093 | l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); | |
3094 | if (!l) { | |
3095 | mutex_unlock(&cgrp->pidlist_mutex); | |
72a8cb30 BB |
3096 | return l; |
3097 | } | |
3098 | init_rwsem(&l->mutex); | |
3099 | down_write(&l->mutex); | |
3100 | l->key.type = type; | |
b70cc5fd | 3101 | l->key.ns = get_pid_ns(ns); |
72a8cb30 BB |
3102 | l->use_count = 0; /* don't increment here */ |
3103 | l->list = NULL; | |
3104 | l->owner = cgrp; | |
3105 | list_add(&l->links, &cgrp->pidlists); | |
3106 | mutex_unlock(&cgrp->pidlist_mutex); | |
3107 | return l; | |
3108 | } | |
3109 | ||
102a775e BB |
3110 | /* |
3111 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3112 | */ | |
72a8cb30 BB |
3113 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3114 | struct cgroup_pidlist **lp) | |
102a775e BB |
3115 | { |
3116 | pid_t *array; | |
3117 | int length; | |
3118 | int pid, n = 0; /* used for populating the array */ | |
817929ec PM |
3119 | struct cgroup_iter it; |
3120 | struct task_struct *tsk; | |
102a775e BB |
3121 | struct cgroup_pidlist *l; |
3122 | ||
3123 | /* | |
3124 | * If cgroup gets more users after we read count, we won't have | |
3125 | * enough space - tough. This race is indistinguishable to the | |
3126 | * caller from the case that the additional cgroup users didn't | |
3127 | * show up until sometime later on. | |
3128 | */ | |
3129 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3130 | array = pidlist_allocate(length); |
102a775e BB |
3131 | if (!array) |
3132 | return -ENOMEM; | |
3133 | /* now, populate the array */ | |
bd89aabc PM |
3134 | cgroup_iter_start(cgrp, &it); |
3135 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
102a775e | 3136 | if (unlikely(n == length)) |
817929ec | 3137 | break; |
102a775e | 3138 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3139 | if (type == CGROUP_FILE_PROCS) |
3140 | pid = task_tgid_vnr(tsk); | |
3141 | else | |
3142 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3143 | if (pid > 0) /* make sure to only use valid results */ |
3144 | array[n++] = pid; | |
817929ec | 3145 | } |
bd89aabc | 3146 | cgroup_iter_end(cgrp, &it); |
102a775e BB |
3147 | length = n; |
3148 | /* now sort & (if procs) strip out duplicates */ | |
3149 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3150 | if (type == CGROUP_FILE_PROCS) |
102a775e | 3151 | length = pidlist_uniq(&array, length); |
72a8cb30 BB |
3152 | l = cgroup_pidlist_find(cgrp, type); |
3153 | if (!l) { | |
d1d9fd33 | 3154 | pidlist_free(array); |
72a8cb30 | 3155 | return -ENOMEM; |
102a775e | 3156 | } |
72a8cb30 | 3157 | /* store array, freeing old if necessary - lock already held */ |
d1d9fd33 | 3158 | pidlist_free(l->list); |
102a775e BB |
3159 | l->list = array; |
3160 | l->length = length; | |
3161 | l->use_count++; | |
3162 | up_write(&l->mutex); | |
72a8cb30 | 3163 | *lp = l; |
102a775e | 3164 | return 0; |
bbcb81d0 PM |
3165 | } |
3166 | ||
846c7bb0 | 3167 | /** |
a043e3b2 | 3168 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3169 | * @stats: cgroupstats to fill information into |
3170 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3171 | * been requested. | |
a043e3b2 LZ |
3172 | * |
3173 | * Build and fill cgroupstats so that taskstats can export it to user | |
3174 | * space. | |
846c7bb0 BS |
3175 | */ |
3176 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3177 | { | |
3178 | int ret = -EINVAL; | |
bd89aabc | 3179 | struct cgroup *cgrp; |
846c7bb0 BS |
3180 | struct cgroup_iter it; |
3181 | struct task_struct *tsk; | |
33d283be | 3182 | |
846c7bb0 | 3183 | /* |
33d283be LZ |
3184 | * Validate dentry by checking the superblock operations, |
3185 | * and make sure it's a directory. | |
846c7bb0 | 3186 | */ |
33d283be LZ |
3187 | if (dentry->d_sb->s_op != &cgroup_ops || |
3188 | !S_ISDIR(dentry->d_inode->i_mode)) | |
846c7bb0 BS |
3189 | goto err; |
3190 | ||
3191 | ret = 0; | |
bd89aabc | 3192 | cgrp = dentry->d_fsdata; |
846c7bb0 | 3193 | |
bd89aabc PM |
3194 | cgroup_iter_start(cgrp, &it); |
3195 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
846c7bb0 BS |
3196 | switch (tsk->state) { |
3197 | case TASK_RUNNING: | |
3198 | stats->nr_running++; | |
3199 | break; | |
3200 | case TASK_INTERRUPTIBLE: | |
3201 | stats->nr_sleeping++; | |
3202 | break; | |
3203 | case TASK_UNINTERRUPTIBLE: | |
3204 | stats->nr_uninterruptible++; | |
3205 | break; | |
3206 | case TASK_STOPPED: | |
3207 | stats->nr_stopped++; | |
3208 | break; | |
3209 | default: | |
3210 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3211 | stats->nr_io_wait++; | |
3212 | break; | |
3213 | } | |
3214 | } | |
bd89aabc | 3215 | cgroup_iter_end(cgrp, &it); |
846c7bb0 | 3216 | |
846c7bb0 BS |
3217 | err: |
3218 | return ret; | |
3219 | } | |
3220 | ||
8f3ff208 | 3221 | |
bbcb81d0 | 3222 | /* |
102a775e | 3223 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3224 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3225 | * in the cgroup->l->list array. |
bbcb81d0 | 3226 | */ |
cc31edce | 3227 | |
102a775e | 3228 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3229 | { |
cc31edce PM |
3230 | /* |
3231 | * Initially we receive a position value that corresponds to | |
3232 | * one more than the last pid shown (or 0 on the first call or | |
3233 | * after a seek to the start). Use a binary-search to find the | |
3234 | * next pid to display, if any | |
3235 | */ | |
102a775e | 3236 | struct cgroup_pidlist *l = s->private; |
cc31edce PM |
3237 | int index = 0, pid = *pos; |
3238 | int *iter; | |
3239 | ||
102a775e | 3240 | down_read(&l->mutex); |
cc31edce | 3241 | if (pid) { |
102a775e | 3242 | int end = l->length; |
20777766 | 3243 | |
cc31edce PM |
3244 | while (index < end) { |
3245 | int mid = (index + end) / 2; | |
102a775e | 3246 | if (l->list[mid] == pid) { |
cc31edce PM |
3247 | index = mid; |
3248 | break; | |
102a775e | 3249 | } else if (l->list[mid] <= pid) |
cc31edce PM |
3250 | index = mid + 1; |
3251 | else | |
3252 | end = mid; | |
3253 | } | |
3254 | } | |
3255 | /* If we're off the end of the array, we're done */ | |
102a775e | 3256 | if (index >= l->length) |
cc31edce PM |
3257 | return NULL; |
3258 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3259 | iter = l->list + index; |
cc31edce PM |
3260 | *pos = *iter; |
3261 | return iter; | |
3262 | } | |
3263 | ||
102a775e | 3264 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3265 | { |
102a775e BB |
3266 | struct cgroup_pidlist *l = s->private; |
3267 | up_read(&l->mutex); | |
cc31edce PM |
3268 | } |
3269 | ||
102a775e | 3270 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3271 | { |
102a775e BB |
3272 | struct cgroup_pidlist *l = s->private; |
3273 | pid_t *p = v; | |
3274 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3275 | /* |
3276 | * Advance to the next pid in the array. If this goes off the | |
3277 | * end, we're done | |
3278 | */ | |
3279 | p++; | |
3280 | if (p >= end) { | |
3281 | return NULL; | |
3282 | } else { | |
3283 | *pos = *p; | |
3284 | return p; | |
3285 | } | |
3286 | } | |
3287 | ||
102a775e | 3288 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3289 | { |
3290 | return seq_printf(s, "%d\n", *(int *)v); | |
3291 | } | |
bbcb81d0 | 3292 | |
102a775e BB |
3293 | /* |
3294 | * seq_operations functions for iterating on pidlists through seq_file - | |
3295 | * independent of whether it's tasks or procs | |
3296 | */ | |
3297 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3298 | .start = cgroup_pidlist_start, | |
3299 | .stop = cgroup_pidlist_stop, | |
3300 | .next = cgroup_pidlist_next, | |
3301 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3302 | }; |
3303 | ||
102a775e | 3304 | static void cgroup_release_pid_array(struct cgroup_pidlist *l) |
cc31edce | 3305 | { |
72a8cb30 BB |
3306 | /* |
3307 | * the case where we're the last user of this particular pidlist will | |
3308 | * have us remove it from the cgroup's list, which entails taking the | |
3309 | * mutex. since in pidlist_find the pidlist->lock depends on cgroup-> | |
3310 | * pidlist_mutex, we have to take pidlist_mutex first. | |
3311 | */ | |
3312 | mutex_lock(&l->owner->pidlist_mutex); | |
102a775e BB |
3313 | down_write(&l->mutex); |
3314 | BUG_ON(!l->use_count); | |
3315 | if (!--l->use_count) { | |
72a8cb30 BB |
3316 | /* we're the last user if refcount is 0; remove and free */ |
3317 | list_del(&l->links); | |
3318 | mutex_unlock(&l->owner->pidlist_mutex); | |
d1d9fd33 | 3319 | pidlist_free(l->list); |
72a8cb30 BB |
3320 | put_pid_ns(l->key.ns); |
3321 | up_write(&l->mutex); | |
3322 | kfree(l); | |
3323 | return; | |
cc31edce | 3324 | } |
72a8cb30 | 3325 | mutex_unlock(&l->owner->pidlist_mutex); |
102a775e | 3326 | up_write(&l->mutex); |
bbcb81d0 PM |
3327 | } |
3328 | ||
102a775e | 3329 | static int cgroup_pidlist_release(struct inode *inode, struct file *file) |
cc31edce | 3330 | { |
102a775e | 3331 | struct cgroup_pidlist *l; |
cc31edce PM |
3332 | if (!(file->f_mode & FMODE_READ)) |
3333 | return 0; | |
102a775e BB |
3334 | /* |
3335 | * the seq_file will only be initialized if the file was opened for | |
3336 | * reading; hence we check if it's not null only in that case. | |
3337 | */ | |
3338 | l = ((struct seq_file *)file->private_data)->private; | |
3339 | cgroup_release_pid_array(l); | |
cc31edce PM |
3340 | return seq_release(inode, file); |
3341 | } | |
3342 | ||
102a775e | 3343 | static const struct file_operations cgroup_pidlist_operations = { |
cc31edce PM |
3344 | .read = seq_read, |
3345 | .llseek = seq_lseek, | |
3346 | .write = cgroup_file_write, | |
102a775e | 3347 | .release = cgroup_pidlist_release, |
cc31edce PM |
3348 | }; |
3349 | ||
bbcb81d0 | 3350 | /* |
102a775e BB |
3351 | * The following functions handle opens on a file that displays a pidlist |
3352 | * (tasks or procs). Prepare an array of the process/thread IDs of whoever's | |
3353 | * in the cgroup. | |
bbcb81d0 | 3354 | */ |
102a775e | 3355 | /* helper function for the two below it */ |
72a8cb30 | 3356 | static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type) |
bbcb81d0 | 3357 | { |
bd89aabc | 3358 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
72a8cb30 | 3359 | struct cgroup_pidlist *l; |
cc31edce | 3360 | int retval; |
bbcb81d0 | 3361 | |
cc31edce | 3362 | /* Nothing to do for write-only files */ |
bbcb81d0 PM |
3363 | if (!(file->f_mode & FMODE_READ)) |
3364 | return 0; | |
3365 | ||
102a775e | 3366 | /* have the array populated */ |
72a8cb30 | 3367 | retval = pidlist_array_load(cgrp, type, &l); |
102a775e BB |
3368 | if (retval) |
3369 | return retval; | |
3370 | /* configure file information */ | |
3371 | file->f_op = &cgroup_pidlist_operations; | |
cc31edce | 3372 | |
102a775e | 3373 | retval = seq_open(file, &cgroup_pidlist_seq_operations); |
cc31edce | 3374 | if (retval) { |
102a775e | 3375 | cgroup_release_pid_array(l); |
cc31edce | 3376 | return retval; |
bbcb81d0 | 3377 | } |
102a775e | 3378 | ((struct seq_file *)file->private_data)->private = l; |
bbcb81d0 PM |
3379 | return 0; |
3380 | } | |
102a775e BB |
3381 | static int cgroup_tasks_open(struct inode *unused, struct file *file) |
3382 | { | |
72a8cb30 | 3383 | return cgroup_pidlist_open(file, CGROUP_FILE_TASKS); |
102a775e BB |
3384 | } |
3385 | static int cgroup_procs_open(struct inode *unused, struct file *file) | |
3386 | { | |
72a8cb30 | 3387 | return cgroup_pidlist_open(file, CGROUP_FILE_PROCS); |
102a775e | 3388 | } |
bbcb81d0 | 3389 | |
bd89aabc | 3390 | static u64 cgroup_read_notify_on_release(struct cgroup *cgrp, |
81a6a5cd PM |
3391 | struct cftype *cft) |
3392 | { | |
bd89aabc | 3393 | return notify_on_release(cgrp); |
81a6a5cd PM |
3394 | } |
3395 | ||
6379c106 PM |
3396 | static int cgroup_write_notify_on_release(struct cgroup *cgrp, |
3397 | struct cftype *cft, | |
3398 | u64 val) | |
3399 | { | |
3400 | clear_bit(CGRP_RELEASABLE, &cgrp->flags); | |
3401 | if (val) | |
3402 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3403 | else | |
3404 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3405 | return 0; | |
3406 | } | |
3407 | ||
0dea1168 KS |
3408 | /* |
3409 | * Unregister event and free resources. | |
3410 | * | |
3411 | * Gets called from workqueue. | |
3412 | */ | |
3413 | static void cgroup_event_remove(struct work_struct *work) | |
3414 | { | |
3415 | struct cgroup_event *event = container_of(work, struct cgroup_event, | |
3416 | remove); | |
3417 | struct cgroup *cgrp = event->cgrp; | |
3418 | ||
0dea1168 KS |
3419 | event->cft->unregister_event(cgrp, event->cft, event->eventfd); |
3420 | ||
3421 | eventfd_ctx_put(event->eventfd); | |
0dea1168 | 3422 | kfree(event); |
a0a4db54 | 3423 | dput(cgrp->dentry); |
0dea1168 KS |
3424 | } |
3425 | ||
3426 | /* | |
3427 | * Gets called on POLLHUP on eventfd when user closes it. | |
3428 | * | |
3429 | * Called with wqh->lock held and interrupts disabled. | |
3430 | */ | |
3431 | static int cgroup_event_wake(wait_queue_t *wait, unsigned mode, | |
3432 | int sync, void *key) | |
3433 | { | |
3434 | struct cgroup_event *event = container_of(wait, | |
3435 | struct cgroup_event, wait); | |
3436 | struct cgroup *cgrp = event->cgrp; | |
3437 | unsigned long flags = (unsigned long)key; | |
3438 | ||
3439 | if (flags & POLLHUP) { | |
a93d2f17 | 3440 | __remove_wait_queue(event->wqh, &event->wait); |
0dea1168 KS |
3441 | spin_lock(&cgrp->event_list_lock); |
3442 | list_del(&event->list); | |
3443 | spin_unlock(&cgrp->event_list_lock); | |
3444 | /* | |
3445 | * We are in atomic context, but cgroup_event_remove() may | |
3446 | * sleep, so we have to call it in workqueue. | |
3447 | */ | |
3448 | schedule_work(&event->remove); | |
3449 | } | |
3450 | ||
3451 | return 0; | |
3452 | } | |
3453 | ||
3454 | static void cgroup_event_ptable_queue_proc(struct file *file, | |
3455 | wait_queue_head_t *wqh, poll_table *pt) | |
3456 | { | |
3457 | struct cgroup_event *event = container_of(pt, | |
3458 | struct cgroup_event, pt); | |
3459 | ||
3460 | event->wqh = wqh; | |
3461 | add_wait_queue(wqh, &event->wait); | |
3462 | } | |
3463 | ||
3464 | /* | |
3465 | * Parse input and register new cgroup event handler. | |
3466 | * | |
3467 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3468 | * Interpretation of args is defined by control file implementation. | |
3469 | */ | |
3470 | static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft, | |
3471 | const char *buffer) | |
3472 | { | |
3473 | struct cgroup_event *event = NULL; | |
3474 | unsigned int efd, cfd; | |
3475 | struct file *efile = NULL; | |
3476 | struct file *cfile = NULL; | |
3477 | char *endp; | |
3478 | int ret; | |
3479 | ||
3480 | efd = simple_strtoul(buffer, &endp, 10); | |
3481 | if (*endp != ' ') | |
3482 | return -EINVAL; | |
3483 | buffer = endp + 1; | |
3484 | ||
3485 | cfd = simple_strtoul(buffer, &endp, 10); | |
3486 | if ((*endp != ' ') && (*endp != '\0')) | |
3487 | return -EINVAL; | |
3488 | buffer = endp + 1; | |
3489 | ||
3490 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3491 | if (!event) | |
3492 | return -ENOMEM; | |
3493 | event->cgrp = cgrp; | |
3494 | INIT_LIST_HEAD(&event->list); | |
3495 | init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc); | |
3496 | init_waitqueue_func_entry(&event->wait, cgroup_event_wake); | |
3497 | INIT_WORK(&event->remove, cgroup_event_remove); | |
3498 | ||
3499 | efile = eventfd_fget(efd); | |
3500 | if (IS_ERR(efile)) { | |
3501 | ret = PTR_ERR(efile); | |
3502 | goto fail; | |
3503 | } | |
3504 | ||
3505 | event->eventfd = eventfd_ctx_fileget(efile); | |
3506 | if (IS_ERR(event->eventfd)) { | |
3507 | ret = PTR_ERR(event->eventfd); | |
3508 | goto fail; | |
3509 | } | |
3510 | ||
3511 | cfile = fget(cfd); | |
3512 | if (!cfile) { | |
3513 | ret = -EBADF; | |
3514 | goto fail; | |
3515 | } | |
3516 | ||
3517 | /* the process need read permission on control file */ | |
3bfa784a AV |
3518 | /* AV: shouldn't we check that it's been opened for read instead? */ |
3519 | ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ); | |
0dea1168 KS |
3520 | if (ret < 0) |
3521 | goto fail; | |
3522 | ||
3523 | event->cft = __file_cft(cfile); | |
3524 | if (IS_ERR(event->cft)) { | |
3525 | ret = PTR_ERR(event->cft); | |
3526 | goto fail; | |
3527 | } | |
3528 | ||
3529 | if (!event->cft->register_event || !event->cft->unregister_event) { | |
3530 | ret = -EINVAL; | |
3531 | goto fail; | |
3532 | } | |
3533 | ||
3534 | ret = event->cft->register_event(cgrp, event->cft, | |
3535 | event->eventfd, buffer); | |
3536 | if (ret) | |
3537 | goto fail; | |
3538 | ||
3539 | if (efile->f_op->poll(efile, &event->pt) & POLLHUP) { | |
3540 | event->cft->unregister_event(cgrp, event->cft, event->eventfd); | |
3541 | ret = 0; | |
3542 | goto fail; | |
3543 | } | |
3544 | ||
a0a4db54 KS |
3545 | /* |
3546 | * Events should be removed after rmdir of cgroup directory, but before | |
3547 | * destroying subsystem state objects. Let's take reference to cgroup | |
3548 | * directory dentry to do that. | |
3549 | */ | |
3550 | dget(cgrp->dentry); | |
3551 | ||
0dea1168 KS |
3552 | spin_lock(&cgrp->event_list_lock); |
3553 | list_add(&event->list, &cgrp->event_list); | |
3554 | spin_unlock(&cgrp->event_list_lock); | |
3555 | ||
3556 | fput(cfile); | |
3557 | fput(efile); | |
3558 | ||
3559 | return 0; | |
3560 | ||
3561 | fail: | |
3562 | if (cfile) | |
3563 | fput(cfile); | |
3564 | ||
3565 | if (event && event->eventfd && !IS_ERR(event->eventfd)) | |
3566 | eventfd_ctx_put(event->eventfd); | |
3567 | ||
3568 | if (!IS_ERR_OR_NULL(efile)) | |
3569 | fput(efile); | |
3570 | ||
3571 | kfree(event); | |
3572 | ||
3573 | return ret; | |
3574 | } | |
3575 | ||
97978e6d DL |
3576 | static u64 cgroup_clone_children_read(struct cgroup *cgrp, |
3577 | struct cftype *cft) | |
3578 | { | |
3579 | return clone_children(cgrp); | |
3580 | } | |
3581 | ||
3582 | static int cgroup_clone_children_write(struct cgroup *cgrp, | |
3583 | struct cftype *cft, | |
3584 | u64 val) | |
3585 | { | |
3586 | if (val) | |
3587 | set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); | |
3588 | else | |
3589 | clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); | |
3590 | return 0; | |
3591 | } | |
3592 | ||
bbcb81d0 PM |
3593 | /* |
3594 | * for the common functions, 'private' gives the type of file | |
3595 | */ | |
102a775e BB |
3596 | /* for hysterical raisins, we can't put this on the older files */ |
3597 | #define CGROUP_FILE_GENERIC_PREFIX "cgroup." | |
81a6a5cd PM |
3598 | static struct cftype files[] = { |
3599 | { | |
3600 | .name = "tasks", | |
3601 | .open = cgroup_tasks_open, | |
af351026 | 3602 | .write_u64 = cgroup_tasks_write, |
102a775e | 3603 | .release = cgroup_pidlist_release, |
099fca32 | 3604 | .mode = S_IRUGO | S_IWUSR, |
81a6a5cd | 3605 | }, |
102a775e BB |
3606 | { |
3607 | .name = CGROUP_FILE_GENERIC_PREFIX "procs", | |
3608 | .open = cgroup_procs_open, | |
74a1166d | 3609 | .write_u64 = cgroup_procs_write, |
102a775e | 3610 | .release = cgroup_pidlist_release, |
74a1166d | 3611 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3612 | }, |
81a6a5cd PM |
3613 | { |
3614 | .name = "notify_on_release", | |
f4c753b7 | 3615 | .read_u64 = cgroup_read_notify_on_release, |
6379c106 | 3616 | .write_u64 = cgroup_write_notify_on_release, |
81a6a5cd | 3617 | }, |
0dea1168 KS |
3618 | { |
3619 | .name = CGROUP_FILE_GENERIC_PREFIX "event_control", | |
3620 | .write_string = cgroup_write_event_control, | |
3621 | .mode = S_IWUGO, | |
3622 | }, | |
97978e6d DL |
3623 | { |
3624 | .name = "cgroup.clone_children", | |
3625 | .read_u64 = cgroup_clone_children_read, | |
3626 | .write_u64 = cgroup_clone_children_write, | |
3627 | }, | |
81a6a5cd PM |
3628 | }; |
3629 | ||
3630 | static struct cftype cft_release_agent = { | |
3631 | .name = "release_agent", | |
e788e066 PM |
3632 | .read_seq_string = cgroup_release_agent_show, |
3633 | .write_string = cgroup_release_agent_write, | |
3634 | .max_write_len = PATH_MAX, | |
bbcb81d0 PM |
3635 | }; |
3636 | ||
bd89aabc | 3637 | static int cgroup_populate_dir(struct cgroup *cgrp) |
ddbcc7e8 PM |
3638 | { |
3639 | int err; | |
3640 | struct cgroup_subsys *ss; | |
3641 | ||
3642 | /* First clear out any existing files */ | |
bd89aabc | 3643 | cgroup_clear_directory(cgrp->dentry); |
ddbcc7e8 | 3644 | |
bd89aabc | 3645 | err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files)); |
bbcb81d0 PM |
3646 | if (err < 0) |
3647 | return err; | |
3648 | ||
bd89aabc PM |
3649 | if (cgrp == cgrp->top_cgroup) { |
3650 | if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0) | |
81a6a5cd PM |
3651 | return err; |
3652 | } | |
3653 | ||
bd89aabc PM |
3654 | for_each_subsys(cgrp->root, ss) { |
3655 | if (ss->populate && (err = ss->populate(ss, cgrp)) < 0) | |
ddbcc7e8 PM |
3656 | return err; |
3657 | } | |
38460b48 KH |
3658 | /* This cgroup is ready now */ |
3659 | for_each_subsys(cgrp->root, ss) { | |
3660 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3661 | /* | |
3662 | * Update id->css pointer and make this css visible from | |
3663 | * CSS ID functions. This pointer will be dereferened | |
3664 | * from RCU-read-side without locks. | |
3665 | */ | |
3666 | if (css->id) | |
3667 | rcu_assign_pointer(css->id->css, css); | |
3668 | } | |
ddbcc7e8 PM |
3669 | |
3670 | return 0; | |
3671 | } | |
3672 | ||
3673 | static void init_cgroup_css(struct cgroup_subsys_state *css, | |
3674 | struct cgroup_subsys *ss, | |
bd89aabc | 3675 | struct cgroup *cgrp) |
ddbcc7e8 | 3676 | { |
bd89aabc | 3677 | css->cgroup = cgrp; |
e7c5ec91 | 3678 | atomic_set(&css->refcnt, 1); |
ddbcc7e8 | 3679 | css->flags = 0; |
38460b48 | 3680 | css->id = NULL; |
bd89aabc | 3681 | if (cgrp == dummytop) |
ddbcc7e8 | 3682 | set_bit(CSS_ROOT, &css->flags); |
bd89aabc PM |
3683 | BUG_ON(cgrp->subsys[ss->subsys_id]); |
3684 | cgrp->subsys[ss->subsys_id] = css; | |
ddbcc7e8 PM |
3685 | } |
3686 | ||
999cd8a4 PM |
3687 | static void cgroup_lock_hierarchy(struct cgroupfs_root *root) |
3688 | { | |
3689 | /* We need to take each hierarchy_mutex in a consistent order */ | |
3690 | int i; | |
3691 | ||
aae8aab4 BB |
3692 | /* |
3693 | * No worry about a race with rebind_subsystems that might mess up the | |
3694 | * locking order, since both parties are under cgroup_mutex. | |
3695 | */ | |
999cd8a4 PM |
3696 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
3697 | struct cgroup_subsys *ss = subsys[i]; | |
aae8aab4 BB |
3698 | if (ss == NULL) |
3699 | continue; | |
999cd8a4 | 3700 | if (ss->root == root) |
cfebe563 | 3701 | mutex_lock(&ss->hierarchy_mutex); |
999cd8a4 PM |
3702 | } |
3703 | } | |
3704 | ||
3705 | static void cgroup_unlock_hierarchy(struct cgroupfs_root *root) | |
3706 | { | |
3707 | int i; | |
3708 | ||
3709 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
3710 | struct cgroup_subsys *ss = subsys[i]; | |
aae8aab4 BB |
3711 | if (ss == NULL) |
3712 | continue; | |
999cd8a4 PM |
3713 | if (ss->root == root) |
3714 | mutex_unlock(&ss->hierarchy_mutex); | |
3715 | } | |
3716 | } | |
3717 | ||
ddbcc7e8 | 3718 | /* |
a043e3b2 LZ |
3719 | * cgroup_create - create a cgroup |
3720 | * @parent: cgroup that will be parent of the new cgroup | |
3721 | * @dentry: dentry of the new cgroup | |
3722 | * @mode: mode to set on new inode | |
ddbcc7e8 | 3723 | * |
a043e3b2 | 3724 | * Must be called with the mutex on the parent inode held |
ddbcc7e8 | 3725 | */ |
ddbcc7e8 | 3726 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, |
a5e7ed32 | 3727 | umode_t mode) |
ddbcc7e8 | 3728 | { |
bd89aabc | 3729 | struct cgroup *cgrp; |
ddbcc7e8 PM |
3730 | struct cgroupfs_root *root = parent->root; |
3731 | int err = 0; | |
3732 | struct cgroup_subsys *ss; | |
3733 | struct super_block *sb = root->sb; | |
3734 | ||
bd89aabc PM |
3735 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
3736 | if (!cgrp) | |
ddbcc7e8 PM |
3737 | return -ENOMEM; |
3738 | ||
3739 | /* Grab a reference on the superblock so the hierarchy doesn't | |
3740 | * get deleted on unmount if there are child cgroups. This | |
3741 | * can be done outside cgroup_mutex, since the sb can't | |
3742 | * disappear while someone has an open control file on the | |
3743 | * fs */ | |
3744 | atomic_inc(&sb->s_active); | |
3745 | ||
3746 | mutex_lock(&cgroup_mutex); | |
3747 | ||
cc31edce | 3748 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3749 | |
bd89aabc PM |
3750 | cgrp->parent = parent; |
3751 | cgrp->root = parent->root; | |
3752 | cgrp->top_cgroup = parent->top_cgroup; | |
ddbcc7e8 | 3753 | |
b6abdb0e LZ |
3754 | if (notify_on_release(parent)) |
3755 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3756 | ||
97978e6d DL |
3757 | if (clone_children(parent)) |
3758 | set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); | |
3759 | ||
ddbcc7e8 | 3760 | for_each_subsys(root, ss) { |
761b3ef5 | 3761 | struct cgroup_subsys_state *css = ss->create(cgrp); |
4528fd05 | 3762 | |
ddbcc7e8 PM |
3763 | if (IS_ERR(css)) { |
3764 | err = PTR_ERR(css); | |
3765 | goto err_destroy; | |
3766 | } | |
bd89aabc | 3767 | init_cgroup_css(css, ss, cgrp); |
4528fd05 LZ |
3768 | if (ss->use_id) { |
3769 | err = alloc_css_id(ss, parent, cgrp); | |
3770 | if (err) | |
38460b48 | 3771 | goto err_destroy; |
4528fd05 | 3772 | } |
38460b48 | 3773 | /* At error, ->destroy() callback has to free assigned ID. */ |
97978e6d | 3774 | if (clone_children(parent) && ss->post_clone) |
761b3ef5 | 3775 | ss->post_clone(cgrp); |
ddbcc7e8 PM |
3776 | } |
3777 | ||
999cd8a4 | 3778 | cgroup_lock_hierarchy(root); |
bd89aabc | 3779 | list_add(&cgrp->sibling, &cgrp->parent->children); |
999cd8a4 | 3780 | cgroup_unlock_hierarchy(root); |
ddbcc7e8 PM |
3781 | root->number_of_cgroups++; |
3782 | ||
bd89aabc | 3783 | err = cgroup_create_dir(cgrp, dentry, mode); |
ddbcc7e8 PM |
3784 | if (err < 0) |
3785 | goto err_remove; | |
3786 | ||
3787 | /* The cgroup directory was pre-locked for us */ | |
bd89aabc | 3788 | BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex)); |
ddbcc7e8 | 3789 | |
bd89aabc | 3790 | err = cgroup_populate_dir(cgrp); |
ddbcc7e8 PM |
3791 | /* If err < 0, we have a half-filled directory - oh well ;) */ |
3792 | ||
3793 | mutex_unlock(&cgroup_mutex); | |
bd89aabc | 3794 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
3795 | |
3796 | return 0; | |
3797 | ||
3798 | err_remove: | |
3799 | ||
baef99a0 | 3800 | cgroup_lock_hierarchy(root); |
bd89aabc | 3801 | list_del(&cgrp->sibling); |
baef99a0 | 3802 | cgroup_unlock_hierarchy(root); |
ddbcc7e8 PM |
3803 | root->number_of_cgroups--; |
3804 | ||
3805 | err_destroy: | |
3806 | ||
3807 | for_each_subsys(root, ss) { | |
bd89aabc | 3808 | if (cgrp->subsys[ss->subsys_id]) |
761b3ef5 | 3809 | ss->destroy(cgrp); |
ddbcc7e8 PM |
3810 | } |
3811 | ||
3812 | mutex_unlock(&cgroup_mutex); | |
3813 | ||
3814 | /* Release the reference count that we took on the superblock */ | |
3815 | deactivate_super(sb); | |
3816 | ||
bd89aabc | 3817 | kfree(cgrp); |
ddbcc7e8 PM |
3818 | return err; |
3819 | } | |
3820 | ||
18bb1db3 | 3821 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
ddbcc7e8 PM |
3822 | { |
3823 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
3824 | ||
3825 | /* the vfs holds inode->i_mutex already */ | |
3826 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
3827 | } | |
3828 | ||
55b6fd01 | 3829 | static int cgroup_has_css_refs(struct cgroup *cgrp) |
81a6a5cd PM |
3830 | { |
3831 | /* Check the reference count on each subsystem. Since we | |
3832 | * already established that there are no tasks in the | |
e7c5ec91 | 3833 | * cgroup, if the css refcount is also 1, then there should |
81a6a5cd PM |
3834 | * be no outstanding references, so the subsystem is safe to |
3835 | * destroy. We scan across all subsystems rather than using | |
3836 | * the per-hierarchy linked list of mounted subsystems since | |
3837 | * we can be called via check_for_release() with no | |
3838 | * synchronization other than RCU, and the subsystem linked | |
3839 | * list isn't RCU-safe */ | |
3840 | int i; | |
aae8aab4 BB |
3841 | /* |
3842 | * We won't need to lock the subsys array, because the subsystems | |
3843 | * we're concerned about aren't going anywhere since our cgroup root | |
3844 | * has a reference on them. | |
3845 | */ | |
81a6a5cd PM |
3846 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
3847 | struct cgroup_subsys *ss = subsys[i]; | |
3848 | struct cgroup_subsys_state *css; | |
aae8aab4 BB |
3849 | /* Skip subsystems not present or not in this hierarchy */ |
3850 | if (ss == NULL || ss->root != cgrp->root) | |
81a6a5cd | 3851 | continue; |
bd89aabc | 3852 | css = cgrp->subsys[ss->subsys_id]; |
81a6a5cd PM |
3853 | /* When called from check_for_release() it's possible |
3854 | * that by this point the cgroup has been removed | |
3855 | * and the css deleted. But a false-positive doesn't | |
3856 | * matter, since it can only happen if the cgroup | |
3857 | * has been deleted and hence no longer needs the | |
3858 | * release agent to be called anyway. */ | |
e7c5ec91 | 3859 | if (css && (atomic_read(&css->refcnt) > 1)) |
81a6a5cd | 3860 | return 1; |
81a6a5cd PM |
3861 | } |
3862 | return 0; | |
3863 | } | |
3864 | ||
e7c5ec91 PM |
3865 | /* |
3866 | * Atomically mark all (or else none) of the cgroup's CSS objects as | |
3867 | * CSS_REMOVED. Return true on success, or false if the cgroup has | |
3868 | * busy subsystems. Call with cgroup_mutex held | |
3869 | */ | |
3870 | ||
3871 | static int cgroup_clear_css_refs(struct cgroup *cgrp) | |
3872 | { | |
3873 | struct cgroup_subsys *ss; | |
3874 | unsigned long flags; | |
3875 | bool failed = false; | |
3876 | local_irq_save(flags); | |
3877 | for_each_subsys(cgrp->root, ss) { | |
3878 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3879 | int refcnt; | |
804b3c28 | 3880 | while (1) { |
e7c5ec91 PM |
3881 | /* We can only remove a CSS with a refcnt==1 */ |
3882 | refcnt = atomic_read(&css->refcnt); | |
3883 | if (refcnt > 1) { | |
3884 | failed = true; | |
3885 | goto done; | |
3886 | } | |
3887 | BUG_ON(!refcnt); | |
3888 | /* | |
3889 | * Drop the refcnt to 0 while we check other | |
3890 | * subsystems. This will cause any racing | |
3891 | * css_tryget() to spin until we set the | |
3892 | * CSS_REMOVED bits or abort | |
3893 | */ | |
804b3c28 PM |
3894 | if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt) |
3895 | break; | |
3896 | cpu_relax(); | |
3897 | } | |
e7c5ec91 PM |
3898 | } |
3899 | done: | |
3900 | for_each_subsys(cgrp->root, ss) { | |
3901 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3902 | if (failed) { | |
3903 | /* | |
3904 | * Restore old refcnt if we previously managed | |
3905 | * to clear it from 1 to 0 | |
3906 | */ | |
3907 | if (!atomic_read(&css->refcnt)) | |
3908 | atomic_set(&css->refcnt, 1); | |
3909 | } else { | |
3910 | /* Commit the fact that the CSS is removed */ | |
3911 | set_bit(CSS_REMOVED, &css->flags); | |
3912 | } | |
3913 | } | |
3914 | local_irq_restore(flags); | |
3915 | return !failed; | |
3916 | } | |
3917 | ||
ddbcc7e8 PM |
3918 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) |
3919 | { | |
bd89aabc | 3920 | struct cgroup *cgrp = dentry->d_fsdata; |
ddbcc7e8 PM |
3921 | struct dentry *d; |
3922 | struct cgroup *parent; | |
ec64f515 | 3923 | DEFINE_WAIT(wait); |
4ab78683 | 3924 | struct cgroup_event *event, *tmp; |
ec64f515 | 3925 | int ret; |
ddbcc7e8 PM |
3926 | |
3927 | /* the vfs holds both inode->i_mutex already */ | |
ec64f515 | 3928 | again: |
ddbcc7e8 | 3929 | mutex_lock(&cgroup_mutex); |
bd89aabc | 3930 | if (atomic_read(&cgrp->count) != 0) { |
ddbcc7e8 PM |
3931 | mutex_unlock(&cgroup_mutex); |
3932 | return -EBUSY; | |
3933 | } | |
bd89aabc | 3934 | if (!list_empty(&cgrp->children)) { |
ddbcc7e8 PM |
3935 | mutex_unlock(&cgroup_mutex); |
3936 | return -EBUSY; | |
3937 | } | |
3fa59dfb | 3938 | mutex_unlock(&cgroup_mutex); |
a043e3b2 | 3939 | |
88703267 KH |
3940 | /* |
3941 | * In general, subsystem has no css->refcnt after pre_destroy(). But | |
3942 | * in racy cases, subsystem may have to get css->refcnt after | |
3943 | * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes | |
3944 | * make rmdir return -EBUSY too often. To avoid that, we use waitqueue | |
3945 | * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir | |
3946 | * and subsystem's reference count handling. Please see css_get/put | |
3947 | * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation. | |
3948 | */ | |
3949 | set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
3950 | ||
4fca88c8 | 3951 | /* |
a043e3b2 LZ |
3952 | * Call pre_destroy handlers of subsys. Notify subsystems |
3953 | * that rmdir() request comes. | |
4fca88c8 | 3954 | */ |
ec64f515 | 3955 | ret = cgroup_call_pre_destroy(cgrp); |
88703267 KH |
3956 | if (ret) { |
3957 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
ec64f515 | 3958 | return ret; |
88703267 | 3959 | } |
ddbcc7e8 | 3960 | |
3fa59dfb KH |
3961 | mutex_lock(&cgroup_mutex); |
3962 | parent = cgrp->parent; | |
ec64f515 | 3963 | if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) { |
88703267 | 3964 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); |
ddbcc7e8 PM |
3965 | mutex_unlock(&cgroup_mutex); |
3966 | return -EBUSY; | |
3967 | } | |
ec64f515 | 3968 | prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE); |
ec64f515 KH |
3969 | if (!cgroup_clear_css_refs(cgrp)) { |
3970 | mutex_unlock(&cgroup_mutex); | |
88703267 KH |
3971 | /* |
3972 | * Because someone may call cgroup_wakeup_rmdir_waiter() before | |
3973 | * prepare_to_wait(), we need to check this flag. | |
3974 | */ | |
3975 | if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)) | |
3976 | schedule(); | |
ec64f515 KH |
3977 | finish_wait(&cgroup_rmdir_waitq, &wait); |
3978 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
3979 | if (signal_pending(current)) | |
3980 | return -EINTR; | |
3981 | goto again; | |
3982 | } | |
3983 | /* NO css_tryget() can success after here. */ | |
3984 | finish_wait(&cgroup_rmdir_waitq, &wait); | |
3985 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
ddbcc7e8 | 3986 | |
cdcc136f | 3987 | raw_spin_lock(&release_list_lock); |
bd89aabc PM |
3988 | set_bit(CGRP_REMOVED, &cgrp->flags); |
3989 | if (!list_empty(&cgrp->release_list)) | |
8d258797 | 3990 | list_del_init(&cgrp->release_list); |
cdcc136f | 3991 | raw_spin_unlock(&release_list_lock); |
999cd8a4 PM |
3992 | |
3993 | cgroup_lock_hierarchy(cgrp->root); | |
3994 | /* delete this cgroup from parent->children */ | |
8d258797 | 3995 | list_del_init(&cgrp->sibling); |
999cd8a4 PM |
3996 | cgroup_unlock_hierarchy(cgrp->root); |
3997 | ||
bd89aabc | 3998 | d = dget(cgrp->dentry); |
ddbcc7e8 PM |
3999 | |
4000 | cgroup_d_remove_dir(d); | |
4001 | dput(d); | |
ddbcc7e8 | 4002 | |
bd89aabc | 4003 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd PM |
4004 | check_for_release(parent); |
4005 | ||
4ab78683 KS |
4006 | /* |
4007 | * Unregister events and notify userspace. | |
4008 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4009 | * directory to avoid race between userspace and kernelspace | |
4010 | */ | |
4011 | spin_lock(&cgrp->event_list_lock); | |
4012 | list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) { | |
4013 | list_del(&event->list); | |
4014 | remove_wait_queue(event->wqh, &event->wait); | |
4015 | eventfd_signal(event->eventfd, 1); | |
4016 | schedule_work(&event->remove); | |
4017 | } | |
4018 | spin_unlock(&cgrp->event_list_lock); | |
4019 | ||
ddbcc7e8 | 4020 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
4021 | return 0; |
4022 | } | |
4023 | ||
06a11920 | 4024 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4025 | { |
ddbcc7e8 | 4026 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4027 | |
4028 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 PM |
4029 | |
4030 | /* Create the top cgroup state for this subsystem */ | |
33a68ac1 | 4031 | list_add(&ss->sibling, &rootnode.subsys_list); |
ddbcc7e8 | 4032 | ss->root = &rootnode; |
761b3ef5 | 4033 | css = ss->create(dummytop); |
ddbcc7e8 PM |
4034 | /* We don't handle early failures gracefully */ |
4035 | BUG_ON(IS_ERR(css)); | |
4036 | init_cgroup_css(css, ss, dummytop); | |
4037 | ||
e8d55fde | 4038 | /* Update the init_css_set to contain a subsys |
817929ec | 4039 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
4040 | * newly registered, all tasks and hence the |
4041 | * init_css_set is in the subsystem's top cgroup. */ | |
4042 | init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id]; | |
ddbcc7e8 PM |
4043 | |
4044 | need_forkexit_callback |= ss->fork || ss->exit; | |
4045 | ||
e8d55fde LZ |
4046 | /* At system boot, before all subsystems have been |
4047 | * registered, no tasks have been forked, so we don't | |
4048 | * need to invoke fork callbacks here. */ | |
4049 | BUG_ON(!list_empty(&init_task.tasks)); | |
4050 | ||
999cd8a4 | 4051 | mutex_init(&ss->hierarchy_mutex); |
cfebe563 | 4052 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); |
ddbcc7e8 | 4053 | ss->active = 1; |
e6a1105b BB |
4054 | |
4055 | /* this function shouldn't be used with modular subsystems, since they | |
4056 | * need to register a subsys_id, among other things */ | |
4057 | BUG_ON(ss->module); | |
4058 | } | |
4059 | ||
4060 | /** | |
4061 | * cgroup_load_subsys: load and register a modular subsystem at runtime | |
4062 | * @ss: the subsystem to load | |
4063 | * | |
4064 | * This function should be called in a modular subsystem's initcall. If the | |
88393161 | 4065 | * subsystem is built as a module, it will be assigned a new subsys_id and set |
e6a1105b BB |
4066 | * up for use. If the subsystem is built-in anyway, work is delegated to the |
4067 | * simpler cgroup_init_subsys. | |
4068 | */ | |
4069 | int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss) | |
4070 | { | |
4071 | int i; | |
4072 | struct cgroup_subsys_state *css; | |
4073 | ||
4074 | /* check name and function validity */ | |
4075 | if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN || | |
4076 | ss->create == NULL || ss->destroy == NULL) | |
4077 | return -EINVAL; | |
4078 | ||
4079 | /* | |
4080 | * we don't support callbacks in modular subsystems. this check is | |
4081 | * before the ss->module check for consistency; a subsystem that could | |
4082 | * be a module should still have no callbacks even if the user isn't | |
4083 | * compiling it as one. | |
4084 | */ | |
4085 | if (ss->fork || ss->exit) | |
4086 | return -EINVAL; | |
4087 | ||
4088 | /* | |
4089 | * an optionally modular subsystem is built-in: we want to do nothing, | |
4090 | * since cgroup_init_subsys will have already taken care of it. | |
4091 | */ | |
4092 | if (ss->module == NULL) { | |
4093 | /* a few sanity checks */ | |
4094 | BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT); | |
4095 | BUG_ON(subsys[ss->subsys_id] != ss); | |
4096 | return 0; | |
4097 | } | |
4098 | ||
4099 | /* | |
4100 | * need to register a subsys id before anything else - for example, | |
4101 | * init_cgroup_css needs it. | |
4102 | */ | |
4103 | mutex_lock(&cgroup_mutex); | |
4104 | /* find the first empty slot in the array */ | |
4105 | for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) { | |
4106 | if (subsys[i] == NULL) | |
4107 | break; | |
4108 | } | |
4109 | if (i == CGROUP_SUBSYS_COUNT) { | |
4110 | /* maximum number of subsystems already registered! */ | |
4111 | mutex_unlock(&cgroup_mutex); | |
4112 | return -EBUSY; | |
4113 | } | |
4114 | /* assign ourselves the subsys_id */ | |
4115 | ss->subsys_id = i; | |
4116 | subsys[i] = ss; | |
4117 | ||
4118 | /* | |
4119 | * no ss->create seems to need anything important in the ss struct, so | |
4120 | * this can happen first (i.e. before the rootnode attachment). | |
4121 | */ | |
761b3ef5 | 4122 | css = ss->create(dummytop); |
e6a1105b BB |
4123 | if (IS_ERR(css)) { |
4124 | /* failure case - need to deassign the subsys[] slot. */ | |
4125 | subsys[i] = NULL; | |
4126 | mutex_unlock(&cgroup_mutex); | |
4127 | return PTR_ERR(css); | |
4128 | } | |
4129 | ||
4130 | list_add(&ss->sibling, &rootnode.subsys_list); | |
4131 | ss->root = &rootnode; | |
4132 | ||
4133 | /* our new subsystem will be attached to the dummy hierarchy. */ | |
4134 | init_cgroup_css(css, ss, dummytop); | |
4135 | /* init_idr must be after init_cgroup_css because it sets css->id. */ | |
4136 | if (ss->use_id) { | |
4137 | int ret = cgroup_init_idr(ss, css); | |
4138 | if (ret) { | |
4139 | dummytop->subsys[ss->subsys_id] = NULL; | |
761b3ef5 | 4140 | ss->destroy(dummytop); |
e6a1105b BB |
4141 | subsys[i] = NULL; |
4142 | mutex_unlock(&cgroup_mutex); | |
4143 | return ret; | |
4144 | } | |
4145 | } | |
4146 | ||
4147 | /* | |
4148 | * Now we need to entangle the css into the existing css_sets. unlike | |
4149 | * in cgroup_init_subsys, there are now multiple css_sets, so each one | |
4150 | * will need a new pointer to it; done by iterating the css_set_table. | |
4151 | * furthermore, modifying the existing css_sets will corrupt the hash | |
4152 | * table state, so each changed css_set will need its hash recomputed. | |
4153 | * this is all done under the css_set_lock. | |
4154 | */ | |
4155 | write_lock(&css_set_lock); | |
4156 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { | |
4157 | struct css_set *cg; | |
4158 | struct hlist_node *node, *tmp; | |
4159 | struct hlist_head *bucket = &css_set_table[i], *new_bucket; | |
4160 | ||
4161 | hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) { | |
4162 | /* skip entries that we already rehashed */ | |
4163 | if (cg->subsys[ss->subsys_id]) | |
4164 | continue; | |
4165 | /* remove existing entry */ | |
4166 | hlist_del(&cg->hlist); | |
4167 | /* set new value */ | |
4168 | cg->subsys[ss->subsys_id] = css; | |
4169 | /* recompute hash and restore entry */ | |
4170 | new_bucket = css_set_hash(cg->subsys); | |
4171 | hlist_add_head(&cg->hlist, new_bucket); | |
4172 | } | |
4173 | } | |
4174 | write_unlock(&css_set_lock); | |
4175 | ||
4176 | mutex_init(&ss->hierarchy_mutex); | |
4177 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); | |
4178 | ss->active = 1; | |
4179 | ||
e6a1105b BB |
4180 | /* success! */ |
4181 | mutex_unlock(&cgroup_mutex); | |
4182 | return 0; | |
ddbcc7e8 | 4183 | } |
e6a1105b | 4184 | EXPORT_SYMBOL_GPL(cgroup_load_subsys); |
ddbcc7e8 | 4185 | |
cf5d5941 BB |
4186 | /** |
4187 | * cgroup_unload_subsys: unload a modular subsystem | |
4188 | * @ss: the subsystem to unload | |
4189 | * | |
4190 | * This function should be called in a modular subsystem's exitcall. When this | |
4191 | * function is invoked, the refcount on the subsystem's module will be 0, so | |
4192 | * the subsystem will not be attached to any hierarchy. | |
4193 | */ | |
4194 | void cgroup_unload_subsys(struct cgroup_subsys *ss) | |
4195 | { | |
4196 | struct cg_cgroup_link *link; | |
4197 | struct hlist_head *hhead; | |
4198 | ||
4199 | BUG_ON(ss->module == NULL); | |
4200 | ||
4201 | /* | |
4202 | * we shouldn't be called if the subsystem is in use, and the use of | |
4203 | * try_module_get in parse_cgroupfs_options should ensure that it | |
4204 | * doesn't start being used while we're killing it off. | |
4205 | */ | |
4206 | BUG_ON(ss->root != &rootnode); | |
4207 | ||
4208 | mutex_lock(&cgroup_mutex); | |
4209 | /* deassign the subsys_id */ | |
4210 | BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT); | |
4211 | subsys[ss->subsys_id] = NULL; | |
4212 | ||
4213 | /* remove subsystem from rootnode's list of subsystems */ | |
8d258797 | 4214 | list_del_init(&ss->sibling); |
cf5d5941 BB |
4215 | |
4216 | /* | |
4217 | * disentangle the css from all css_sets attached to the dummytop. as | |
4218 | * in loading, we need to pay our respects to the hashtable gods. | |
4219 | */ | |
4220 | write_lock(&css_set_lock); | |
4221 | list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) { | |
4222 | struct css_set *cg = link->cg; | |
4223 | ||
4224 | hlist_del(&cg->hlist); | |
4225 | BUG_ON(!cg->subsys[ss->subsys_id]); | |
4226 | cg->subsys[ss->subsys_id] = NULL; | |
4227 | hhead = css_set_hash(cg->subsys); | |
4228 | hlist_add_head(&cg->hlist, hhead); | |
4229 | } | |
4230 | write_unlock(&css_set_lock); | |
4231 | ||
4232 | /* | |
4233 | * remove subsystem's css from the dummytop and free it - need to free | |
4234 | * before marking as null because ss->destroy needs the cgrp->subsys | |
4235 | * pointer to find their state. note that this also takes care of | |
4236 | * freeing the css_id. | |
4237 | */ | |
761b3ef5 | 4238 | ss->destroy(dummytop); |
cf5d5941 BB |
4239 | dummytop->subsys[ss->subsys_id] = NULL; |
4240 | ||
4241 | mutex_unlock(&cgroup_mutex); | |
4242 | } | |
4243 | EXPORT_SYMBOL_GPL(cgroup_unload_subsys); | |
4244 | ||
ddbcc7e8 | 4245 | /** |
a043e3b2 LZ |
4246 | * cgroup_init_early - cgroup initialization at system boot |
4247 | * | |
4248 | * Initialize cgroups at system boot, and initialize any | |
4249 | * subsystems that request early init. | |
ddbcc7e8 PM |
4250 | */ |
4251 | int __init cgroup_init_early(void) | |
4252 | { | |
4253 | int i; | |
146aa1bd | 4254 | atomic_set(&init_css_set.refcount, 1); |
817929ec PM |
4255 | INIT_LIST_HEAD(&init_css_set.cg_links); |
4256 | INIT_LIST_HEAD(&init_css_set.tasks); | |
472b1053 | 4257 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 4258 | css_set_count = 1; |
ddbcc7e8 | 4259 | init_cgroup_root(&rootnode); |
817929ec PM |
4260 | root_count = 1; |
4261 | init_task.cgroups = &init_css_set; | |
4262 | ||
4263 | init_css_set_link.cg = &init_css_set; | |
7717f7ba | 4264 | init_css_set_link.cgrp = dummytop; |
bd89aabc | 4265 | list_add(&init_css_set_link.cgrp_link_list, |
817929ec PM |
4266 | &rootnode.top_cgroup.css_sets); |
4267 | list_add(&init_css_set_link.cg_link_list, | |
4268 | &init_css_set.cg_links); | |
ddbcc7e8 | 4269 | |
472b1053 LZ |
4270 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) |
4271 | INIT_HLIST_HEAD(&css_set_table[i]); | |
4272 | ||
aae8aab4 BB |
4273 | /* at bootup time, we don't worry about modular subsystems */ |
4274 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
ddbcc7e8 PM |
4275 | struct cgroup_subsys *ss = subsys[i]; |
4276 | ||
4277 | BUG_ON(!ss->name); | |
4278 | BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); | |
4279 | BUG_ON(!ss->create); | |
4280 | BUG_ON(!ss->destroy); | |
4281 | if (ss->subsys_id != i) { | |
cfe36bde | 4282 | printk(KERN_ERR "cgroup: Subsys %s id == %d\n", |
ddbcc7e8 PM |
4283 | ss->name, ss->subsys_id); |
4284 | BUG(); | |
4285 | } | |
4286 | ||
4287 | if (ss->early_init) | |
4288 | cgroup_init_subsys(ss); | |
4289 | } | |
4290 | return 0; | |
4291 | } | |
4292 | ||
4293 | /** | |
a043e3b2 LZ |
4294 | * cgroup_init - cgroup initialization |
4295 | * | |
4296 | * Register cgroup filesystem and /proc file, and initialize | |
4297 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4298 | */ |
4299 | int __init cgroup_init(void) | |
4300 | { | |
4301 | int err; | |
4302 | int i; | |
472b1053 | 4303 | struct hlist_head *hhead; |
a424316c PM |
4304 | |
4305 | err = bdi_init(&cgroup_backing_dev_info); | |
4306 | if (err) | |
4307 | return err; | |
ddbcc7e8 | 4308 | |
aae8aab4 BB |
4309 | /* at bootup time, we don't worry about modular subsystems */ |
4310 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
ddbcc7e8 PM |
4311 | struct cgroup_subsys *ss = subsys[i]; |
4312 | if (!ss->early_init) | |
4313 | cgroup_init_subsys(ss); | |
38460b48 | 4314 | if (ss->use_id) |
e6a1105b | 4315 | cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]); |
ddbcc7e8 PM |
4316 | } |
4317 | ||
472b1053 LZ |
4318 | /* Add init_css_set to the hash table */ |
4319 | hhead = css_set_hash(init_css_set.subsys); | |
4320 | hlist_add_head(&init_css_set.hlist, hhead); | |
2c6ab6d2 | 4321 | BUG_ON(!init_root_id(&rootnode)); |
676db4af GK |
4322 | |
4323 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); | |
4324 | if (!cgroup_kobj) { | |
4325 | err = -ENOMEM; | |
4326 | goto out; | |
4327 | } | |
4328 | ||
ddbcc7e8 | 4329 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
4330 | if (err < 0) { |
4331 | kobject_put(cgroup_kobj); | |
ddbcc7e8 | 4332 | goto out; |
676db4af | 4333 | } |
ddbcc7e8 | 4334 | |
46ae220b | 4335 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
a424316c | 4336 | |
ddbcc7e8 | 4337 | out: |
a424316c PM |
4338 | if (err) |
4339 | bdi_destroy(&cgroup_backing_dev_info); | |
4340 | ||
ddbcc7e8 PM |
4341 | return err; |
4342 | } | |
b4f48b63 | 4343 | |
a424316c PM |
4344 | /* |
4345 | * proc_cgroup_show() | |
4346 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4347 | * - Used for /proc/<pid>/cgroup. | |
4348 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4349 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4350 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4351 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4352 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4353 | * cgroup to top_cgroup. | |
4354 | */ | |
4355 | ||
4356 | /* TODO: Use a proper seq_file iterator */ | |
4357 | static int proc_cgroup_show(struct seq_file *m, void *v) | |
4358 | { | |
4359 | struct pid *pid; | |
4360 | struct task_struct *tsk; | |
4361 | char *buf; | |
4362 | int retval; | |
4363 | struct cgroupfs_root *root; | |
4364 | ||
4365 | retval = -ENOMEM; | |
4366 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4367 | if (!buf) | |
4368 | goto out; | |
4369 | ||
4370 | retval = -ESRCH; | |
4371 | pid = m->private; | |
4372 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4373 | if (!tsk) | |
4374 | goto out_free; | |
4375 | ||
4376 | retval = 0; | |
4377 | ||
4378 | mutex_lock(&cgroup_mutex); | |
4379 | ||
e5f6a860 | 4380 | for_each_active_root(root) { |
a424316c | 4381 | struct cgroup_subsys *ss; |
bd89aabc | 4382 | struct cgroup *cgrp; |
a424316c PM |
4383 | int count = 0; |
4384 | ||
2c6ab6d2 | 4385 | seq_printf(m, "%d:", root->hierarchy_id); |
a424316c PM |
4386 | for_each_subsys(root, ss) |
4387 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4388 | if (strlen(root->name)) |
4389 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4390 | root->name); | |
a424316c | 4391 | seq_putc(m, ':'); |
7717f7ba | 4392 | cgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 4393 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); |
a424316c PM |
4394 | if (retval < 0) |
4395 | goto out_unlock; | |
4396 | seq_puts(m, buf); | |
4397 | seq_putc(m, '\n'); | |
4398 | } | |
4399 | ||
4400 | out_unlock: | |
4401 | mutex_unlock(&cgroup_mutex); | |
4402 | put_task_struct(tsk); | |
4403 | out_free: | |
4404 | kfree(buf); | |
4405 | out: | |
4406 | return retval; | |
4407 | } | |
4408 | ||
4409 | static int cgroup_open(struct inode *inode, struct file *file) | |
4410 | { | |
4411 | struct pid *pid = PROC_I(inode)->pid; | |
4412 | return single_open(file, proc_cgroup_show, pid); | |
4413 | } | |
4414 | ||
828c0950 | 4415 | const struct file_operations proc_cgroup_operations = { |
a424316c PM |
4416 | .open = cgroup_open, |
4417 | .read = seq_read, | |
4418 | .llseek = seq_lseek, | |
4419 | .release = single_release, | |
4420 | }; | |
4421 | ||
4422 | /* Display information about each subsystem and each hierarchy */ | |
4423 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4424 | { | |
4425 | int i; | |
a424316c | 4426 | |
8bab8dde | 4427 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4428 | /* |
4429 | * ideally we don't want subsystems moving around while we do this. | |
4430 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4431 | * subsys/hierarchy state. | |
4432 | */ | |
a424316c | 4433 | mutex_lock(&cgroup_mutex); |
a424316c PM |
4434 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
4435 | struct cgroup_subsys *ss = subsys[i]; | |
aae8aab4 BB |
4436 | if (ss == NULL) |
4437 | continue; | |
2c6ab6d2 PM |
4438 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4439 | ss->name, ss->root->hierarchy_id, | |
8bab8dde | 4440 | ss->root->number_of_cgroups, !ss->disabled); |
a424316c PM |
4441 | } |
4442 | mutex_unlock(&cgroup_mutex); | |
4443 | return 0; | |
4444 | } | |
4445 | ||
4446 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4447 | { | |
9dce07f1 | 4448 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4449 | } |
4450 | ||
828c0950 | 4451 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4452 | .open = cgroupstats_open, |
4453 | .read = seq_read, | |
4454 | .llseek = seq_lseek, | |
4455 | .release = single_release, | |
4456 | }; | |
4457 | ||
b4f48b63 PM |
4458 | /** |
4459 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4460 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4461 | * |
4462 | * Description: A task inherits its parent's cgroup at fork(). | |
4463 | * | |
4464 | * A pointer to the shared css_set was automatically copied in | |
4465 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
7e381b0e FW |
4466 | * it was not made under the protection of RCU, cgroup_mutex or |
4467 | * threadgroup_change_begin(), so it might no longer be a valid | |
4468 | * cgroup pointer. cgroup_attach_task() might have already changed | |
4469 | * current->cgroups, allowing the previously referenced cgroup | |
4470 | * group to be removed and freed. | |
4471 | * | |
4472 | * Outside the pointer validity we also need to process the css_set | |
4473 | * inheritance between threadgoup_change_begin() and | |
4474 | * threadgoup_change_end(), this way there is no leak in any process | |
4475 | * wide migration performed by cgroup_attach_proc() that could otherwise | |
4476 | * miss a thread because it is too early or too late in the fork stage. | |
b4f48b63 PM |
4477 | * |
4478 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4479 | * task, and the passed argument 'child' points to the child task. | |
4480 | */ | |
4481 | void cgroup_fork(struct task_struct *child) | |
4482 | { | |
7e381b0e FW |
4483 | /* |
4484 | * We don't need to task_lock() current because current->cgroups | |
4485 | * can't be changed concurrently here. The parent obviously hasn't | |
4486 | * exited and called cgroup_exit(), and we are synchronized against | |
4487 | * cgroup migration through threadgroup_change_begin(). | |
4488 | */ | |
817929ec PM |
4489 | child->cgroups = current->cgroups; |
4490 | get_css_set(child->cgroups); | |
817929ec | 4491 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4492 | } |
4493 | ||
4494 | /** | |
a043e3b2 LZ |
4495 | * cgroup_fork_callbacks - run fork callbacks |
4496 | * @child: the new task | |
4497 | * | |
4498 | * Called on a new task very soon before adding it to the | |
4499 | * tasklist. No need to take any locks since no-one can | |
4500 | * be operating on this task. | |
b4f48b63 PM |
4501 | */ |
4502 | void cgroup_fork_callbacks(struct task_struct *child) | |
4503 | { | |
4504 | if (need_forkexit_callback) { | |
4505 | int i; | |
aae8aab4 BB |
4506 | /* |
4507 | * forkexit callbacks are only supported for builtin | |
4508 | * subsystems, and the builtin section of the subsys array is | |
4509 | * immutable, so we don't need to lock the subsys array here. | |
4510 | */ | |
4511 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
b4f48b63 PM |
4512 | struct cgroup_subsys *ss = subsys[i]; |
4513 | if (ss->fork) | |
761b3ef5 | 4514 | ss->fork(child); |
b4f48b63 PM |
4515 | } |
4516 | } | |
4517 | } | |
4518 | ||
817929ec | 4519 | /** |
a043e3b2 LZ |
4520 | * cgroup_post_fork - called on a new task after adding it to the task list |
4521 | * @child: the task in question | |
4522 | * | |
4523 | * Adds the task to the list running through its css_set if necessary. | |
4524 | * Has to be after the task is visible on the task list in case we race | |
4525 | * with the first call to cgroup_iter_start() - to guarantee that the | |
4526 | * new task ends up on its list. | |
4527 | */ | |
817929ec PM |
4528 | void cgroup_post_fork(struct task_struct *child) |
4529 | { | |
3ce3230a FW |
4530 | /* |
4531 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
4532 | * under the tasklist_lock and we read it here after we added the child | |
4533 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
4534 | * yet in the tasklist when we walked through it from | |
4535 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
4536 | * should be visible now due to the paired locking and barriers implied | |
4537 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
4538 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
4539 | * lock on fork. | |
4540 | */ | |
817929ec PM |
4541 | if (use_task_css_set_links) { |
4542 | write_lock(&css_set_lock); | |
7e3aa30a FW |
4543 | if (list_empty(&child->cg_list)) { |
4544 | /* | |
4545 | * It's safe to use child->cgroups without task_lock() | |
4546 | * here because we are protected through | |
4547 | * threadgroup_change_begin() against concurrent | |
4548 | * css_set change in cgroup_task_migrate(). Also | |
4549 | * the task can't exit at that point until | |
4550 | * wake_up_new_task() is called, so we are protected | |
4551 | * against cgroup_exit() setting child->cgroup to | |
4552 | * init_css_set. | |
4553 | */ | |
817929ec | 4554 | list_add(&child->cg_list, &child->cgroups->tasks); |
7e3aa30a | 4555 | } |
817929ec PM |
4556 | write_unlock(&css_set_lock); |
4557 | } | |
4558 | } | |
b4f48b63 PM |
4559 | /** |
4560 | * cgroup_exit - detach cgroup from exiting task | |
4561 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4562 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4563 | * |
4564 | * Description: Detach cgroup from @tsk and release it. | |
4565 | * | |
4566 | * Note that cgroups marked notify_on_release force every task in | |
4567 | * them to take the global cgroup_mutex mutex when exiting. | |
4568 | * This could impact scaling on very large systems. Be reluctant to | |
4569 | * use notify_on_release cgroups where very high task exit scaling | |
4570 | * is required on large systems. | |
4571 | * | |
4572 | * the_top_cgroup_hack: | |
4573 | * | |
4574 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4575 | * | |
4576 | * We call cgroup_exit() while the task is still competent to | |
4577 | * handle notify_on_release(), then leave the task attached to the | |
4578 | * root cgroup in each hierarchy for the remainder of its exit. | |
4579 | * | |
4580 | * To do this properly, we would increment the reference count on | |
4581 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4582 | * code we would add a second cgroup function call, to drop that | |
4583 | * reference. This would just create an unnecessary hot spot on | |
4584 | * the top_cgroup reference count, to no avail. | |
4585 | * | |
4586 | * Normally, holding a reference to a cgroup without bumping its | |
4587 | * count is unsafe. The cgroup could go away, or someone could | |
4588 | * attach us to a different cgroup, decrementing the count on | |
4589 | * the first cgroup that we never incremented. But in this case, | |
4590 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4591 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4592 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4593 | */ |
4594 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4595 | { | |
817929ec | 4596 | struct css_set *cg; |
d41d5a01 | 4597 | int i; |
817929ec PM |
4598 | |
4599 | /* | |
4600 | * Unlink from the css_set task list if necessary. | |
4601 | * Optimistically check cg_list before taking | |
4602 | * css_set_lock | |
4603 | */ | |
4604 | if (!list_empty(&tsk->cg_list)) { | |
4605 | write_lock(&css_set_lock); | |
4606 | if (!list_empty(&tsk->cg_list)) | |
8d258797 | 4607 | list_del_init(&tsk->cg_list); |
817929ec PM |
4608 | write_unlock(&css_set_lock); |
4609 | } | |
4610 | ||
b4f48b63 PM |
4611 | /* Reassign the task to the init_css_set. */ |
4612 | task_lock(tsk); | |
817929ec PM |
4613 | cg = tsk->cgroups; |
4614 | tsk->cgroups = &init_css_set; | |
d41d5a01 PZ |
4615 | |
4616 | if (run_callbacks && need_forkexit_callback) { | |
4617 | /* | |
4618 | * modular subsystems can't use callbacks, so no need to lock | |
4619 | * the subsys array | |
4620 | */ | |
4621 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
4622 | struct cgroup_subsys *ss = subsys[i]; | |
4623 | if (ss->exit) { | |
4624 | struct cgroup *old_cgrp = | |
4625 | rcu_dereference_raw(cg->subsys[i])->cgroup; | |
4626 | struct cgroup *cgrp = task_cgroup(tsk, i); | |
761b3ef5 | 4627 | ss->exit(cgrp, old_cgrp, tsk); |
d41d5a01 PZ |
4628 | } |
4629 | } | |
4630 | } | |
b4f48b63 | 4631 | task_unlock(tsk); |
d41d5a01 | 4632 | |
817929ec | 4633 | if (cg) |
81a6a5cd | 4634 | put_css_set_taskexit(cg); |
b4f48b63 | 4635 | } |
697f4161 | 4636 | |
a043e3b2 | 4637 | /** |
313e924c | 4638 | * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp |
a043e3b2 | 4639 | * @cgrp: the cgroup in question |
313e924c | 4640 | * @task: the task in question |
a043e3b2 | 4641 | * |
313e924c GN |
4642 | * See if @cgrp is a descendant of @task's cgroup in the appropriate |
4643 | * hierarchy. | |
697f4161 PM |
4644 | * |
4645 | * If we are sending in dummytop, then presumably we are creating | |
4646 | * the top cgroup in the subsystem. | |
4647 | * | |
4648 | * Called only by the ns (nsproxy) cgroup. | |
4649 | */ | |
313e924c | 4650 | int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task) |
697f4161 PM |
4651 | { |
4652 | int ret; | |
4653 | struct cgroup *target; | |
697f4161 | 4654 | |
bd89aabc | 4655 | if (cgrp == dummytop) |
697f4161 PM |
4656 | return 1; |
4657 | ||
7717f7ba | 4658 | target = task_cgroup_from_root(task, cgrp->root); |
bd89aabc PM |
4659 | while (cgrp != target && cgrp!= cgrp->top_cgroup) |
4660 | cgrp = cgrp->parent; | |
4661 | ret = (cgrp == target); | |
697f4161 PM |
4662 | return ret; |
4663 | } | |
81a6a5cd | 4664 | |
bd89aabc | 4665 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd PM |
4666 | { |
4667 | /* All of these checks rely on RCU to keep the cgroup | |
4668 | * structure alive */ | |
bd89aabc PM |
4669 | if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count) |
4670 | && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) { | |
81a6a5cd PM |
4671 | /* Control Group is currently removeable. If it's not |
4672 | * already queued for a userspace notification, queue | |
4673 | * it now */ | |
4674 | int need_schedule_work = 0; | |
cdcc136f | 4675 | raw_spin_lock(&release_list_lock); |
bd89aabc PM |
4676 | if (!cgroup_is_removed(cgrp) && |
4677 | list_empty(&cgrp->release_list)) { | |
4678 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4679 | need_schedule_work = 1; |
4680 | } | |
cdcc136f | 4681 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4682 | if (need_schedule_work) |
4683 | schedule_work(&release_agent_work); | |
4684 | } | |
4685 | } | |
4686 | ||
d7b9fff7 DN |
4687 | /* Caller must verify that the css is not for root cgroup */ |
4688 | void __css_put(struct cgroup_subsys_state *css, int count) | |
81a6a5cd | 4689 | { |
bd89aabc | 4690 | struct cgroup *cgrp = css->cgroup; |
3dece834 | 4691 | int val; |
81a6a5cd | 4692 | rcu_read_lock(); |
d7b9fff7 | 4693 | val = atomic_sub_return(count, &css->refcnt); |
3dece834 | 4694 | if (val == 1) { |
ec64f515 KH |
4695 | if (notify_on_release(cgrp)) { |
4696 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
4697 | check_for_release(cgrp); | |
4698 | } | |
88703267 | 4699 | cgroup_wakeup_rmdir_waiter(cgrp); |
81a6a5cd PM |
4700 | } |
4701 | rcu_read_unlock(); | |
3dece834 | 4702 | WARN_ON_ONCE(val < 1); |
81a6a5cd | 4703 | } |
67523c48 | 4704 | EXPORT_SYMBOL_GPL(__css_put); |
81a6a5cd PM |
4705 | |
4706 | /* | |
4707 | * Notify userspace when a cgroup is released, by running the | |
4708 | * configured release agent with the name of the cgroup (path | |
4709 | * relative to the root of cgroup file system) as the argument. | |
4710 | * | |
4711 | * Most likely, this user command will try to rmdir this cgroup. | |
4712 | * | |
4713 | * This races with the possibility that some other task will be | |
4714 | * attached to this cgroup before it is removed, or that some other | |
4715 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4716 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4717 | * unused, and this cgroup will be reprieved from its death sentence, | |
4718 | * to continue to serve a useful existence. Next time it's released, | |
4719 | * we will get notified again, if it still has 'notify_on_release' set. | |
4720 | * | |
4721 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4722 | * means only wait until the task is successfully execve()'d. The | |
4723 | * separate release agent task is forked by call_usermodehelper(), | |
4724 | * then control in this thread returns here, without waiting for the | |
4725 | * release agent task. We don't bother to wait because the caller of | |
4726 | * this routine has no use for the exit status of the release agent | |
4727 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4728 | */ |
81a6a5cd PM |
4729 | static void cgroup_release_agent(struct work_struct *work) |
4730 | { | |
4731 | BUG_ON(work != &release_agent_work); | |
4732 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4733 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4734 | while (!list_empty(&release_list)) { |
4735 | char *argv[3], *envp[3]; | |
4736 | int i; | |
e788e066 | 4737 | char *pathbuf = NULL, *agentbuf = NULL; |
bd89aabc | 4738 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4739 | struct cgroup, |
4740 | release_list); | |
bd89aabc | 4741 | list_del_init(&cgrp->release_list); |
cdcc136f | 4742 | raw_spin_unlock(&release_list_lock); |
81a6a5cd | 4743 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
e788e066 PM |
4744 | if (!pathbuf) |
4745 | goto continue_free; | |
4746 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
4747 | goto continue_free; | |
4748 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4749 | if (!agentbuf) | |
4750 | goto continue_free; | |
81a6a5cd PM |
4751 | |
4752 | i = 0; | |
e788e066 PM |
4753 | argv[i++] = agentbuf; |
4754 | argv[i++] = pathbuf; | |
81a6a5cd PM |
4755 | argv[i] = NULL; |
4756 | ||
4757 | i = 0; | |
4758 | /* minimal command environment */ | |
4759 | envp[i++] = "HOME=/"; | |
4760 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4761 | envp[i] = NULL; | |
4762 | ||
4763 | /* Drop the lock while we invoke the usermode helper, | |
4764 | * since the exec could involve hitting disk and hence | |
4765 | * be a slow process */ | |
4766 | mutex_unlock(&cgroup_mutex); | |
4767 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4768 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4769 | continue_free: |
4770 | kfree(pathbuf); | |
4771 | kfree(agentbuf); | |
cdcc136f | 4772 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4773 | } |
cdcc136f | 4774 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4775 | mutex_unlock(&cgroup_mutex); |
4776 | } | |
8bab8dde PM |
4777 | |
4778 | static int __init cgroup_disable(char *str) | |
4779 | { | |
4780 | int i; | |
4781 | char *token; | |
4782 | ||
4783 | while ((token = strsep(&str, ",")) != NULL) { | |
4784 | if (!*token) | |
4785 | continue; | |
aae8aab4 BB |
4786 | /* |
4787 | * cgroup_disable, being at boot time, can't know about module | |
4788 | * subsystems, so we don't worry about them. | |
4789 | */ | |
4790 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
8bab8dde PM |
4791 | struct cgroup_subsys *ss = subsys[i]; |
4792 | ||
4793 | if (!strcmp(token, ss->name)) { | |
4794 | ss->disabled = 1; | |
4795 | printk(KERN_INFO "Disabling %s control group" | |
4796 | " subsystem\n", ss->name); | |
4797 | break; | |
4798 | } | |
4799 | } | |
4800 | } | |
4801 | return 1; | |
4802 | } | |
4803 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 KH |
4804 | |
4805 | /* | |
4806 | * Functons for CSS ID. | |
4807 | */ | |
4808 | ||
4809 | /* | |
4810 | *To get ID other than 0, this should be called when !cgroup_is_removed(). | |
4811 | */ | |
4812 | unsigned short css_id(struct cgroup_subsys_state *css) | |
4813 | { | |
7f0f1546 KH |
4814 | struct css_id *cssid; |
4815 | ||
4816 | /* | |
4817 | * This css_id() can return correct value when somone has refcnt | |
4818 | * on this or this is under rcu_read_lock(). Once css->id is allocated, | |
4819 | * it's unchanged until freed. | |
4820 | */ | |
d8bf4ca9 | 4821 | cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt)); |
38460b48 KH |
4822 | |
4823 | if (cssid) | |
4824 | return cssid->id; | |
4825 | return 0; | |
4826 | } | |
67523c48 | 4827 | EXPORT_SYMBOL_GPL(css_id); |
38460b48 KH |
4828 | |
4829 | unsigned short css_depth(struct cgroup_subsys_state *css) | |
4830 | { | |
7f0f1546 KH |
4831 | struct css_id *cssid; |
4832 | ||
d8bf4ca9 | 4833 | cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt)); |
38460b48 KH |
4834 | |
4835 | if (cssid) | |
4836 | return cssid->depth; | |
4837 | return 0; | |
4838 | } | |
67523c48 | 4839 | EXPORT_SYMBOL_GPL(css_depth); |
38460b48 | 4840 | |
747388d7 KH |
4841 | /** |
4842 | * css_is_ancestor - test "root" css is an ancestor of "child" | |
4843 | * @child: the css to be tested. | |
4844 | * @root: the css supporsed to be an ancestor of the child. | |
4845 | * | |
4846 | * Returns true if "root" is an ancestor of "child" in its hierarchy. Because | |
4847 | * this function reads css->id, this use rcu_dereference() and rcu_read_lock(). | |
4848 | * But, considering usual usage, the csses should be valid objects after test. | |
4849 | * Assuming that the caller will do some action to the child if this returns | |
4850 | * returns true, the caller must take "child";s reference count. | |
4851 | * If "child" is valid object and this returns true, "root" is valid, too. | |
4852 | */ | |
4853 | ||
38460b48 | 4854 | bool css_is_ancestor(struct cgroup_subsys_state *child, |
0b7f569e | 4855 | const struct cgroup_subsys_state *root) |
38460b48 | 4856 | { |
747388d7 KH |
4857 | struct css_id *child_id; |
4858 | struct css_id *root_id; | |
4859 | bool ret = true; | |
38460b48 | 4860 | |
747388d7 KH |
4861 | rcu_read_lock(); |
4862 | child_id = rcu_dereference(child->id); | |
4863 | root_id = rcu_dereference(root->id); | |
4864 | if (!child_id | |
4865 | || !root_id | |
4866 | || (child_id->depth < root_id->depth) | |
4867 | || (child_id->stack[root_id->depth] != root_id->id)) | |
4868 | ret = false; | |
4869 | rcu_read_unlock(); | |
4870 | return ret; | |
38460b48 KH |
4871 | } |
4872 | ||
38460b48 KH |
4873 | void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css) |
4874 | { | |
4875 | struct css_id *id = css->id; | |
4876 | /* When this is called before css_id initialization, id can be NULL */ | |
4877 | if (!id) | |
4878 | return; | |
4879 | ||
4880 | BUG_ON(!ss->use_id); | |
4881 | ||
4882 | rcu_assign_pointer(id->css, NULL); | |
4883 | rcu_assign_pointer(css->id, NULL); | |
42aee6c4 | 4884 | spin_lock(&ss->id_lock); |
38460b48 | 4885 | idr_remove(&ss->idr, id->id); |
42aee6c4 | 4886 | spin_unlock(&ss->id_lock); |
025cea99 | 4887 | kfree_rcu(id, rcu_head); |
38460b48 | 4888 | } |
67523c48 | 4889 | EXPORT_SYMBOL_GPL(free_css_id); |
38460b48 KH |
4890 | |
4891 | /* | |
4892 | * This is called by init or create(). Then, calls to this function are | |
4893 | * always serialized (By cgroup_mutex() at create()). | |
4894 | */ | |
4895 | ||
4896 | static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth) | |
4897 | { | |
4898 | struct css_id *newid; | |
4899 | int myid, error, size; | |
4900 | ||
4901 | BUG_ON(!ss->use_id); | |
4902 | ||
4903 | size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1); | |
4904 | newid = kzalloc(size, GFP_KERNEL); | |
4905 | if (!newid) | |
4906 | return ERR_PTR(-ENOMEM); | |
4907 | /* get id */ | |
4908 | if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) { | |
4909 | error = -ENOMEM; | |
4910 | goto err_out; | |
4911 | } | |
42aee6c4 | 4912 | spin_lock(&ss->id_lock); |
38460b48 KH |
4913 | /* Don't use 0. allocates an ID of 1-65535 */ |
4914 | error = idr_get_new_above(&ss->idr, newid, 1, &myid); | |
42aee6c4 | 4915 | spin_unlock(&ss->id_lock); |
38460b48 KH |
4916 | |
4917 | /* Returns error when there are no free spaces for new ID.*/ | |
4918 | if (error) { | |
4919 | error = -ENOSPC; | |
4920 | goto err_out; | |
4921 | } | |
4922 | if (myid > CSS_ID_MAX) | |
4923 | goto remove_idr; | |
4924 | ||
4925 | newid->id = myid; | |
4926 | newid->depth = depth; | |
4927 | return newid; | |
4928 | remove_idr: | |
4929 | error = -ENOSPC; | |
42aee6c4 | 4930 | spin_lock(&ss->id_lock); |
38460b48 | 4931 | idr_remove(&ss->idr, myid); |
42aee6c4 | 4932 | spin_unlock(&ss->id_lock); |
38460b48 KH |
4933 | err_out: |
4934 | kfree(newid); | |
4935 | return ERR_PTR(error); | |
4936 | ||
4937 | } | |
4938 | ||
e6a1105b BB |
4939 | static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss, |
4940 | struct cgroup_subsys_state *rootcss) | |
38460b48 KH |
4941 | { |
4942 | struct css_id *newid; | |
38460b48 | 4943 | |
42aee6c4 | 4944 | spin_lock_init(&ss->id_lock); |
38460b48 KH |
4945 | idr_init(&ss->idr); |
4946 | ||
38460b48 KH |
4947 | newid = get_new_cssid(ss, 0); |
4948 | if (IS_ERR(newid)) | |
4949 | return PTR_ERR(newid); | |
4950 | ||
4951 | newid->stack[0] = newid->id; | |
4952 | newid->css = rootcss; | |
4953 | rootcss->id = newid; | |
4954 | return 0; | |
4955 | } | |
4956 | ||
4957 | static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent, | |
4958 | struct cgroup *child) | |
4959 | { | |
4960 | int subsys_id, i, depth = 0; | |
4961 | struct cgroup_subsys_state *parent_css, *child_css; | |
fae9c791 | 4962 | struct css_id *child_id, *parent_id; |
38460b48 KH |
4963 | |
4964 | subsys_id = ss->subsys_id; | |
4965 | parent_css = parent->subsys[subsys_id]; | |
4966 | child_css = child->subsys[subsys_id]; | |
38460b48 | 4967 | parent_id = parent_css->id; |
94b3dd0f | 4968 | depth = parent_id->depth + 1; |
38460b48 KH |
4969 | |
4970 | child_id = get_new_cssid(ss, depth); | |
4971 | if (IS_ERR(child_id)) | |
4972 | return PTR_ERR(child_id); | |
4973 | ||
4974 | for (i = 0; i < depth; i++) | |
4975 | child_id->stack[i] = parent_id->stack[i]; | |
4976 | child_id->stack[depth] = child_id->id; | |
4977 | /* | |
4978 | * child_id->css pointer will be set after this cgroup is available | |
4979 | * see cgroup_populate_dir() | |
4980 | */ | |
4981 | rcu_assign_pointer(child_css->id, child_id); | |
4982 | ||
4983 | return 0; | |
4984 | } | |
4985 | ||
4986 | /** | |
4987 | * css_lookup - lookup css by id | |
4988 | * @ss: cgroup subsys to be looked into. | |
4989 | * @id: the id | |
4990 | * | |
4991 | * Returns pointer to cgroup_subsys_state if there is valid one with id. | |
4992 | * NULL if not. Should be called under rcu_read_lock() | |
4993 | */ | |
4994 | struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id) | |
4995 | { | |
4996 | struct css_id *cssid = NULL; | |
4997 | ||
4998 | BUG_ON(!ss->use_id); | |
4999 | cssid = idr_find(&ss->idr, id); | |
5000 | ||
5001 | if (unlikely(!cssid)) | |
5002 | return NULL; | |
5003 | ||
5004 | return rcu_dereference(cssid->css); | |
5005 | } | |
67523c48 | 5006 | EXPORT_SYMBOL_GPL(css_lookup); |
38460b48 KH |
5007 | |
5008 | /** | |
5009 | * css_get_next - lookup next cgroup under specified hierarchy. | |
5010 | * @ss: pointer to subsystem | |
5011 | * @id: current position of iteration. | |
5012 | * @root: pointer to css. search tree under this. | |
5013 | * @foundid: position of found object. | |
5014 | * | |
5015 | * Search next css under the specified hierarchy of rootid. Calling under | |
5016 | * rcu_read_lock() is necessary. Returns NULL if it reaches the end. | |
5017 | */ | |
5018 | struct cgroup_subsys_state * | |
5019 | css_get_next(struct cgroup_subsys *ss, int id, | |
5020 | struct cgroup_subsys_state *root, int *foundid) | |
5021 | { | |
5022 | struct cgroup_subsys_state *ret = NULL; | |
5023 | struct css_id *tmp; | |
5024 | int tmpid; | |
5025 | int rootid = css_id(root); | |
5026 | int depth = css_depth(root); | |
5027 | ||
5028 | if (!rootid) | |
5029 | return NULL; | |
5030 | ||
5031 | BUG_ON(!ss->use_id); | |
ca464d69 HD |
5032 | WARN_ON_ONCE(!rcu_read_lock_held()); |
5033 | ||
38460b48 KH |
5034 | /* fill start point for scan */ |
5035 | tmpid = id; | |
5036 | while (1) { | |
5037 | /* | |
5038 | * scan next entry from bitmap(tree), tmpid is updated after | |
5039 | * idr_get_next(). | |
5040 | */ | |
38460b48 | 5041 | tmp = idr_get_next(&ss->idr, &tmpid); |
38460b48 KH |
5042 | if (!tmp) |
5043 | break; | |
5044 | if (tmp->depth >= depth && tmp->stack[depth] == rootid) { | |
5045 | ret = rcu_dereference(tmp->css); | |
5046 | if (ret) { | |
5047 | *foundid = tmpid; | |
5048 | break; | |
5049 | } | |
5050 | } | |
5051 | /* continue to scan from next id */ | |
5052 | tmpid = tmpid + 1; | |
5053 | } | |
5054 | return ret; | |
5055 | } | |
5056 | ||
e5d1367f SE |
5057 | /* |
5058 | * get corresponding css from file open on cgroupfs directory | |
5059 | */ | |
5060 | struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id) | |
5061 | { | |
5062 | struct cgroup *cgrp; | |
5063 | struct inode *inode; | |
5064 | struct cgroup_subsys_state *css; | |
5065 | ||
5066 | inode = f->f_dentry->d_inode; | |
5067 | /* check in cgroup filesystem dir */ | |
5068 | if (inode->i_op != &cgroup_dir_inode_operations) | |
5069 | return ERR_PTR(-EBADF); | |
5070 | ||
5071 | if (id < 0 || id >= CGROUP_SUBSYS_COUNT) | |
5072 | return ERR_PTR(-EINVAL); | |
5073 | ||
5074 | /* get cgroup */ | |
5075 | cgrp = __d_cgrp(f->f_dentry); | |
5076 | css = cgrp->subsys[id]; | |
5077 | return css ? css : ERR_PTR(-ENOENT); | |
5078 | } | |
5079 | ||
fe693435 | 5080 | #ifdef CONFIG_CGROUP_DEBUG |
761b3ef5 | 5081 | static struct cgroup_subsys_state *debug_create(struct cgroup *cont) |
fe693435 PM |
5082 | { |
5083 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5084 | ||
5085 | if (!css) | |
5086 | return ERR_PTR(-ENOMEM); | |
5087 | ||
5088 | return css; | |
5089 | } | |
5090 | ||
761b3ef5 | 5091 | static void debug_destroy(struct cgroup *cont) |
fe693435 PM |
5092 | { |
5093 | kfree(cont->subsys[debug_subsys_id]); | |
5094 | } | |
5095 | ||
5096 | static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft) | |
5097 | { | |
5098 | return atomic_read(&cont->count); | |
5099 | } | |
5100 | ||
5101 | static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft) | |
5102 | { | |
5103 | return cgroup_task_count(cont); | |
5104 | } | |
5105 | ||
5106 | static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft) | |
5107 | { | |
5108 | return (u64)(unsigned long)current->cgroups; | |
5109 | } | |
5110 | ||
5111 | static u64 current_css_set_refcount_read(struct cgroup *cont, | |
5112 | struct cftype *cft) | |
5113 | { | |
5114 | u64 count; | |
5115 | ||
5116 | rcu_read_lock(); | |
5117 | count = atomic_read(¤t->cgroups->refcount); | |
5118 | rcu_read_unlock(); | |
5119 | return count; | |
5120 | } | |
5121 | ||
7717f7ba PM |
5122 | static int current_css_set_cg_links_read(struct cgroup *cont, |
5123 | struct cftype *cft, | |
5124 | struct seq_file *seq) | |
5125 | { | |
5126 | struct cg_cgroup_link *link; | |
5127 | struct css_set *cg; | |
5128 | ||
5129 | read_lock(&css_set_lock); | |
5130 | rcu_read_lock(); | |
5131 | cg = rcu_dereference(current->cgroups); | |
5132 | list_for_each_entry(link, &cg->cg_links, cg_link_list) { | |
5133 | struct cgroup *c = link->cgrp; | |
5134 | const char *name; | |
5135 | ||
5136 | if (c->dentry) | |
5137 | name = c->dentry->d_name.name; | |
5138 | else | |
5139 | name = "?"; | |
2c6ab6d2 PM |
5140 | seq_printf(seq, "Root %d group %s\n", |
5141 | c->root->hierarchy_id, name); | |
7717f7ba PM |
5142 | } |
5143 | rcu_read_unlock(); | |
5144 | read_unlock(&css_set_lock); | |
5145 | return 0; | |
5146 | } | |
5147 | ||
5148 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
5149 | static int cgroup_css_links_read(struct cgroup *cont, | |
5150 | struct cftype *cft, | |
5151 | struct seq_file *seq) | |
5152 | { | |
5153 | struct cg_cgroup_link *link; | |
5154 | ||
5155 | read_lock(&css_set_lock); | |
5156 | list_for_each_entry(link, &cont->css_sets, cgrp_link_list) { | |
5157 | struct css_set *cg = link->cg; | |
5158 | struct task_struct *task; | |
5159 | int count = 0; | |
5160 | seq_printf(seq, "css_set %p\n", cg); | |
5161 | list_for_each_entry(task, &cg->tasks, cg_list) { | |
5162 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { | |
5163 | seq_puts(seq, " ...\n"); | |
5164 | break; | |
5165 | } else { | |
5166 | seq_printf(seq, " task %d\n", | |
5167 | task_pid_vnr(task)); | |
5168 | } | |
5169 | } | |
5170 | } | |
5171 | read_unlock(&css_set_lock); | |
5172 | return 0; | |
5173 | } | |
5174 | ||
fe693435 PM |
5175 | static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft) |
5176 | { | |
5177 | return test_bit(CGRP_RELEASABLE, &cgrp->flags); | |
5178 | } | |
5179 | ||
5180 | static struct cftype debug_files[] = { | |
5181 | { | |
5182 | .name = "cgroup_refcount", | |
5183 | .read_u64 = cgroup_refcount_read, | |
5184 | }, | |
5185 | { | |
5186 | .name = "taskcount", | |
5187 | .read_u64 = debug_taskcount_read, | |
5188 | }, | |
5189 | ||
5190 | { | |
5191 | .name = "current_css_set", | |
5192 | .read_u64 = current_css_set_read, | |
5193 | }, | |
5194 | ||
5195 | { | |
5196 | .name = "current_css_set_refcount", | |
5197 | .read_u64 = current_css_set_refcount_read, | |
5198 | }, | |
5199 | ||
7717f7ba PM |
5200 | { |
5201 | .name = "current_css_set_cg_links", | |
5202 | .read_seq_string = current_css_set_cg_links_read, | |
5203 | }, | |
5204 | ||
5205 | { | |
5206 | .name = "cgroup_css_links", | |
5207 | .read_seq_string = cgroup_css_links_read, | |
5208 | }, | |
5209 | ||
fe693435 PM |
5210 | { |
5211 | .name = "releasable", | |
5212 | .read_u64 = releasable_read, | |
5213 | }, | |
5214 | }; | |
5215 | ||
5216 | static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
5217 | { | |
5218 | return cgroup_add_files(cont, ss, debug_files, | |
5219 | ARRAY_SIZE(debug_files)); | |
5220 | } | |
5221 | ||
5222 | struct cgroup_subsys debug_subsys = { | |
5223 | .name = "debug", | |
5224 | .create = debug_create, | |
5225 | .destroy = debug_destroy, | |
5226 | .populate = debug_populate, | |
5227 | .subsys_id = debug_subsys_id, | |
5228 | }; | |
5229 | #endif /* CONFIG_CGROUP_DEBUG */ |