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 | ||
780 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int 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 | ||
ddbcc7e8 PM |
795 | static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb) |
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 KH |
820 | if (ss->pre_destroy) { |
821 | ret = ss->pre_destroy(ss, cgrp); | |
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 LZ |
848 | for_each_subsys(cgrp->root, ss) |
849 | ss->destroy(ss, 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 KH |
940 | */ |
941 | DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq); | |
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) |
bd89aabc | 1018 | ss->bind(ss, 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 PM |
1027 | if (ss->bind) |
1028 | ss->bind(ss, dummytop); | |
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 | ||
1059 | static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs) | |
1060 | { | |
1061 | struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info; | |
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, | |
1196 | * otherwise 'all, 'none' and a subsystem name options were not | |
1197 | * specified, let's default to 'all' | |
1198 | */ | |
1199 | if (all_ss || (!all_ss && !one_ss && !opts->none)) { | |
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); | |
1475 | struct dentry *dentry; | |
1476 | ||
1477 | if (!inode) | |
1478 | return -ENOMEM; | |
1479 | ||
ddbcc7e8 PM |
1480 | inode->i_fop = &simple_dir_operations; |
1481 | inode->i_op = &cgroup_dir_inode_operations; | |
1482 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
1483 | inc_nlink(inode); | |
1484 | dentry = d_alloc_root(inode); | |
1485 | if (!dentry) { | |
1486 | iput(inode); | |
1487 | return -ENOMEM; | |
1488 | } | |
1489 | sb->s_root = dentry; | |
0df6a63f AV |
1490 | /* for everything else we want ->d_op set */ |
1491 | sb->s_d_op = &cgroup_dops; | |
ddbcc7e8 PM |
1492 | return 0; |
1493 | } | |
1494 | ||
f7e83571 | 1495 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1496 | int flags, const char *unused_dev_name, |
f7e83571 | 1497 | void *data) |
ddbcc7e8 PM |
1498 | { |
1499 | struct cgroup_sb_opts opts; | |
c6d57f33 | 1500 | struct cgroupfs_root *root; |
ddbcc7e8 PM |
1501 | int ret = 0; |
1502 | struct super_block *sb; | |
c6d57f33 | 1503 | struct cgroupfs_root *new_root; |
e25e2cbb | 1504 | struct inode *inode; |
ddbcc7e8 PM |
1505 | |
1506 | /* First find the desired set of subsystems */ | |
aae8aab4 | 1507 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1508 | ret = parse_cgroupfs_options(data, &opts); |
aae8aab4 | 1509 | mutex_unlock(&cgroup_mutex); |
c6d57f33 PM |
1510 | if (ret) |
1511 | goto out_err; | |
ddbcc7e8 | 1512 | |
c6d57f33 PM |
1513 | /* |
1514 | * Allocate a new cgroup root. We may not need it if we're | |
1515 | * reusing an existing hierarchy. | |
1516 | */ | |
1517 | new_root = cgroup_root_from_opts(&opts); | |
1518 | if (IS_ERR(new_root)) { | |
1519 | ret = PTR_ERR(new_root); | |
cf5d5941 | 1520 | goto drop_modules; |
81a6a5cd | 1521 | } |
c6d57f33 | 1522 | opts.new_root = new_root; |
ddbcc7e8 | 1523 | |
c6d57f33 PM |
1524 | /* Locate an existing or new sb for this hierarchy */ |
1525 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts); | |
ddbcc7e8 | 1526 | if (IS_ERR(sb)) { |
c6d57f33 | 1527 | ret = PTR_ERR(sb); |
2c6ab6d2 | 1528 | cgroup_drop_root(opts.new_root); |
cf5d5941 | 1529 | goto drop_modules; |
ddbcc7e8 PM |
1530 | } |
1531 | ||
c6d57f33 PM |
1532 | root = sb->s_fs_info; |
1533 | BUG_ON(!root); | |
1534 | if (root == opts.new_root) { | |
1535 | /* We used the new root structure, so this is a new hierarchy */ | |
1536 | struct list_head tmp_cg_links; | |
c12f65d4 | 1537 | struct cgroup *root_cgrp = &root->top_cgroup; |
c6d57f33 | 1538 | struct cgroupfs_root *existing_root; |
2ce9738b | 1539 | const struct cred *cred; |
28fd5dfc | 1540 | int i; |
ddbcc7e8 PM |
1541 | |
1542 | BUG_ON(sb->s_root != NULL); | |
1543 | ||
1544 | ret = cgroup_get_rootdir(sb); | |
1545 | if (ret) | |
1546 | goto drop_new_super; | |
817929ec | 1547 | inode = sb->s_root->d_inode; |
ddbcc7e8 | 1548 | |
817929ec | 1549 | mutex_lock(&inode->i_mutex); |
ddbcc7e8 | 1550 | mutex_lock(&cgroup_mutex); |
e25e2cbb | 1551 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 | 1552 | |
e25e2cbb TH |
1553 | /* Check for name clashes with existing mounts */ |
1554 | ret = -EBUSY; | |
1555 | if (strlen(root->name)) | |
1556 | for_each_active_root(existing_root) | |
1557 | if (!strcmp(existing_root->name, root->name)) | |
1558 | goto unlock_drop; | |
c6d57f33 | 1559 | |
817929ec PM |
1560 | /* |
1561 | * We're accessing css_set_count without locking | |
1562 | * css_set_lock here, but that's OK - it can only be | |
1563 | * increased by someone holding cgroup_lock, and | |
1564 | * that's us. The worst that can happen is that we | |
1565 | * have some link structures left over | |
1566 | */ | |
1567 | ret = allocate_cg_links(css_set_count, &tmp_cg_links); | |
e25e2cbb TH |
1568 | if (ret) |
1569 | goto unlock_drop; | |
817929ec | 1570 | |
ddbcc7e8 PM |
1571 | ret = rebind_subsystems(root, root->subsys_bits); |
1572 | if (ret == -EBUSY) { | |
c6d57f33 | 1573 | free_cg_links(&tmp_cg_links); |
e25e2cbb | 1574 | goto unlock_drop; |
ddbcc7e8 | 1575 | } |
cf5d5941 BB |
1576 | /* |
1577 | * There must be no failure case after here, since rebinding | |
1578 | * takes care of subsystems' refcounts, which are explicitly | |
1579 | * dropped in the failure exit path. | |
1580 | */ | |
ddbcc7e8 PM |
1581 | |
1582 | /* EBUSY should be the only error here */ | |
1583 | BUG_ON(ret); | |
1584 | ||
1585 | list_add(&root->root_list, &roots); | |
817929ec | 1586 | root_count++; |
ddbcc7e8 | 1587 | |
c12f65d4 | 1588 | sb->s_root->d_fsdata = root_cgrp; |
ddbcc7e8 PM |
1589 | root->top_cgroup.dentry = sb->s_root; |
1590 | ||
817929ec PM |
1591 | /* Link the top cgroup in this hierarchy into all |
1592 | * the css_set objects */ | |
1593 | write_lock(&css_set_lock); | |
28fd5dfc LZ |
1594 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { |
1595 | struct hlist_head *hhead = &css_set_table[i]; | |
1596 | struct hlist_node *node; | |
817929ec | 1597 | struct css_set *cg; |
28fd5dfc | 1598 | |
c12f65d4 LZ |
1599 | hlist_for_each_entry(cg, node, hhead, hlist) |
1600 | link_css_set(&tmp_cg_links, cg, root_cgrp); | |
28fd5dfc | 1601 | } |
817929ec PM |
1602 | write_unlock(&css_set_lock); |
1603 | ||
1604 | free_cg_links(&tmp_cg_links); | |
1605 | ||
c12f65d4 LZ |
1606 | BUG_ON(!list_empty(&root_cgrp->sibling)); |
1607 | BUG_ON(!list_empty(&root_cgrp->children)); | |
ddbcc7e8 PM |
1608 | BUG_ON(root->number_of_cgroups != 1); |
1609 | ||
2ce9738b | 1610 | cred = override_creds(&init_cred); |
c12f65d4 | 1611 | cgroup_populate_dir(root_cgrp); |
2ce9738b | 1612 | revert_creds(cred); |
e25e2cbb | 1613 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 | 1614 | mutex_unlock(&cgroup_mutex); |
34f77a90 | 1615 | mutex_unlock(&inode->i_mutex); |
c6d57f33 PM |
1616 | } else { |
1617 | /* | |
1618 | * We re-used an existing hierarchy - the new root (if | |
1619 | * any) is not needed | |
1620 | */ | |
2c6ab6d2 | 1621 | cgroup_drop_root(opts.new_root); |
cf5d5941 BB |
1622 | /* no subsys rebinding, so refcounts don't change */ |
1623 | drop_parsed_module_refcounts(opts.subsys_bits); | |
ddbcc7e8 PM |
1624 | } |
1625 | ||
c6d57f33 PM |
1626 | kfree(opts.release_agent); |
1627 | kfree(opts.name); | |
f7e83571 | 1628 | return dget(sb->s_root); |
ddbcc7e8 | 1629 | |
e25e2cbb TH |
1630 | unlock_drop: |
1631 | mutex_unlock(&cgroup_root_mutex); | |
1632 | mutex_unlock(&cgroup_mutex); | |
1633 | mutex_unlock(&inode->i_mutex); | |
ddbcc7e8 | 1634 | drop_new_super: |
6f5bbff9 | 1635 | deactivate_locked_super(sb); |
cf5d5941 BB |
1636 | drop_modules: |
1637 | drop_parsed_module_refcounts(opts.subsys_bits); | |
c6d57f33 PM |
1638 | out_err: |
1639 | kfree(opts.release_agent); | |
1640 | kfree(opts.name); | |
f7e83571 | 1641 | return ERR_PTR(ret); |
ddbcc7e8 PM |
1642 | } |
1643 | ||
1644 | static void cgroup_kill_sb(struct super_block *sb) { | |
1645 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1646 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 | 1647 | int ret; |
71cbb949 KM |
1648 | struct cg_cgroup_link *link; |
1649 | struct cg_cgroup_link *saved_link; | |
ddbcc7e8 PM |
1650 | |
1651 | BUG_ON(!root); | |
1652 | ||
1653 | BUG_ON(root->number_of_cgroups != 1); | |
bd89aabc PM |
1654 | BUG_ON(!list_empty(&cgrp->children)); |
1655 | BUG_ON(!list_empty(&cgrp->sibling)); | |
ddbcc7e8 PM |
1656 | |
1657 | mutex_lock(&cgroup_mutex); | |
e25e2cbb | 1658 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1659 | |
1660 | /* Rebind all subsystems back to the default hierarchy */ | |
1661 | ret = rebind_subsystems(root, 0); | |
1662 | /* Shouldn't be able to fail ... */ | |
1663 | BUG_ON(ret); | |
1664 | ||
817929ec PM |
1665 | /* |
1666 | * Release all the links from css_sets to this hierarchy's | |
1667 | * root cgroup | |
1668 | */ | |
1669 | write_lock(&css_set_lock); | |
71cbb949 KM |
1670 | |
1671 | list_for_each_entry_safe(link, saved_link, &cgrp->css_sets, | |
1672 | cgrp_link_list) { | |
817929ec | 1673 | list_del(&link->cg_link_list); |
bd89aabc | 1674 | list_del(&link->cgrp_link_list); |
817929ec PM |
1675 | kfree(link); |
1676 | } | |
1677 | write_unlock(&css_set_lock); | |
1678 | ||
839ec545 PM |
1679 | if (!list_empty(&root->root_list)) { |
1680 | list_del(&root->root_list); | |
1681 | root_count--; | |
1682 | } | |
e5f6a860 | 1683 | |
e25e2cbb | 1684 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1685 | mutex_unlock(&cgroup_mutex); |
1686 | ||
ddbcc7e8 | 1687 | kill_litter_super(sb); |
2c6ab6d2 | 1688 | cgroup_drop_root(root); |
ddbcc7e8 PM |
1689 | } |
1690 | ||
1691 | static struct file_system_type cgroup_fs_type = { | |
1692 | .name = "cgroup", | |
f7e83571 | 1693 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1694 | .kill_sb = cgroup_kill_sb, |
1695 | }; | |
1696 | ||
676db4af GK |
1697 | static struct kobject *cgroup_kobj; |
1698 | ||
bd89aabc | 1699 | static inline struct cgroup *__d_cgrp(struct dentry *dentry) |
ddbcc7e8 PM |
1700 | { |
1701 | return dentry->d_fsdata; | |
1702 | } | |
1703 | ||
1704 | static inline struct cftype *__d_cft(struct dentry *dentry) | |
1705 | { | |
1706 | return dentry->d_fsdata; | |
1707 | } | |
1708 | ||
a043e3b2 LZ |
1709 | /** |
1710 | * cgroup_path - generate the path of a cgroup | |
1711 | * @cgrp: the cgroup in question | |
1712 | * @buf: the buffer to write the path into | |
1713 | * @buflen: the length of the buffer | |
1714 | * | |
a47295e6 PM |
1715 | * Called with cgroup_mutex held or else with an RCU-protected cgroup |
1716 | * reference. Writes path of cgroup into buf. Returns 0 on success, | |
1717 | * -errno on error. | |
ddbcc7e8 | 1718 | */ |
bd89aabc | 1719 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
ddbcc7e8 PM |
1720 | { |
1721 | char *start; | |
9a9686b6 | 1722 | struct dentry *dentry = rcu_dereference_check(cgrp->dentry, |
9a9686b6 | 1723 | cgroup_lock_is_held()); |
ddbcc7e8 | 1724 | |
a47295e6 | 1725 | if (!dentry || cgrp == dummytop) { |
ddbcc7e8 PM |
1726 | /* |
1727 | * Inactive subsystems have no dentry for their root | |
1728 | * cgroup | |
1729 | */ | |
1730 | strcpy(buf, "/"); | |
1731 | return 0; | |
1732 | } | |
1733 | ||
1734 | start = buf + buflen; | |
1735 | ||
1736 | *--start = '\0'; | |
1737 | for (;;) { | |
a47295e6 | 1738 | int len = dentry->d_name.len; |
9a9686b6 | 1739 | |
ddbcc7e8 PM |
1740 | if ((start -= len) < buf) |
1741 | return -ENAMETOOLONG; | |
9a9686b6 | 1742 | memcpy(start, dentry->d_name.name, len); |
bd89aabc PM |
1743 | cgrp = cgrp->parent; |
1744 | if (!cgrp) | |
ddbcc7e8 | 1745 | break; |
9a9686b6 LZ |
1746 | |
1747 | dentry = rcu_dereference_check(cgrp->dentry, | |
9a9686b6 | 1748 | cgroup_lock_is_held()); |
bd89aabc | 1749 | if (!cgrp->parent) |
ddbcc7e8 PM |
1750 | continue; |
1751 | if (--start < buf) | |
1752 | return -ENAMETOOLONG; | |
1753 | *start = '/'; | |
1754 | } | |
1755 | memmove(buf, start, buf + buflen - start); | |
1756 | return 0; | |
1757 | } | |
67523c48 | 1758 | EXPORT_SYMBOL_GPL(cgroup_path); |
ddbcc7e8 | 1759 | |
134d3373 TH |
1760 | struct task_and_cgroup { |
1761 | struct task_struct *task; | |
1762 | struct cgroup *cgrp; | |
1763 | }; | |
1764 | ||
74a1166d BB |
1765 | /* |
1766 | * cgroup_task_migrate - move a task from one cgroup to another. | |
1767 | * | |
1768 | * 'guarantee' is set if the caller promises that a new css_set for the task | |
1769 | * will already exist. If not set, this function might sleep, and can fail with | |
cd3d0952 | 1770 | * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked. |
74a1166d BB |
1771 | */ |
1772 | static int cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp, | |
1773 | struct task_struct *tsk, bool guarantee) | |
1774 | { | |
1775 | struct css_set *oldcg; | |
1776 | struct css_set *newcg; | |
1777 | ||
1778 | /* | |
1779 | * get old css_set. we need to take task_lock and refcount it, because | |
1780 | * an exiting task can change its css_set to init_css_set and drop its | |
1781 | * old one without taking cgroup_mutex. | |
1782 | */ | |
1783 | task_lock(tsk); | |
1784 | oldcg = tsk->cgroups; | |
1785 | get_css_set(oldcg); | |
1786 | task_unlock(tsk); | |
1787 | ||
1788 | /* locate or allocate a new css_set for this task. */ | |
1789 | if (guarantee) { | |
1790 | /* we know the css_set we want already exists. */ | |
1791 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT]; | |
1792 | read_lock(&css_set_lock); | |
1793 | newcg = find_existing_css_set(oldcg, cgrp, template); | |
1794 | BUG_ON(!newcg); | |
1795 | get_css_set(newcg); | |
1796 | read_unlock(&css_set_lock); | |
1797 | } else { | |
1798 | might_sleep(); | |
1799 | /* find_css_set will give us newcg already referenced. */ | |
1800 | newcg = find_css_set(oldcg, cgrp); | |
1801 | if (!newcg) { | |
1802 | put_css_set(oldcg); | |
1803 | return -ENOMEM; | |
1804 | } | |
1805 | } | |
1806 | put_css_set(oldcg); | |
1807 | ||
cd3d0952 | 1808 | /* @tsk can't exit as its threadgroup is locked */ |
74a1166d | 1809 | task_lock(tsk); |
cd3d0952 | 1810 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
74a1166d BB |
1811 | rcu_assign_pointer(tsk->cgroups, newcg); |
1812 | task_unlock(tsk); | |
1813 | ||
1814 | /* Update the css_set linked lists if we're using them */ | |
1815 | write_lock(&css_set_lock); | |
1816 | if (!list_empty(&tsk->cg_list)) | |
1817 | list_move(&tsk->cg_list, &newcg->tasks); | |
1818 | write_unlock(&css_set_lock); | |
1819 | ||
1820 | /* | |
1821 | * We just gained a reference on oldcg by taking it from the task. As | |
1822 | * trading it for newcg is protected by cgroup_mutex, we're safe to drop | |
1823 | * it here; it will be freed under RCU. | |
1824 | */ | |
1825 | put_css_set(oldcg); | |
1826 | ||
1827 | set_bit(CGRP_RELEASABLE, &oldcgrp->flags); | |
1828 | return 0; | |
1829 | } | |
1830 | ||
a043e3b2 LZ |
1831 | /** |
1832 | * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp' | |
1833 | * @cgrp: the cgroup the task is attaching to | |
1834 | * @tsk: the task to be attached | |
bbcb81d0 | 1835 | * |
cd3d0952 TH |
1836 | * Call with cgroup_mutex and threadgroup locked. May take task_lock of |
1837 | * @tsk during call. | |
bbcb81d0 | 1838 | */ |
956db3ca | 1839 | int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
bbcb81d0 | 1840 | { |
74a1166d | 1841 | int retval; |
2468c723 | 1842 | struct cgroup_subsys *ss, *failed_ss = NULL; |
bd89aabc | 1843 | struct cgroup *oldcgrp; |
bd89aabc | 1844 | struct cgroupfs_root *root = cgrp->root; |
bbcb81d0 | 1845 | |
cd3d0952 TH |
1846 | /* @tsk either already exited or can't exit until the end */ |
1847 | if (tsk->flags & PF_EXITING) | |
1848 | return -ESRCH; | |
1849 | ||
bbcb81d0 | 1850 | /* Nothing to do if the task is already in that cgroup */ |
7717f7ba | 1851 | oldcgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 1852 | if (cgrp == oldcgrp) |
bbcb81d0 PM |
1853 | return 0; |
1854 | ||
1855 | for_each_subsys(root, ss) { | |
1856 | if (ss->can_attach) { | |
f780bdb7 | 1857 | retval = ss->can_attach(ss, cgrp, tsk); |
2468c723 DN |
1858 | if (retval) { |
1859 | /* | |
1860 | * Remember on which subsystem the can_attach() | |
1861 | * failed, so that we only call cancel_attach() | |
1862 | * against the subsystems whose can_attach() | |
1863 | * succeeded. (See below) | |
1864 | */ | |
1865 | failed_ss = ss; | |
1866 | goto out; | |
1867 | } | |
bbcb81d0 | 1868 | } |
f780bdb7 BB |
1869 | if (ss->can_attach_task) { |
1870 | retval = ss->can_attach_task(cgrp, tsk); | |
1871 | if (retval) { | |
1872 | failed_ss = ss; | |
1873 | goto out; | |
1874 | } | |
1875 | } | |
bbcb81d0 PM |
1876 | } |
1877 | ||
74a1166d BB |
1878 | retval = cgroup_task_migrate(cgrp, oldcgrp, tsk, false); |
1879 | if (retval) | |
2468c723 | 1880 | goto out; |
817929ec | 1881 | |
bbcb81d0 | 1882 | for_each_subsys(root, ss) { |
f780bdb7 BB |
1883 | if (ss->pre_attach) |
1884 | ss->pre_attach(cgrp); | |
1885 | if (ss->attach_task) | |
1886 | ss->attach_task(cgrp, tsk); | |
e18f6318 | 1887 | if (ss->attach) |
f780bdb7 | 1888 | ss->attach(ss, cgrp, oldcgrp, tsk); |
bbcb81d0 | 1889 | } |
74a1166d | 1890 | |
bbcb81d0 | 1891 | synchronize_rcu(); |
ec64f515 KH |
1892 | |
1893 | /* | |
1894 | * wake up rmdir() waiter. the rmdir should fail since the cgroup | |
1895 | * is no longer empty. | |
1896 | */ | |
88703267 | 1897 | cgroup_wakeup_rmdir_waiter(cgrp); |
2468c723 DN |
1898 | out: |
1899 | if (retval) { | |
1900 | for_each_subsys(root, ss) { | |
1901 | if (ss == failed_ss) | |
1902 | /* | |
1903 | * This subsystem was the one that failed the | |
1904 | * can_attach() check earlier, so we don't need | |
1905 | * to call cancel_attach() against it or any | |
1906 | * remaining subsystems. | |
1907 | */ | |
1908 | break; | |
1909 | if (ss->cancel_attach) | |
f780bdb7 | 1910 | ss->cancel_attach(ss, cgrp, tsk); |
2468c723 DN |
1911 | } |
1912 | } | |
1913 | return retval; | |
bbcb81d0 PM |
1914 | } |
1915 | ||
d7926ee3 | 1916 | /** |
31583bb0 MT |
1917 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' |
1918 | * @from: attach to all cgroups of a given task | |
d7926ee3 SS |
1919 | * @tsk: the task to be attached |
1920 | */ | |
31583bb0 | 1921 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) |
d7926ee3 SS |
1922 | { |
1923 | struct cgroupfs_root *root; | |
d7926ee3 SS |
1924 | int retval = 0; |
1925 | ||
1926 | cgroup_lock(); | |
1927 | for_each_active_root(root) { | |
31583bb0 MT |
1928 | struct cgroup *from_cg = task_cgroup_from_root(from, root); |
1929 | ||
1930 | retval = cgroup_attach_task(from_cg, tsk); | |
d7926ee3 SS |
1931 | if (retval) |
1932 | break; | |
1933 | } | |
1934 | cgroup_unlock(); | |
1935 | ||
1936 | return retval; | |
1937 | } | |
31583bb0 | 1938 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); |
d7926ee3 | 1939 | |
bbcb81d0 | 1940 | /* |
74a1166d BB |
1941 | * cgroup_attach_proc works in two stages, the first of which prefetches all |
1942 | * new css_sets needed (to make sure we have enough memory before committing | |
1943 | * to the move) and stores them in a list of entries of the following type. | |
1944 | * TODO: possible optimization: use css_set->rcu_head for chaining instead | |
1945 | */ | |
1946 | struct cg_list_entry { | |
1947 | struct css_set *cg; | |
1948 | struct list_head links; | |
1949 | }; | |
1950 | ||
1951 | static bool css_set_check_fetched(struct cgroup *cgrp, | |
1952 | struct task_struct *tsk, struct css_set *cg, | |
1953 | struct list_head *newcg_list) | |
1954 | { | |
1955 | struct css_set *newcg; | |
1956 | struct cg_list_entry *cg_entry; | |
1957 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT]; | |
1958 | ||
1959 | read_lock(&css_set_lock); | |
1960 | newcg = find_existing_css_set(cg, cgrp, template); | |
1961 | if (newcg) | |
1962 | get_css_set(newcg); | |
1963 | read_unlock(&css_set_lock); | |
1964 | ||
1965 | /* doesn't exist at all? */ | |
1966 | if (!newcg) | |
1967 | return false; | |
1968 | /* see if it's already in the list */ | |
1969 | list_for_each_entry(cg_entry, newcg_list, links) { | |
1970 | if (cg_entry->cg == newcg) { | |
1971 | put_css_set(newcg); | |
1972 | return true; | |
1973 | } | |
1974 | } | |
1975 | ||
1976 | /* not found */ | |
1977 | put_css_set(newcg); | |
1978 | return false; | |
1979 | } | |
1980 | ||
1981 | /* | |
1982 | * Find the new css_set and store it in the list in preparation for moving the | |
1983 | * given task to the given cgroup. Returns 0 or -ENOMEM. | |
1984 | */ | |
1985 | static int css_set_prefetch(struct cgroup *cgrp, struct css_set *cg, | |
1986 | struct list_head *newcg_list) | |
1987 | { | |
1988 | struct css_set *newcg; | |
1989 | struct cg_list_entry *cg_entry; | |
1990 | ||
1991 | /* ensure a new css_set will exist for this thread */ | |
1992 | newcg = find_css_set(cg, cgrp); | |
1993 | if (!newcg) | |
1994 | return -ENOMEM; | |
1995 | /* add it to the list */ | |
1996 | cg_entry = kmalloc(sizeof(struct cg_list_entry), GFP_KERNEL); | |
1997 | if (!cg_entry) { | |
1998 | put_css_set(newcg); | |
1999 | return -ENOMEM; | |
2000 | } | |
2001 | cg_entry->cg = newcg; | |
2002 | list_add(&cg_entry->links, newcg_list); | |
2003 | return 0; | |
2004 | } | |
2005 | ||
2006 | /** | |
2007 | * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup | |
2008 | * @cgrp: the cgroup to attach to | |
2009 | * @leader: the threadgroup leader task_struct of the group to be attached | |
2010 | * | |
257058ae TH |
2011 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
2012 | * task_lock of each thread in leader's threadgroup individually in turn. | |
74a1166d BB |
2013 | */ |
2014 | int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader) | |
2015 | { | |
134d3373 | 2016 | int retval, i, group_size, nr_migrating_tasks; |
74a1166d BB |
2017 | struct cgroup_subsys *ss, *failed_ss = NULL; |
2018 | bool cancel_failed_ss = false; | |
2019 | /* guaranteed to be initialized later, but the compiler needs this */ | |
74a1166d BB |
2020 | struct css_set *oldcg; |
2021 | struct cgroupfs_root *root = cgrp->root; | |
2022 | /* threadgroup list cursor and array */ | |
2023 | struct task_struct *tsk; | |
134d3373 | 2024 | struct task_and_cgroup *tc; |
d846687d | 2025 | struct flex_array *group; |
74a1166d BB |
2026 | /* |
2027 | * we need to make sure we have css_sets for all the tasks we're | |
2028 | * going to move -before- we actually start moving them, so that in | |
2029 | * case we get an ENOMEM we can bail out before making any changes. | |
2030 | */ | |
2031 | struct list_head newcg_list; | |
2032 | struct cg_list_entry *cg_entry, *temp_nobe; | |
2033 | ||
2034 | /* | |
2035 | * step 0: in order to do expensive, possibly blocking operations for | |
2036 | * every thread, we cannot iterate the thread group list, since it needs | |
2037 | * rcu or tasklist locked. instead, build an array of all threads in the | |
257058ae TH |
2038 | * group - group_rwsem prevents new threads from appearing, and if |
2039 | * threads exit, this will just be an over-estimate. | |
74a1166d BB |
2040 | */ |
2041 | group_size = get_nr_threads(leader); | |
d846687d | 2042 | /* flex_array supports very large thread-groups better than kmalloc. */ |
134d3373 | 2043 | group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL); |
74a1166d BB |
2044 | if (!group) |
2045 | return -ENOMEM; | |
d846687d BB |
2046 | /* pre-allocate to guarantee space while iterating in rcu read-side. */ |
2047 | retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL); | |
2048 | if (retval) | |
2049 | goto out_free_group_list; | |
74a1166d BB |
2050 | |
2051 | /* prevent changes to the threadgroup list while we take a snapshot. */ | |
33ef6b69 | 2052 | read_lock(&tasklist_lock); |
74a1166d BB |
2053 | if (!thread_group_leader(leader)) { |
2054 | /* | |
2055 | * a race with de_thread from another thread's exec() may strip | |
2056 | * us of our leadership, making while_each_thread unsafe to use | |
2057 | * on this task. if this happens, there is no choice but to | |
2058 | * throw this task away and try again (from cgroup_procs_write); | |
2059 | * this is "double-double-toil-and-trouble-check locking". | |
2060 | */ | |
33ef6b69 | 2061 | read_unlock(&tasklist_lock); |
74a1166d BB |
2062 | retval = -EAGAIN; |
2063 | goto out_free_group_list; | |
2064 | } | |
2065 | /* take a reference on each task in the group to go in the array. */ | |
2066 | tsk = leader; | |
134d3373 | 2067 | i = nr_migrating_tasks = 0; |
74a1166d | 2068 | do { |
134d3373 TH |
2069 | struct task_and_cgroup ent; |
2070 | ||
cd3d0952 TH |
2071 | /* @tsk either already exited or can't exit until the end */ |
2072 | if (tsk->flags & PF_EXITING) | |
2073 | continue; | |
2074 | ||
74a1166d BB |
2075 | /* as per above, nr_threads may decrease, but not increase. */ |
2076 | BUG_ON(i >= group_size); | |
2077 | get_task_struct(tsk); | |
d846687d BB |
2078 | /* |
2079 | * saying GFP_ATOMIC has no effect here because we did prealloc | |
2080 | * earlier, but it's good form to communicate our expectations. | |
2081 | */ | |
134d3373 TH |
2082 | ent.task = tsk; |
2083 | ent.cgrp = task_cgroup_from_root(tsk, root); | |
2084 | retval = flex_array_put(group, i, &ent, GFP_ATOMIC); | |
d846687d | 2085 | BUG_ON(retval != 0); |
74a1166d | 2086 | i++; |
134d3373 TH |
2087 | if (ent.cgrp != cgrp) |
2088 | nr_migrating_tasks++; | |
74a1166d BB |
2089 | } while_each_thread(leader, tsk); |
2090 | /* remember the number of threads in the array for later. */ | |
2091 | group_size = i; | |
33ef6b69 | 2092 | read_unlock(&tasklist_lock); |
74a1166d | 2093 | |
134d3373 TH |
2094 | /* methods shouldn't be called if no task is actually migrating */ |
2095 | retval = 0; | |
2096 | if (!nr_migrating_tasks) | |
2097 | goto out_put_tasks; | |
2098 | ||
74a1166d BB |
2099 | /* |
2100 | * step 1: check that we can legitimately attach to the cgroup. | |
2101 | */ | |
2102 | for_each_subsys(root, ss) { | |
2103 | if (ss->can_attach) { | |
2104 | retval = ss->can_attach(ss, cgrp, leader); | |
2105 | if (retval) { | |
2106 | failed_ss = ss; | |
2107 | goto out_cancel_attach; | |
2108 | } | |
2109 | } | |
2110 | /* a callback to be run on every thread in the threadgroup. */ | |
2111 | if (ss->can_attach_task) { | |
2112 | /* run on each task in the threadgroup. */ | |
2113 | for (i = 0; i < group_size; i++) { | |
134d3373 TH |
2114 | tc = flex_array_get(group, i); |
2115 | if (tc->cgrp == cgrp) | |
2116 | continue; | |
2117 | retval = ss->can_attach_task(cgrp, tc->task); | |
74a1166d BB |
2118 | if (retval) { |
2119 | failed_ss = ss; | |
2120 | cancel_failed_ss = true; | |
2121 | goto out_cancel_attach; | |
2122 | } | |
2123 | } | |
2124 | } | |
2125 | } | |
2126 | ||
2127 | /* | |
2128 | * step 2: make sure css_sets exist for all threads to be migrated. | |
2129 | * we use find_css_set, which allocates a new one if necessary. | |
2130 | */ | |
2131 | INIT_LIST_HEAD(&newcg_list); | |
2132 | for (i = 0; i < group_size; i++) { | |
134d3373 | 2133 | tc = flex_array_get(group, i); |
74a1166d | 2134 | /* nothing to do if this task is already in the cgroup */ |
134d3373 | 2135 | if (tc->cgrp == cgrp) |
74a1166d BB |
2136 | continue; |
2137 | /* get old css_set pointer */ | |
134d3373 TH |
2138 | task_lock(tc->task); |
2139 | oldcg = tc->task->cgroups; | |
74a1166d | 2140 | get_css_set(oldcg); |
134d3373 | 2141 | task_unlock(tc->task); |
74a1166d | 2142 | /* see if the new one for us is already in the list? */ |
134d3373 | 2143 | if (css_set_check_fetched(cgrp, tc->task, oldcg, &newcg_list)) { |
74a1166d BB |
2144 | /* was already there, nothing to do. */ |
2145 | put_css_set(oldcg); | |
2146 | } else { | |
2147 | /* we don't already have it. get new one. */ | |
2148 | retval = css_set_prefetch(cgrp, oldcg, &newcg_list); | |
2149 | put_css_set(oldcg); | |
2150 | if (retval) | |
2151 | goto out_list_teardown; | |
2152 | } | |
2153 | } | |
2154 | ||
2155 | /* | |
2156 | * step 3: now that we're guaranteed success wrt the css_sets, proceed | |
2157 | * to move all tasks to the new cgroup, calling ss->attach_task for each | |
2158 | * one along the way. there are no failure cases after here, so this is | |
2159 | * the commit point. | |
2160 | */ | |
2161 | for_each_subsys(root, ss) { | |
2162 | if (ss->pre_attach) | |
2163 | ss->pre_attach(cgrp); | |
2164 | } | |
2165 | for (i = 0; i < group_size; i++) { | |
134d3373 | 2166 | tc = flex_array_get(group, i); |
74a1166d | 2167 | /* leave current thread as it is if it's already there */ |
134d3373 | 2168 | if (tc->cgrp == cgrp) |
74a1166d | 2169 | continue; |
134d3373 | 2170 | retval = cgroup_task_migrate(cgrp, tc->cgrp, tc->task, true); |
cd3d0952 TH |
2171 | BUG_ON(retval); |
2172 | /* attach each task to each subsystem */ | |
2173 | for_each_subsys(root, ss) { | |
2174 | if (ss->attach_task) | |
134d3373 | 2175 | ss->attach_task(cgrp, tc->task); |
77ceab8e | 2176 | } |
74a1166d BB |
2177 | } |
2178 | /* nothing is sensitive to fork() after this point. */ | |
2179 | ||
2180 | /* | |
2181 | * step 4: do expensive, non-thread-specific subsystem callbacks. | |
2182 | * TODO: if ever a subsystem needs to know the oldcgrp for each task | |
2183 | * being moved, this call will need to be reworked to communicate that. | |
2184 | */ | |
2185 | for_each_subsys(root, ss) { | |
134d3373 TH |
2186 | if (ss->attach) { |
2187 | tc = flex_array_get(group, 0); | |
2188 | ss->attach(ss, cgrp, tc->cgrp, tc->task); | |
2189 | } | |
74a1166d BB |
2190 | } |
2191 | ||
2192 | /* | |
2193 | * step 5: success! and cleanup | |
2194 | */ | |
2195 | synchronize_rcu(); | |
2196 | cgroup_wakeup_rmdir_waiter(cgrp); | |
2197 | retval = 0; | |
2198 | out_list_teardown: | |
2199 | /* clean up the list of prefetched css_sets. */ | |
2200 | list_for_each_entry_safe(cg_entry, temp_nobe, &newcg_list, links) { | |
2201 | list_del(&cg_entry->links); | |
2202 | put_css_set(cg_entry->cg); | |
2203 | kfree(cg_entry); | |
2204 | } | |
2205 | out_cancel_attach: | |
2206 | /* same deal as in cgroup_attach_task */ | |
2207 | if (retval) { | |
2208 | for_each_subsys(root, ss) { | |
2209 | if (ss == failed_ss) { | |
2210 | if (cancel_failed_ss && ss->cancel_attach) | |
2211 | ss->cancel_attach(ss, cgrp, leader); | |
2212 | break; | |
2213 | } | |
2214 | if (ss->cancel_attach) | |
2215 | ss->cancel_attach(ss, cgrp, leader); | |
2216 | } | |
2217 | } | |
134d3373 | 2218 | out_put_tasks: |
74a1166d | 2219 | /* clean up the array of referenced threads in the group. */ |
d846687d | 2220 | for (i = 0; i < group_size; i++) { |
134d3373 TH |
2221 | tc = flex_array_get(group, i); |
2222 | put_task_struct(tc->task); | |
d846687d | 2223 | } |
74a1166d | 2224 | out_free_group_list: |
d846687d | 2225 | flex_array_free(group); |
74a1166d BB |
2226 | return retval; |
2227 | } | |
2228 | ||
2229 | /* | |
2230 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
2231 | * function to attach either it or all tasks in its threadgroup. Will lock |
2232 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 2233 | */ |
74a1166d | 2234 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2235 | { |
bbcb81d0 | 2236 | struct task_struct *tsk; |
c69e8d9c | 2237 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2238 | int ret; |
2239 | ||
74a1166d BB |
2240 | if (!cgroup_lock_live_group(cgrp)) |
2241 | return -ENODEV; | |
2242 | ||
bbcb81d0 PM |
2243 | if (pid) { |
2244 | rcu_read_lock(); | |
73507f33 | 2245 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2246 | if (!tsk) { |
2247 | rcu_read_unlock(); | |
2248 | cgroup_unlock(); | |
2249 | return -ESRCH; | |
2250 | } | |
2251 | if (threadgroup) { | |
2252 | /* | |
2253 | * RCU protects this access, since tsk was found in the | |
2254 | * tid map. a race with de_thread may cause group_leader | |
2255 | * to stop being the leader, but cgroup_attach_proc will | |
2256 | * detect it later. | |
2257 | */ | |
2258 | tsk = tsk->group_leader; | |
bbcb81d0 | 2259 | } |
74a1166d BB |
2260 | /* |
2261 | * even if we're attaching all tasks in the thread group, we | |
2262 | * only need to check permissions on one of them. | |
2263 | */ | |
c69e8d9c DH |
2264 | tcred = __task_cred(tsk); |
2265 | if (cred->euid && | |
2266 | cred->euid != tcred->uid && | |
2267 | cred->euid != tcred->suid) { | |
2268 | rcu_read_unlock(); | |
74a1166d | 2269 | cgroup_unlock(); |
bbcb81d0 PM |
2270 | return -EACCES; |
2271 | } | |
c69e8d9c DH |
2272 | get_task_struct(tsk); |
2273 | rcu_read_unlock(); | |
bbcb81d0 | 2274 | } else { |
74a1166d BB |
2275 | if (threadgroup) |
2276 | tsk = current->group_leader; | |
2277 | else | |
2278 | tsk = current; | |
bbcb81d0 PM |
2279 | get_task_struct(tsk); |
2280 | } | |
2281 | ||
cd3d0952 TH |
2282 | threadgroup_lock(tsk); |
2283 | ||
2284 | if (threadgroup) | |
74a1166d | 2285 | ret = cgroup_attach_proc(cgrp, tsk); |
cd3d0952 | 2286 | else |
74a1166d | 2287 | ret = cgroup_attach_task(cgrp, tsk); |
cd3d0952 TH |
2288 | |
2289 | threadgroup_unlock(tsk); | |
2290 | ||
bbcb81d0 | 2291 | put_task_struct(tsk); |
74a1166d | 2292 | cgroup_unlock(); |
bbcb81d0 PM |
2293 | return ret; |
2294 | } | |
2295 | ||
af351026 | 2296 | static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid) |
74a1166d BB |
2297 | { |
2298 | return attach_task_by_pid(cgrp, pid, false); | |
2299 | } | |
2300 | ||
2301 | static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid) | |
af351026 PM |
2302 | { |
2303 | int ret; | |
74a1166d BB |
2304 | do { |
2305 | /* | |
2306 | * attach_proc fails with -EAGAIN if threadgroup leadership | |
2307 | * changes in the middle of the operation, in which case we need | |
2308 | * to find the task_struct for the new leader and start over. | |
2309 | */ | |
2310 | ret = attach_task_by_pid(cgrp, tgid, true); | |
2311 | } while (ret == -EAGAIN); | |
af351026 PM |
2312 | return ret; |
2313 | } | |
2314 | ||
e788e066 PM |
2315 | /** |
2316 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
2317 | * @cgrp: the cgroup to be checked for liveness | |
2318 | * | |
84eea842 PM |
2319 | * On success, returns true; the lock should be later released with |
2320 | * cgroup_unlock(). On failure returns false with no lock held. | |
e788e066 | 2321 | */ |
84eea842 | 2322 | bool cgroup_lock_live_group(struct cgroup *cgrp) |
e788e066 PM |
2323 | { |
2324 | mutex_lock(&cgroup_mutex); | |
2325 | if (cgroup_is_removed(cgrp)) { | |
2326 | mutex_unlock(&cgroup_mutex); | |
2327 | return false; | |
2328 | } | |
2329 | return true; | |
2330 | } | |
67523c48 | 2331 | EXPORT_SYMBOL_GPL(cgroup_lock_live_group); |
e788e066 PM |
2332 | |
2333 | static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft, | |
2334 | const char *buffer) | |
2335 | { | |
2336 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
f4a2589f EK |
2337 | if (strlen(buffer) >= PATH_MAX) |
2338 | return -EINVAL; | |
e788e066 PM |
2339 | if (!cgroup_lock_live_group(cgrp)) |
2340 | return -ENODEV; | |
e25e2cbb | 2341 | mutex_lock(&cgroup_root_mutex); |
e788e066 | 2342 | strcpy(cgrp->root->release_agent_path, buffer); |
e25e2cbb | 2343 | mutex_unlock(&cgroup_root_mutex); |
84eea842 | 2344 | cgroup_unlock(); |
e788e066 PM |
2345 | return 0; |
2346 | } | |
2347 | ||
2348 | static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft, | |
2349 | struct seq_file *seq) | |
2350 | { | |
2351 | if (!cgroup_lock_live_group(cgrp)) | |
2352 | return -ENODEV; | |
2353 | seq_puts(seq, cgrp->root->release_agent_path); | |
2354 | seq_putc(seq, '\n'); | |
84eea842 | 2355 | cgroup_unlock(); |
e788e066 PM |
2356 | return 0; |
2357 | } | |
2358 | ||
84eea842 PM |
2359 | /* A buffer size big enough for numbers or short strings */ |
2360 | #define CGROUP_LOCAL_BUFFER_SIZE 64 | |
2361 | ||
e73d2c61 | 2362 | static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft, |
f4c753b7 PM |
2363 | struct file *file, |
2364 | const char __user *userbuf, | |
2365 | size_t nbytes, loff_t *unused_ppos) | |
355e0c48 | 2366 | { |
84eea842 | 2367 | char buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
355e0c48 | 2368 | int retval = 0; |
355e0c48 PM |
2369 | char *end; |
2370 | ||
2371 | if (!nbytes) | |
2372 | return -EINVAL; | |
2373 | if (nbytes >= sizeof(buffer)) | |
2374 | return -E2BIG; | |
2375 | if (copy_from_user(buffer, userbuf, nbytes)) | |
2376 | return -EFAULT; | |
2377 | ||
2378 | buffer[nbytes] = 0; /* nul-terminate */ | |
e73d2c61 | 2379 | if (cft->write_u64) { |
478988d3 | 2380 | u64 val = simple_strtoull(strstrip(buffer), &end, 0); |
e73d2c61 PM |
2381 | if (*end) |
2382 | return -EINVAL; | |
2383 | retval = cft->write_u64(cgrp, cft, val); | |
2384 | } else { | |
478988d3 | 2385 | s64 val = simple_strtoll(strstrip(buffer), &end, 0); |
e73d2c61 PM |
2386 | if (*end) |
2387 | return -EINVAL; | |
2388 | retval = cft->write_s64(cgrp, cft, val); | |
2389 | } | |
355e0c48 PM |
2390 | if (!retval) |
2391 | retval = nbytes; | |
2392 | return retval; | |
2393 | } | |
2394 | ||
db3b1497 PM |
2395 | static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft, |
2396 | struct file *file, | |
2397 | const char __user *userbuf, | |
2398 | size_t nbytes, loff_t *unused_ppos) | |
2399 | { | |
84eea842 | 2400 | char local_buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
db3b1497 PM |
2401 | int retval = 0; |
2402 | size_t max_bytes = cft->max_write_len; | |
2403 | char *buffer = local_buffer; | |
2404 | ||
2405 | if (!max_bytes) | |
2406 | max_bytes = sizeof(local_buffer) - 1; | |
2407 | if (nbytes >= max_bytes) | |
2408 | return -E2BIG; | |
2409 | /* Allocate a dynamic buffer if we need one */ | |
2410 | if (nbytes >= sizeof(local_buffer)) { | |
2411 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); | |
2412 | if (buffer == NULL) | |
2413 | return -ENOMEM; | |
2414 | } | |
5a3eb9f6 LZ |
2415 | if (nbytes && copy_from_user(buffer, userbuf, nbytes)) { |
2416 | retval = -EFAULT; | |
2417 | goto out; | |
2418 | } | |
db3b1497 PM |
2419 | |
2420 | buffer[nbytes] = 0; /* nul-terminate */ | |
478988d3 | 2421 | retval = cft->write_string(cgrp, cft, strstrip(buffer)); |
db3b1497 PM |
2422 | if (!retval) |
2423 | retval = nbytes; | |
5a3eb9f6 | 2424 | out: |
db3b1497 PM |
2425 | if (buffer != local_buffer) |
2426 | kfree(buffer); | |
2427 | return retval; | |
2428 | } | |
2429 | ||
ddbcc7e8 PM |
2430 | static ssize_t cgroup_file_write(struct file *file, const char __user *buf, |
2431 | size_t nbytes, loff_t *ppos) | |
2432 | { | |
2433 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 2434 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 2435 | |
75139b82 | 2436 | if (cgroup_is_removed(cgrp)) |
ddbcc7e8 | 2437 | return -ENODEV; |
355e0c48 | 2438 | if (cft->write) |
bd89aabc | 2439 | return cft->write(cgrp, cft, file, buf, nbytes, ppos); |
e73d2c61 PM |
2440 | if (cft->write_u64 || cft->write_s64) |
2441 | return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos); | |
db3b1497 PM |
2442 | if (cft->write_string) |
2443 | return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos); | |
d447ea2f PE |
2444 | if (cft->trigger) { |
2445 | int ret = cft->trigger(cgrp, (unsigned int)cft->private); | |
2446 | return ret ? ret : nbytes; | |
2447 | } | |
355e0c48 | 2448 | return -EINVAL; |
ddbcc7e8 PM |
2449 | } |
2450 | ||
f4c753b7 PM |
2451 | static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft, |
2452 | struct file *file, | |
2453 | char __user *buf, size_t nbytes, | |
2454 | loff_t *ppos) | |
ddbcc7e8 | 2455 | { |
84eea842 | 2456 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
f4c753b7 | 2457 | u64 val = cft->read_u64(cgrp, cft); |
ddbcc7e8 PM |
2458 | int len = sprintf(tmp, "%llu\n", (unsigned long long) val); |
2459 | ||
2460 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
2461 | } | |
2462 | ||
e73d2c61 PM |
2463 | static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft, |
2464 | struct file *file, | |
2465 | char __user *buf, size_t nbytes, | |
2466 | loff_t *ppos) | |
2467 | { | |
84eea842 | 2468 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
e73d2c61 PM |
2469 | s64 val = cft->read_s64(cgrp, cft); |
2470 | int len = sprintf(tmp, "%lld\n", (long long) val); | |
2471 | ||
2472 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
2473 | } | |
2474 | ||
ddbcc7e8 PM |
2475 | static ssize_t cgroup_file_read(struct file *file, char __user *buf, |
2476 | size_t nbytes, loff_t *ppos) | |
2477 | { | |
2478 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 2479 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 2480 | |
75139b82 | 2481 | if (cgroup_is_removed(cgrp)) |
ddbcc7e8 PM |
2482 | return -ENODEV; |
2483 | ||
2484 | if (cft->read) | |
bd89aabc | 2485 | return cft->read(cgrp, cft, file, buf, nbytes, ppos); |
f4c753b7 PM |
2486 | if (cft->read_u64) |
2487 | return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos); | |
e73d2c61 PM |
2488 | if (cft->read_s64) |
2489 | return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos); | |
ddbcc7e8 PM |
2490 | return -EINVAL; |
2491 | } | |
2492 | ||
91796569 PM |
2493 | /* |
2494 | * seqfile ops/methods for returning structured data. Currently just | |
2495 | * supports string->u64 maps, but can be extended in future. | |
2496 | */ | |
2497 | ||
2498 | struct cgroup_seqfile_state { | |
2499 | struct cftype *cft; | |
2500 | struct cgroup *cgroup; | |
2501 | }; | |
2502 | ||
2503 | static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value) | |
2504 | { | |
2505 | struct seq_file *sf = cb->state; | |
2506 | return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value); | |
2507 | } | |
2508 | ||
2509 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) | |
2510 | { | |
2511 | struct cgroup_seqfile_state *state = m->private; | |
2512 | struct cftype *cft = state->cft; | |
29486df3 SH |
2513 | if (cft->read_map) { |
2514 | struct cgroup_map_cb cb = { | |
2515 | .fill = cgroup_map_add, | |
2516 | .state = m, | |
2517 | }; | |
2518 | return cft->read_map(state->cgroup, cft, &cb); | |
2519 | } | |
2520 | return cft->read_seq_string(state->cgroup, cft, m); | |
91796569 PM |
2521 | } |
2522 | ||
96930a63 | 2523 | static int cgroup_seqfile_release(struct inode *inode, struct file *file) |
91796569 PM |
2524 | { |
2525 | struct seq_file *seq = file->private_data; | |
2526 | kfree(seq->private); | |
2527 | return single_release(inode, file); | |
2528 | } | |
2529 | ||
828c0950 | 2530 | static const struct file_operations cgroup_seqfile_operations = { |
91796569 | 2531 | .read = seq_read, |
e788e066 | 2532 | .write = cgroup_file_write, |
91796569 PM |
2533 | .llseek = seq_lseek, |
2534 | .release = cgroup_seqfile_release, | |
2535 | }; | |
2536 | ||
ddbcc7e8 PM |
2537 | static int cgroup_file_open(struct inode *inode, struct file *file) |
2538 | { | |
2539 | int err; | |
2540 | struct cftype *cft; | |
2541 | ||
2542 | err = generic_file_open(inode, file); | |
2543 | if (err) | |
2544 | return err; | |
ddbcc7e8 | 2545 | cft = __d_cft(file->f_dentry); |
75139b82 | 2546 | |
29486df3 | 2547 | if (cft->read_map || cft->read_seq_string) { |
91796569 PM |
2548 | struct cgroup_seqfile_state *state = |
2549 | kzalloc(sizeof(*state), GFP_USER); | |
2550 | if (!state) | |
2551 | return -ENOMEM; | |
2552 | state->cft = cft; | |
2553 | state->cgroup = __d_cgrp(file->f_dentry->d_parent); | |
2554 | file->f_op = &cgroup_seqfile_operations; | |
2555 | err = single_open(file, cgroup_seqfile_show, state); | |
2556 | if (err < 0) | |
2557 | kfree(state); | |
2558 | } else if (cft->open) | |
ddbcc7e8 PM |
2559 | err = cft->open(inode, file); |
2560 | else | |
2561 | err = 0; | |
2562 | ||
2563 | return err; | |
2564 | } | |
2565 | ||
2566 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
2567 | { | |
2568 | struct cftype *cft = __d_cft(file->f_dentry); | |
2569 | if (cft->release) | |
2570 | return cft->release(inode, file); | |
2571 | return 0; | |
2572 | } | |
2573 | ||
2574 | /* | |
2575 | * cgroup_rename - Only allow simple rename of directories in place. | |
2576 | */ | |
2577 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
2578 | struct inode *new_dir, struct dentry *new_dentry) | |
2579 | { | |
2580 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) | |
2581 | return -ENOTDIR; | |
2582 | if (new_dentry->d_inode) | |
2583 | return -EEXIST; | |
2584 | if (old_dir != new_dir) | |
2585 | return -EIO; | |
2586 | return simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
2587 | } | |
2588 | ||
828c0950 | 2589 | static const struct file_operations cgroup_file_operations = { |
ddbcc7e8 PM |
2590 | .read = cgroup_file_read, |
2591 | .write = cgroup_file_write, | |
2592 | .llseek = generic_file_llseek, | |
2593 | .open = cgroup_file_open, | |
2594 | .release = cgroup_file_release, | |
2595 | }; | |
2596 | ||
6e1d5dcc | 2597 | static const struct inode_operations cgroup_dir_inode_operations = { |
c72a04e3 | 2598 | .lookup = cgroup_lookup, |
ddbcc7e8 PM |
2599 | .mkdir = cgroup_mkdir, |
2600 | .rmdir = cgroup_rmdir, | |
2601 | .rename = cgroup_rename, | |
2602 | }; | |
2603 | ||
c72a04e3 AV |
2604 | static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) |
2605 | { | |
2606 | if (dentry->d_name.len > NAME_MAX) | |
2607 | return ERR_PTR(-ENAMETOOLONG); | |
2608 | d_add(dentry, NULL); | |
2609 | return NULL; | |
2610 | } | |
2611 | ||
0dea1168 KS |
2612 | /* |
2613 | * Check if a file is a control file | |
2614 | */ | |
2615 | static inline struct cftype *__file_cft(struct file *file) | |
2616 | { | |
2617 | if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations) | |
2618 | return ERR_PTR(-EINVAL); | |
2619 | return __d_cft(file->f_dentry); | |
2620 | } | |
2621 | ||
5adcee1d NP |
2622 | static int cgroup_create_file(struct dentry *dentry, mode_t mode, |
2623 | struct super_block *sb) | |
2624 | { | |
ddbcc7e8 PM |
2625 | struct inode *inode; |
2626 | ||
2627 | if (!dentry) | |
2628 | return -ENOENT; | |
2629 | if (dentry->d_inode) | |
2630 | return -EEXIST; | |
2631 | ||
2632 | inode = cgroup_new_inode(mode, sb); | |
2633 | if (!inode) | |
2634 | return -ENOMEM; | |
2635 | ||
2636 | if (S_ISDIR(mode)) { | |
2637 | inode->i_op = &cgroup_dir_inode_operations; | |
2638 | inode->i_fop = &simple_dir_operations; | |
2639 | ||
2640 | /* start off with i_nlink == 2 (for "." entry) */ | |
2641 | inc_nlink(inode); | |
2642 | ||
2643 | /* start with the directory inode held, so that we can | |
2644 | * populate it without racing with another mkdir */ | |
817929ec | 2645 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); |
ddbcc7e8 PM |
2646 | } else if (S_ISREG(mode)) { |
2647 | inode->i_size = 0; | |
2648 | inode->i_fop = &cgroup_file_operations; | |
2649 | } | |
ddbcc7e8 PM |
2650 | d_instantiate(dentry, inode); |
2651 | dget(dentry); /* Extra count - pin the dentry in core */ | |
2652 | return 0; | |
2653 | } | |
2654 | ||
2655 | /* | |
a043e3b2 LZ |
2656 | * cgroup_create_dir - create a directory for an object. |
2657 | * @cgrp: the cgroup we create the directory for. It must have a valid | |
2658 | * ->parent field. And we are going to fill its ->dentry field. | |
2659 | * @dentry: dentry of the new cgroup | |
2660 | * @mode: mode to set on new directory. | |
ddbcc7e8 | 2661 | */ |
bd89aabc | 2662 | static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry, |
099fca32 | 2663 | mode_t mode) |
ddbcc7e8 PM |
2664 | { |
2665 | struct dentry *parent; | |
2666 | int error = 0; | |
2667 | ||
bd89aabc PM |
2668 | parent = cgrp->parent->dentry; |
2669 | error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb); | |
ddbcc7e8 | 2670 | if (!error) { |
bd89aabc | 2671 | dentry->d_fsdata = cgrp; |
ddbcc7e8 | 2672 | inc_nlink(parent->d_inode); |
a47295e6 | 2673 | rcu_assign_pointer(cgrp->dentry, dentry); |
ddbcc7e8 PM |
2674 | dget(dentry); |
2675 | } | |
2676 | dput(dentry); | |
2677 | ||
2678 | return error; | |
2679 | } | |
2680 | ||
099fca32 LZ |
2681 | /** |
2682 | * cgroup_file_mode - deduce file mode of a control file | |
2683 | * @cft: the control file in question | |
2684 | * | |
2685 | * returns cft->mode if ->mode is not 0 | |
2686 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
2687 | * returns S_IRUGO if it has only a read handler | |
2688 | * returns S_IWUSR if it has only a write hander | |
2689 | */ | |
2690 | static mode_t cgroup_file_mode(const struct cftype *cft) | |
2691 | { | |
2692 | mode_t mode = 0; | |
2693 | ||
2694 | if (cft->mode) | |
2695 | return cft->mode; | |
2696 | ||
2697 | if (cft->read || cft->read_u64 || cft->read_s64 || | |
2698 | cft->read_map || cft->read_seq_string) | |
2699 | mode |= S_IRUGO; | |
2700 | ||
2701 | if (cft->write || cft->write_u64 || cft->write_s64 || | |
2702 | cft->write_string || cft->trigger) | |
2703 | mode |= S_IWUSR; | |
2704 | ||
2705 | return mode; | |
2706 | } | |
2707 | ||
bd89aabc | 2708 | int cgroup_add_file(struct cgroup *cgrp, |
ddbcc7e8 PM |
2709 | struct cgroup_subsys *subsys, |
2710 | const struct cftype *cft) | |
2711 | { | |
bd89aabc | 2712 | struct dentry *dir = cgrp->dentry; |
ddbcc7e8 PM |
2713 | struct dentry *dentry; |
2714 | int error; | |
099fca32 | 2715 | mode_t mode; |
ddbcc7e8 PM |
2716 | |
2717 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; | |
bd89aabc | 2718 | if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) { |
ddbcc7e8 PM |
2719 | strcpy(name, subsys->name); |
2720 | strcat(name, "."); | |
2721 | } | |
2722 | strcat(name, cft->name); | |
2723 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); | |
2724 | dentry = lookup_one_len(name, dir, strlen(name)); | |
2725 | if (!IS_ERR(dentry)) { | |
099fca32 LZ |
2726 | mode = cgroup_file_mode(cft); |
2727 | error = cgroup_create_file(dentry, mode | S_IFREG, | |
bd89aabc | 2728 | cgrp->root->sb); |
ddbcc7e8 PM |
2729 | if (!error) |
2730 | dentry->d_fsdata = (void *)cft; | |
2731 | dput(dentry); | |
2732 | } else | |
2733 | error = PTR_ERR(dentry); | |
2734 | return error; | |
2735 | } | |
e6a1105b | 2736 | EXPORT_SYMBOL_GPL(cgroup_add_file); |
ddbcc7e8 | 2737 | |
bd89aabc | 2738 | int cgroup_add_files(struct cgroup *cgrp, |
ddbcc7e8 PM |
2739 | struct cgroup_subsys *subsys, |
2740 | const struct cftype cft[], | |
2741 | int count) | |
2742 | { | |
2743 | int i, err; | |
2744 | for (i = 0; i < count; i++) { | |
bd89aabc | 2745 | err = cgroup_add_file(cgrp, subsys, &cft[i]); |
ddbcc7e8 PM |
2746 | if (err) |
2747 | return err; | |
2748 | } | |
2749 | return 0; | |
2750 | } | |
e6a1105b | 2751 | EXPORT_SYMBOL_GPL(cgroup_add_files); |
ddbcc7e8 | 2752 | |
a043e3b2 LZ |
2753 | /** |
2754 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2755 | * @cgrp: the cgroup in question | |
2756 | * | |
2757 | * Return the number of tasks in the cgroup. | |
2758 | */ | |
bd89aabc | 2759 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2760 | { |
2761 | int count = 0; | |
71cbb949 | 2762 | struct cg_cgroup_link *link; |
817929ec PM |
2763 | |
2764 | read_lock(&css_set_lock); | |
71cbb949 | 2765 | list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) { |
146aa1bd | 2766 | count += atomic_read(&link->cg->refcount); |
817929ec PM |
2767 | } |
2768 | read_unlock(&css_set_lock); | |
bbcb81d0 PM |
2769 | return count; |
2770 | } | |
2771 | ||
817929ec PM |
2772 | /* |
2773 | * Advance a list_head iterator. The iterator should be positioned at | |
2774 | * the start of a css_set | |
2775 | */ | |
bd89aabc | 2776 | static void cgroup_advance_iter(struct cgroup *cgrp, |
7717f7ba | 2777 | struct cgroup_iter *it) |
817929ec PM |
2778 | { |
2779 | struct list_head *l = it->cg_link; | |
2780 | struct cg_cgroup_link *link; | |
2781 | struct css_set *cg; | |
2782 | ||
2783 | /* Advance to the next non-empty css_set */ | |
2784 | do { | |
2785 | l = l->next; | |
bd89aabc | 2786 | if (l == &cgrp->css_sets) { |
817929ec PM |
2787 | it->cg_link = NULL; |
2788 | return; | |
2789 | } | |
bd89aabc | 2790 | link = list_entry(l, struct cg_cgroup_link, cgrp_link_list); |
817929ec PM |
2791 | cg = link->cg; |
2792 | } while (list_empty(&cg->tasks)); | |
2793 | it->cg_link = l; | |
2794 | it->task = cg->tasks.next; | |
2795 | } | |
2796 | ||
31a7df01 CW |
2797 | /* |
2798 | * To reduce the fork() overhead for systems that are not actually | |
2799 | * using their cgroups capability, we don't maintain the lists running | |
2800 | * through each css_set to its tasks until we see the list actually | |
2801 | * used - in other words after the first call to cgroup_iter_start(). | |
2802 | * | |
2803 | * The tasklist_lock is not held here, as do_each_thread() and | |
2804 | * while_each_thread() are protected by RCU. | |
2805 | */ | |
3df91fe3 | 2806 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2807 | { |
2808 | struct task_struct *p, *g; | |
2809 | write_lock(&css_set_lock); | |
2810 | use_task_css_set_links = 1; | |
2811 | do_each_thread(g, p) { | |
2812 | task_lock(p); | |
0e04388f LZ |
2813 | /* |
2814 | * We should check if the process is exiting, otherwise | |
2815 | * it will race with cgroup_exit() in that the list | |
2816 | * entry won't be deleted though the process has exited. | |
2817 | */ | |
2818 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
31a7df01 CW |
2819 | list_add(&p->cg_list, &p->cgroups->tasks); |
2820 | task_unlock(p); | |
2821 | } while_each_thread(g, p); | |
2822 | write_unlock(&css_set_lock); | |
2823 | } | |
2824 | ||
bd89aabc | 2825 | void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it) |
817929ec PM |
2826 | { |
2827 | /* | |
2828 | * The first time anyone tries to iterate across a cgroup, | |
2829 | * we need to enable the list linking each css_set to its | |
2830 | * tasks, and fix up all existing tasks. | |
2831 | */ | |
31a7df01 CW |
2832 | if (!use_task_css_set_links) |
2833 | cgroup_enable_task_cg_lists(); | |
2834 | ||
817929ec | 2835 | read_lock(&css_set_lock); |
bd89aabc PM |
2836 | it->cg_link = &cgrp->css_sets; |
2837 | cgroup_advance_iter(cgrp, it); | |
817929ec PM |
2838 | } |
2839 | ||
bd89aabc | 2840 | struct task_struct *cgroup_iter_next(struct cgroup *cgrp, |
817929ec PM |
2841 | struct cgroup_iter *it) |
2842 | { | |
2843 | struct task_struct *res; | |
2844 | struct list_head *l = it->task; | |
2019f634 | 2845 | struct cg_cgroup_link *link; |
817929ec PM |
2846 | |
2847 | /* If the iterator cg is NULL, we have no tasks */ | |
2848 | if (!it->cg_link) | |
2849 | return NULL; | |
2850 | res = list_entry(l, struct task_struct, cg_list); | |
2851 | /* Advance iterator to find next entry */ | |
2852 | l = l->next; | |
2019f634 LJ |
2853 | link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list); |
2854 | if (l == &link->cg->tasks) { | |
817929ec PM |
2855 | /* We reached the end of this task list - move on to |
2856 | * the next cg_cgroup_link */ | |
bd89aabc | 2857 | cgroup_advance_iter(cgrp, it); |
817929ec PM |
2858 | } else { |
2859 | it->task = l; | |
2860 | } | |
2861 | return res; | |
2862 | } | |
2863 | ||
bd89aabc | 2864 | void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it) |
817929ec PM |
2865 | { |
2866 | read_unlock(&css_set_lock); | |
2867 | } | |
2868 | ||
31a7df01 CW |
2869 | static inline int started_after_time(struct task_struct *t1, |
2870 | struct timespec *time, | |
2871 | struct task_struct *t2) | |
2872 | { | |
2873 | int start_diff = timespec_compare(&t1->start_time, time); | |
2874 | if (start_diff > 0) { | |
2875 | return 1; | |
2876 | } else if (start_diff < 0) { | |
2877 | return 0; | |
2878 | } else { | |
2879 | /* | |
2880 | * Arbitrarily, if two processes started at the same | |
2881 | * time, we'll say that the lower pointer value | |
2882 | * started first. Note that t2 may have exited by now | |
2883 | * so this may not be a valid pointer any longer, but | |
2884 | * that's fine - it still serves to distinguish | |
2885 | * between two tasks started (effectively) simultaneously. | |
2886 | */ | |
2887 | return t1 > t2; | |
2888 | } | |
2889 | } | |
2890 | ||
2891 | /* | |
2892 | * This function is a callback from heap_insert() and is used to order | |
2893 | * the heap. | |
2894 | * In this case we order the heap in descending task start time. | |
2895 | */ | |
2896 | static inline int started_after(void *p1, void *p2) | |
2897 | { | |
2898 | struct task_struct *t1 = p1; | |
2899 | struct task_struct *t2 = p2; | |
2900 | return started_after_time(t1, &t2->start_time, t2); | |
2901 | } | |
2902 | ||
2903 | /** | |
2904 | * cgroup_scan_tasks - iterate though all the tasks in a cgroup | |
2905 | * @scan: struct cgroup_scanner containing arguments for the scan | |
2906 | * | |
2907 | * Arguments include pointers to callback functions test_task() and | |
2908 | * process_task(). | |
2909 | * Iterate through all the tasks in a cgroup, calling test_task() for each, | |
2910 | * and if it returns true, call process_task() for it also. | |
2911 | * The test_task pointer may be NULL, meaning always true (select all tasks). | |
2912 | * Effectively duplicates cgroup_iter_{start,next,end}() | |
2913 | * but does not lock css_set_lock for the call to process_task(). | |
2914 | * The struct cgroup_scanner may be embedded in any structure of the caller's | |
2915 | * creation. | |
2916 | * It is guaranteed that process_task() will act on every task that | |
2917 | * is a member of the cgroup for the duration of this call. This | |
2918 | * function may or may not call process_task() for tasks that exit | |
2919 | * or move to a different cgroup during the call, or are forked or | |
2920 | * move into the cgroup during the call. | |
2921 | * | |
2922 | * Note that test_task() may be called with locks held, and may in some | |
2923 | * situations be called multiple times for the same task, so it should | |
2924 | * be cheap. | |
2925 | * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been | |
2926 | * pre-allocated and will be used for heap operations (and its "gt" member will | |
2927 | * be overwritten), else a temporary heap will be used (allocation of which | |
2928 | * may cause this function to fail). | |
2929 | */ | |
2930 | int cgroup_scan_tasks(struct cgroup_scanner *scan) | |
2931 | { | |
2932 | int retval, i; | |
2933 | struct cgroup_iter it; | |
2934 | struct task_struct *p, *dropped; | |
2935 | /* Never dereference latest_task, since it's not refcounted */ | |
2936 | struct task_struct *latest_task = NULL; | |
2937 | struct ptr_heap tmp_heap; | |
2938 | struct ptr_heap *heap; | |
2939 | struct timespec latest_time = { 0, 0 }; | |
2940 | ||
2941 | if (scan->heap) { | |
2942 | /* The caller supplied our heap and pre-allocated its memory */ | |
2943 | heap = scan->heap; | |
2944 | heap->gt = &started_after; | |
2945 | } else { | |
2946 | /* We need to allocate our own heap memory */ | |
2947 | heap = &tmp_heap; | |
2948 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
2949 | if (retval) | |
2950 | /* cannot allocate the heap */ | |
2951 | return retval; | |
2952 | } | |
2953 | ||
2954 | again: | |
2955 | /* | |
2956 | * Scan tasks in the cgroup, using the scanner's "test_task" callback | |
2957 | * to determine which are of interest, and using the scanner's | |
2958 | * "process_task" callback to process any of them that need an update. | |
2959 | * Since we don't want to hold any locks during the task updates, | |
2960 | * gather tasks to be processed in a heap structure. | |
2961 | * The heap is sorted by descending task start time. | |
2962 | * If the statically-sized heap fills up, we overflow tasks that | |
2963 | * started later, and in future iterations only consider tasks that | |
2964 | * started after the latest task in the previous pass. This | |
2965 | * guarantees forward progress and that we don't miss any tasks. | |
2966 | */ | |
2967 | heap->size = 0; | |
2968 | cgroup_iter_start(scan->cg, &it); | |
2969 | while ((p = cgroup_iter_next(scan->cg, &it))) { | |
2970 | /* | |
2971 | * Only affect tasks that qualify per the caller's callback, | |
2972 | * if he provided one | |
2973 | */ | |
2974 | if (scan->test_task && !scan->test_task(p, scan)) | |
2975 | continue; | |
2976 | /* | |
2977 | * Only process tasks that started after the last task | |
2978 | * we processed | |
2979 | */ | |
2980 | if (!started_after_time(p, &latest_time, latest_task)) | |
2981 | continue; | |
2982 | dropped = heap_insert(heap, p); | |
2983 | if (dropped == NULL) { | |
2984 | /* | |
2985 | * The new task was inserted; the heap wasn't | |
2986 | * previously full | |
2987 | */ | |
2988 | get_task_struct(p); | |
2989 | } else if (dropped != p) { | |
2990 | /* | |
2991 | * The new task was inserted, and pushed out a | |
2992 | * different task | |
2993 | */ | |
2994 | get_task_struct(p); | |
2995 | put_task_struct(dropped); | |
2996 | } | |
2997 | /* | |
2998 | * Else the new task was newer than anything already in | |
2999 | * the heap and wasn't inserted | |
3000 | */ | |
3001 | } | |
3002 | cgroup_iter_end(scan->cg, &it); | |
3003 | ||
3004 | if (heap->size) { | |
3005 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 3006 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 3007 | if (i == 0) { |
4fe91d51 PJ |
3008 | latest_time = q->start_time; |
3009 | latest_task = q; | |
31a7df01 CW |
3010 | } |
3011 | /* Process the task per the caller's callback */ | |
4fe91d51 PJ |
3012 | scan->process_task(q, scan); |
3013 | put_task_struct(q); | |
31a7df01 CW |
3014 | } |
3015 | /* | |
3016 | * If we had to process any tasks at all, scan again | |
3017 | * in case some of them were in the middle of forking | |
3018 | * children that didn't get processed. | |
3019 | * Not the most efficient way to do it, but it avoids | |
3020 | * having to take callback_mutex in the fork path | |
3021 | */ | |
3022 | goto again; | |
3023 | } | |
3024 | if (heap == &tmp_heap) | |
3025 | heap_free(&tmp_heap); | |
3026 | return 0; | |
3027 | } | |
3028 | ||
bbcb81d0 | 3029 | /* |
102a775e | 3030 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3031 | * |
3032 | * Reading this file can return large amounts of data if a cgroup has | |
3033 | * *lots* of attached tasks. So it may need several calls to read(), | |
3034 | * but we cannot guarantee that the information we produce is correct | |
3035 | * unless we produce it entirely atomically. | |
3036 | * | |
bbcb81d0 | 3037 | */ |
bbcb81d0 | 3038 | |
d1d9fd33 BB |
3039 | /* |
3040 | * The following two functions "fix" the issue where there are more pids | |
3041 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3042 | * TODO: replace with a kernel-wide solution to this problem | |
3043 | */ | |
3044 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3045 | static void *pidlist_allocate(int count) | |
3046 | { | |
3047 | if (PIDLIST_TOO_LARGE(count)) | |
3048 | return vmalloc(count * sizeof(pid_t)); | |
3049 | else | |
3050 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3051 | } | |
3052 | static void pidlist_free(void *p) | |
3053 | { | |
3054 | if (is_vmalloc_addr(p)) | |
3055 | vfree(p); | |
3056 | else | |
3057 | kfree(p); | |
3058 | } | |
3059 | static void *pidlist_resize(void *p, int newcount) | |
3060 | { | |
3061 | void *newlist; | |
3062 | /* note: if new alloc fails, old p will still be valid either way */ | |
3063 | if (is_vmalloc_addr(p)) { | |
3064 | newlist = vmalloc(newcount * sizeof(pid_t)); | |
3065 | if (!newlist) | |
3066 | return NULL; | |
3067 | memcpy(newlist, p, newcount * sizeof(pid_t)); | |
3068 | vfree(p); | |
3069 | } else { | |
3070 | newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL); | |
3071 | } | |
3072 | return newlist; | |
3073 | } | |
3074 | ||
bbcb81d0 | 3075 | /* |
102a775e BB |
3076 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
3077 | * If the new stripped list is sufficiently smaller and there's enough memory | |
3078 | * to allocate a new buffer, will let go of the unneeded memory. Returns the | |
3079 | * number of unique elements. | |
bbcb81d0 | 3080 | */ |
102a775e BB |
3081 | /* is the size difference enough that we should re-allocate the array? */ |
3082 | #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new)) | |
3083 | static int pidlist_uniq(pid_t **p, int length) | |
bbcb81d0 | 3084 | { |
102a775e BB |
3085 | int src, dest = 1; |
3086 | pid_t *list = *p; | |
3087 | pid_t *newlist; | |
3088 | ||
3089 | /* | |
3090 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3091 | * edge cases first; no work needs to be done for either | |
3092 | */ | |
3093 | if (length == 0 || length == 1) | |
3094 | return length; | |
3095 | /* src and dest walk down the list; dest counts unique elements */ | |
3096 | for (src = 1; src < length; src++) { | |
3097 | /* find next unique element */ | |
3098 | while (list[src] == list[src-1]) { | |
3099 | src++; | |
3100 | if (src == length) | |
3101 | goto after; | |
3102 | } | |
3103 | /* dest always points to where the next unique element goes */ | |
3104 | list[dest] = list[src]; | |
3105 | dest++; | |
3106 | } | |
3107 | after: | |
3108 | /* | |
3109 | * if the length difference is large enough, we want to allocate a | |
3110 | * smaller buffer to save memory. if this fails due to out of memory, | |
3111 | * we'll just stay with what we've got. | |
3112 | */ | |
3113 | if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) { | |
d1d9fd33 | 3114 | newlist = pidlist_resize(list, dest); |
102a775e BB |
3115 | if (newlist) |
3116 | *p = newlist; | |
3117 | } | |
3118 | return dest; | |
3119 | } | |
3120 | ||
3121 | static int cmppid(const void *a, const void *b) | |
3122 | { | |
3123 | return *(pid_t *)a - *(pid_t *)b; | |
3124 | } | |
3125 | ||
72a8cb30 BB |
3126 | /* |
3127 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3128 | * returns with the lock on that pidlist already held, and takes care | |
3129 | * of the use count, or returns NULL with no locks held if we're out of | |
3130 | * memory. | |
3131 | */ | |
3132 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
3133 | enum cgroup_filetype type) | |
3134 | { | |
3135 | struct cgroup_pidlist *l; | |
3136 | /* don't need task_nsproxy() if we're looking at ourself */ | |
b70cc5fd LZ |
3137 | struct pid_namespace *ns = current->nsproxy->pid_ns; |
3138 | ||
72a8cb30 BB |
3139 | /* |
3140 | * We can't drop the pidlist_mutex before taking the l->mutex in case | |
3141 | * the last ref-holder is trying to remove l from the list at the same | |
3142 | * time. Holding the pidlist_mutex precludes somebody taking whichever | |
3143 | * list we find out from under us - compare release_pid_array(). | |
3144 | */ | |
3145 | mutex_lock(&cgrp->pidlist_mutex); | |
3146 | list_for_each_entry(l, &cgrp->pidlists, links) { | |
3147 | if (l->key.type == type && l->key.ns == ns) { | |
72a8cb30 BB |
3148 | /* make sure l doesn't vanish out from under us */ |
3149 | down_write(&l->mutex); | |
3150 | mutex_unlock(&cgrp->pidlist_mutex); | |
72a8cb30 BB |
3151 | return l; |
3152 | } | |
3153 | } | |
3154 | /* entry not found; create a new one */ | |
3155 | l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); | |
3156 | if (!l) { | |
3157 | mutex_unlock(&cgrp->pidlist_mutex); | |
72a8cb30 BB |
3158 | return l; |
3159 | } | |
3160 | init_rwsem(&l->mutex); | |
3161 | down_write(&l->mutex); | |
3162 | l->key.type = type; | |
b70cc5fd | 3163 | l->key.ns = get_pid_ns(ns); |
72a8cb30 BB |
3164 | l->use_count = 0; /* don't increment here */ |
3165 | l->list = NULL; | |
3166 | l->owner = cgrp; | |
3167 | list_add(&l->links, &cgrp->pidlists); | |
3168 | mutex_unlock(&cgrp->pidlist_mutex); | |
3169 | return l; | |
3170 | } | |
3171 | ||
102a775e BB |
3172 | /* |
3173 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3174 | */ | |
72a8cb30 BB |
3175 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3176 | struct cgroup_pidlist **lp) | |
102a775e BB |
3177 | { |
3178 | pid_t *array; | |
3179 | int length; | |
3180 | int pid, n = 0; /* used for populating the array */ | |
817929ec PM |
3181 | struct cgroup_iter it; |
3182 | struct task_struct *tsk; | |
102a775e BB |
3183 | struct cgroup_pidlist *l; |
3184 | ||
3185 | /* | |
3186 | * If cgroup gets more users after we read count, we won't have | |
3187 | * enough space - tough. This race is indistinguishable to the | |
3188 | * caller from the case that the additional cgroup users didn't | |
3189 | * show up until sometime later on. | |
3190 | */ | |
3191 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3192 | array = pidlist_allocate(length); |
102a775e BB |
3193 | if (!array) |
3194 | return -ENOMEM; | |
3195 | /* now, populate the array */ | |
bd89aabc PM |
3196 | cgroup_iter_start(cgrp, &it); |
3197 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
102a775e | 3198 | if (unlikely(n == length)) |
817929ec | 3199 | break; |
102a775e | 3200 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3201 | if (type == CGROUP_FILE_PROCS) |
3202 | pid = task_tgid_vnr(tsk); | |
3203 | else | |
3204 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3205 | if (pid > 0) /* make sure to only use valid results */ |
3206 | array[n++] = pid; | |
817929ec | 3207 | } |
bd89aabc | 3208 | cgroup_iter_end(cgrp, &it); |
102a775e BB |
3209 | length = n; |
3210 | /* now sort & (if procs) strip out duplicates */ | |
3211 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3212 | if (type == CGROUP_FILE_PROCS) |
102a775e | 3213 | length = pidlist_uniq(&array, length); |
72a8cb30 BB |
3214 | l = cgroup_pidlist_find(cgrp, type); |
3215 | if (!l) { | |
d1d9fd33 | 3216 | pidlist_free(array); |
72a8cb30 | 3217 | return -ENOMEM; |
102a775e | 3218 | } |
72a8cb30 | 3219 | /* store array, freeing old if necessary - lock already held */ |
d1d9fd33 | 3220 | pidlist_free(l->list); |
102a775e BB |
3221 | l->list = array; |
3222 | l->length = length; | |
3223 | l->use_count++; | |
3224 | up_write(&l->mutex); | |
72a8cb30 | 3225 | *lp = l; |
102a775e | 3226 | return 0; |
bbcb81d0 PM |
3227 | } |
3228 | ||
846c7bb0 | 3229 | /** |
a043e3b2 | 3230 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3231 | * @stats: cgroupstats to fill information into |
3232 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3233 | * been requested. | |
a043e3b2 LZ |
3234 | * |
3235 | * Build and fill cgroupstats so that taskstats can export it to user | |
3236 | * space. | |
846c7bb0 BS |
3237 | */ |
3238 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3239 | { | |
3240 | int ret = -EINVAL; | |
bd89aabc | 3241 | struct cgroup *cgrp; |
846c7bb0 BS |
3242 | struct cgroup_iter it; |
3243 | struct task_struct *tsk; | |
33d283be | 3244 | |
846c7bb0 | 3245 | /* |
33d283be LZ |
3246 | * Validate dentry by checking the superblock operations, |
3247 | * and make sure it's a directory. | |
846c7bb0 | 3248 | */ |
33d283be LZ |
3249 | if (dentry->d_sb->s_op != &cgroup_ops || |
3250 | !S_ISDIR(dentry->d_inode->i_mode)) | |
846c7bb0 BS |
3251 | goto err; |
3252 | ||
3253 | ret = 0; | |
bd89aabc | 3254 | cgrp = dentry->d_fsdata; |
846c7bb0 | 3255 | |
bd89aabc PM |
3256 | cgroup_iter_start(cgrp, &it); |
3257 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
846c7bb0 BS |
3258 | switch (tsk->state) { |
3259 | case TASK_RUNNING: | |
3260 | stats->nr_running++; | |
3261 | break; | |
3262 | case TASK_INTERRUPTIBLE: | |
3263 | stats->nr_sleeping++; | |
3264 | break; | |
3265 | case TASK_UNINTERRUPTIBLE: | |
3266 | stats->nr_uninterruptible++; | |
3267 | break; | |
3268 | case TASK_STOPPED: | |
3269 | stats->nr_stopped++; | |
3270 | break; | |
3271 | default: | |
3272 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3273 | stats->nr_io_wait++; | |
3274 | break; | |
3275 | } | |
3276 | } | |
bd89aabc | 3277 | cgroup_iter_end(cgrp, &it); |
846c7bb0 | 3278 | |
846c7bb0 BS |
3279 | err: |
3280 | return ret; | |
3281 | } | |
3282 | ||
8f3ff208 | 3283 | |
bbcb81d0 | 3284 | /* |
102a775e | 3285 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3286 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3287 | * in the cgroup->l->list array. |
bbcb81d0 | 3288 | */ |
cc31edce | 3289 | |
102a775e | 3290 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3291 | { |
cc31edce PM |
3292 | /* |
3293 | * Initially we receive a position value that corresponds to | |
3294 | * one more than the last pid shown (or 0 on the first call or | |
3295 | * after a seek to the start). Use a binary-search to find the | |
3296 | * next pid to display, if any | |
3297 | */ | |
102a775e | 3298 | struct cgroup_pidlist *l = s->private; |
cc31edce PM |
3299 | int index = 0, pid = *pos; |
3300 | int *iter; | |
3301 | ||
102a775e | 3302 | down_read(&l->mutex); |
cc31edce | 3303 | if (pid) { |
102a775e | 3304 | int end = l->length; |
20777766 | 3305 | |
cc31edce PM |
3306 | while (index < end) { |
3307 | int mid = (index + end) / 2; | |
102a775e | 3308 | if (l->list[mid] == pid) { |
cc31edce PM |
3309 | index = mid; |
3310 | break; | |
102a775e | 3311 | } else if (l->list[mid] <= pid) |
cc31edce PM |
3312 | index = mid + 1; |
3313 | else | |
3314 | end = mid; | |
3315 | } | |
3316 | } | |
3317 | /* If we're off the end of the array, we're done */ | |
102a775e | 3318 | if (index >= l->length) |
cc31edce PM |
3319 | return NULL; |
3320 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3321 | iter = l->list + index; |
cc31edce PM |
3322 | *pos = *iter; |
3323 | return iter; | |
3324 | } | |
3325 | ||
102a775e | 3326 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3327 | { |
102a775e BB |
3328 | struct cgroup_pidlist *l = s->private; |
3329 | up_read(&l->mutex); | |
cc31edce PM |
3330 | } |
3331 | ||
102a775e | 3332 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3333 | { |
102a775e BB |
3334 | struct cgroup_pidlist *l = s->private; |
3335 | pid_t *p = v; | |
3336 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3337 | /* |
3338 | * Advance to the next pid in the array. If this goes off the | |
3339 | * end, we're done | |
3340 | */ | |
3341 | p++; | |
3342 | if (p >= end) { | |
3343 | return NULL; | |
3344 | } else { | |
3345 | *pos = *p; | |
3346 | return p; | |
3347 | } | |
3348 | } | |
3349 | ||
102a775e | 3350 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3351 | { |
3352 | return seq_printf(s, "%d\n", *(int *)v); | |
3353 | } | |
bbcb81d0 | 3354 | |
102a775e BB |
3355 | /* |
3356 | * seq_operations functions for iterating on pidlists through seq_file - | |
3357 | * independent of whether it's tasks or procs | |
3358 | */ | |
3359 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3360 | .start = cgroup_pidlist_start, | |
3361 | .stop = cgroup_pidlist_stop, | |
3362 | .next = cgroup_pidlist_next, | |
3363 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3364 | }; |
3365 | ||
102a775e | 3366 | static void cgroup_release_pid_array(struct cgroup_pidlist *l) |
cc31edce | 3367 | { |
72a8cb30 BB |
3368 | /* |
3369 | * the case where we're the last user of this particular pidlist will | |
3370 | * have us remove it from the cgroup's list, which entails taking the | |
3371 | * mutex. since in pidlist_find the pidlist->lock depends on cgroup-> | |
3372 | * pidlist_mutex, we have to take pidlist_mutex first. | |
3373 | */ | |
3374 | mutex_lock(&l->owner->pidlist_mutex); | |
102a775e BB |
3375 | down_write(&l->mutex); |
3376 | BUG_ON(!l->use_count); | |
3377 | if (!--l->use_count) { | |
72a8cb30 BB |
3378 | /* we're the last user if refcount is 0; remove and free */ |
3379 | list_del(&l->links); | |
3380 | mutex_unlock(&l->owner->pidlist_mutex); | |
d1d9fd33 | 3381 | pidlist_free(l->list); |
72a8cb30 BB |
3382 | put_pid_ns(l->key.ns); |
3383 | up_write(&l->mutex); | |
3384 | kfree(l); | |
3385 | return; | |
cc31edce | 3386 | } |
72a8cb30 | 3387 | mutex_unlock(&l->owner->pidlist_mutex); |
102a775e | 3388 | up_write(&l->mutex); |
bbcb81d0 PM |
3389 | } |
3390 | ||
102a775e | 3391 | static int cgroup_pidlist_release(struct inode *inode, struct file *file) |
cc31edce | 3392 | { |
102a775e | 3393 | struct cgroup_pidlist *l; |
cc31edce PM |
3394 | if (!(file->f_mode & FMODE_READ)) |
3395 | return 0; | |
102a775e BB |
3396 | /* |
3397 | * the seq_file will only be initialized if the file was opened for | |
3398 | * reading; hence we check if it's not null only in that case. | |
3399 | */ | |
3400 | l = ((struct seq_file *)file->private_data)->private; | |
3401 | cgroup_release_pid_array(l); | |
cc31edce PM |
3402 | return seq_release(inode, file); |
3403 | } | |
3404 | ||
102a775e | 3405 | static const struct file_operations cgroup_pidlist_operations = { |
cc31edce PM |
3406 | .read = seq_read, |
3407 | .llseek = seq_lseek, | |
3408 | .write = cgroup_file_write, | |
102a775e | 3409 | .release = cgroup_pidlist_release, |
cc31edce PM |
3410 | }; |
3411 | ||
bbcb81d0 | 3412 | /* |
102a775e BB |
3413 | * The following functions handle opens on a file that displays a pidlist |
3414 | * (tasks or procs). Prepare an array of the process/thread IDs of whoever's | |
3415 | * in the cgroup. | |
bbcb81d0 | 3416 | */ |
102a775e | 3417 | /* helper function for the two below it */ |
72a8cb30 | 3418 | static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type) |
bbcb81d0 | 3419 | { |
bd89aabc | 3420 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
72a8cb30 | 3421 | struct cgroup_pidlist *l; |
cc31edce | 3422 | int retval; |
bbcb81d0 | 3423 | |
cc31edce | 3424 | /* Nothing to do for write-only files */ |
bbcb81d0 PM |
3425 | if (!(file->f_mode & FMODE_READ)) |
3426 | return 0; | |
3427 | ||
102a775e | 3428 | /* have the array populated */ |
72a8cb30 | 3429 | retval = pidlist_array_load(cgrp, type, &l); |
102a775e BB |
3430 | if (retval) |
3431 | return retval; | |
3432 | /* configure file information */ | |
3433 | file->f_op = &cgroup_pidlist_operations; | |
cc31edce | 3434 | |
102a775e | 3435 | retval = seq_open(file, &cgroup_pidlist_seq_operations); |
cc31edce | 3436 | if (retval) { |
102a775e | 3437 | cgroup_release_pid_array(l); |
cc31edce | 3438 | return retval; |
bbcb81d0 | 3439 | } |
102a775e | 3440 | ((struct seq_file *)file->private_data)->private = l; |
bbcb81d0 PM |
3441 | return 0; |
3442 | } | |
102a775e BB |
3443 | static int cgroup_tasks_open(struct inode *unused, struct file *file) |
3444 | { | |
72a8cb30 | 3445 | return cgroup_pidlist_open(file, CGROUP_FILE_TASKS); |
102a775e BB |
3446 | } |
3447 | static int cgroup_procs_open(struct inode *unused, struct file *file) | |
3448 | { | |
72a8cb30 | 3449 | return cgroup_pidlist_open(file, CGROUP_FILE_PROCS); |
102a775e | 3450 | } |
bbcb81d0 | 3451 | |
bd89aabc | 3452 | static u64 cgroup_read_notify_on_release(struct cgroup *cgrp, |
81a6a5cd PM |
3453 | struct cftype *cft) |
3454 | { | |
bd89aabc | 3455 | return notify_on_release(cgrp); |
81a6a5cd PM |
3456 | } |
3457 | ||
6379c106 PM |
3458 | static int cgroup_write_notify_on_release(struct cgroup *cgrp, |
3459 | struct cftype *cft, | |
3460 | u64 val) | |
3461 | { | |
3462 | clear_bit(CGRP_RELEASABLE, &cgrp->flags); | |
3463 | if (val) | |
3464 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3465 | else | |
3466 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3467 | return 0; | |
3468 | } | |
3469 | ||
0dea1168 KS |
3470 | /* |
3471 | * Unregister event and free resources. | |
3472 | * | |
3473 | * Gets called from workqueue. | |
3474 | */ | |
3475 | static void cgroup_event_remove(struct work_struct *work) | |
3476 | { | |
3477 | struct cgroup_event *event = container_of(work, struct cgroup_event, | |
3478 | remove); | |
3479 | struct cgroup *cgrp = event->cgrp; | |
3480 | ||
0dea1168 KS |
3481 | event->cft->unregister_event(cgrp, event->cft, event->eventfd); |
3482 | ||
3483 | eventfd_ctx_put(event->eventfd); | |
0dea1168 | 3484 | kfree(event); |
a0a4db54 | 3485 | dput(cgrp->dentry); |
0dea1168 KS |
3486 | } |
3487 | ||
3488 | /* | |
3489 | * Gets called on POLLHUP on eventfd when user closes it. | |
3490 | * | |
3491 | * Called with wqh->lock held and interrupts disabled. | |
3492 | */ | |
3493 | static int cgroup_event_wake(wait_queue_t *wait, unsigned mode, | |
3494 | int sync, void *key) | |
3495 | { | |
3496 | struct cgroup_event *event = container_of(wait, | |
3497 | struct cgroup_event, wait); | |
3498 | struct cgroup *cgrp = event->cgrp; | |
3499 | unsigned long flags = (unsigned long)key; | |
3500 | ||
3501 | if (flags & POLLHUP) { | |
a93d2f17 | 3502 | __remove_wait_queue(event->wqh, &event->wait); |
0dea1168 KS |
3503 | spin_lock(&cgrp->event_list_lock); |
3504 | list_del(&event->list); | |
3505 | spin_unlock(&cgrp->event_list_lock); | |
3506 | /* | |
3507 | * We are in atomic context, but cgroup_event_remove() may | |
3508 | * sleep, so we have to call it in workqueue. | |
3509 | */ | |
3510 | schedule_work(&event->remove); | |
3511 | } | |
3512 | ||
3513 | return 0; | |
3514 | } | |
3515 | ||
3516 | static void cgroup_event_ptable_queue_proc(struct file *file, | |
3517 | wait_queue_head_t *wqh, poll_table *pt) | |
3518 | { | |
3519 | struct cgroup_event *event = container_of(pt, | |
3520 | struct cgroup_event, pt); | |
3521 | ||
3522 | event->wqh = wqh; | |
3523 | add_wait_queue(wqh, &event->wait); | |
3524 | } | |
3525 | ||
3526 | /* | |
3527 | * Parse input and register new cgroup event handler. | |
3528 | * | |
3529 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3530 | * Interpretation of args is defined by control file implementation. | |
3531 | */ | |
3532 | static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft, | |
3533 | const char *buffer) | |
3534 | { | |
3535 | struct cgroup_event *event = NULL; | |
3536 | unsigned int efd, cfd; | |
3537 | struct file *efile = NULL; | |
3538 | struct file *cfile = NULL; | |
3539 | char *endp; | |
3540 | int ret; | |
3541 | ||
3542 | efd = simple_strtoul(buffer, &endp, 10); | |
3543 | if (*endp != ' ') | |
3544 | return -EINVAL; | |
3545 | buffer = endp + 1; | |
3546 | ||
3547 | cfd = simple_strtoul(buffer, &endp, 10); | |
3548 | if ((*endp != ' ') && (*endp != '\0')) | |
3549 | return -EINVAL; | |
3550 | buffer = endp + 1; | |
3551 | ||
3552 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3553 | if (!event) | |
3554 | return -ENOMEM; | |
3555 | event->cgrp = cgrp; | |
3556 | INIT_LIST_HEAD(&event->list); | |
3557 | init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc); | |
3558 | init_waitqueue_func_entry(&event->wait, cgroup_event_wake); | |
3559 | INIT_WORK(&event->remove, cgroup_event_remove); | |
3560 | ||
3561 | efile = eventfd_fget(efd); | |
3562 | if (IS_ERR(efile)) { | |
3563 | ret = PTR_ERR(efile); | |
3564 | goto fail; | |
3565 | } | |
3566 | ||
3567 | event->eventfd = eventfd_ctx_fileget(efile); | |
3568 | if (IS_ERR(event->eventfd)) { | |
3569 | ret = PTR_ERR(event->eventfd); | |
3570 | goto fail; | |
3571 | } | |
3572 | ||
3573 | cfile = fget(cfd); | |
3574 | if (!cfile) { | |
3575 | ret = -EBADF; | |
3576 | goto fail; | |
3577 | } | |
3578 | ||
3579 | /* the process need read permission on control file */ | |
3bfa784a AV |
3580 | /* AV: shouldn't we check that it's been opened for read instead? */ |
3581 | ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ); | |
0dea1168 KS |
3582 | if (ret < 0) |
3583 | goto fail; | |
3584 | ||
3585 | event->cft = __file_cft(cfile); | |
3586 | if (IS_ERR(event->cft)) { | |
3587 | ret = PTR_ERR(event->cft); | |
3588 | goto fail; | |
3589 | } | |
3590 | ||
3591 | if (!event->cft->register_event || !event->cft->unregister_event) { | |
3592 | ret = -EINVAL; | |
3593 | goto fail; | |
3594 | } | |
3595 | ||
3596 | ret = event->cft->register_event(cgrp, event->cft, | |
3597 | event->eventfd, buffer); | |
3598 | if (ret) | |
3599 | goto fail; | |
3600 | ||
3601 | if (efile->f_op->poll(efile, &event->pt) & POLLHUP) { | |
3602 | event->cft->unregister_event(cgrp, event->cft, event->eventfd); | |
3603 | ret = 0; | |
3604 | goto fail; | |
3605 | } | |
3606 | ||
a0a4db54 KS |
3607 | /* |
3608 | * Events should be removed after rmdir of cgroup directory, but before | |
3609 | * destroying subsystem state objects. Let's take reference to cgroup | |
3610 | * directory dentry to do that. | |
3611 | */ | |
3612 | dget(cgrp->dentry); | |
3613 | ||
0dea1168 KS |
3614 | spin_lock(&cgrp->event_list_lock); |
3615 | list_add(&event->list, &cgrp->event_list); | |
3616 | spin_unlock(&cgrp->event_list_lock); | |
3617 | ||
3618 | fput(cfile); | |
3619 | fput(efile); | |
3620 | ||
3621 | return 0; | |
3622 | ||
3623 | fail: | |
3624 | if (cfile) | |
3625 | fput(cfile); | |
3626 | ||
3627 | if (event && event->eventfd && !IS_ERR(event->eventfd)) | |
3628 | eventfd_ctx_put(event->eventfd); | |
3629 | ||
3630 | if (!IS_ERR_OR_NULL(efile)) | |
3631 | fput(efile); | |
3632 | ||
3633 | kfree(event); | |
3634 | ||
3635 | return ret; | |
3636 | } | |
3637 | ||
97978e6d DL |
3638 | static u64 cgroup_clone_children_read(struct cgroup *cgrp, |
3639 | struct cftype *cft) | |
3640 | { | |
3641 | return clone_children(cgrp); | |
3642 | } | |
3643 | ||
3644 | static int cgroup_clone_children_write(struct cgroup *cgrp, | |
3645 | struct cftype *cft, | |
3646 | u64 val) | |
3647 | { | |
3648 | if (val) | |
3649 | set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); | |
3650 | else | |
3651 | clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); | |
3652 | return 0; | |
3653 | } | |
3654 | ||
bbcb81d0 PM |
3655 | /* |
3656 | * for the common functions, 'private' gives the type of file | |
3657 | */ | |
102a775e BB |
3658 | /* for hysterical raisins, we can't put this on the older files */ |
3659 | #define CGROUP_FILE_GENERIC_PREFIX "cgroup." | |
81a6a5cd PM |
3660 | static struct cftype files[] = { |
3661 | { | |
3662 | .name = "tasks", | |
3663 | .open = cgroup_tasks_open, | |
af351026 | 3664 | .write_u64 = cgroup_tasks_write, |
102a775e | 3665 | .release = cgroup_pidlist_release, |
099fca32 | 3666 | .mode = S_IRUGO | S_IWUSR, |
81a6a5cd | 3667 | }, |
102a775e BB |
3668 | { |
3669 | .name = CGROUP_FILE_GENERIC_PREFIX "procs", | |
3670 | .open = cgroup_procs_open, | |
74a1166d | 3671 | .write_u64 = cgroup_procs_write, |
102a775e | 3672 | .release = cgroup_pidlist_release, |
74a1166d | 3673 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3674 | }, |
81a6a5cd PM |
3675 | { |
3676 | .name = "notify_on_release", | |
f4c753b7 | 3677 | .read_u64 = cgroup_read_notify_on_release, |
6379c106 | 3678 | .write_u64 = cgroup_write_notify_on_release, |
81a6a5cd | 3679 | }, |
0dea1168 KS |
3680 | { |
3681 | .name = CGROUP_FILE_GENERIC_PREFIX "event_control", | |
3682 | .write_string = cgroup_write_event_control, | |
3683 | .mode = S_IWUGO, | |
3684 | }, | |
97978e6d DL |
3685 | { |
3686 | .name = "cgroup.clone_children", | |
3687 | .read_u64 = cgroup_clone_children_read, | |
3688 | .write_u64 = cgroup_clone_children_write, | |
3689 | }, | |
81a6a5cd PM |
3690 | }; |
3691 | ||
3692 | static struct cftype cft_release_agent = { | |
3693 | .name = "release_agent", | |
e788e066 PM |
3694 | .read_seq_string = cgroup_release_agent_show, |
3695 | .write_string = cgroup_release_agent_write, | |
3696 | .max_write_len = PATH_MAX, | |
bbcb81d0 PM |
3697 | }; |
3698 | ||
bd89aabc | 3699 | static int cgroup_populate_dir(struct cgroup *cgrp) |
ddbcc7e8 PM |
3700 | { |
3701 | int err; | |
3702 | struct cgroup_subsys *ss; | |
3703 | ||
3704 | /* First clear out any existing files */ | |
bd89aabc | 3705 | cgroup_clear_directory(cgrp->dentry); |
ddbcc7e8 | 3706 | |
bd89aabc | 3707 | err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files)); |
bbcb81d0 PM |
3708 | if (err < 0) |
3709 | return err; | |
3710 | ||
bd89aabc PM |
3711 | if (cgrp == cgrp->top_cgroup) { |
3712 | if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0) | |
81a6a5cd PM |
3713 | return err; |
3714 | } | |
3715 | ||
bd89aabc PM |
3716 | for_each_subsys(cgrp->root, ss) { |
3717 | if (ss->populate && (err = ss->populate(ss, cgrp)) < 0) | |
ddbcc7e8 PM |
3718 | return err; |
3719 | } | |
38460b48 KH |
3720 | /* This cgroup is ready now */ |
3721 | for_each_subsys(cgrp->root, ss) { | |
3722 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3723 | /* | |
3724 | * Update id->css pointer and make this css visible from | |
3725 | * CSS ID functions. This pointer will be dereferened | |
3726 | * from RCU-read-side without locks. | |
3727 | */ | |
3728 | if (css->id) | |
3729 | rcu_assign_pointer(css->id->css, css); | |
3730 | } | |
ddbcc7e8 PM |
3731 | |
3732 | return 0; | |
3733 | } | |
3734 | ||
3735 | static void init_cgroup_css(struct cgroup_subsys_state *css, | |
3736 | struct cgroup_subsys *ss, | |
bd89aabc | 3737 | struct cgroup *cgrp) |
ddbcc7e8 | 3738 | { |
bd89aabc | 3739 | css->cgroup = cgrp; |
e7c5ec91 | 3740 | atomic_set(&css->refcnt, 1); |
ddbcc7e8 | 3741 | css->flags = 0; |
38460b48 | 3742 | css->id = NULL; |
bd89aabc | 3743 | if (cgrp == dummytop) |
ddbcc7e8 | 3744 | set_bit(CSS_ROOT, &css->flags); |
bd89aabc PM |
3745 | BUG_ON(cgrp->subsys[ss->subsys_id]); |
3746 | cgrp->subsys[ss->subsys_id] = css; | |
ddbcc7e8 PM |
3747 | } |
3748 | ||
999cd8a4 PM |
3749 | static void cgroup_lock_hierarchy(struct cgroupfs_root *root) |
3750 | { | |
3751 | /* We need to take each hierarchy_mutex in a consistent order */ | |
3752 | int i; | |
3753 | ||
aae8aab4 BB |
3754 | /* |
3755 | * No worry about a race with rebind_subsystems that might mess up the | |
3756 | * locking order, since both parties are under cgroup_mutex. | |
3757 | */ | |
999cd8a4 PM |
3758 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
3759 | struct cgroup_subsys *ss = subsys[i]; | |
aae8aab4 BB |
3760 | if (ss == NULL) |
3761 | continue; | |
999cd8a4 | 3762 | if (ss->root == root) |
cfebe563 | 3763 | mutex_lock(&ss->hierarchy_mutex); |
999cd8a4 PM |
3764 | } |
3765 | } | |
3766 | ||
3767 | static void cgroup_unlock_hierarchy(struct cgroupfs_root *root) | |
3768 | { | |
3769 | int i; | |
3770 | ||
3771 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
3772 | struct cgroup_subsys *ss = subsys[i]; | |
aae8aab4 BB |
3773 | if (ss == NULL) |
3774 | continue; | |
999cd8a4 PM |
3775 | if (ss->root == root) |
3776 | mutex_unlock(&ss->hierarchy_mutex); | |
3777 | } | |
3778 | } | |
3779 | ||
ddbcc7e8 | 3780 | /* |
a043e3b2 LZ |
3781 | * cgroup_create - create a cgroup |
3782 | * @parent: cgroup that will be parent of the new cgroup | |
3783 | * @dentry: dentry of the new cgroup | |
3784 | * @mode: mode to set on new inode | |
ddbcc7e8 | 3785 | * |
a043e3b2 | 3786 | * Must be called with the mutex on the parent inode held |
ddbcc7e8 | 3787 | */ |
ddbcc7e8 | 3788 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, |
099fca32 | 3789 | mode_t mode) |
ddbcc7e8 | 3790 | { |
bd89aabc | 3791 | struct cgroup *cgrp; |
ddbcc7e8 PM |
3792 | struct cgroupfs_root *root = parent->root; |
3793 | int err = 0; | |
3794 | struct cgroup_subsys *ss; | |
3795 | struct super_block *sb = root->sb; | |
3796 | ||
bd89aabc PM |
3797 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
3798 | if (!cgrp) | |
ddbcc7e8 PM |
3799 | return -ENOMEM; |
3800 | ||
3801 | /* Grab a reference on the superblock so the hierarchy doesn't | |
3802 | * get deleted on unmount if there are child cgroups. This | |
3803 | * can be done outside cgroup_mutex, since the sb can't | |
3804 | * disappear while someone has an open control file on the | |
3805 | * fs */ | |
3806 | atomic_inc(&sb->s_active); | |
3807 | ||
3808 | mutex_lock(&cgroup_mutex); | |
3809 | ||
cc31edce | 3810 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3811 | |
bd89aabc PM |
3812 | cgrp->parent = parent; |
3813 | cgrp->root = parent->root; | |
3814 | cgrp->top_cgroup = parent->top_cgroup; | |
ddbcc7e8 | 3815 | |
b6abdb0e LZ |
3816 | if (notify_on_release(parent)) |
3817 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3818 | ||
97978e6d DL |
3819 | if (clone_children(parent)) |
3820 | set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); | |
3821 | ||
ddbcc7e8 | 3822 | for_each_subsys(root, ss) { |
bd89aabc | 3823 | struct cgroup_subsys_state *css = ss->create(ss, cgrp); |
4528fd05 | 3824 | |
ddbcc7e8 PM |
3825 | if (IS_ERR(css)) { |
3826 | err = PTR_ERR(css); | |
3827 | goto err_destroy; | |
3828 | } | |
bd89aabc | 3829 | init_cgroup_css(css, ss, cgrp); |
4528fd05 LZ |
3830 | if (ss->use_id) { |
3831 | err = alloc_css_id(ss, parent, cgrp); | |
3832 | if (err) | |
38460b48 | 3833 | goto err_destroy; |
4528fd05 | 3834 | } |
38460b48 | 3835 | /* At error, ->destroy() callback has to free assigned ID. */ |
97978e6d DL |
3836 | if (clone_children(parent) && ss->post_clone) |
3837 | ss->post_clone(ss, cgrp); | |
ddbcc7e8 PM |
3838 | } |
3839 | ||
999cd8a4 | 3840 | cgroup_lock_hierarchy(root); |
bd89aabc | 3841 | list_add(&cgrp->sibling, &cgrp->parent->children); |
999cd8a4 | 3842 | cgroup_unlock_hierarchy(root); |
ddbcc7e8 PM |
3843 | root->number_of_cgroups++; |
3844 | ||
bd89aabc | 3845 | err = cgroup_create_dir(cgrp, dentry, mode); |
ddbcc7e8 PM |
3846 | if (err < 0) |
3847 | goto err_remove; | |
3848 | ||
3849 | /* The cgroup directory was pre-locked for us */ | |
bd89aabc | 3850 | BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex)); |
ddbcc7e8 | 3851 | |
bd89aabc | 3852 | err = cgroup_populate_dir(cgrp); |
ddbcc7e8 PM |
3853 | /* If err < 0, we have a half-filled directory - oh well ;) */ |
3854 | ||
3855 | mutex_unlock(&cgroup_mutex); | |
bd89aabc | 3856 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
3857 | |
3858 | return 0; | |
3859 | ||
3860 | err_remove: | |
3861 | ||
baef99a0 | 3862 | cgroup_lock_hierarchy(root); |
bd89aabc | 3863 | list_del(&cgrp->sibling); |
baef99a0 | 3864 | cgroup_unlock_hierarchy(root); |
ddbcc7e8 PM |
3865 | root->number_of_cgroups--; |
3866 | ||
3867 | err_destroy: | |
3868 | ||
3869 | for_each_subsys(root, ss) { | |
bd89aabc PM |
3870 | if (cgrp->subsys[ss->subsys_id]) |
3871 | ss->destroy(ss, cgrp); | |
ddbcc7e8 PM |
3872 | } |
3873 | ||
3874 | mutex_unlock(&cgroup_mutex); | |
3875 | ||
3876 | /* Release the reference count that we took on the superblock */ | |
3877 | deactivate_super(sb); | |
3878 | ||
bd89aabc | 3879 | kfree(cgrp); |
ddbcc7e8 PM |
3880 | return err; |
3881 | } | |
3882 | ||
3883 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
3884 | { | |
3885 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
3886 | ||
3887 | /* the vfs holds inode->i_mutex already */ | |
3888 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
3889 | } | |
3890 | ||
55b6fd01 | 3891 | static int cgroup_has_css_refs(struct cgroup *cgrp) |
81a6a5cd PM |
3892 | { |
3893 | /* Check the reference count on each subsystem. Since we | |
3894 | * already established that there are no tasks in the | |
e7c5ec91 | 3895 | * cgroup, if the css refcount is also 1, then there should |
81a6a5cd PM |
3896 | * be no outstanding references, so the subsystem is safe to |
3897 | * destroy. We scan across all subsystems rather than using | |
3898 | * the per-hierarchy linked list of mounted subsystems since | |
3899 | * we can be called via check_for_release() with no | |
3900 | * synchronization other than RCU, and the subsystem linked | |
3901 | * list isn't RCU-safe */ | |
3902 | int i; | |
aae8aab4 BB |
3903 | /* |
3904 | * We won't need to lock the subsys array, because the subsystems | |
3905 | * we're concerned about aren't going anywhere since our cgroup root | |
3906 | * has a reference on them. | |
3907 | */ | |
81a6a5cd PM |
3908 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
3909 | struct cgroup_subsys *ss = subsys[i]; | |
3910 | struct cgroup_subsys_state *css; | |
aae8aab4 BB |
3911 | /* Skip subsystems not present or not in this hierarchy */ |
3912 | if (ss == NULL || ss->root != cgrp->root) | |
81a6a5cd | 3913 | continue; |
bd89aabc | 3914 | css = cgrp->subsys[ss->subsys_id]; |
81a6a5cd PM |
3915 | /* When called from check_for_release() it's possible |
3916 | * that by this point the cgroup has been removed | |
3917 | * and the css deleted. But a false-positive doesn't | |
3918 | * matter, since it can only happen if the cgroup | |
3919 | * has been deleted and hence no longer needs the | |
3920 | * release agent to be called anyway. */ | |
e7c5ec91 | 3921 | if (css && (atomic_read(&css->refcnt) > 1)) |
81a6a5cd | 3922 | return 1; |
81a6a5cd PM |
3923 | } |
3924 | return 0; | |
3925 | } | |
3926 | ||
e7c5ec91 PM |
3927 | /* |
3928 | * Atomically mark all (or else none) of the cgroup's CSS objects as | |
3929 | * CSS_REMOVED. Return true on success, or false if the cgroup has | |
3930 | * busy subsystems. Call with cgroup_mutex held | |
3931 | */ | |
3932 | ||
3933 | static int cgroup_clear_css_refs(struct cgroup *cgrp) | |
3934 | { | |
3935 | struct cgroup_subsys *ss; | |
3936 | unsigned long flags; | |
3937 | bool failed = false; | |
3938 | local_irq_save(flags); | |
3939 | for_each_subsys(cgrp->root, ss) { | |
3940 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3941 | int refcnt; | |
804b3c28 | 3942 | while (1) { |
e7c5ec91 PM |
3943 | /* We can only remove a CSS with a refcnt==1 */ |
3944 | refcnt = atomic_read(&css->refcnt); | |
3945 | if (refcnt > 1) { | |
3946 | failed = true; | |
3947 | goto done; | |
3948 | } | |
3949 | BUG_ON(!refcnt); | |
3950 | /* | |
3951 | * Drop the refcnt to 0 while we check other | |
3952 | * subsystems. This will cause any racing | |
3953 | * css_tryget() to spin until we set the | |
3954 | * CSS_REMOVED bits or abort | |
3955 | */ | |
804b3c28 PM |
3956 | if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt) |
3957 | break; | |
3958 | cpu_relax(); | |
3959 | } | |
e7c5ec91 PM |
3960 | } |
3961 | done: | |
3962 | for_each_subsys(cgrp->root, ss) { | |
3963 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3964 | if (failed) { | |
3965 | /* | |
3966 | * Restore old refcnt if we previously managed | |
3967 | * to clear it from 1 to 0 | |
3968 | */ | |
3969 | if (!atomic_read(&css->refcnt)) | |
3970 | atomic_set(&css->refcnt, 1); | |
3971 | } else { | |
3972 | /* Commit the fact that the CSS is removed */ | |
3973 | set_bit(CSS_REMOVED, &css->flags); | |
3974 | } | |
3975 | } | |
3976 | local_irq_restore(flags); | |
3977 | return !failed; | |
3978 | } | |
3979 | ||
ddbcc7e8 PM |
3980 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) |
3981 | { | |
bd89aabc | 3982 | struct cgroup *cgrp = dentry->d_fsdata; |
ddbcc7e8 PM |
3983 | struct dentry *d; |
3984 | struct cgroup *parent; | |
ec64f515 | 3985 | DEFINE_WAIT(wait); |
4ab78683 | 3986 | struct cgroup_event *event, *tmp; |
ec64f515 | 3987 | int ret; |
ddbcc7e8 PM |
3988 | |
3989 | /* the vfs holds both inode->i_mutex already */ | |
ec64f515 | 3990 | again: |
ddbcc7e8 | 3991 | mutex_lock(&cgroup_mutex); |
bd89aabc | 3992 | if (atomic_read(&cgrp->count) != 0) { |
ddbcc7e8 PM |
3993 | mutex_unlock(&cgroup_mutex); |
3994 | return -EBUSY; | |
3995 | } | |
bd89aabc | 3996 | if (!list_empty(&cgrp->children)) { |
ddbcc7e8 PM |
3997 | mutex_unlock(&cgroup_mutex); |
3998 | return -EBUSY; | |
3999 | } | |
3fa59dfb | 4000 | mutex_unlock(&cgroup_mutex); |
a043e3b2 | 4001 | |
88703267 KH |
4002 | /* |
4003 | * In general, subsystem has no css->refcnt after pre_destroy(). But | |
4004 | * in racy cases, subsystem may have to get css->refcnt after | |
4005 | * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes | |
4006 | * make rmdir return -EBUSY too often. To avoid that, we use waitqueue | |
4007 | * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir | |
4008 | * and subsystem's reference count handling. Please see css_get/put | |
4009 | * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation. | |
4010 | */ | |
4011 | set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
4012 | ||
4fca88c8 | 4013 | /* |
a043e3b2 LZ |
4014 | * Call pre_destroy handlers of subsys. Notify subsystems |
4015 | * that rmdir() request comes. | |
4fca88c8 | 4016 | */ |
ec64f515 | 4017 | ret = cgroup_call_pre_destroy(cgrp); |
88703267 KH |
4018 | if (ret) { |
4019 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
ec64f515 | 4020 | return ret; |
88703267 | 4021 | } |
ddbcc7e8 | 4022 | |
3fa59dfb KH |
4023 | mutex_lock(&cgroup_mutex); |
4024 | parent = cgrp->parent; | |
ec64f515 | 4025 | if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) { |
88703267 | 4026 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); |
ddbcc7e8 PM |
4027 | mutex_unlock(&cgroup_mutex); |
4028 | return -EBUSY; | |
4029 | } | |
ec64f515 | 4030 | prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE); |
ec64f515 KH |
4031 | if (!cgroup_clear_css_refs(cgrp)) { |
4032 | mutex_unlock(&cgroup_mutex); | |
88703267 KH |
4033 | /* |
4034 | * Because someone may call cgroup_wakeup_rmdir_waiter() before | |
4035 | * prepare_to_wait(), we need to check this flag. | |
4036 | */ | |
4037 | if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)) | |
4038 | schedule(); | |
ec64f515 KH |
4039 | finish_wait(&cgroup_rmdir_waitq, &wait); |
4040 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
4041 | if (signal_pending(current)) | |
4042 | return -EINTR; | |
4043 | goto again; | |
4044 | } | |
4045 | /* NO css_tryget() can success after here. */ | |
4046 | finish_wait(&cgroup_rmdir_waitq, &wait); | |
4047 | clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags); | |
ddbcc7e8 | 4048 | |
cdcc136f | 4049 | raw_spin_lock(&release_list_lock); |
bd89aabc PM |
4050 | set_bit(CGRP_REMOVED, &cgrp->flags); |
4051 | if (!list_empty(&cgrp->release_list)) | |
8d258797 | 4052 | list_del_init(&cgrp->release_list); |
cdcc136f | 4053 | raw_spin_unlock(&release_list_lock); |
999cd8a4 PM |
4054 | |
4055 | cgroup_lock_hierarchy(cgrp->root); | |
4056 | /* delete this cgroup from parent->children */ | |
8d258797 | 4057 | list_del_init(&cgrp->sibling); |
999cd8a4 PM |
4058 | cgroup_unlock_hierarchy(cgrp->root); |
4059 | ||
bd89aabc | 4060 | d = dget(cgrp->dentry); |
ddbcc7e8 PM |
4061 | |
4062 | cgroup_d_remove_dir(d); | |
4063 | dput(d); | |
ddbcc7e8 | 4064 | |
bd89aabc | 4065 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd PM |
4066 | check_for_release(parent); |
4067 | ||
4ab78683 KS |
4068 | /* |
4069 | * Unregister events and notify userspace. | |
4070 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4071 | * directory to avoid race between userspace and kernelspace | |
4072 | */ | |
4073 | spin_lock(&cgrp->event_list_lock); | |
4074 | list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) { | |
4075 | list_del(&event->list); | |
4076 | remove_wait_queue(event->wqh, &event->wait); | |
4077 | eventfd_signal(event->eventfd, 1); | |
4078 | schedule_work(&event->remove); | |
4079 | } | |
4080 | spin_unlock(&cgrp->event_list_lock); | |
4081 | ||
ddbcc7e8 | 4082 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
4083 | return 0; |
4084 | } | |
4085 | ||
06a11920 | 4086 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4087 | { |
ddbcc7e8 | 4088 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4089 | |
4090 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 PM |
4091 | |
4092 | /* Create the top cgroup state for this subsystem */ | |
33a68ac1 | 4093 | list_add(&ss->sibling, &rootnode.subsys_list); |
ddbcc7e8 PM |
4094 | ss->root = &rootnode; |
4095 | css = ss->create(ss, dummytop); | |
4096 | /* We don't handle early failures gracefully */ | |
4097 | BUG_ON(IS_ERR(css)); | |
4098 | init_cgroup_css(css, ss, dummytop); | |
4099 | ||
e8d55fde | 4100 | /* Update the init_css_set to contain a subsys |
817929ec | 4101 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
4102 | * newly registered, all tasks and hence the |
4103 | * init_css_set is in the subsystem's top cgroup. */ | |
4104 | init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id]; | |
ddbcc7e8 PM |
4105 | |
4106 | need_forkexit_callback |= ss->fork || ss->exit; | |
4107 | ||
e8d55fde LZ |
4108 | /* At system boot, before all subsystems have been |
4109 | * registered, no tasks have been forked, so we don't | |
4110 | * need to invoke fork callbacks here. */ | |
4111 | BUG_ON(!list_empty(&init_task.tasks)); | |
4112 | ||
999cd8a4 | 4113 | mutex_init(&ss->hierarchy_mutex); |
cfebe563 | 4114 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); |
ddbcc7e8 | 4115 | ss->active = 1; |
e6a1105b BB |
4116 | |
4117 | /* this function shouldn't be used with modular subsystems, since they | |
4118 | * need to register a subsys_id, among other things */ | |
4119 | BUG_ON(ss->module); | |
4120 | } | |
4121 | ||
4122 | /** | |
4123 | * cgroup_load_subsys: load and register a modular subsystem at runtime | |
4124 | * @ss: the subsystem to load | |
4125 | * | |
4126 | * This function should be called in a modular subsystem's initcall. If the | |
88393161 | 4127 | * subsystem is built as a module, it will be assigned a new subsys_id and set |
e6a1105b BB |
4128 | * up for use. If the subsystem is built-in anyway, work is delegated to the |
4129 | * simpler cgroup_init_subsys. | |
4130 | */ | |
4131 | int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss) | |
4132 | { | |
4133 | int i; | |
4134 | struct cgroup_subsys_state *css; | |
4135 | ||
4136 | /* check name and function validity */ | |
4137 | if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN || | |
4138 | ss->create == NULL || ss->destroy == NULL) | |
4139 | return -EINVAL; | |
4140 | ||
4141 | /* | |
4142 | * we don't support callbacks in modular subsystems. this check is | |
4143 | * before the ss->module check for consistency; a subsystem that could | |
4144 | * be a module should still have no callbacks even if the user isn't | |
4145 | * compiling it as one. | |
4146 | */ | |
4147 | if (ss->fork || ss->exit) | |
4148 | return -EINVAL; | |
4149 | ||
4150 | /* | |
4151 | * an optionally modular subsystem is built-in: we want to do nothing, | |
4152 | * since cgroup_init_subsys will have already taken care of it. | |
4153 | */ | |
4154 | if (ss->module == NULL) { | |
4155 | /* a few sanity checks */ | |
4156 | BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT); | |
4157 | BUG_ON(subsys[ss->subsys_id] != ss); | |
4158 | return 0; | |
4159 | } | |
4160 | ||
4161 | /* | |
4162 | * need to register a subsys id before anything else - for example, | |
4163 | * init_cgroup_css needs it. | |
4164 | */ | |
4165 | mutex_lock(&cgroup_mutex); | |
4166 | /* find the first empty slot in the array */ | |
4167 | for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) { | |
4168 | if (subsys[i] == NULL) | |
4169 | break; | |
4170 | } | |
4171 | if (i == CGROUP_SUBSYS_COUNT) { | |
4172 | /* maximum number of subsystems already registered! */ | |
4173 | mutex_unlock(&cgroup_mutex); | |
4174 | return -EBUSY; | |
4175 | } | |
4176 | /* assign ourselves the subsys_id */ | |
4177 | ss->subsys_id = i; | |
4178 | subsys[i] = ss; | |
4179 | ||
4180 | /* | |
4181 | * no ss->create seems to need anything important in the ss struct, so | |
4182 | * this can happen first (i.e. before the rootnode attachment). | |
4183 | */ | |
4184 | css = ss->create(ss, dummytop); | |
4185 | if (IS_ERR(css)) { | |
4186 | /* failure case - need to deassign the subsys[] slot. */ | |
4187 | subsys[i] = NULL; | |
4188 | mutex_unlock(&cgroup_mutex); | |
4189 | return PTR_ERR(css); | |
4190 | } | |
4191 | ||
4192 | list_add(&ss->sibling, &rootnode.subsys_list); | |
4193 | ss->root = &rootnode; | |
4194 | ||
4195 | /* our new subsystem will be attached to the dummy hierarchy. */ | |
4196 | init_cgroup_css(css, ss, dummytop); | |
4197 | /* init_idr must be after init_cgroup_css because it sets css->id. */ | |
4198 | if (ss->use_id) { | |
4199 | int ret = cgroup_init_idr(ss, css); | |
4200 | if (ret) { | |
4201 | dummytop->subsys[ss->subsys_id] = NULL; | |
4202 | ss->destroy(ss, dummytop); | |
4203 | subsys[i] = NULL; | |
4204 | mutex_unlock(&cgroup_mutex); | |
4205 | return ret; | |
4206 | } | |
4207 | } | |
4208 | ||
4209 | /* | |
4210 | * Now we need to entangle the css into the existing css_sets. unlike | |
4211 | * in cgroup_init_subsys, there are now multiple css_sets, so each one | |
4212 | * will need a new pointer to it; done by iterating the css_set_table. | |
4213 | * furthermore, modifying the existing css_sets will corrupt the hash | |
4214 | * table state, so each changed css_set will need its hash recomputed. | |
4215 | * this is all done under the css_set_lock. | |
4216 | */ | |
4217 | write_lock(&css_set_lock); | |
4218 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { | |
4219 | struct css_set *cg; | |
4220 | struct hlist_node *node, *tmp; | |
4221 | struct hlist_head *bucket = &css_set_table[i], *new_bucket; | |
4222 | ||
4223 | hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) { | |
4224 | /* skip entries that we already rehashed */ | |
4225 | if (cg->subsys[ss->subsys_id]) | |
4226 | continue; | |
4227 | /* remove existing entry */ | |
4228 | hlist_del(&cg->hlist); | |
4229 | /* set new value */ | |
4230 | cg->subsys[ss->subsys_id] = css; | |
4231 | /* recompute hash and restore entry */ | |
4232 | new_bucket = css_set_hash(cg->subsys); | |
4233 | hlist_add_head(&cg->hlist, new_bucket); | |
4234 | } | |
4235 | } | |
4236 | write_unlock(&css_set_lock); | |
4237 | ||
4238 | mutex_init(&ss->hierarchy_mutex); | |
4239 | lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key); | |
4240 | ss->active = 1; | |
4241 | ||
e6a1105b BB |
4242 | /* success! */ |
4243 | mutex_unlock(&cgroup_mutex); | |
4244 | return 0; | |
ddbcc7e8 | 4245 | } |
e6a1105b | 4246 | EXPORT_SYMBOL_GPL(cgroup_load_subsys); |
ddbcc7e8 | 4247 | |
cf5d5941 BB |
4248 | /** |
4249 | * cgroup_unload_subsys: unload a modular subsystem | |
4250 | * @ss: the subsystem to unload | |
4251 | * | |
4252 | * This function should be called in a modular subsystem's exitcall. When this | |
4253 | * function is invoked, the refcount on the subsystem's module will be 0, so | |
4254 | * the subsystem will not be attached to any hierarchy. | |
4255 | */ | |
4256 | void cgroup_unload_subsys(struct cgroup_subsys *ss) | |
4257 | { | |
4258 | struct cg_cgroup_link *link; | |
4259 | struct hlist_head *hhead; | |
4260 | ||
4261 | BUG_ON(ss->module == NULL); | |
4262 | ||
4263 | /* | |
4264 | * we shouldn't be called if the subsystem is in use, and the use of | |
4265 | * try_module_get in parse_cgroupfs_options should ensure that it | |
4266 | * doesn't start being used while we're killing it off. | |
4267 | */ | |
4268 | BUG_ON(ss->root != &rootnode); | |
4269 | ||
4270 | mutex_lock(&cgroup_mutex); | |
4271 | /* deassign the subsys_id */ | |
4272 | BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT); | |
4273 | subsys[ss->subsys_id] = NULL; | |
4274 | ||
4275 | /* remove subsystem from rootnode's list of subsystems */ | |
8d258797 | 4276 | list_del_init(&ss->sibling); |
cf5d5941 BB |
4277 | |
4278 | /* | |
4279 | * disentangle the css from all css_sets attached to the dummytop. as | |
4280 | * in loading, we need to pay our respects to the hashtable gods. | |
4281 | */ | |
4282 | write_lock(&css_set_lock); | |
4283 | list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) { | |
4284 | struct css_set *cg = link->cg; | |
4285 | ||
4286 | hlist_del(&cg->hlist); | |
4287 | BUG_ON(!cg->subsys[ss->subsys_id]); | |
4288 | cg->subsys[ss->subsys_id] = NULL; | |
4289 | hhead = css_set_hash(cg->subsys); | |
4290 | hlist_add_head(&cg->hlist, hhead); | |
4291 | } | |
4292 | write_unlock(&css_set_lock); | |
4293 | ||
4294 | /* | |
4295 | * remove subsystem's css from the dummytop and free it - need to free | |
4296 | * before marking as null because ss->destroy needs the cgrp->subsys | |
4297 | * pointer to find their state. note that this also takes care of | |
4298 | * freeing the css_id. | |
4299 | */ | |
4300 | ss->destroy(ss, dummytop); | |
4301 | dummytop->subsys[ss->subsys_id] = NULL; | |
4302 | ||
4303 | mutex_unlock(&cgroup_mutex); | |
4304 | } | |
4305 | EXPORT_SYMBOL_GPL(cgroup_unload_subsys); | |
4306 | ||
ddbcc7e8 | 4307 | /** |
a043e3b2 LZ |
4308 | * cgroup_init_early - cgroup initialization at system boot |
4309 | * | |
4310 | * Initialize cgroups at system boot, and initialize any | |
4311 | * subsystems that request early init. | |
ddbcc7e8 PM |
4312 | */ |
4313 | int __init cgroup_init_early(void) | |
4314 | { | |
4315 | int i; | |
146aa1bd | 4316 | atomic_set(&init_css_set.refcount, 1); |
817929ec PM |
4317 | INIT_LIST_HEAD(&init_css_set.cg_links); |
4318 | INIT_LIST_HEAD(&init_css_set.tasks); | |
472b1053 | 4319 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 4320 | css_set_count = 1; |
ddbcc7e8 | 4321 | init_cgroup_root(&rootnode); |
817929ec PM |
4322 | root_count = 1; |
4323 | init_task.cgroups = &init_css_set; | |
4324 | ||
4325 | init_css_set_link.cg = &init_css_set; | |
7717f7ba | 4326 | init_css_set_link.cgrp = dummytop; |
bd89aabc | 4327 | list_add(&init_css_set_link.cgrp_link_list, |
817929ec PM |
4328 | &rootnode.top_cgroup.css_sets); |
4329 | list_add(&init_css_set_link.cg_link_list, | |
4330 | &init_css_set.cg_links); | |
ddbcc7e8 | 4331 | |
472b1053 LZ |
4332 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) |
4333 | INIT_HLIST_HEAD(&css_set_table[i]); | |
4334 | ||
aae8aab4 BB |
4335 | /* at bootup time, we don't worry about modular subsystems */ |
4336 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
ddbcc7e8 PM |
4337 | struct cgroup_subsys *ss = subsys[i]; |
4338 | ||
4339 | BUG_ON(!ss->name); | |
4340 | BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); | |
4341 | BUG_ON(!ss->create); | |
4342 | BUG_ON(!ss->destroy); | |
4343 | if (ss->subsys_id != i) { | |
cfe36bde | 4344 | printk(KERN_ERR "cgroup: Subsys %s id == %d\n", |
ddbcc7e8 PM |
4345 | ss->name, ss->subsys_id); |
4346 | BUG(); | |
4347 | } | |
4348 | ||
4349 | if (ss->early_init) | |
4350 | cgroup_init_subsys(ss); | |
4351 | } | |
4352 | return 0; | |
4353 | } | |
4354 | ||
4355 | /** | |
a043e3b2 LZ |
4356 | * cgroup_init - cgroup initialization |
4357 | * | |
4358 | * Register cgroup filesystem and /proc file, and initialize | |
4359 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4360 | */ |
4361 | int __init cgroup_init(void) | |
4362 | { | |
4363 | int err; | |
4364 | int i; | |
472b1053 | 4365 | struct hlist_head *hhead; |
a424316c PM |
4366 | |
4367 | err = bdi_init(&cgroup_backing_dev_info); | |
4368 | if (err) | |
4369 | return err; | |
ddbcc7e8 | 4370 | |
aae8aab4 BB |
4371 | /* at bootup time, we don't worry about modular subsystems */ |
4372 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
ddbcc7e8 PM |
4373 | struct cgroup_subsys *ss = subsys[i]; |
4374 | if (!ss->early_init) | |
4375 | cgroup_init_subsys(ss); | |
38460b48 | 4376 | if (ss->use_id) |
e6a1105b | 4377 | cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]); |
ddbcc7e8 PM |
4378 | } |
4379 | ||
472b1053 LZ |
4380 | /* Add init_css_set to the hash table */ |
4381 | hhead = css_set_hash(init_css_set.subsys); | |
4382 | hlist_add_head(&init_css_set.hlist, hhead); | |
2c6ab6d2 | 4383 | BUG_ON(!init_root_id(&rootnode)); |
676db4af GK |
4384 | |
4385 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); | |
4386 | if (!cgroup_kobj) { | |
4387 | err = -ENOMEM; | |
4388 | goto out; | |
4389 | } | |
4390 | ||
ddbcc7e8 | 4391 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
4392 | if (err < 0) { |
4393 | kobject_put(cgroup_kobj); | |
ddbcc7e8 | 4394 | goto out; |
676db4af | 4395 | } |
ddbcc7e8 | 4396 | |
46ae220b | 4397 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
a424316c | 4398 | |
ddbcc7e8 | 4399 | out: |
a424316c PM |
4400 | if (err) |
4401 | bdi_destroy(&cgroup_backing_dev_info); | |
4402 | ||
ddbcc7e8 PM |
4403 | return err; |
4404 | } | |
b4f48b63 | 4405 | |
a424316c PM |
4406 | /* |
4407 | * proc_cgroup_show() | |
4408 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4409 | * - Used for /proc/<pid>/cgroup. | |
4410 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4411 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4412 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4413 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4414 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4415 | * cgroup to top_cgroup. | |
4416 | */ | |
4417 | ||
4418 | /* TODO: Use a proper seq_file iterator */ | |
4419 | static int proc_cgroup_show(struct seq_file *m, void *v) | |
4420 | { | |
4421 | struct pid *pid; | |
4422 | struct task_struct *tsk; | |
4423 | char *buf; | |
4424 | int retval; | |
4425 | struct cgroupfs_root *root; | |
4426 | ||
4427 | retval = -ENOMEM; | |
4428 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4429 | if (!buf) | |
4430 | goto out; | |
4431 | ||
4432 | retval = -ESRCH; | |
4433 | pid = m->private; | |
4434 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4435 | if (!tsk) | |
4436 | goto out_free; | |
4437 | ||
4438 | retval = 0; | |
4439 | ||
4440 | mutex_lock(&cgroup_mutex); | |
4441 | ||
e5f6a860 | 4442 | for_each_active_root(root) { |
a424316c | 4443 | struct cgroup_subsys *ss; |
bd89aabc | 4444 | struct cgroup *cgrp; |
a424316c PM |
4445 | int count = 0; |
4446 | ||
2c6ab6d2 | 4447 | seq_printf(m, "%d:", root->hierarchy_id); |
a424316c PM |
4448 | for_each_subsys(root, ss) |
4449 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4450 | if (strlen(root->name)) |
4451 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4452 | root->name); | |
a424316c | 4453 | seq_putc(m, ':'); |
7717f7ba | 4454 | cgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 4455 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); |
a424316c PM |
4456 | if (retval < 0) |
4457 | goto out_unlock; | |
4458 | seq_puts(m, buf); | |
4459 | seq_putc(m, '\n'); | |
4460 | } | |
4461 | ||
4462 | out_unlock: | |
4463 | mutex_unlock(&cgroup_mutex); | |
4464 | put_task_struct(tsk); | |
4465 | out_free: | |
4466 | kfree(buf); | |
4467 | out: | |
4468 | return retval; | |
4469 | } | |
4470 | ||
4471 | static int cgroup_open(struct inode *inode, struct file *file) | |
4472 | { | |
4473 | struct pid *pid = PROC_I(inode)->pid; | |
4474 | return single_open(file, proc_cgroup_show, pid); | |
4475 | } | |
4476 | ||
828c0950 | 4477 | const struct file_operations proc_cgroup_operations = { |
a424316c PM |
4478 | .open = cgroup_open, |
4479 | .read = seq_read, | |
4480 | .llseek = seq_lseek, | |
4481 | .release = single_release, | |
4482 | }; | |
4483 | ||
4484 | /* Display information about each subsystem and each hierarchy */ | |
4485 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4486 | { | |
4487 | int i; | |
a424316c | 4488 | |
8bab8dde | 4489 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4490 | /* |
4491 | * ideally we don't want subsystems moving around while we do this. | |
4492 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4493 | * subsys/hierarchy state. | |
4494 | */ | |
a424316c | 4495 | mutex_lock(&cgroup_mutex); |
a424316c PM |
4496 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
4497 | struct cgroup_subsys *ss = subsys[i]; | |
aae8aab4 BB |
4498 | if (ss == NULL) |
4499 | continue; | |
2c6ab6d2 PM |
4500 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4501 | ss->name, ss->root->hierarchy_id, | |
8bab8dde | 4502 | ss->root->number_of_cgroups, !ss->disabled); |
a424316c PM |
4503 | } |
4504 | mutex_unlock(&cgroup_mutex); | |
4505 | return 0; | |
4506 | } | |
4507 | ||
4508 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4509 | { | |
9dce07f1 | 4510 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4511 | } |
4512 | ||
828c0950 | 4513 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4514 | .open = cgroupstats_open, |
4515 | .read = seq_read, | |
4516 | .llseek = seq_lseek, | |
4517 | .release = single_release, | |
4518 | }; | |
4519 | ||
b4f48b63 PM |
4520 | /** |
4521 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4522 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4523 | * |
4524 | * Description: A task inherits its parent's cgroup at fork(). | |
4525 | * | |
4526 | * A pointer to the shared css_set was automatically copied in | |
4527 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
4528 | * it was not made under the protection of RCU or cgroup_mutex, so | |
956db3ca | 4529 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might |
817929ec PM |
4530 | * have already changed current->cgroups, allowing the previously |
4531 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
4532 | * |
4533 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4534 | * task, and the passed argument 'child' points to the child task. | |
4535 | */ | |
4536 | void cgroup_fork(struct task_struct *child) | |
4537 | { | |
817929ec PM |
4538 | task_lock(current); |
4539 | child->cgroups = current->cgroups; | |
4540 | get_css_set(child->cgroups); | |
4541 | task_unlock(current); | |
4542 | INIT_LIST_HEAD(&child->cg_list); | |
b4f48b63 PM |
4543 | } |
4544 | ||
4545 | /** | |
a043e3b2 LZ |
4546 | * cgroup_fork_callbacks - run fork callbacks |
4547 | * @child: the new task | |
4548 | * | |
4549 | * Called on a new task very soon before adding it to the | |
4550 | * tasklist. No need to take any locks since no-one can | |
4551 | * be operating on this task. | |
b4f48b63 PM |
4552 | */ |
4553 | void cgroup_fork_callbacks(struct task_struct *child) | |
4554 | { | |
4555 | if (need_forkexit_callback) { | |
4556 | int i; | |
aae8aab4 BB |
4557 | /* |
4558 | * forkexit callbacks are only supported for builtin | |
4559 | * subsystems, and the builtin section of the subsys array is | |
4560 | * immutable, so we don't need to lock the subsys array here. | |
4561 | */ | |
4562 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
b4f48b63 PM |
4563 | struct cgroup_subsys *ss = subsys[i]; |
4564 | if (ss->fork) | |
4565 | ss->fork(ss, child); | |
4566 | } | |
4567 | } | |
4568 | } | |
4569 | ||
817929ec | 4570 | /** |
a043e3b2 LZ |
4571 | * cgroup_post_fork - called on a new task after adding it to the task list |
4572 | * @child: the task in question | |
4573 | * | |
4574 | * Adds the task to the list running through its css_set if necessary. | |
4575 | * Has to be after the task is visible on the task list in case we race | |
4576 | * with the first call to cgroup_iter_start() - to guarantee that the | |
4577 | * new task ends up on its list. | |
4578 | */ | |
817929ec PM |
4579 | void cgroup_post_fork(struct task_struct *child) |
4580 | { | |
4581 | if (use_task_css_set_links) { | |
4582 | write_lock(&css_set_lock); | |
b12b533f | 4583 | task_lock(child); |
817929ec PM |
4584 | if (list_empty(&child->cg_list)) |
4585 | list_add(&child->cg_list, &child->cgroups->tasks); | |
b12b533f | 4586 | task_unlock(child); |
817929ec PM |
4587 | write_unlock(&css_set_lock); |
4588 | } | |
4589 | } | |
b4f48b63 PM |
4590 | /** |
4591 | * cgroup_exit - detach cgroup from exiting task | |
4592 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4593 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4594 | * |
4595 | * Description: Detach cgroup from @tsk and release it. | |
4596 | * | |
4597 | * Note that cgroups marked notify_on_release force every task in | |
4598 | * them to take the global cgroup_mutex mutex when exiting. | |
4599 | * This could impact scaling on very large systems. Be reluctant to | |
4600 | * use notify_on_release cgroups where very high task exit scaling | |
4601 | * is required on large systems. | |
4602 | * | |
4603 | * the_top_cgroup_hack: | |
4604 | * | |
4605 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4606 | * | |
4607 | * We call cgroup_exit() while the task is still competent to | |
4608 | * handle notify_on_release(), then leave the task attached to the | |
4609 | * root cgroup in each hierarchy for the remainder of its exit. | |
4610 | * | |
4611 | * To do this properly, we would increment the reference count on | |
4612 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4613 | * code we would add a second cgroup function call, to drop that | |
4614 | * reference. This would just create an unnecessary hot spot on | |
4615 | * the top_cgroup reference count, to no avail. | |
4616 | * | |
4617 | * Normally, holding a reference to a cgroup without bumping its | |
4618 | * count is unsafe. The cgroup could go away, or someone could | |
4619 | * attach us to a different cgroup, decrementing the count on | |
4620 | * the first cgroup that we never incremented. But in this case, | |
4621 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4622 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4623 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4624 | */ |
4625 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4626 | { | |
817929ec | 4627 | struct css_set *cg; |
d41d5a01 | 4628 | int i; |
817929ec PM |
4629 | |
4630 | /* | |
4631 | * Unlink from the css_set task list if necessary. | |
4632 | * Optimistically check cg_list before taking | |
4633 | * css_set_lock | |
4634 | */ | |
4635 | if (!list_empty(&tsk->cg_list)) { | |
4636 | write_lock(&css_set_lock); | |
4637 | if (!list_empty(&tsk->cg_list)) | |
8d258797 | 4638 | list_del_init(&tsk->cg_list); |
817929ec PM |
4639 | write_unlock(&css_set_lock); |
4640 | } | |
4641 | ||
b4f48b63 PM |
4642 | /* Reassign the task to the init_css_set. */ |
4643 | task_lock(tsk); | |
817929ec PM |
4644 | cg = tsk->cgroups; |
4645 | tsk->cgroups = &init_css_set; | |
d41d5a01 PZ |
4646 | |
4647 | if (run_callbacks && need_forkexit_callback) { | |
4648 | /* | |
4649 | * modular subsystems can't use callbacks, so no need to lock | |
4650 | * the subsys array | |
4651 | */ | |
4652 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
4653 | struct cgroup_subsys *ss = subsys[i]; | |
4654 | if (ss->exit) { | |
4655 | struct cgroup *old_cgrp = | |
4656 | rcu_dereference_raw(cg->subsys[i])->cgroup; | |
4657 | struct cgroup *cgrp = task_cgroup(tsk, i); | |
4658 | ss->exit(ss, cgrp, old_cgrp, tsk); | |
4659 | } | |
4660 | } | |
4661 | } | |
b4f48b63 | 4662 | task_unlock(tsk); |
d41d5a01 | 4663 | |
817929ec | 4664 | if (cg) |
81a6a5cd | 4665 | put_css_set_taskexit(cg); |
b4f48b63 | 4666 | } |
697f4161 | 4667 | |
a043e3b2 | 4668 | /** |
313e924c | 4669 | * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp |
a043e3b2 | 4670 | * @cgrp: the cgroup in question |
313e924c | 4671 | * @task: the task in question |
a043e3b2 | 4672 | * |
313e924c GN |
4673 | * See if @cgrp is a descendant of @task's cgroup in the appropriate |
4674 | * hierarchy. | |
697f4161 PM |
4675 | * |
4676 | * If we are sending in dummytop, then presumably we are creating | |
4677 | * the top cgroup in the subsystem. | |
4678 | * | |
4679 | * Called only by the ns (nsproxy) cgroup. | |
4680 | */ | |
313e924c | 4681 | int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task) |
697f4161 PM |
4682 | { |
4683 | int ret; | |
4684 | struct cgroup *target; | |
697f4161 | 4685 | |
bd89aabc | 4686 | if (cgrp == dummytop) |
697f4161 PM |
4687 | return 1; |
4688 | ||
7717f7ba | 4689 | target = task_cgroup_from_root(task, cgrp->root); |
bd89aabc PM |
4690 | while (cgrp != target && cgrp!= cgrp->top_cgroup) |
4691 | cgrp = cgrp->parent; | |
4692 | ret = (cgrp == target); | |
697f4161 PM |
4693 | return ret; |
4694 | } | |
81a6a5cd | 4695 | |
bd89aabc | 4696 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd PM |
4697 | { |
4698 | /* All of these checks rely on RCU to keep the cgroup | |
4699 | * structure alive */ | |
bd89aabc PM |
4700 | if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count) |
4701 | && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) { | |
81a6a5cd PM |
4702 | /* Control Group is currently removeable. If it's not |
4703 | * already queued for a userspace notification, queue | |
4704 | * it now */ | |
4705 | int need_schedule_work = 0; | |
cdcc136f | 4706 | raw_spin_lock(&release_list_lock); |
bd89aabc PM |
4707 | if (!cgroup_is_removed(cgrp) && |
4708 | list_empty(&cgrp->release_list)) { | |
4709 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4710 | need_schedule_work = 1; |
4711 | } | |
cdcc136f | 4712 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4713 | if (need_schedule_work) |
4714 | schedule_work(&release_agent_work); | |
4715 | } | |
4716 | } | |
4717 | ||
d7b9fff7 DN |
4718 | /* Caller must verify that the css is not for root cgroup */ |
4719 | void __css_put(struct cgroup_subsys_state *css, int count) | |
81a6a5cd | 4720 | { |
bd89aabc | 4721 | struct cgroup *cgrp = css->cgroup; |
3dece834 | 4722 | int val; |
81a6a5cd | 4723 | rcu_read_lock(); |
d7b9fff7 | 4724 | val = atomic_sub_return(count, &css->refcnt); |
3dece834 | 4725 | if (val == 1) { |
ec64f515 KH |
4726 | if (notify_on_release(cgrp)) { |
4727 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
4728 | check_for_release(cgrp); | |
4729 | } | |
88703267 | 4730 | cgroup_wakeup_rmdir_waiter(cgrp); |
81a6a5cd PM |
4731 | } |
4732 | rcu_read_unlock(); | |
3dece834 | 4733 | WARN_ON_ONCE(val < 1); |
81a6a5cd | 4734 | } |
67523c48 | 4735 | EXPORT_SYMBOL_GPL(__css_put); |
81a6a5cd PM |
4736 | |
4737 | /* | |
4738 | * Notify userspace when a cgroup is released, by running the | |
4739 | * configured release agent with the name of the cgroup (path | |
4740 | * relative to the root of cgroup file system) as the argument. | |
4741 | * | |
4742 | * Most likely, this user command will try to rmdir this cgroup. | |
4743 | * | |
4744 | * This races with the possibility that some other task will be | |
4745 | * attached to this cgroup before it is removed, or that some other | |
4746 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4747 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4748 | * unused, and this cgroup will be reprieved from its death sentence, | |
4749 | * to continue to serve a useful existence. Next time it's released, | |
4750 | * we will get notified again, if it still has 'notify_on_release' set. | |
4751 | * | |
4752 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4753 | * means only wait until the task is successfully execve()'d. The | |
4754 | * separate release agent task is forked by call_usermodehelper(), | |
4755 | * then control in this thread returns here, without waiting for the | |
4756 | * release agent task. We don't bother to wait because the caller of | |
4757 | * this routine has no use for the exit status of the release agent | |
4758 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4759 | */ |
81a6a5cd PM |
4760 | static void cgroup_release_agent(struct work_struct *work) |
4761 | { | |
4762 | BUG_ON(work != &release_agent_work); | |
4763 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4764 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4765 | while (!list_empty(&release_list)) { |
4766 | char *argv[3], *envp[3]; | |
4767 | int i; | |
e788e066 | 4768 | char *pathbuf = NULL, *agentbuf = NULL; |
bd89aabc | 4769 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4770 | struct cgroup, |
4771 | release_list); | |
bd89aabc | 4772 | list_del_init(&cgrp->release_list); |
cdcc136f | 4773 | raw_spin_unlock(&release_list_lock); |
81a6a5cd | 4774 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
e788e066 PM |
4775 | if (!pathbuf) |
4776 | goto continue_free; | |
4777 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
4778 | goto continue_free; | |
4779 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4780 | if (!agentbuf) | |
4781 | goto continue_free; | |
81a6a5cd PM |
4782 | |
4783 | i = 0; | |
e788e066 PM |
4784 | argv[i++] = agentbuf; |
4785 | argv[i++] = pathbuf; | |
81a6a5cd PM |
4786 | argv[i] = NULL; |
4787 | ||
4788 | i = 0; | |
4789 | /* minimal command environment */ | |
4790 | envp[i++] = "HOME=/"; | |
4791 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4792 | envp[i] = NULL; | |
4793 | ||
4794 | /* Drop the lock while we invoke the usermode helper, | |
4795 | * since the exec could involve hitting disk and hence | |
4796 | * be a slow process */ | |
4797 | mutex_unlock(&cgroup_mutex); | |
4798 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4799 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4800 | continue_free: |
4801 | kfree(pathbuf); | |
4802 | kfree(agentbuf); | |
cdcc136f | 4803 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4804 | } |
cdcc136f | 4805 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4806 | mutex_unlock(&cgroup_mutex); |
4807 | } | |
8bab8dde PM |
4808 | |
4809 | static int __init cgroup_disable(char *str) | |
4810 | { | |
4811 | int i; | |
4812 | char *token; | |
4813 | ||
4814 | while ((token = strsep(&str, ",")) != NULL) { | |
4815 | if (!*token) | |
4816 | continue; | |
aae8aab4 BB |
4817 | /* |
4818 | * cgroup_disable, being at boot time, can't know about module | |
4819 | * subsystems, so we don't worry about them. | |
4820 | */ | |
4821 | for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) { | |
8bab8dde PM |
4822 | struct cgroup_subsys *ss = subsys[i]; |
4823 | ||
4824 | if (!strcmp(token, ss->name)) { | |
4825 | ss->disabled = 1; | |
4826 | printk(KERN_INFO "Disabling %s control group" | |
4827 | " subsystem\n", ss->name); | |
4828 | break; | |
4829 | } | |
4830 | } | |
4831 | } | |
4832 | return 1; | |
4833 | } | |
4834 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 KH |
4835 | |
4836 | /* | |
4837 | * Functons for CSS ID. | |
4838 | */ | |
4839 | ||
4840 | /* | |
4841 | *To get ID other than 0, this should be called when !cgroup_is_removed(). | |
4842 | */ | |
4843 | unsigned short css_id(struct cgroup_subsys_state *css) | |
4844 | { | |
7f0f1546 KH |
4845 | struct css_id *cssid; |
4846 | ||
4847 | /* | |
4848 | * This css_id() can return correct value when somone has refcnt | |
4849 | * on this or this is under rcu_read_lock(). Once css->id is allocated, | |
4850 | * it's unchanged until freed. | |
4851 | */ | |
d8bf4ca9 | 4852 | cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt)); |
38460b48 KH |
4853 | |
4854 | if (cssid) | |
4855 | return cssid->id; | |
4856 | return 0; | |
4857 | } | |
67523c48 | 4858 | EXPORT_SYMBOL_GPL(css_id); |
38460b48 KH |
4859 | |
4860 | unsigned short css_depth(struct cgroup_subsys_state *css) | |
4861 | { | |
7f0f1546 KH |
4862 | struct css_id *cssid; |
4863 | ||
d8bf4ca9 | 4864 | cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt)); |
38460b48 KH |
4865 | |
4866 | if (cssid) | |
4867 | return cssid->depth; | |
4868 | return 0; | |
4869 | } | |
67523c48 | 4870 | EXPORT_SYMBOL_GPL(css_depth); |
38460b48 | 4871 | |
747388d7 KH |
4872 | /** |
4873 | * css_is_ancestor - test "root" css is an ancestor of "child" | |
4874 | * @child: the css to be tested. | |
4875 | * @root: the css supporsed to be an ancestor of the child. | |
4876 | * | |
4877 | * Returns true if "root" is an ancestor of "child" in its hierarchy. Because | |
4878 | * this function reads css->id, this use rcu_dereference() and rcu_read_lock(). | |
4879 | * But, considering usual usage, the csses should be valid objects after test. | |
4880 | * Assuming that the caller will do some action to the child if this returns | |
4881 | * returns true, the caller must take "child";s reference count. | |
4882 | * If "child" is valid object and this returns true, "root" is valid, too. | |
4883 | */ | |
4884 | ||
38460b48 | 4885 | bool css_is_ancestor(struct cgroup_subsys_state *child, |
0b7f569e | 4886 | const struct cgroup_subsys_state *root) |
38460b48 | 4887 | { |
747388d7 KH |
4888 | struct css_id *child_id; |
4889 | struct css_id *root_id; | |
4890 | bool ret = true; | |
38460b48 | 4891 | |
747388d7 KH |
4892 | rcu_read_lock(); |
4893 | child_id = rcu_dereference(child->id); | |
4894 | root_id = rcu_dereference(root->id); | |
4895 | if (!child_id | |
4896 | || !root_id | |
4897 | || (child_id->depth < root_id->depth) | |
4898 | || (child_id->stack[root_id->depth] != root_id->id)) | |
4899 | ret = false; | |
4900 | rcu_read_unlock(); | |
4901 | return ret; | |
38460b48 KH |
4902 | } |
4903 | ||
38460b48 KH |
4904 | void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css) |
4905 | { | |
4906 | struct css_id *id = css->id; | |
4907 | /* When this is called before css_id initialization, id can be NULL */ | |
4908 | if (!id) | |
4909 | return; | |
4910 | ||
4911 | BUG_ON(!ss->use_id); | |
4912 | ||
4913 | rcu_assign_pointer(id->css, NULL); | |
4914 | rcu_assign_pointer(css->id, NULL); | |
c1e2ee2d | 4915 | write_lock(&ss->id_lock); |
38460b48 | 4916 | idr_remove(&ss->idr, id->id); |
c1e2ee2d | 4917 | write_unlock(&ss->id_lock); |
025cea99 | 4918 | kfree_rcu(id, rcu_head); |
38460b48 | 4919 | } |
67523c48 | 4920 | EXPORT_SYMBOL_GPL(free_css_id); |
38460b48 KH |
4921 | |
4922 | /* | |
4923 | * This is called by init or create(). Then, calls to this function are | |
4924 | * always serialized (By cgroup_mutex() at create()). | |
4925 | */ | |
4926 | ||
4927 | static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth) | |
4928 | { | |
4929 | struct css_id *newid; | |
4930 | int myid, error, size; | |
4931 | ||
4932 | BUG_ON(!ss->use_id); | |
4933 | ||
4934 | size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1); | |
4935 | newid = kzalloc(size, GFP_KERNEL); | |
4936 | if (!newid) | |
4937 | return ERR_PTR(-ENOMEM); | |
4938 | /* get id */ | |
4939 | if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) { | |
4940 | error = -ENOMEM; | |
4941 | goto err_out; | |
4942 | } | |
c1e2ee2d | 4943 | write_lock(&ss->id_lock); |
38460b48 KH |
4944 | /* Don't use 0. allocates an ID of 1-65535 */ |
4945 | error = idr_get_new_above(&ss->idr, newid, 1, &myid); | |
c1e2ee2d | 4946 | write_unlock(&ss->id_lock); |
38460b48 KH |
4947 | |
4948 | /* Returns error when there are no free spaces for new ID.*/ | |
4949 | if (error) { | |
4950 | error = -ENOSPC; | |
4951 | goto err_out; | |
4952 | } | |
4953 | if (myid > CSS_ID_MAX) | |
4954 | goto remove_idr; | |
4955 | ||
4956 | newid->id = myid; | |
4957 | newid->depth = depth; | |
4958 | return newid; | |
4959 | remove_idr: | |
4960 | error = -ENOSPC; | |
c1e2ee2d | 4961 | write_lock(&ss->id_lock); |
38460b48 | 4962 | idr_remove(&ss->idr, myid); |
c1e2ee2d | 4963 | write_unlock(&ss->id_lock); |
38460b48 KH |
4964 | err_out: |
4965 | kfree(newid); | |
4966 | return ERR_PTR(error); | |
4967 | ||
4968 | } | |
4969 | ||
e6a1105b BB |
4970 | static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss, |
4971 | struct cgroup_subsys_state *rootcss) | |
38460b48 KH |
4972 | { |
4973 | struct css_id *newid; | |
38460b48 | 4974 | |
c1e2ee2d | 4975 | rwlock_init(&ss->id_lock); |
38460b48 KH |
4976 | idr_init(&ss->idr); |
4977 | ||
38460b48 KH |
4978 | newid = get_new_cssid(ss, 0); |
4979 | if (IS_ERR(newid)) | |
4980 | return PTR_ERR(newid); | |
4981 | ||
4982 | newid->stack[0] = newid->id; | |
4983 | newid->css = rootcss; | |
4984 | rootcss->id = newid; | |
4985 | return 0; | |
4986 | } | |
4987 | ||
4988 | static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent, | |
4989 | struct cgroup *child) | |
4990 | { | |
4991 | int subsys_id, i, depth = 0; | |
4992 | struct cgroup_subsys_state *parent_css, *child_css; | |
fae9c791 | 4993 | struct css_id *child_id, *parent_id; |
38460b48 KH |
4994 | |
4995 | subsys_id = ss->subsys_id; | |
4996 | parent_css = parent->subsys[subsys_id]; | |
4997 | child_css = child->subsys[subsys_id]; | |
38460b48 | 4998 | parent_id = parent_css->id; |
94b3dd0f | 4999 | depth = parent_id->depth + 1; |
38460b48 KH |
5000 | |
5001 | child_id = get_new_cssid(ss, depth); | |
5002 | if (IS_ERR(child_id)) | |
5003 | return PTR_ERR(child_id); | |
5004 | ||
5005 | for (i = 0; i < depth; i++) | |
5006 | child_id->stack[i] = parent_id->stack[i]; | |
5007 | child_id->stack[depth] = child_id->id; | |
5008 | /* | |
5009 | * child_id->css pointer will be set after this cgroup is available | |
5010 | * see cgroup_populate_dir() | |
5011 | */ | |
5012 | rcu_assign_pointer(child_css->id, child_id); | |
5013 | ||
5014 | return 0; | |
5015 | } | |
5016 | ||
5017 | /** | |
5018 | * css_lookup - lookup css by id | |
5019 | * @ss: cgroup subsys to be looked into. | |
5020 | * @id: the id | |
5021 | * | |
5022 | * Returns pointer to cgroup_subsys_state if there is valid one with id. | |
5023 | * NULL if not. Should be called under rcu_read_lock() | |
5024 | */ | |
5025 | struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id) | |
5026 | { | |
5027 | struct css_id *cssid = NULL; | |
5028 | ||
5029 | BUG_ON(!ss->use_id); | |
5030 | cssid = idr_find(&ss->idr, id); | |
5031 | ||
5032 | if (unlikely(!cssid)) | |
5033 | return NULL; | |
5034 | ||
5035 | return rcu_dereference(cssid->css); | |
5036 | } | |
67523c48 | 5037 | EXPORT_SYMBOL_GPL(css_lookup); |
38460b48 KH |
5038 | |
5039 | /** | |
5040 | * css_get_next - lookup next cgroup under specified hierarchy. | |
5041 | * @ss: pointer to subsystem | |
5042 | * @id: current position of iteration. | |
5043 | * @root: pointer to css. search tree under this. | |
5044 | * @foundid: position of found object. | |
5045 | * | |
5046 | * Search next css under the specified hierarchy of rootid. Calling under | |
5047 | * rcu_read_lock() is necessary. Returns NULL if it reaches the end. | |
5048 | */ | |
5049 | struct cgroup_subsys_state * | |
5050 | css_get_next(struct cgroup_subsys *ss, int id, | |
5051 | struct cgroup_subsys_state *root, int *foundid) | |
5052 | { | |
5053 | struct cgroup_subsys_state *ret = NULL; | |
5054 | struct css_id *tmp; | |
5055 | int tmpid; | |
5056 | int rootid = css_id(root); | |
5057 | int depth = css_depth(root); | |
5058 | ||
5059 | if (!rootid) | |
5060 | return NULL; | |
5061 | ||
5062 | BUG_ON(!ss->use_id); | |
5063 | /* fill start point for scan */ | |
5064 | tmpid = id; | |
5065 | while (1) { | |
5066 | /* | |
5067 | * scan next entry from bitmap(tree), tmpid is updated after | |
5068 | * idr_get_next(). | |
5069 | */ | |
c1e2ee2d | 5070 | read_lock(&ss->id_lock); |
38460b48 | 5071 | tmp = idr_get_next(&ss->idr, &tmpid); |
c1e2ee2d | 5072 | read_unlock(&ss->id_lock); |
38460b48 KH |
5073 | |
5074 | if (!tmp) | |
5075 | break; | |
5076 | if (tmp->depth >= depth && tmp->stack[depth] == rootid) { | |
5077 | ret = rcu_dereference(tmp->css); | |
5078 | if (ret) { | |
5079 | *foundid = tmpid; | |
5080 | break; | |
5081 | } | |
5082 | } | |
5083 | /* continue to scan from next id */ | |
5084 | tmpid = tmpid + 1; | |
5085 | } | |
5086 | return ret; | |
5087 | } | |
5088 | ||
e5d1367f SE |
5089 | /* |
5090 | * get corresponding css from file open on cgroupfs directory | |
5091 | */ | |
5092 | struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id) | |
5093 | { | |
5094 | struct cgroup *cgrp; | |
5095 | struct inode *inode; | |
5096 | struct cgroup_subsys_state *css; | |
5097 | ||
5098 | inode = f->f_dentry->d_inode; | |
5099 | /* check in cgroup filesystem dir */ | |
5100 | if (inode->i_op != &cgroup_dir_inode_operations) | |
5101 | return ERR_PTR(-EBADF); | |
5102 | ||
5103 | if (id < 0 || id >= CGROUP_SUBSYS_COUNT) | |
5104 | return ERR_PTR(-EINVAL); | |
5105 | ||
5106 | /* get cgroup */ | |
5107 | cgrp = __d_cgrp(f->f_dentry); | |
5108 | css = cgrp->subsys[id]; | |
5109 | return css ? css : ERR_PTR(-ENOENT); | |
5110 | } | |
5111 | ||
fe693435 PM |
5112 | #ifdef CONFIG_CGROUP_DEBUG |
5113 | static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss, | |
5114 | struct cgroup *cont) | |
5115 | { | |
5116 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5117 | ||
5118 | if (!css) | |
5119 | return ERR_PTR(-ENOMEM); | |
5120 | ||
5121 | return css; | |
5122 | } | |
5123 | ||
5124 | static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
5125 | { | |
5126 | kfree(cont->subsys[debug_subsys_id]); | |
5127 | } | |
5128 | ||
5129 | static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft) | |
5130 | { | |
5131 | return atomic_read(&cont->count); | |
5132 | } | |
5133 | ||
5134 | static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft) | |
5135 | { | |
5136 | return cgroup_task_count(cont); | |
5137 | } | |
5138 | ||
5139 | static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft) | |
5140 | { | |
5141 | return (u64)(unsigned long)current->cgroups; | |
5142 | } | |
5143 | ||
5144 | static u64 current_css_set_refcount_read(struct cgroup *cont, | |
5145 | struct cftype *cft) | |
5146 | { | |
5147 | u64 count; | |
5148 | ||
5149 | rcu_read_lock(); | |
5150 | count = atomic_read(¤t->cgroups->refcount); | |
5151 | rcu_read_unlock(); | |
5152 | return count; | |
5153 | } | |
5154 | ||
7717f7ba PM |
5155 | static int current_css_set_cg_links_read(struct cgroup *cont, |
5156 | struct cftype *cft, | |
5157 | struct seq_file *seq) | |
5158 | { | |
5159 | struct cg_cgroup_link *link; | |
5160 | struct css_set *cg; | |
5161 | ||
5162 | read_lock(&css_set_lock); | |
5163 | rcu_read_lock(); | |
5164 | cg = rcu_dereference(current->cgroups); | |
5165 | list_for_each_entry(link, &cg->cg_links, cg_link_list) { | |
5166 | struct cgroup *c = link->cgrp; | |
5167 | const char *name; | |
5168 | ||
5169 | if (c->dentry) | |
5170 | name = c->dentry->d_name.name; | |
5171 | else | |
5172 | name = "?"; | |
2c6ab6d2 PM |
5173 | seq_printf(seq, "Root %d group %s\n", |
5174 | c->root->hierarchy_id, name); | |
7717f7ba PM |
5175 | } |
5176 | rcu_read_unlock(); | |
5177 | read_unlock(&css_set_lock); | |
5178 | return 0; | |
5179 | } | |
5180 | ||
5181 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
5182 | static int cgroup_css_links_read(struct cgroup *cont, | |
5183 | struct cftype *cft, | |
5184 | struct seq_file *seq) | |
5185 | { | |
5186 | struct cg_cgroup_link *link; | |
5187 | ||
5188 | read_lock(&css_set_lock); | |
5189 | list_for_each_entry(link, &cont->css_sets, cgrp_link_list) { | |
5190 | struct css_set *cg = link->cg; | |
5191 | struct task_struct *task; | |
5192 | int count = 0; | |
5193 | seq_printf(seq, "css_set %p\n", cg); | |
5194 | list_for_each_entry(task, &cg->tasks, cg_list) { | |
5195 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { | |
5196 | seq_puts(seq, " ...\n"); | |
5197 | break; | |
5198 | } else { | |
5199 | seq_printf(seq, " task %d\n", | |
5200 | task_pid_vnr(task)); | |
5201 | } | |
5202 | } | |
5203 | } | |
5204 | read_unlock(&css_set_lock); | |
5205 | return 0; | |
5206 | } | |
5207 | ||
fe693435 PM |
5208 | static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft) |
5209 | { | |
5210 | return test_bit(CGRP_RELEASABLE, &cgrp->flags); | |
5211 | } | |
5212 | ||
5213 | static struct cftype debug_files[] = { | |
5214 | { | |
5215 | .name = "cgroup_refcount", | |
5216 | .read_u64 = cgroup_refcount_read, | |
5217 | }, | |
5218 | { | |
5219 | .name = "taskcount", | |
5220 | .read_u64 = debug_taskcount_read, | |
5221 | }, | |
5222 | ||
5223 | { | |
5224 | .name = "current_css_set", | |
5225 | .read_u64 = current_css_set_read, | |
5226 | }, | |
5227 | ||
5228 | { | |
5229 | .name = "current_css_set_refcount", | |
5230 | .read_u64 = current_css_set_refcount_read, | |
5231 | }, | |
5232 | ||
7717f7ba PM |
5233 | { |
5234 | .name = "current_css_set_cg_links", | |
5235 | .read_seq_string = current_css_set_cg_links_read, | |
5236 | }, | |
5237 | ||
5238 | { | |
5239 | .name = "cgroup_css_links", | |
5240 | .read_seq_string = cgroup_css_links_read, | |
5241 | }, | |
5242 | ||
fe693435 PM |
5243 | { |
5244 | .name = "releasable", | |
5245 | .read_u64 = releasable_read, | |
5246 | }, | |
5247 | }; | |
5248 | ||
5249 | static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
5250 | { | |
5251 | return cgroup_add_files(cont, ss, debug_files, | |
5252 | ARRAY_SIZE(debug_files)); | |
5253 | } | |
5254 | ||
5255 | struct cgroup_subsys debug_subsys = { | |
5256 | .name = "debug", | |
5257 | .create = debug_create, | |
5258 | .destroy = debug_destroy, | |
5259 | .populate = debug_populate, | |
5260 | .subsys_id = debug_subsys_id, | |
5261 | }; | |
5262 | #endif /* CONFIG_CGROUP_DEBUG */ |