]>
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 */ |
c4c27fbd | 63 | #include <linux/kthread.h> |
846c7bb0 | 64 | |
60063497 | 65 | #include <linux/atomic.h> |
ddbcc7e8 | 66 | |
28b4c27b TH |
67 | /* css deactivation bias, makes css->refcnt negative to deny new trygets */ |
68 | #define CSS_DEACT_BIAS INT_MIN | |
69 | ||
e25e2cbb TH |
70 | /* |
71 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
72 | * hierarchy must be performed while holding it. | |
73 | * | |
74 | * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify | |
75 | * cgroupfs_root of any cgroup hierarchy - subsys list, flags, | |
76 | * release_agent_path and so on. Modifying requires both cgroup_mutex and | |
77 | * cgroup_root_mutex. Readers can acquire either of the two. This is to | |
78 | * break the following locking order cycle. | |
79 | * | |
80 | * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem | |
81 | * B. namespace_sem -> cgroup_mutex | |
82 | * | |
83 | * B happens only through cgroup_show_options() and using cgroup_root_mutex | |
84 | * breaks it. | |
85 | */ | |
81a6a5cd | 86 | static DEFINE_MUTEX(cgroup_mutex); |
e25e2cbb | 87 | static DEFINE_MUTEX(cgroup_root_mutex); |
81a6a5cd | 88 | |
aae8aab4 BB |
89 | /* |
90 | * Generate an array of cgroup subsystem pointers. At boot time, this is | |
be45c900 | 91 | * populated with the built in subsystems, and modular subsystems are |
aae8aab4 BB |
92 | * registered after that. The mutable section of this array is protected by |
93 | * cgroup_mutex. | |
94 | */ | |
80f4c877 | 95 | #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys, |
5fc0b025 | 96 | #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option) |
aae8aab4 | 97 | static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = { |
ddbcc7e8 PM |
98 | #include <linux/cgroup_subsys.h> |
99 | }; | |
100 | ||
c6d57f33 PM |
101 | #define MAX_CGROUP_ROOT_NAMELEN 64 |
102 | ||
ddbcc7e8 PM |
103 | /* |
104 | * A cgroupfs_root represents the root of a cgroup hierarchy, | |
105 | * and may be associated with a superblock to form an active | |
106 | * hierarchy | |
107 | */ | |
108 | struct cgroupfs_root { | |
109 | struct super_block *sb; | |
110 | ||
111 | /* | |
112 | * The bitmask of subsystems intended to be attached to this | |
113 | * hierarchy | |
114 | */ | |
a1a71b45 | 115 | unsigned long subsys_mask; |
ddbcc7e8 | 116 | |
2c6ab6d2 PM |
117 | /* Unique id for this hierarchy. */ |
118 | int hierarchy_id; | |
119 | ||
ddbcc7e8 | 120 | /* The bitmask of subsystems currently attached to this hierarchy */ |
a1a71b45 | 121 | unsigned long actual_subsys_mask; |
ddbcc7e8 PM |
122 | |
123 | /* A list running through the attached subsystems */ | |
124 | struct list_head subsys_list; | |
125 | ||
126 | /* The root cgroup for this hierarchy */ | |
127 | struct cgroup top_cgroup; | |
128 | ||
129 | /* Tracks how many cgroups are currently defined in hierarchy.*/ | |
130 | int number_of_cgroups; | |
131 | ||
e5f6a860 | 132 | /* A list running through the active hierarchies */ |
ddbcc7e8 PM |
133 | struct list_head root_list; |
134 | ||
b0ca5a84 TH |
135 | /* All cgroups on this root, cgroup_mutex protected */ |
136 | struct list_head allcg_list; | |
137 | ||
ddbcc7e8 PM |
138 | /* Hierarchy-specific flags */ |
139 | unsigned long flags; | |
81a6a5cd | 140 | |
e788e066 | 141 | /* The path to use for release notifications. */ |
81a6a5cd | 142 | char release_agent_path[PATH_MAX]; |
c6d57f33 PM |
143 | |
144 | /* The name for this hierarchy - may be empty */ | |
145 | char name[MAX_CGROUP_ROOT_NAMELEN]; | |
ddbcc7e8 PM |
146 | }; |
147 | ||
ddbcc7e8 PM |
148 | /* |
149 | * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the | |
150 | * subsystems that are otherwise unattached - it never has more than a | |
151 | * single cgroup, and all tasks are part of that cgroup. | |
152 | */ | |
153 | static struct cgroupfs_root rootnode; | |
154 | ||
05ef1d7c TH |
155 | /* |
156 | * cgroupfs file entry, pointed to from leaf dentry->d_fsdata. | |
157 | */ | |
158 | struct cfent { | |
159 | struct list_head node; | |
160 | struct dentry *dentry; | |
161 | struct cftype *type; | |
162 | }; | |
163 | ||
38460b48 KH |
164 | /* |
165 | * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when | |
166 | * cgroup_subsys->use_id != 0. | |
167 | */ | |
168 | #define CSS_ID_MAX (65535) | |
169 | struct css_id { | |
170 | /* | |
171 | * The css to which this ID points. This pointer is set to valid value | |
172 | * after cgroup is populated. If cgroup is removed, this will be NULL. | |
173 | * This pointer is expected to be RCU-safe because destroy() | |
e9316080 TH |
174 | * is called after synchronize_rcu(). But for safe use, css_tryget() |
175 | * should be used for avoiding race. | |
38460b48 | 176 | */ |
2c392b8c | 177 | struct cgroup_subsys_state __rcu *css; |
38460b48 KH |
178 | /* |
179 | * ID of this css. | |
180 | */ | |
181 | unsigned short id; | |
182 | /* | |
183 | * Depth in hierarchy which this ID belongs to. | |
184 | */ | |
185 | unsigned short depth; | |
186 | /* | |
187 | * ID is freed by RCU. (and lookup routine is RCU safe.) | |
188 | */ | |
189 | struct rcu_head rcu_head; | |
190 | /* | |
191 | * Hierarchy of CSS ID belongs to. | |
192 | */ | |
193 | unsigned short stack[0]; /* Array of Length (depth+1) */ | |
194 | }; | |
195 | ||
0dea1168 | 196 | /* |
25985edc | 197 | * cgroup_event represents events which userspace want to receive. |
0dea1168 KS |
198 | */ |
199 | struct cgroup_event { | |
200 | /* | |
201 | * Cgroup which the event belongs to. | |
202 | */ | |
203 | struct cgroup *cgrp; | |
204 | /* | |
205 | * Control file which the event associated. | |
206 | */ | |
207 | struct cftype *cft; | |
208 | /* | |
209 | * eventfd to signal userspace about the event. | |
210 | */ | |
211 | struct eventfd_ctx *eventfd; | |
212 | /* | |
213 | * Each of these stored in a list by the cgroup. | |
214 | */ | |
215 | struct list_head list; | |
216 | /* | |
217 | * All fields below needed to unregister event when | |
218 | * userspace closes eventfd. | |
219 | */ | |
220 | poll_table pt; | |
221 | wait_queue_head_t *wqh; | |
222 | wait_queue_t wait; | |
223 | struct work_struct remove; | |
224 | }; | |
38460b48 | 225 | |
ddbcc7e8 PM |
226 | /* The list of hierarchy roots */ |
227 | ||
228 | static LIST_HEAD(roots); | |
817929ec | 229 | static int root_count; |
ddbcc7e8 | 230 | |
2c6ab6d2 PM |
231 | static DEFINE_IDA(hierarchy_ida); |
232 | static int next_hierarchy_id; | |
233 | static DEFINE_SPINLOCK(hierarchy_id_lock); | |
234 | ||
ddbcc7e8 PM |
235 | /* dummytop is a shorthand for the dummy hierarchy's top cgroup */ |
236 | #define dummytop (&rootnode.top_cgroup) | |
237 | ||
238 | /* This flag indicates whether tasks in the fork and exit paths should | |
a043e3b2 LZ |
239 | * check for fork/exit handlers to call. This avoids us having to do |
240 | * extra work in the fork/exit path if none of the subsystems need to | |
241 | * be called. | |
ddbcc7e8 | 242 | */ |
8947f9d5 | 243 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 244 | |
42809dd4 TH |
245 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
246 | ||
d11c563d PM |
247 | #ifdef CONFIG_PROVE_LOCKING |
248 | int cgroup_lock_is_held(void) | |
249 | { | |
250 | return lockdep_is_held(&cgroup_mutex); | |
251 | } | |
252 | #else /* #ifdef CONFIG_PROVE_LOCKING */ | |
253 | int cgroup_lock_is_held(void) | |
254 | { | |
255 | return mutex_is_locked(&cgroup_mutex); | |
256 | } | |
257 | #endif /* #else #ifdef CONFIG_PROVE_LOCKING */ | |
258 | ||
259 | EXPORT_SYMBOL_GPL(cgroup_lock_is_held); | |
260 | ||
8e3bbf42 SQ |
261 | static int css_unbias_refcnt(int refcnt) |
262 | { | |
263 | return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS; | |
264 | } | |
265 | ||
28b4c27b TH |
266 | /* the current nr of refs, always >= 0 whether @css is deactivated or not */ |
267 | static int css_refcnt(struct cgroup_subsys_state *css) | |
268 | { | |
269 | int v = atomic_read(&css->refcnt); | |
270 | ||
8e3bbf42 | 271 | return css_unbias_refcnt(v); |
28b4c27b TH |
272 | } |
273 | ||
ddbcc7e8 | 274 | /* convenient tests for these bits */ |
bd89aabc | 275 | inline int cgroup_is_removed(const struct cgroup *cgrp) |
ddbcc7e8 | 276 | { |
bd89aabc | 277 | return test_bit(CGRP_REMOVED, &cgrp->flags); |
ddbcc7e8 PM |
278 | } |
279 | ||
280 | /* bits in struct cgroupfs_root flags field */ | |
281 | enum { | |
03b1cde6 AR |
282 | ROOT_NOPREFIX, /* mounted subsystems have no named prefix */ |
283 | ROOT_XATTR, /* supports extended attributes */ | |
ddbcc7e8 PM |
284 | }; |
285 | ||
e9685a03 | 286 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
287 | { |
288 | const int bits = | |
bd89aabc PM |
289 | (1 << CGRP_RELEASABLE) | |
290 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
291 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
292 | } |
293 | ||
e9685a03 | 294 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 295 | { |
bd89aabc | 296 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
297 | } |
298 | ||
97978e6d DL |
299 | static int clone_children(const struct cgroup *cgrp) |
300 | { | |
301 | return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); | |
302 | } | |
303 | ||
ddbcc7e8 PM |
304 | /* |
305 | * for_each_subsys() allows you to iterate on each subsystem attached to | |
306 | * an active hierarchy | |
307 | */ | |
308 | #define for_each_subsys(_root, _ss) \ | |
309 | list_for_each_entry(_ss, &_root->subsys_list, sibling) | |
310 | ||
e5f6a860 LZ |
311 | /* for_each_active_root() allows you to iterate across the active hierarchies */ |
312 | #define for_each_active_root(_root) \ | |
ddbcc7e8 PM |
313 | list_for_each_entry(_root, &roots, root_list) |
314 | ||
f6ea9372 TH |
315 | static inline struct cgroup *__d_cgrp(struct dentry *dentry) |
316 | { | |
317 | return dentry->d_fsdata; | |
318 | } | |
319 | ||
05ef1d7c | 320 | static inline struct cfent *__d_cfe(struct dentry *dentry) |
f6ea9372 TH |
321 | { |
322 | return dentry->d_fsdata; | |
323 | } | |
324 | ||
05ef1d7c TH |
325 | static inline struct cftype *__d_cft(struct dentry *dentry) |
326 | { | |
327 | return __d_cfe(dentry)->type; | |
328 | } | |
329 | ||
81a6a5cd PM |
330 | /* the list of cgroups eligible for automatic release. Protected by |
331 | * release_list_lock */ | |
332 | static LIST_HEAD(release_list); | |
cdcc136f | 333 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
334 | static void cgroup_release_agent(struct work_struct *work); |
335 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 336 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 337 | |
817929ec PM |
338 | /* Link structure for associating css_set objects with cgroups */ |
339 | struct cg_cgroup_link { | |
340 | /* | |
341 | * List running through cg_cgroup_links associated with a | |
342 | * cgroup, anchored on cgroup->css_sets | |
343 | */ | |
bd89aabc | 344 | struct list_head cgrp_link_list; |
7717f7ba | 345 | struct cgroup *cgrp; |
817929ec PM |
346 | /* |
347 | * List running through cg_cgroup_links pointing at a | |
348 | * single css_set object, anchored on css_set->cg_links | |
349 | */ | |
350 | struct list_head cg_link_list; | |
351 | struct css_set *cg; | |
352 | }; | |
353 | ||
354 | /* The default css_set - used by init and its children prior to any | |
355 | * hierarchies being mounted. It contains a pointer to the root state | |
356 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
357 | * reference-counted, to improve performance when child cgroups | |
358 | * haven't been created. | |
359 | */ | |
360 | ||
361 | static struct css_set init_css_set; | |
362 | static struct cg_cgroup_link init_css_set_link; | |
363 | ||
e6a1105b BB |
364 | static int cgroup_init_idr(struct cgroup_subsys *ss, |
365 | struct cgroup_subsys_state *css); | |
38460b48 | 366 | |
817929ec PM |
367 | /* css_set_lock protects the list of css_set objects, and the |
368 | * chain of tasks off each css_set. Nests outside task->alloc_lock | |
369 | * due to cgroup_iter_start() */ | |
370 | static DEFINE_RWLOCK(css_set_lock); | |
371 | static int css_set_count; | |
372 | ||
7717f7ba PM |
373 | /* |
374 | * hash table for cgroup groups. This improves the performance to find | |
375 | * an existing css_set. This hash doesn't (currently) take into | |
376 | * account cgroups in empty hierarchies. | |
377 | */ | |
472b1053 LZ |
378 | #define CSS_SET_HASH_BITS 7 |
379 | #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS) | |
380 | static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE]; | |
381 | ||
382 | static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[]) | |
383 | { | |
384 | int i; | |
385 | int index; | |
386 | unsigned long tmp = 0UL; | |
387 | ||
388 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) | |
389 | tmp += (unsigned long)css[i]; | |
390 | tmp = (tmp >> 16) ^ tmp; | |
391 | ||
392 | index = hash_long(tmp, CSS_SET_HASH_BITS); | |
393 | ||
394 | return &css_set_table[index]; | |
395 | } | |
396 | ||
817929ec PM |
397 | /* We don't maintain the lists running through each css_set to its |
398 | * task until after the first call to cgroup_iter_start(). This | |
399 | * reduces the fork()/exit() overhead for people who have cgroups | |
400 | * compiled into their kernel but not actually in use */ | |
8947f9d5 | 401 | static int use_task_css_set_links __read_mostly; |
817929ec | 402 | |
2c6ab6d2 | 403 | static void __put_css_set(struct css_set *cg, int taskexit) |
b4f48b63 | 404 | { |
71cbb949 KM |
405 | struct cg_cgroup_link *link; |
406 | struct cg_cgroup_link *saved_link; | |
146aa1bd LJ |
407 | /* |
408 | * Ensure that the refcount doesn't hit zero while any readers | |
409 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
410 | * rwlock | |
411 | */ | |
412 | if (atomic_add_unless(&cg->refcount, -1, 1)) | |
413 | return; | |
414 | write_lock(&css_set_lock); | |
415 | if (!atomic_dec_and_test(&cg->refcount)) { | |
416 | write_unlock(&css_set_lock); | |
417 | return; | |
418 | } | |
81a6a5cd | 419 | |
2c6ab6d2 PM |
420 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
421 | hlist_del(&cg->hlist); | |
422 | css_set_count--; | |
423 | ||
424 | list_for_each_entry_safe(link, saved_link, &cg->cg_links, | |
425 | cg_link_list) { | |
426 | struct cgroup *cgrp = link->cgrp; | |
427 | list_del(&link->cg_link_list); | |
428 | list_del(&link->cgrp_link_list); | |
bd89aabc PM |
429 | if (atomic_dec_and_test(&cgrp->count) && |
430 | notify_on_release(cgrp)) { | |
81a6a5cd | 431 | if (taskexit) |
bd89aabc PM |
432 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
433 | check_for_release(cgrp); | |
81a6a5cd | 434 | } |
2c6ab6d2 PM |
435 | |
436 | kfree(link); | |
81a6a5cd | 437 | } |
2c6ab6d2 PM |
438 | |
439 | write_unlock(&css_set_lock); | |
30088ad8 | 440 | kfree_rcu(cg, rcu_head); |
b4f48b63 PM |
441 | } |
442 | ||
817929ec PM |
443 | /* |
444 | * refcounted get/put for css_set objects | |
445 | */ | |
446 | static inline void get_css_set(struct css_set *cg) | |
447 | { | |
146aa1bd | 448 | atomic_inc(&cg->refcount); |
817929ec PM |
449 | } |
450 | ||
451 | static inline void put_css_set(struct css_set *cg) | |
452 | { | |
146aa1bd | 453 | __put_css_set(cg, 0); |
817929ec PM |
454 | } |
455 | ||
81a6a5cd PM |
456 | static inline void put_css_set_taskexit(struct css_set *cg) |
457 | { | |
146aa1bd | 458 | __put_css_set(cg, 1); |
81a6a5cd PM |
459 | } |
460 | ||
7717f7ba PM |
461 | /* |
462 | * compare_css_sets - helper function for find_existing_css_set(). | |
463 | * @cg: candidate css_set being tested | |
464 | * @old_cg: existing css_set for a task | |
465 | * @new_cgrp: cgroup that's being entered by the task | |
466 | * @template: desired set of css pointers in css_set (pre-calculated) | |
467 | * | |
468 | * Returns true if "cg" matches "old_cg" except for the hierarchy | |
469 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". | |
470 | */ | |
471 | static bool compare_css_sets(struct css_set *cg, | |
472 | struct css_set *old_cg, | |
473 | struct cgroup *new_cgrp, | |
474 | struct cgroup_subsys_state *template[]) | |
475 | { | |
476 | struct list_head *l1, *l2; | |
477 | ||
478 | if (memcmp(template, cg->subsys, sizeof(cg->subsys))) { | |
479 | /* Not all subsystems matched */ | |
480 | return false; | |
481 | } | |
482 | ||
483 | /* | |
484 | * Compare cgroup pointers in order to distinguish between | |
485 | * different cgroups in heirarchies with no subsystems. We | |
486 | * could get by with just this check alone (and skip the | |
487 | * memcmp above) but on most setups the memcmp check will | |
488 | * avoid the need for this more expensive check on almost all | |
489 | * candidates. | |
490 | */ | |
491 | ||
492 | l1 = &cg->cg_links; | |
493 | l2 = &old_cg->cg_links; | |
494 | while (1) { | |
495 | struct cg_cgroup_link *cgl1, *cgl2; | |
496 | struct cgroup *cg1, *cg2; | |
497 | ||
498 | l1 = l1->next; | |
499 | l2 = l2->next; | |
500 | /* See if we reached the end - both lists are equal length. */ | |
501 | if (l1 == &cg->cg_links) { | |
502 | BUG_ON(l2 != &old_cg->cg_links); | |
503 | break; | |
504 | } else { | |
505 | BUG_ON(l2 == &old_cg->cg_links); | |
506 | } | |
507 | /* Locate the cgroups associated with these links. */ | |
508 | cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list); | |
509 | cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list); | |
510 | cg1 = cgl1->cgrp; | |
511 | cg2 = cgl2->cgrp; | |
512 | /* Hierarchies should be linked in the same order. */ | |
513 | BUG_ON(cg1->root != cg2->root); | |
514 | ||
515 | /* | |
516 | * If this hierarchy is the hierarchy of the cgroup | |
517 | * that's changing, then we need to check that this | |
518 | * css_set points to the new cgroup; if it's any other | |
519 | * hierarchy, then this css_set should point to the | |
520 | * same cgroup as the old css_set. | |
521 | */ | |
522 | if (cg1->root == new_cgrp->root) { | |
523 | if (cg1 != new_cgrp) | |
524 | return false; | |
525 | } else { | |
526 | if (cg1 != cg2) | |
527 | return false; | |
528 | } | |
529 | } | |
530 | return true; | |
531 | } | |
532 | ||
817929ec PM |
533 | /* |
534 | * find_existing_css_set() is a helper for | |
535 | * find_css_set(), and checks to see whether an existing | |
472b1053 | 536 | * css_set is suitable. |
817929ec PM |
537 | * |
538 | * oldcg: the cgroup group that we're using before the cgroup | |
539 | * transition | |
540 | * | |
bd89aabc | 541 | * cgrp: the cgroup that we're moving into |
817929ec PM |
542 | * |
543 | * template: location in which to build the desired set of subsystem | |
544 | * state objects for the new cgroup group | |
545 | */ | |
817929ec PM |
546 | static struct css_set *find_existing_css_set( |
547 | struct css_set *oldcg, | |
bd89aabc | 548 | struct cgroup *cgrp, |
817929ec | 549 | struct cgroup_subsys_state *template[]) |
b4f48b63 PM |
550 | { |
551 | int i; | |
bd89aabc | 552 | struct cgroupfs_root *root = cgrp->root; |
472b1053 LZ |
553 | struct hlist_head *hhead; |
554 | struct hlist_node *node; | |
555 | struct css_set *cg; | |
817929ec | 556 | |
aae8aab4 BB |
557 | /* |
558 | * Build the set of subsystem state objects that we want to see in the | |
559 | * new css_set. while subsystems can change globally, the entries here | |
560 | * won't change, so no need for locking. | |
561 | */ | |
817929ec | 562 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
a1a71b45 | 563 | if (root->subsys_mask & (1UL << i)) { |
817929ec PM |
564 | /* Subsystem is in this hierarchy. So we want |
565 | * the subsystem state from the new | |
566 | * cgroup */ | |
bd89aabc | 567 | template[i] = cgrp->subsys[i]; |
817929ec PM |
568 | } else { |
569 | /* Subsystem is not in this hierarchy, so we | |
570 | * don't want to change the subsystem state */ | |
571 | template[i] = oldcg->subsys[i]; | |
572 | } | |
573 | } | |
574 | ||
472b1053 LZ |
575 | hhead = css_set_hash(template); |
576 | hlist_for_each_entry(cg, node, hhead, hlist) { | |
7717f7ba PM |
577 | if (!compare_css_sets(cg, oldcg, cgrp, template)) |
578 | continue; | |
579 | ||
580 | /* This css_set matches what we need */ | |
581 | return cg; | |
472b1053 | 582 | } |
817929ec PM |
583 | |
584 | /* No existing cgroup group matched */ | |
585 | return NULL; | |
586 | } | |
587 | ||
36553434 LZ |
588 | static void free_cg_links(struct list_head *tmp) |
589 | { | |
590 | struct cg_cgroup_link *link; | |
591 | struct cg_cgroup_link *saved_link; | |
592 | ||
593 | list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) { | |
594 | list_del(&link->cgrp_link_list); | |
595 | kfree(link); | |
596 | } | |
597 | } | |
598 | ||
817929ec PM |
599 | /* |
600 | * allocate_cg_links() allocates "count" cg_cgroup_link structures | |
bd89aabc | 601 | * and chains them on tmp through their cgrp_link_list fields. Returns 0 on |
817929ec PM |
602 | * success or a negative error |
603 | */ | |
817929ec PM |
604 | static int allocate_cg_links(int count, struct list_head *tmp) |
605 | { | |
606 | struct cg_cgroup_link *link; | |
607 | int i; | |
608 | INIT_LIST_HEAD(tmp); | |
609 | for (i = 0; i < count; i++) { | |
610 | link = kmalloc(sizeof(*link), GFP_KERNEL); | |
611 | if (!link) { | |
36553434 | 612 | free_cg_links(tmp); |
817929ec PM |
613 | return -ENOMEM; |
614 | } | |
bd89aabc | 615 | list_add(&link->cgrp_link_list, tmp); |
817929ec PM |
616 | } |
617 | return 0; | |
618 | } | |
619 | ||
c12f65d4 LZ |
620 | /** |
621 | * link_css_set - a helper function to link a css_set to a cgroup | |
622 | * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links() | |
623 | * @cg: the css_set to be linked | |
624 | * @cgrp: the destination cgroup | |
625 | */ | |
626 | static void link_css_set(struct list_head *tmp_cg_links, | |
627 | struct css_set *cg, struct cgroup *cgrp) | |
628 | { | |
629 | struct cg_cgroup_link *link; | |
630 | ||
631 | BUG_ON(list_empty(tmp_cg_links)); | |
632 | link = list_first_entry(tmp_cg_links, struct cg_cgroup_link, | |
633 | cgrp_link_list); | |
634 | link->cg = cg; | |
7717f7ba | 635 | link->cgrp = cgrp; |
2c6ab6d2 | 636 | atomic_inc(&cgrp->count); |
c12f65d4 | 637 | list_move(&link->cgrp_link_list, &cgrp->css_sets); |
7717f7ba PM |
638 | /* |
639 | * Always add links to the tail of the list so that the list | |
640 | * is sorted by order of hierarchy creation | |
641 | */ | |
642 | list_add_tail(&link->cg_link_list, &cg->cg_links); | |
c12f65d4 LZ |
643 | } |
644 | ||
817929ec PM |
645 | /* |
646 | * find_css_set() takes an existing cgroup group and a | |
647 | * cgroup object, and returns a css_set object that's | |
648 | * equivalent to the old group, but with the given cgroup | |
649 | * substituted into the appropriate hierarchy. Must be called with | |
650 | * cgroup_mutex held | |
651 | */ | |
817929ec | 652 | static struct css_set *find_css_set( |
bd89aabc | 653 | struct css_set *oldcg, struct cgroup *cgrp) |
817929ec PM |
654 | { |
655 | struct css_set *res; | |
656 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT]; | |
817929ec PM |
657 | |
658 | struct list_head tmp_cg_links; | |
817929ec | 659 | |
472b1053 | 660 | struct hlist_head *hhead; |
7717f7ba | 661 | struct cg_cgroup_link *link; |
472b1053 | 662 | |
817929ec PM |
663 | /* First see if we already have a cgroup group that matches |
664 | * the desired set */ | |
7e9abd89 | 665 | read_lock(&css_set_lock); |
bd89aabc | 666 | res = find_existing_css_set(oldcg, cgrp, template); |
817929ec PM |
667 | if (res) |
668 | get_css_set(res); | |
7e9abd89 | 669 | read_unlock(&css_set_lock); |
817929ec PM |
670 | |
671 | if (res) | |
672 | return res; | |
673 | ||
674 | res = kmalloc(sizeof(*res), GFP_KERNEL); | |
675 | if (!res) | |
676 | return NULL; | |
677 | ||
678 | /* Allocate all the cg_cgroup_link objects that we'll need */ | |
679 | if (allocate_cg_links(root_count, &tmp_cg_links) < 0) { | |
680 | kfree(res); | |
681 | return NULL; | |
682 | } | |
683 | ||
146aa1bd | 684 | atomic_set(&res->refcount, 1); |
817929ec PM |
685 | INIT_LIST_HEAD(&res->cg_links); |
686 | INIT_LIST_HEAD(&res->tasks); | |
472b1053 | 687 | INIT_HLIST_NODE(&res->hlist); |
817929ec PM |
688 | |
689 | /* Copy the set of subsystem state objects generated in | |
690 | * find_existing_css_set() */ | |
691 | memcpy(res->subsys, template, sizeof(res->subsys)); | |
692 | ||
693 | write_lock(&css_set_lock); | |
694 | /* Add reference counts and links from the new css_set. */ | |
7717f7ba PM |
695 | list_for_each_entry(link, &oldcg->cg_links, cg_link_list) { |
696 | struct cgroup *c = link->cgrp; | |
697 | if (c->root == cgrp->root) | |
698 | c = cgrp; | |
699 | link_css_set(&tmp_cg_links, res, c); | |
700 | } | |
817929ec PM |
701 | |
702 | BUG_ON(!list_empty(&tmp_cg_links)); | |
703 | ||
817929ec | 704 | css_set_count++; |
472b1053 LZ |
705 | |
706 | /* Add this cgroup group to the hash table */ | |
707 | hhead = css_set_hash(res->subsys); | |
708 | hlist_add_head(&res->hlist, hhead); | |
709 | ||
817929ec PM |
710 | write_unlock(&css_set_lock); |
711 | ||
712 | return res; | |
b4f48b63 PM |
713 | } |
714 | ||
7717f7ba PM |
715 | /* |
716 | * Return the cgroup for "task" from the given hierarchy. Must be | |
717 | * called with cgroup_mutex held. | |
718 | */ | |
719 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
720 | struct cgroupfs_root *root) | |
721 | { | |
722 | struct css_set *css; | |
723 | struct cgroup *res = NULL; | |
724 | ||
725 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
726 | read_lock(&css_set_lock); | |
727 | /* | |
728 | * No need to lock the task - since we hold cgroup_mutex the | |
729 | * task can't change groups, so the only thing that can happen | |
730 | * is that it exits and its css is set back to init_css_set. | |
731 | */ | |
732 | css = task->cgroups; | |
733 | if (css == &init_css_set) { | |
734 | res = &root->top_cgroup; | |
735 | } else { | |
736 | struct cg_cgroup_link *link; | |
737 | list_for_each_entry(link, &css->cg_links, cg_link_list) { | |
738 | struct cgroup *c = link->cgrp; | |
739 | if (c->root == root) { | |
740 | res = c; | |
741 | break; | |
742 | } | |
743 | } | |
744 | } | |
745 | read_unlock(&css_set_lock); | |
746 | BUG_ON(!res); | |
747 | return res; | |
748 | } | |
749 | ||
ddbcc7e8 PM |
750 | /* |
751 | * There is one global cgroup mutex. We also require taking | |
752 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
753 | * See "The task_lock() exception", at the end of this comment. | |
754 | * | |
755 | * A task must hold cgroup_mutex to modify cgroups. | |
756 | * | |
757 | * Any task can increment and decrement the count field without lock. | |
758 | * So in general, code holding cgroup_mutex can't rely on the count | |
759 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 760 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
761 | * means that no tasks are currently attached, therefore there is no |
762 | * way a task attached to that cgroup can fork (the other way to | |
763 | * increment the count). So code holding cgroup_mutex can safely | |
764 | * assume that if the count is zero, it will stay zero. Similarly, if | |
765 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
766 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
767 | * needs that mutex. | |
768 | * | |
ddbcc7e8 PM |
769 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
770 | * (usually) take cgroup_mutex. These are the two most performance | |
771 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
772 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
773 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
774 | * to the release agent with the name of the cgroup (path relative to |
775 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
776 | * |
777 | * A cgroup can only be deleted if both its 'count' of using tasks | |
778 | * is zero, and its list of 'children' cgroups is empty. Since all | |
779 | * tasks in the system use _some_ cgroup, and since there is always at | |
780 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
781 | * always has either children cgroups and/or using tasks. So we don't | |
782 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
783 | * | |
784 | * The task_lock() exception | |
785 | * | |
786 | * The need for this exception arises from the action of | |
956db3ca | 787 | * cgroup_attach_task(), which overwrites one tasks cgroup pointer with |
a043e3b2 | 788 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
789 | * several performance critical places that need to reference |
790 | * task->cgroup without the expense of grabbing a system global | |
791 | * mutex. Therefore except as noted below, when dereferencing or, as | |
956db3ca | 792 | * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use |
ddbcc7e8 PM |
793 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
794 | * the task_struct routinely used for such matters. | |
795 | * | |
796 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 797 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
798 | */ |
799 | ||
ddbcc7e8 PM |
800 | /** |
801 | * cgroup_lock - lock out any changes to cgroup structures | |
802 | * | |
803 | */ | |
ddbcc7e8 PM |
804 | void cgroup_lock(void) |
805 | { | |
806 | mutex_lock(&cgroup_mutex); | |
807 | } | |
67523c48 | 808 | EXPORT_SYMBOL_GPL(cgroup_lock); |
ddbcc7e8 PM |
809 | |
810 | /** | |
811 | * cgroup_unlock - release lock on cgroup changes | |
812 | * | |
813 | * Undo the lock taken in a previous cgroup_lock() call. | |
814 | */ | |
ddbcc7e8 PM |
815 | void cgroup_unlock(void) |
816 | { | |
817 | mutex_unlock(&cgroup_mutex); | |
818 | } | |
67523c48 | 819 | EXPORT_SYMBOL_GPL(cgroup_unlock); |
ddbcc7e8 PM |
820 | |
821 | /* | |
822 | * A couple of forward declarations required, due to cyclic reference loop: | |
823 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> | |
824 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations | |
825 | * -> cgroup_mkdir. | |
826 | */ | |
827 | ||
18bb1db3 | 828 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode); |
00cd8dd3 | 829 | static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int); |
ddbcc7e8 | 830 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); |
13af07df AR |
831 | static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files, |
832 | unsigned long subsys_mask); | |
6e1d5dcc | 833 | static const struct inode_operations cgroup_dir_inode_operations; |
828c0950 | 834 | static const struct file_operations proc_cgroupstats_operations; |
a424316c PM |
835 | |
836 | static struct backing_dev_info cgroup_backing_dev_info = { | |
d993831f | 837 | .name = "cgroup", |
e4ad08fe | 838 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
a424316c | 839 | }; |
ddbcc7e8 | 840 | |
38460b48 KH |
841 | static int alloc_css_id(struct cgroup_subsys *ss, |
842 | struct cgroup *parent, struct cgroup *child); | |
843 | ||
a5e7ed32 | 844 | static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb) |
ddbcc7e8 PM |
845 | { |
846 | struct inode *inode = new_inode(sb); | |
ddbcc7e8 PM |
847 | |
848 | if (inode) { | |
85fe4025 | 849 | inode->i_ino = get_next_ino(); |
ddbcc7e8 | 850 | inode->i_mode = mode; |
76aac0e9 DH |
851 | inode->i_uid = current_fsuid(); |
852 | inode->i_gid = current_fsgid(); | |
ddbcc7e8 PM |
853 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
854 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; | |
855 | } | |
856 | return inode; | |
857 | } | |
858 | ||
859 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) | |
860 | { | |
861 | /* is dentry a directory ? if so, kfree() associated cgroup */ | |
862 | if (S_ISDIR(inode->i_mode)) { | |
bd89aabc | 863 | struct cgroup *cgrp = dentry->d_fsdata; |
8dc4f3e1 | 864 | struct cgroup_subsys *ss; |
bd89aabc | 865 | BUG_ON(!(cgroup_is_removed(cgrp))); |
81a6a5cd PM |
866 | /* It's possible for external users to be holding css |
867 | * reference counts on a cgroup; css_put() needs to | |
868 | * be able to access the cgroup after decrementing | |
869 | * the reference count in order to know if it needs to | |
870 | * queue the cgroup to be handled by the release | |
871 | * agent */ | |
872 | synchronize_rcu(); | |
8dc4f3e1 PM |
873 | |
874 | mutex_lock(&cgroup_mutex); | |
875 | /* | |
876 | * Release the subsystem state objects. | |
877 | */ | |
75139b82 | 878 | for_each_subsys(cgrp->root, ss) |
761b3ef5 | 879 | ss->destroy(cgrp); |
8dc4f3e1 PM |
880 | |
881 | cgrp->root->number_of_cgroups--; | |
882 | mutex_unlock(&cgroup_mutex); | |
883 | ||
a47295e6 | 884 | /* |
7db5b3ca TH |
885 | * Drop the active superblock reference that we took when we |
886 | * created the cgroup | |
a47295e6 | 887 | */ |
7db5b3ca | 888 | deactivate_super(cgrp->root->sb); |
8dc4f3e1 | 889 | |
72a8cb30 BB |
890 | /* |
891 | * if we're getting rid of the cgroup, refcount should ensure | |
892 | * that there are no pidlists left. | |
893 | */ | |
894 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
895 | ||
03b1cde6 AR |
896 | simple_xattrs_free(&cgrp->xattrs); |
897 | ||
f2da1c40 | 898 | kfree_rcu(cgrp, rcu_head); |
05ef1d7c TH |
899 | } else { |
900 | struct cfent *cfe = __d_cfe(dentry); | |
901 | struct cgroup *cgrp = dentry->d_parent->d_fsdata; | |
03b1cde6 | 902 | struct cftype *cft = cfe->type; |
05ef1d7c TH |
903 | |
904 | WARN_ONCE(!list_empty(&cfe->node) && | |
905 | cgrp != &cgrp->root->top_cgroup, | |
906 | "cfe still linked for %s\n", cfe->type->name); | |
907 | kfree(cfe); | |
03b1cde6 | 908 | simple_xattrs_free(&cft->xattrs); |
ddbcc7e8 PM |
909 | } |
910 | iput(inode); | |
911 | } | |
912 | ||
c72a04e3 AV |
913 | static int cgroup_delete(const struct dentry *d) |
914 | { | |
915 | return 1; | |
916 | } | |
917 | ||
ddbcc7e8 PM |
918 | static void remove_dir(struct dentry *d) |
919 | { | |
920 | struct dentry *parent = dget(d->d_parent); | |
921 | ||
922 | d_delete(d); | |
923 | simple_rmdir(parent->d_inode, d); | |
924 | dput(parent); | |
925 | } | |
926 | ||
05ef1d7c TH |
927 | static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
928 | { | |
929 | struct cfent *cfe; | |
930 | ||
931 | lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex); | |
932 | lockdep_assert_held(&cgroup_mutex); | |
933 | ||
934 | list_for_each_entry(cfe, &cgrp->files, node) { | |
935 | struct dentry *d = cfe->dentry; | |
936 | ||
937 | if (cft && cfe->type != cft) | |
938 | continue; | |
939 | ||
940 | dget(d); | |
941 | d_delete(d); | |
ce27e317 | 942 | simple_unlink(cgrp->dentry->d_inode, d); |
05ef1d7c TH |
943 | list_del_init(&cfe->node); |
944 | dput(d); | |
945 | ||
946 | return 0; | |
ddbcc7e8 | 947 | } |
05ef1d7c TH |
948 | return -ENOENT; |
949 | } | |
950 | ||
13af07df AR |
951 | /** |
952 | * cgroup_clear_directory - selective removal of base and subsystem files | |
953 | * @dir: directory containing the files | |
954 | * @base_files: true if the base files should be removed | |
955 | * @subsys_mask: mask of the subsystem ids whose files should be removed | |
956 | */ | |
957 | static void cgroup_clear_directory(struct dentry *dir, bool base_files, | |
958 | unsigned long subsys_mask) | |
05ef1d7c TH |
959 | { |
960 | struct cgroup *cgrp = __d_cgrp(dir); | |
13af07df | 961 | struct cgroup_subsys *ss; |
05ef1d7c | 962 | |
13af07df AR |
963 | for_each_subsys(cgrp->root, ss) { |
964 | struct cftype_set *set; | |
965 | if (!test_bit(ss->subsys_id, &subsys_mask)) | |
966 | continue; | |
967 | list_for_each_entry(set, &ss->cftsets, node) | |
968 | cgroup_rm_file(cgrp, set->cfts); | |
969 | } | |
970 | if (base_files) { | |
971 | while (!list_empty(&cgrp->files)) | |
972 | cgroup_rm_file(cgrp, NULL); | |
973 | } | |
ddbcc7e8 PM |
974 | } |
975 | ||
976 | /* | |
977 | * NOTE : the dentry must have been dget()'ed | |
978 | */ | |
979 | static void cgroup_d_remove_dir(struct dentry *dentry) | |
980 | { | |
2fd6b7f5 | 981 | struct dentry *parent; |
13af07df | 982 | struct cgroupfs_root *root = dentry->d_sb->s_fs_info; |
2fd6b7f5 | 983 | |
a1a71b45 | 984 | cgroup_clear_directory(dentry, true, root->subsys_mask); |
ddbcc7e8 | 985 | |
2fd6b7f5 NP |
986 | parent = dentry->d_parent; |
987 | spin_lock(&parent->d_lock); | |
3ec762ad | 988 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
ddbcc7e8 | 989 | list_del_init(&dentry->d_u.d_child); |
2fd6b7f5 NP |
990 | spin_unlock(&dentry->d_lock); |
991 | spin_unlock(&parent->d_lock); | |
ddbcc7e8 PM |
992 | remove_dir(dentry); |
993 | } | |
994 | ||
aae8aab4 | 995 | /* |
cf5d5941 BB |
996 | * Call with cgroup_mutex held. Drops reference counts on modules, including |
997 | * any duplicate ones that parse_cgroupfs_options took. If this function | |
998 | * returns an error, no reference counts are touched. | |
aae8aab4 | 999 | */ |
ddbcc7e8 | 1000 | static int rebind_subsystems(struct cgroupfs_root *root, |
a1a71b45 | 1001 | unsigned long final_subsys_mask) |
ddbcc7e8 | 1002 | { |
a1a71b45 | 1003 | unsigned long added_mask, removed_mask; |
bd89aabc | 1004 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 PM |
1005 | int i; |
1006 | ||
aae8aab4 | 1007 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
e25e2cbb | 1008 | BUG_ON(!mutex_is_locked(&cgroup_root_mutex)); |
aae8aab4 | 1009 | |
a1a71b45 AR |
1010 | removed_mask = root->actual_subsys_mask & ~final_subsys_mask; |
1011 | added_mask = final_subsys_mask & ~root->actual_subsys_mask; | |
ddbcc7e8 PM |
1012 | /* Check that any added subsystems are currently free */ |
1013 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
8d53d55d | 1014 | unsigned long bit = 1UL << i; |
ddbcc7e8 | 1015 | struct cgroup_subsys *ss = subsys[i]; |
a1a71b45 | 1016 | if (!(bit & added_mask)) |
ddbcc7e8 | 1017 | continue; |
aae8aab4 BB |
1018 | /* |
1019 | * Nobody should tell us to do a subsys that doesn't exist: | |
1020 | * parse_cgroupfs_options should catch that case and refcounts | |
1021 | * ensure that subsystems won't disappear once selected. | |
1022 | */ | |
1023 | BUG_ON(ss == NULL); | |
ddbcc7e8 PM |
1024 | if (ss->root != &rootnode) { |
1025 | /* Subsystem isn't free */ | |
1026 | return -EBUSY; | |
1027 | } | |
1028 | } | |
1029 | ||
1030 | /* Currently we don't handle adding/removing subsystems when | |
1031 | * any child cgroups exist. This is theoretically supportable | |
1032 | * but involves complex error handling, so it's being left until | |
1033 | * later */ | |
307257cf | 1034 | if (root->number_of_cgroups > 1) |
ddbcc7e8 PM |
1035 | return -EBUSY; |
1036 | ||
1037 | /* Process each subsystem */ | |
1038 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1039 | struct cgroup_subsys *ss = subsys[i]; | |
1040 | unsigned long bit = 1UL << i; | |
a1a71b45 | 1041 | if (bit & added_mask) { |
ddbcc7e8 | 1042 | /* We're binding this subsystem to this hierarchy */ |
aae8aab4 | 1043 | BUG_ON(ss == NULL); |
bd89aabc | 1044 | BUG_ON(cgrp->subsys[i]); |
ddbcc7e8 PM |
1045 | BUG_ON(!dummytop->subsys[i]); |
1046 | BUG_ON(dummytop->subsys[i]->cgroup != dummytop); | |
bd89aabc PM |
1047 | cgrp->subsys[i] = dummytop->subsys[i]; |
1048 | cgrp->subsys[i]->cgroup = cgrp; | |
33a68ac1 | 1049 | list_move(&ss->sibling, &root->subsys_list); |
b2aa30f7 | 1050 | ss->root = root; |
ddbcc7e8 | 1051 | if (ss->bind) |
761b3ef5 | 1052 | ss->bind(cgrp); |
cf5d5941 | 1053 | /* refcount was already taken, and we're keeping it */ |
a1a71b45 | 1054 | } else if (bit & removed_mask) { |
ddbcc7e8 | 1055 | /* We're removing this subsystem */ |
aae8aab4 | 1056 | BUG_ON(ss == NULL); |
bd89aabc PM |
1057 | BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]); |
1058 | BUG_ON(cgrp->subsys[i]->cgroup != cgrp); | |
ddbcc7e8 | 1059 | if (ss->bind) |
761b3ef5 | 1060 | ss->bind(dummytop); |
ddbcc7e8 | 1061 | dummytop->subsys[i]->cgroup = dummytop; |
bd89aabc | 1062 | cgrp->subsys[i] = NULL; |
b2aa30f7 | 1063 | subsys[i]->root = &rootnode; |
33a68ac1 | 1064 | list_move(&ss->sibling, &rootnode.subsys_list); |
cf5d5941 BB |
1065 | /* subsystem is now free - drop reference on module */ |
1066 | module_put(ss->module); | |
a1a71b45 | 1067 | } else if (bit & final_subsys_mask) { |
ddbcc7e8 | 1068 | /* Subsystem state should already exist */ |
aae8aab4 | 1069 | BUG_ON(ss == NULL); |
bd89aabc | 1070 | BUG_ON(!cgrp->subsys[i]); |
cf5d5941 BB |
1071 | /* |
1072 | * a refcount was taken, but we already had one, so | |
1073 | * drop the extra reference. | |
1074 | */ | |
1075 | module_put(ss->module); | |
1076 | #ifdef CONFIG_MODULE_UNLOAD | |
1077 | BUG_ON(ss->module && !module_refcount(ss->module)); | |
1078 | #endif | |
ddbcc7e8 PM |
1079 | } else { |
1080 | /* Subsystem state shouldn't exist */ | |
bd89aabc | 1081 | BUG_ON(cgrp->subsys[i]); |
ddbcc7e8 PM |
1082 | } |
1083 | } | |
a1a71b45 | 1084 | root->subsys_mask = root->actual_subsys_mask = final_subsys_mask; |
ddbcc7e8 PM |
1085 | synchronize_rcu(); |
1086 | ||
1087 | return 0; | |
1088 | } | |
1089 | ||
34c80b1d | 1090 | static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry) |
ddbcc7e8 | 1091 | { |
34c80b1d | 1092 | struct cgroupfs_root *root = dentry->d_sb->s_fs_info; |
ddbcc7e8 PM |
1093 | struct cgroup_subsys *ss; |
1094 | ||
e25e2cbb | 1095 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1096 | for_each_subsys(root, ss) |
1097 | seq_printf(seq, ",%s", ss->name); | |
1098 | if (test_bit(ROOT_NOPREFIX, &root->flags)) | |
1099 | seq_puts(seq, ",noprefix"); | |
03b1cde6 AR |
1100 | if (test_bit(ROOT_XATTR, &root->flags)) |
1101 | seq_puts(seq, ",xattr"); | |
81a6a5cd PM |
1102 | if (strlen(root->release_agent_path)) |
1103 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
97978e6d DL |
1104 | if (clone_children(&root->top_cgroup)) |
1105 | seq_puts(seq, ",clone_children"); | |
c6d57f33 PM |
1106 | if (strlen(root->name)) |
1107 | seq_printf(seq, ",name=%s", root->name); | |
e25e2cbb | 1108 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1109 | return 0; |
1110 | } | |
1111 | ||
1112 | struct cgroup_sb_opts { | |
a1a71b45 | 1113 | unsigned long subsys_mask; |
ddbcc7e8 | 1114 | unsigned long flags; |
81a6a5cd | 1115 | char *release_agent; |
97978e6d | 1116 | bool clone_children; |
c6d57f33 | 1117 | char *name; |
2c6ab6d2 PM |
1118 | /* User explicitly requested empty subsystem */ |
1119 | bool none; | |
c6d57f33 PM |
1120 | |
1121 | struct cgroupfs_root *new_root; | |
2c6ab6d2 | 1122 | |
ddbcc7e8 PM |
1123 | }; |
1124 | ||
aae8aab4 BB |
1125 | /* |
1126 | * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call | |
cf5d5941 BB |
1127 | * with cgroup_mutex held to protect the subsys[] array. This function takes |
1128 | * refcounts on subsystems to be used, unless it returns error, in which case | |
1129 | * no refcounts are taken. | |
aae8aab4 | 1130 | */ |
cf5d5941 | 1131 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1132 | { |
32a8cf23 DL |
1133 | char *token, *o = data; |
1134 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1135 | unsigned long mask = (unsigned long)-1; |
cf5d5941 BB |
1136 | int i; |
1137 | bool module_pin_failed = false; | |
f9ab5b5b | 1138 | |
aae8aab4 BB |
1139 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1140 | ||
f9ab5b5b LZ |
1141 | #ifdef CONFIG_CPUSETS |
1142 | mask = ~(1UL << cpuset_subsys_id); | |
1143 | #endif | |
ddbcc7e8 | 1144 | |
c6d57f33 | 1145 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1146 | |
1147 | while ((token = strsep(&o, ",")) != NULL) { | |
1148 | if (!*token) | |
1149 | return -EINVAL; | |
32a8cf23 | 1150 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1151 | /* Explicitly have no subsystems */ |
1152 | opts->none = true; | |
32a8cf23 DL |
1153 | continue; |
1154 | } | |
1155 | if (!strcmp(token, "all")) { | |
1156 | /* Mutually exclusive option 'all' + subsystem name */ | |
1157 | if (one_ss) | |
1158 | return -EINVAL; | |
1159 | all_ss = true; | |
1160 | continue; | |
1161 | } | |
1162 | if (!strcmp(token, "noprefix")) { | |
ddbcc7e8 | 1163 | set_bit(ROOT_NOPREFIX, &opts->flags); |
32a8cf23 DL |
1164 | continue; |
1165 | } | |
1166 | if (!strcmp(token, "clone_children")) { | |
97978e6d | 1167 | opts->clone_children = true; |
32a8cf23 DL |
1168 | continue; |
1169 | } | |
03b1cde6 AR |
1170 | if (!strcmp(token, "xattr")) { |
1171 | set_bit(ROOT_XATTR, &opts->flags); | |
1172 | continue; | |
1173 | } | |
32a8cf23 | 1174 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1175 | /* Specifying two release agents is forbidden */ |
1176 | if (opts->release_agent) | |
1177 | return -EINVAL; | |
c6d57f33 | 1178 | opts->release_agent = |
e400c285 | 1179 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1180 | if (!opts->release_agent) |
1181 | return -ENOMEM; | |
32a8cf23 DL |
1182 | continue; |
1183 | } | |
1184 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1185 | const char *name = token + 5; |
1186 | /* Can't specify an empty name */ | |
1187 | if (!strlen(name)) | |
1188 | return -EINVAL; | |
1189 | /* Must match [\w.-]+ */ | |
1190 | for (i = 0; i < strlen(name); i++) { | |
1191 | char c = name[i]; | |
1192 | if (isalnum(c)) | |
1193 | continue; | |
1194 | if ((c == '.') || (c == '-') || (c == '_')) | |
1195 | continue; | |
1196 | return -EINVAL; | |
1197 | } | |
1198 | /* Specifying two names is forbidden */ | |
1199 | if (opts->name) | |
1200 | return -EINVAL; | |
1201 | opts->name = kstrndup(name, | |
e400c285 | 1202 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1203 | GFP_KERNEL); |
1204 | if (!opts->name) | |
1205 | return -ENOMEM; | |
32a8cf23 DL |
1206 | |
1207 | continue; | |
1208 | } | |
1209 | ||
1210 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
1211 | struct cgroup_subsys *ss = subsys[i]; | |
1212 | if (ss == NULL) | |
1213 | continue; | |
1214 | if (strcmp(token, ss->name)) | |
1215 | continue; | |
1216 | if (ss->disabled) | |
1217 | continue; | |
1218 | ||
1219 | /* Mutually exclusive option 'all' + subsystem name */ | |
1220 | if (all_ss) | |
1221 | return -EINVAL; | |
a1a71b45 | 1222 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1223 | one_ss = true; |
1224 | ||
1225 | break; | |
1226 | } | |
1227 | if (i == CGROUP_SUBSYS_COUNT) | |
1228 | return -ENOENT; | |
1229 | } | |
1230 | ||
1231 | /* | |
1232 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1233 | * otherwise if 'none', 'name=' and a subsystem name options |
1234 | * were not specified, let's default to 'all' | |
32a8cf23 | 1235 | */ |
0d19ea86 | 1236 | if (all_ss || (!one_ss && !opts->none && !opts->name)) { |
32a8cf23 DL |
1237 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
1238 | struct cgroup_subsys *ss = subsys[i]; | |
1239 | if (ss == NULL) | |
1240 | continue; | |
1241 | if (ss->disabled) | |
1242 | continue; | |
a1a71b45 | 1243 | set_bit(i, &opts->subsys_mask); |
ddbcc7e8 PM |
1244 | } |
1245 | } | |
1246 | ||
2c6ab6d2 PM |
1247 | /* Consistency checks */ |
1248 | ||
f9ab5b5b LZ |
1249 | /* |
1250 | * Option noprefix was introduced just for backward compatibility | |
1251 | * with the old cpuset, so we allow noprefix only if mounting just | |
1252 | * the cpuset subsystem. | |
1253 | */ | |
1254 | if (test_bit(ROOT_NOPREFIX, &opts->flags) && | |
a1a71b45 | 1255 | (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1256 | return -EINVAL; |
1257 | ||
2c6ab6d2 PM |
1258 | |
1259 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1260 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1261 | return -EINVAL; |
1262 | ||
1263 | /* | |
1264 | * We either have to specify by name or by subsystems. (So all | |
1265 | * empty hierarchies must have a name). | |
1266 | */ | |
a1a71b45 | 1267 | if (!opts->subsys_mask && !opts->name) |
ddbcc7e8 PM |
1268 | return -EINVAL; |
1269 | ||
cf5d5941 BB |
1270 | /* |
1271 | * Grab references on all the modules we'll need, so the subsystems | |
1272 | * don't dance around before rebind_subsystems attaches them. This may | |
1273 | * take duplicate reference counts on a subsystem that's already used, | |
1274 | * but rebind_subsystems handles this case. | |
1275 | */ | |
be45c900 | 1276 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
cf5d5941 BB |
1277 | unsigned long bit = 1UL << i; |
1278 | ||
a1a71b45 | 1279 | if (!(bit & opts->subsys_mask)) |
cf5d5941 BB |
1280 | continue; |
1281 | if (!try_module_get(subsys[i]->module)) { | |
1282 | module_pin_failed = true; | |
1283 | break; | |
1284 | } | |
1285 | } | |
1286 | if (module_pin_failed) { | |
1287 | /* | |
1288 | * oops, one of the modules was going away. this means that we | |
1289 | * raced with a module_delete call, and to the user this is | |
1290 | * essentially a "subsystem doesn't exist" case. | |
1291 | */ | |
be45c900 | 1292 | for (i--; i >= 0; i--) { |
cf5d5941 BB |
1293 | /* drop refcounts only on the ones we took */ |
1294 | unsigned long bit = 1UL << i; | |
1295 | ||
a1a71b45 | 1296 | if (!(bit & opts->subsys_mask)) |
cf5d5941 BB |
1297 | continue; |
1298 | module_put(subsys[i]->module); | |
1299 | } | |
1300 | return -ENOENT; | |
1301 | } | |
1302 | ||
ddbcc7e8 PM |
1303 | return 0; |
1304 | } | |
1305 | ||
a1a71b45 | 1306 | static void drop_parsed_module_refcounts(unsigned long subsys_mask) |
cf5d5941 BB |
1307 | { |
1308 | int i; | |
be45c900 | 1309 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
cf5d5941 BB |
1310 | unsigned long bit = 1UL << i; |
1311 | ||
a1a71b45 | 1312 | if (!(bit & subsys_mask)) |
cf5d5941 BB |
1313 | continue; |
1314 | module_put(subsys[i]->module); | |
1315 | } | |
1316 | } | |
1317 | ||
ddbcc7e8 PM |
1318 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) |
1319 | { | |
1320 | int ret = 0; | |
1321 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1322 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 | 1323 | struct cgroup_sb_opts opts; |
a1a71b45 | 1324 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1325 | |
bd89aabc | 1326 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 | 1327 | mutex_lock(&cgroup_mutex); |
e25e2cbb | 1328 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1329 | |
1330 | /* See what subsystems are wanted */ | |
1331 | ret = parse_cgroupfs_options(data, &opts); | |
1332 | if (ret) | |
1333 | goto out_unlock; | |
1334 | ||
8b5a5a9d | 1335 | /* See feature-removal-schedule.txt */ |
a1a71b45 | 1336 | if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1337 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1338 | task_tgid_nr(current), current->comm); | |
1339 | ||
a1a71b45 AR |
1340 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1341 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1342 | |
cf5d5941 BB |
1343 | /* Don't allow flags or name to change at remount */ |
1344 | if (opts.flags != root->flags || | |
1345 | (opts.name && strcmp(opts.name, root->name))) { | |
c6d57f33 | 1346 | ret = -EINVAL; |
a1a71b45 | 1347 | drop_parsed_module_refcounts(opts.subsys_mask); |
c6d57f33 PM |
1348 | goto out_unlock; |
1349 | } | |
1350 | ||
a1a71b45 | 1351 | ret = rebind_subsystems(root, opts.subsys_mask); |
cf5d5941 | 1352 | if (ret) { |
a1a71b45 | 1353 | drop_parsed_module_refcounts(opts.subsys_mask); |
0670e08b | 1354 | goto out_unlock; |
cf5d5941 | 1355 | } |
ddbcc7e8 | 1356 | |
ff4c8d50 | 1357 | /* clear out any existing files and repopulate subsystem files */ |
a1a71b45 | 1358 | cgroup_clear_directory(cgrp->dentry, false, removed_mask); |
13af07df | 1359 | /* re-populate subsystem files */ |
a1a71b45 | 1360 | cgroup_populate_dir(cgrp, false, added_mask); |
ddbcc7e8 | 1361 | |
81a6a5cd PM |
1362 | if (opts.release_agent) |
1363 | strcpy(root->release_agent_path, opts.release_agent); | |
ddbcc7e8 | 1364 | out_unlock: |
66bdc9cf | 1365 | kfree(opts.release_agent); |
c6d57f33 | 1366 | kfree(opts.name); |
e25e2cbb | 1367 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 | 1368 | mutex_unlock(&cgroup_mutex); |
bd89aabc | 1369 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
1370 | return ret; |
1371 | } | |
1372 | ||
b87221de | 1373 | static const struct super_operations cgroup_ops = { |
ddbcc7e8 PM |
1374 | .statfs = simple_statfs, |
1375 | .drop_inode = generic_delete_inode, | |
1376 | .show_options = cgroup_show_options, | |
1377 | .remount_fs = cgroup_remount, | |
1378 | }; | |
1379 | ||
cc31edce PM |
1380 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1381 | { | |
1382 | INIT_LIST_HEAD(&cgrp->sibling); | |
1383 | INIT_LIST_HEAD(&cgrp->children); | |
05ef1d7c | 1384 | INIT_LIST_HEAD(&cgrp->files); |
cc31edce | 1385 | INIT_LIST_HEAD(&cgrp->css_sets); |
2243076a | 1386 | INIT_LIST_HEAD(&cgrp->allcg_node); |
cc31edce | 1387 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1388 | INIT_LIST_HEAD(&cgrp->pidlists); |
1389 | mutex_init(&cgrp->pidlist_mutex); | |
0dea1168 KS |
1390 | INIT_LIST_HEAD(&cgrp->event_list); |
1391 | spin_lock_init(&cgrp->event_list_lock); | |
03b1cde6 | 1392 | simple_xattrs_init(&cgrp->xattrs); |
cc31edce | 1393 | } |
c6d57f33 | 1394 | |
ddbcc7e8 PM |
1395 | static void init_cgroup_root(struct cgroupfs_root *root) |
1396 | { | |
bd89aabc | 1397 | struct cgroup *cgrp = &root->top_cgroup; |
b0ca5a84 | 1398 | |
ddbcc7e8 PM |
1399 | INIT_LIST_HEAD(&root->subsys_list); |
1400 | INIT_LIST_HEAD(&root->root_list); | |
b0ca5a84 | 1401 | INIT_LIST_HEAD(&root->allcg_list); |
ddbcc7e8 | 1402 | root->number_of_cgroups = 1; |
bd89aabc PM |
1403 | cgrp->root = root; |
1404 | cgrp->top_cgroup = cgrp; | |
b0ca5a84 | 1405 | list_add_tail(&cgrp->allcg_node, &root->allcg_list); |
cc31edce | 1406 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 PM |
1407 | } |
1408 | ||
2c6ab6d2 PM |
1409 | static bool init_root_id(struct cgroupfs_root *root) |
1410 | { | |
1411 | int ret = 0; | |
1412 | ||
1413 | do { | |
1414 | if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL)) | |
1415 | return false; | |
1416 | spin_lock(&hierarchy_id_lock); | |
1417 | /* Try to allocate the next unused ID */ | |
1418 | ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id, | |
1419 | &root->hierarchy_id); | |
1420 | if (ret == -ENOSPC) | |
1421 | /* Try again starting from 0 */ | |
1422 | ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id); | |
1423 | if (!ret) { | |
1424 | next_hierarchy_id = root->hierarchy_id + 1; | |
1425 | } else if (ret != -EAGAIN) { | |
1426 | /* Can only get here if the 31-bit IDR is full ... */ | |
1427 | BUG_ON(ret); | |
1428 | } | |
1429 | spin_unlock(&hierarchy_id_lock); | |
1430 | } while (ret); | |
1431 | return true; | |
1432 | } | |
1433 | ||
ddbcc7e8 PM |
1434 | static int cgroup_test_super(struct super_block *sb, void *data) |
1435 | { | |
c6d57f33 | 1436 | struct cgroup_sb_opts *opts = data; |
ddbcc7e8 PM |
1437 | struct cgroupfs_root *root = sb->s_fs_info; |
1438 | ||
c6d57f33 PM |
1439 | /* If we asked for a name then it must match */ |
1440 | if (opts->name && strcmp(opts->name, root->name)) | |
1441 | return 0; | |
ddbcc7e8 | 1442 | |
2c6ab6d2 PM |
1443 | /* |
1444 | * If we asked for subsystems (or explicitly for no | |
1445 | * subsystems) then they must match | |
1446 | */ | |
a1a71b45 AR |
1447 | if ((opts->subsys_mask || opts->none) |
1448 | && (opts->subsys_mask != root->subsys_mask)) | |
ddbcc7e8 PM |
1449 | return 0; |
1450 | ||
1451 | return 1; | |
1452 | } | |
1453 | ||
c6d57f33 PM |
1454 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1455 | { | |
1456 | struct cgroupfs_root *root; | |
1457 | ||
a1a71b45 | 1458 | if (!opts->subsys_mask && !opts->none) |
c6d57f33 PM |
1459 | return NULL; |
1460 | ||
1461 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1462 | if (!root) | |
1463 | return ERR_PTR(-ENOMEM); | |
1464 | ||
2c6ab6d2 PM |
1465 | if (!init_root_id(root)) { |
1466 | kfree(root); | |
1467 | return ERR_PTR(-ENOMEM); | |
1468 | } | |
c6d57f33 | 1469 | init_cgroup_root(root); |
2c6ab6d2 | 1470 | |
a1a71b45 | 1471 | root->subsys_mask = opts->subsys_mask; |
c6d57f33 PM |
1472 | root->flags = opts->flags; |
1473 | if (opts->release_agent) | |
1474 | strcpy(root->release_agent_path, opts->release_agent); | |
1475 | if (opts->name) | |
1476 | strcpy(root->name, opts->name); | |
97978e6d DL |
1477 | if (opts->clone_children) |
1478 | set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1479 | return root; |
1480 | } | |
1481 | ||
2c6ab6d2 PM |
1482 | static void cgroup_drop_root(struct cgroupfs_root *root) |
1483 | { | |
1484 | if (!root) | |
1485 | return; | |
1486 | ||
1487 | BUG_ON(!root->hierarchy_id); | |
1488 | spin_lock(&hierarchy_id_lock); | |
1489 | ida_remove(&hierarchy_ida, root->hierarchy_id); | |
1490 | spin_unlock(&hierarchy_id_lock); | |
1491 | kfree(root); | |
1492 | } | |
1493 | ||
ddbcc7e8 PM |
1494 | static int cgroup_set_super(struct super_block *sb, void *data) |
1495 | { | |
1496 | int ret; | |
c6d57f33 PM |
1497 | struct cgroup_sb_opts *opts = data; |
1498 | ||
1499 | /* If we don't have a new root, we can't set up a new sb */ | |
1500 | if (!opts->new_root) | |
1501 | return -EINVAL; | |
1502 | ||
a1a71b45 | 1503 | BUG_ON(!opts->subsys_mask && !opts->none); |
ddbcc7e8 PM |
1504 | |
1505 | ret = set_anon_super(sb, NULL); | |
1506 | if (ret) | |
1507 | return ret; | |
1508 | ||
c6d57f33 PM |
1509 | sb->s_fs_info = opts->new_root; |
1510 | opts->new_root->sb = sb; | |
ddbcc7e8 PM |
1511 | |
1512 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
1513 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
1514 | sb->s_magic = CGROUP_SUPER_MAGIC; | |
1515 | sb->s_op = &cgroup_ops; | |
1516 | ||
1517 | return 0; | |
1518 | } | |
1519 | ||
1520 | static int cgroup_get_rootdir(struct super_block *sb) | |
1521 | { | |
0df6a63f AV |
1522 | static const struct dentry_operations cgroup_dops = { |
1523 | .d_iput = cgroup_diput, | |
c72a04e3 | 1524 | .d_delete = cgroup_delete, |
0df6a63f AV |
1525 | }; |
1526 | ||
ddbcc7e8 PM |
1527 | struct inode *inode = |
1528 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); | |
ddbcc7e8 PM |
1529 | |
1530 | if (!inode) | |
1531 | return -ENOMEM; | |
1532 | ||
ddbcc7e8 PM |
1533 | inode->i_fop = &simple_dir_operations; |
1534 | inode->i_op = &cgroup_dir_inode_operations; | |
1535 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
1536 | inc_nlink(inode); | |
48fde701 AV |
1537 | sb->s_root = d_make_root(inode); |
1538 | if (!sb->s_root) | |
ddbcc7e8 | 1539 | return -ENOMEM; |
0df6a63f AV |
1540 | /* for everything else we want ->d_op set */ |
1541 | sb->s_d_op = &cgroup_dops; | |
ddbcc7e8 PM |
1542 | return 0; |
1543 | } | |
1544 | ||
f7e83571 | 1545 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1546 | int flags, const char *unused_dev_name, |
f7e83571 | 1547 | void *data) |
ddbcc7e8 PM |
1548 | { |
1549 | struct cgroup_sb_opts opts; | |
c6d57f33 | 1550 | struct cgroupfs_root *root; |
ddbcc7e8 PM |
1551 | int ret = 0; |
1552 | struct super_block *sb; | |
c6d57f33 | 1553 | struct cgroupfs_root *new_root; |
e25e2cbb | 1554 | struct inode *inode; |
ddbcc7e8 PM |
1555 | |
1556 | /* First find the desired set of subsystems */ | |
aae8aab4 | 1557 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1558 | ret = parse_cgroupfs_options(data, &opts); |
aae8aab4 | 1559 | mutex_unlock(&cgroup_mutex); |
c6d57f33 PM |
1560 | if (ret) |
1561 | goto out_err; | |
ddbcc7e8 | 1562 | |
c6d57f33 PM |
1563 | /* |
1564 | * Allocate a new cgroup root. We may not need it if we're | |
1565 | * reusing an existing hierarchy. | |
1566 | */ | |
1567 | new_root = cgroup_root_from_opts(&opts); | |
1568 | if (IS_ERR(new_root)) { | |
1569 | ret = PTR_ERR(new_root); | |
cf5d5941 | 1570 | goto drop_modules; |
81a6a5cd | 1571 | } |
c6d57f33 | 1572 | opts.new_root = new_root; |
ddbcc7e8 | 1573 | |
c6d57f33 | 1574 | /* Locate an existing or new sb for this hierarchy */ |
9249e17f | 1575 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts); |
ddbcc7e8 | 1576 | if (IS_ERR(sb)) { |
c6d57f33 | 1577 | ret = PTR_ERR(sb); |
2c6ab6d2 | 1578 | cgroup_drop_root(opts.new_root); |
cf5d5941 | 1579 | goto drop_modules; |
ddbcc7e8 PM |
1580 | } |
1581 | ||
c6d57f33 PM |
1582 | root = sb->s_fs_info; |
1583 | BUG_ON(!root); | |
1584 | if (root == opts.new_root) { | |
1585 | /* We used the new root structure, so this is a new hierarchy */ | |
1586 | struct list_head tmp_cg_links; | |
c12f65d4 | 1587 | struct cgroup *root_cgrp = &root->top_cgroup; |
c6d57f33 | 1588 | struct cgroupfs_root *existing_root; |
2ce9738b | 1589 | const struct cred *cred; |
28fd5dfc | 1590 | int i; |
ddbcc7e8 PM |
1591 | |
1592 | BUG_ON(sb->s_root != NULL); | |
1593 | ||
1594 | ret = cgroup_get_rootdir(sb); | |
1595 | if (ret) | |
1596 | goto drop_new_super; | |
817929ec | 1597 | inode = sb->s_root->d_inode; |
ddbcc7e8 | 1598 | |
817929ec | 1599 | mutex_lock(&inode->i_mutex); |
ddbcc7e8 | 1600 | mutex_lock(&cgroup_mutex); |
e25e2cbb | 1601 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 | 1602 | |
e25e2cbb TH |
1603 | /* Check for name clashes with existing mounts */ |
1604 | ret = -EBUSY; | |
1605 | if (strlen(root->name)) | |
1606 | for_each_active_root(existing_root) | |
1607 | if (!strcmp(existing_root->name, root->name)) | |
1608 | goto unlock_drop; | |
c6d57f33 | 1609 | |
817929ec PM |
1610 | /* |
1611 | * We're accessing css_set_count without locking | |
1612 | * css_set_lock here, but that's OK - it can only be | |
1613 | * increased by someone holding cgroup_lock, and | |
1614 | * that's us. The worst that can happen is that we | |
1615 | * have some link structures left over | |
1616 | */ | |
1617 | ret = allocate_cg_links(css_set_count, &tmp_cg_links); | |
e25e2cbb TH |
1618 | if (ret) |
1619 | goto unlock_drop; | |
817929ec | 1620 | |
a1a71b45 | 1621 | ret = rebind_subsystems(root, root->subsys_mask); |
ddbcc7e8 | 1622 | if (ret == -EBUSY) { |
c6d57f33 | 1623 | free_cg_links(&tmp_cg_links); |
e25e2cbb | 1624 | goto unlock_drop; |
ddbcc7e8 | 1625 | } |
cf5d5941 BB |
1626 | /* |
1627 | * There must be no failure case after here, since rebinding | |
1628 | * takes care of subsystems' refcounts, which are explicitly | |
1629 | * dropped in the failure exit path. | |
1630 | */ | |
ddbcc7e8 PM |
1631 | |
1632 | /* EBUSY should be the only error here */ | |
1633 | BUG_ON(ret); | |
1634 | ||
1635 | list_add(&root->root_list, &roots); | |
817929ec | 1636 | root_count++; |
ddbcc7e8 | 1637 | |
c12f65d4 | 1638 | sb->s_root->d_fsdata = root_cgrp; |
ddbcc7e8 PM |
1639 | root->top_cgroup.dentry = sb->s_root; |
1640 | ||
817929ec PM |
1641 | /* Link the top cgroup in this hierarchy into all |
1642 | * the css_set objects */ | |
1643 | write_lock(&css_set_lock); | |
28fd5dfc LZ |
1644 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { |
1645 | struct hlist_head *hhead = &css_set_table[i]; | |
1646 | struct hlist_node *node; | |
817929ec | 1647 | struct css_set *cg; |
28fd5dfc | 1648 | |
c12f65d4 LZ |
1649 | hlist_for_each_entry(cg, node, hhead, hlist) |
1650 | link_css_set(&tmp_cg_links, cg, root_cgrp); | |
28fd5dfc | 1651 | } |
817929ec PM |
1652 | write_unlock(&css_set_lock); |
1653 | ||
1654 | free_cg_links(&tmp_cg_links); | |
1655 | ||
c12f65d4 | 1656 | BUG_ON(!list_empty(&root_cgrp->children)); |
ddbcc7e8 PM |
1657 | BUG_ON(root->number_of_cgroups != 1); |
1658 | ||
2ce9738b | 1659 | cred = override_creds(&init_cred); |
a1a71b45 | 1660 | cgroup_populate_dir(root_cgrp, true, root->subsys_mask); |
2ce9738b | 1661 | revert_creds(cred); |
e25e2cbb | 1662 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 | 1663 | mutex_unlock(&cgroup_mutex); |
34f77a90 | 1664 | mutex_unlock(&inode->i_mutex); |
c6d57f33 PM |
1665 | } else { |
1666 | /* | |
1667 | * We re-used an existing hierarchy - the new root (if | |
1668 | * any) is not needed | |
1669 | */ | |
2c6ab6d2 | 1670 | cgroup_drop_root(opts.new_root); |
cf5d5941 | 1671 | /* no subsys rebinding, so refcounts don't change */ |
a1a71b45 | 1672 | drop_parsed_module_refcounts(opts.subsys_mask); |
ddbcc7e8 PM |
1673 | } |
1674 | ||
c6d57f33 PM |
1675 | kfree(opts.release_agent); |
1676 | kfree(opts.name); | |
f7e83571 | 1677 | return dget(sb->s_root); |
ddbcc7e8 | 1678 | |
e25e2cbb TH |
1679 | unlock_drop: |
1680 | mutex_unlock(&cgroup_root_mutex); | |
1681 | mutex_unlock(&cgroup_mutex); | |
1682 | mutex_unlock(&inode->i_mutex); | |
ddbcc7e8 | 1683 | drop_new_super: |
6f5bbff9 | 1684 | deactivate_locked_super(sb); |
cf5d5941 | 1685 | drop_modules: |
a1a71b45 | 1686 | drop_parsed_module_refcounts(opts.subsys_mask); |
c6d57f33 PM |
1687 | out_err: |
1688 | kfree(opts.release_agent); | |
1689 | kfree(opts.name); | |
f7e83571 | 1690 | return ERR_PTR(ret); |
ddbcc7e8 PM |
1691 | } |
1692 | ||
1693 | static void cgroup_kill_sb(struct super_block *sb) { | |
1694 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1695 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 | 1696 | int ret; |
71cbb949 KM |
1697 | struct cg_cgroup_link *link; |
1698 | struct cg_cgroup_link *saved_link; | |
ddbcc7e8 PM |
1699 | |
1700 | BUG_ON(!root); | |
1701 | ||
1702 | BUG_ON(root->number_of_cgroups != 1); | |
bd89aabc | 1703 | BUG_ON(!list_empty(&cgrp->children)); |
ddbcc7e8 PM |
1704 | |
1705 | mutex_lock(&cgroup_mutex); | |
e25e2cbb | 1706 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1707 | |
1708 | /* Rebind all subsystems back to the default hierarchy */ | |
1709 | ret = rebind_subsystems(root, 0); | |
1710 | /* Shouldn't be able to fail ... */ | |
1711 | BUG_ON(ret); | |
1712 | ||
817929ec PM |
1713 | /* |
1714 | * Release all the links from css_sets to this hierarchy's | |
1715 | * root cgroup | |
1716 | */ | |
1717 | write_lock(&css_set_lock); | |
71cbb949 KM |
1718 | |
1719 | list_for_each_entry_safe(link, saved_link, &cgrp->css_sets, | |
1720 | cgrp_link_list) { | |
817929ec | 1721 | list_del(&link->cg_link_list); |
bd89aabc | 1722 | list_del(&link->cgrp_link_list); |
817929ec PM |
1723 | kfree(link); |
1724 | } | |
1725 | write_unlock(&css_set_lock); | |
1726 | ||
839ec545 PM |
1727 | if (!list_empty(&root->root_list)) { |
1728 | list_del(&root->root_list); | |
1729 | root_count--; | |
1730 | } | |
e5f6a860 | 1731 | |
e25e2cbb | 1732 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1733 | mutex_unlock(&cgroup_mutex); |
1734 | ||
03b1cde6 AR |
1735 | simple_xattrs_free(&cgrp->xattrs); |
1736 | ||
ddbcc7e8 | 1737 | kill_litter_super(sb); |
2c6ab6d2 | 1738 | cgroup_drop_root(root); |
ddbcc7e8 PM |
1739 | } |
1740 | ||
1741 | static struct file_system_type cgroup_fs_type = { | |
1742 | .name = "cgroup", | |
f7e83571 | 1743 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1744 | .kill_sb = cgroup_kill_sb, |
1745 | }; | |
1746 | ||
676db4af GKH |
1747 | static struct kobject *cgroup_kobj; |
1748 | ||
a043e3b2 LZ |
1749 | /** |
1750 | * cgroup_path - generate the path of a cgroup | |
1751 | * @cgrp: the cgroup in question | |
1752 | * @buf: the buffer to write the path into | |
1753 | * @buflen: the length of the buffer | |
1754 | * | |
a47295e6 PM |
1755 | * Called with cgroup_mutex held or else with an RCU-protected cgroup |
1756 | * reference. Writes path of cgroup into buf. Returns 0 on success, | |
1757 | * -errno on error. | |
ddbcc7e8 | 1758 | */ |
bd89aabc | 1759 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
ddbcc7e8 | 1760 | { |
febfcef6 | 1761 | struct dentry *dentry = cgrp->dentry; |
ddbcc7e8 | 1762 | char *start; |
febfcef6 TH |
1763 | |
1764 | rcu_lockdep_assert(rcu_read_lock_held() || cgroup_lock_is_held(), | |
1765 | "cgroup_path() called without proper locking"); | |
ddbcc7e8 | 1766 | |
a47295e6 | 1767 | if (!dentry || cgrp == dummytop) { |
ddbcc7e8 PM |
1768 | /* |
1769 | * Inactive subsystems have no dentry for their root | |
1770 | * cgroup | |
1771 | */ | |
1772 | strcpy(buf, "/"); | |
1773 | return 0; | |
1774 | } | |
1775 | ||
316eb661 | 1776 | start = buf + buflen - 1; |
ddbcc7e8 | 1777 | |
316eb661 | 1778 | *start = '\0'; |
ddbcc7e8 | 1779 | for (;;) { |
a47295e6 | 1780 | int len = dentry->d_name.len; |
9a9686b6 | 1781 | |
ddbcc7e8 PM |
1782 | if ((start -= len) < buf) |
1783 | return -ENAMETOOLONG; | |
9a9686b6 | 1784 | memcpy(start, dentry->d_name.name, len); |
bd89aabc PM |
1785 | cgrp = cgrp->parent; |
1786 | if (!cgrp) | |
ddbcc7e8 | 1787 | break; |
9a9686b6 | 1788 | |
febfcef6 | 1789 | dentry = cgrp->dentry; |
bd89aabc | 1790 | if (!cgrp->parent) |
ddbcc7e8 PM |
1791 | continue; |
1792 | if (--start < buf) | |
1793 | return -ENAMETOOLONG; | |
1794 | *start = '/'; | |
1795 | } | |
1796 | memmove(buf, start, buf + buflen - start); | |
1797 | return 0; | |
1798 | } | |
67523c48 | 1799 | EXPORT_SYMBOL_GPL(cgroup_path); |
ddbcc7e8 | 1800 | |
2f7ee569 TH |
1801 | /* |
1802 | * Control Group taskset | |
1803 | */ | |
134d3373 TH |
1804 | struct task_and_cgroup { |
1805 | struct task_struct *task; | |
1806 | struct cgroup *cgrp; | |
61d1d219 | 1807 | struct css_set *cg; |
134d3373 TH |
1808 | }; |
1809 | ||
2f7ee569 TH |
1810 | struct cgroup_taskset { |
1811 | struct task_and_cgroup single; | |
1812 | struct flex_array *tc_array; | |
1813 | int tc_array_len; | |
1814 | int idx; | |
1815 | struct cgroup *cur_cgrp; | |
1816 | }; | |
1817 | ||
1818 | /** | |
1819 | * cgroup_taskset_first - reset taskset and return the first task | |
1820 | * @tset: taskset of interest | |
1821 | * | |
1822 | * @tset iteration is initialized and the first task is returned. | |
1823 | */ | |
1824 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1825 | { | |
1826 | if (tset->tc_array) { | |
1827 | tset->idx = 0; | |
1828 | return cgroup_taskset_next(tset); | |
1829 | } else { | |
1830 | tset->cur_cgrp = tset->single.cgrp; | |
1831 | return tset->single.task; | |
1832 | } | |
1833 | } | |
1834 | EXPORT_SYMBOL_GPL(cgroup_taskset_first); | |
1835 | ||
1836 | /** | |
1837 | * cgroup_taskset_next - iterate to the next task in taskset | |
1838 | * @tset: taskset of interest | |
1839 | * | |
1840 | * Return the next task in @tset. Iteration must have been initialized | |
1841 | * with cgroup_taskset_first(). | |
1842 | */ | |
1843 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1844 | { | |
1845 | struct task_and_cgroup *tc; | |
1846 | ||
1847 | if (!tset->tc_array || tset->idx >= tset->tc_array_len) | |
1848 | return NULL; | |
1849 | ||
1850 | tc = flex_array_get(tset->tc_array, tset->idx++); | |
1851 | tset->cur_cgrp = tc->cgrp; | |
1852 | return tc->task; | |
1853 | } | |
1854 | EXPORT_SYMBOL_GPL(cgroup_taskset_next); | |
1855 | ||
1856 | /** | |
1857 | * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task | |
1858 | * @tset: taskset of interest | |
1859 | * | |
1860 | * Return the cgroup for the current (last returned) task of @tset. This | |
1861 | * function must be preceded by either cgroup_taskset_first() or | |
1862 | * cgroup_taskset_next(). | |
1863 | */ | |
1864 | struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset) | |
1865 | { | |
1866 | return tset->cur_cgrp; | |
1867 | } | |
1868 | EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup); | |
1869 | ||
1870 | /** | |
1871 | * cgroup_taskset_size - return the number of tasks in taskset | |
1872 | * @tset: taskset of interest | |
1873 | */ | |
1874 | int cgroup_taskset_size(struct cgroup_taskset *tset) | |
1875 | { | |
1876 | return tset->tc_array ? tset->tc_array_len : 1; | |
1877 | } | |
1878 | EXPORT_SYMBOL_GPL(cgroup_taskset_size); | |
1879 | ||
1880 | ||
74a1166d BB |
1881 | /* |
1882 | * cgroup_task_migrate - move a task from one cgroup to another. | |
1883 | * | |
1884 | * 'guarantee' is set if the caller promises that a new css_set for the task | |
1885 | * will already exist. If not set, this function might sleep, and can fail with | |
cd3d0952 | 1886 | * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked. |
74a1166d | 1887 | */ |
61d1d219 MSB |
1888 | static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp, |
1889 | struct task_struct *tsk, struct css_set *newcg) | |
74a1166d BB |
1890 | { |
1891 | struct css_set *oldcg; | |
74a1166d BB |
1892 | |
1893 | /* | |
026085ef MSB |
1894 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1895 | * setting such that we can't race against cgroup_exit() changing the | |
1896 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1897 | */ |
c84cdf75 | 1898 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
74a1166d | 1899 | oldcg = tsk->cgroups; |
74a1166d | 1900 | |
74a1166d | 1901 | task_lock(tsk); |
74a1166d BB |
1902 | rcu_assign_pointer(tsk->cgroups, newcg); |
1903 | task_unlock(tsk); | |
1904 | ||
1905 | /* Update the css_set linked lists if we're using them */ | |
1906 | write_lock(&css_set_lock); | |
1907 | if (!list_empty(&tsk->cg_list)) | |
1908 | list_move(&tsk->cg_list, &newcg->tasks); | |
1909 | write_unlock(&css_set_lock); | |
1910 | ||
1911 | /* | |
1912 | * We just gained a reference on oldcg by taking it from the task. As | |
1913 | * trading it for newcg is protected by cgroup_mutex, we're safe to drop | |
1914 | * it here; it will be freed under RCU. | |
1915 | */ | |
74a1166d | 1916 | set_bit(CGRP_RELEASABLE, &oldcgrp->flags); |
1f5320d5 | 1917 | put_css_set(oldcg); |
74a1166d BB |
1918 | } |
1919 | ||
a043e3b2 LZ |
1920 | /** |
1921 | * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp' | |
1922 | * @cgrp: the cgroup the task is attaching to | |
1923 | * @tsk: the task to be attached | |
bbcb81d0 | 1924 | * |
cd3d0952 TH |
1925 | * Call with cgroup_mutex and threadgroup locked. May take task_lock of |
1926 | * @tsk during call. | |
bbcb81d0 | 1927 | */ |
956db3ca | 1928 | int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
bbcb81d0 | 1929 | { |
8f121918 | 1930 | int retval = 0; |
2468c723 | 1931 | struct cgroup_subsys *ss, *failed_ss = NULL; |
bd89aabc | 1932 | struct cgroup *oldcgrp; |
bd89aabc | 1933 | struct cgroupfs_root *root = cgrp->root; |
2f7ee569 | 1934 | struct cgroup_taskset tset = { }; |
61d1d219 | 1935 | struct css_set *newcg; |
bbcb81d0 | 1936 | |
cd3d0952 TH |
1937 | /* @tsk either already exited or can't exit until the end */ |
1938 | if (tsk->flags & PF_EXITING) | |
1939 | return -ESRCH; | |
bbcb81d0 PM |
1940 | |
1941 | /* Nothing to do if the task is already in that cgroup */ | |
7717f7ba | 1942 | oldcgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 1943 | if (cgrp == oldcgrp) |
bbcb81d0 PM |
1944 | return 0; |
1945 | ||
2f7ee569 TH |
1946 | tset.single.task = tsk; |
1947 | tset.single.cgrp = oldcgrp; | |
1948 | ||
bbcb81d0 PM |
1949 | for_each_subsys(root, ss) { |
1950 | if (ss->can_attach) { | |
761b3ef5 | 1951 | retval = ss->can_attach(cgrp, &tset); |
2468c723 DN |
1952 | if (retval) { |
1953 | /* | |
1954 | * Remember on which subsystem the can_attach() | |
1955 | * failed, so that we only call cancel_attach() | |
1956 | * against the subsystems whose can_attach() | |
1957 | * succeeded. (See below) | |
1958 | */ | |
1959 | failed_ss = ss; | |
1960 | goto out; | |
1961 | } | |
bbcb81d0 PM |
1962 | } |
1963 | } | |
1964 | ||
61d1d219 MSB |
1965 | newcg = find_css_set(tsk->cgroups, cgrp); |
1966 | if (!newcg) { | |
1967 | retval = -ENOMEM; | |
2468c723 | 1968 | goto out; |
61d1d219 MSB |
1969 | } |
1970 | ||
1971 | cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg); | |
817929ec | 1972 | |
bbcb81d0 | 1973 | for_each_subsys(root, ss) { |
e18f6318 | 1974 | if (ss->attach) |
761b3ef5 | 1975 | ss->attach(cgrp, &tset); |
bbcb81d0 | 1976 | } |
74a1166d | 1977 | |
bbcb81d0 | 1978 | synchronize_rcu(); |
2468c723 DN |
1979 | out: |
1980 | if (retval) { | |
1981 | for_each_subsys(root, ss) { | |
1982 | if (ss == failed_ss) | |
1983 | /* | |
1984 | * This subsystem was the one that failed the | |
1985 | * can_attach() check earlier, so we don't need | |
1986 | * to call cancel_attach() against it or any | |
1987 | * remaining subsystems. | |
1988 | */ | |
1989 | break; | |
1990 | if (ss->cancel_attach) | |
761b3ef5 | 1991 | ss->cancel_attach(cgrp, &tset); |
2468c723 DN |
1992 | } |
1993 | } | |
1994 | return retval; | |
bbcb81d0 PM |
1995 | } |
1996 | ||
d7926ee3 | 1997 | /** |
31583bb0 MT |
1998 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' |
1999 | * @from: attach to all cgroups of a given task | |
d7926ee3 SS |
2000 | * @tsk: the task to be attached |
2001 | */ | |
31583bb0 | 2002 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) |
d7926ee3 SS |
2003 | { |
2004 | struct cgroupfs_root *root; | |
d7926ee3 SS |
2005 | int retval = 0; |
2006 | ||
2007 | cgroup_lock(); | |
2008 | for_each_active_root(root) { | |
31583bb0 MT |
2009 | struct cgroup *from_cg = task_cgroup_from_root(from, root); |
2010 | ||
2011 | retval = cgroup_attach_task(from_cg, tsk); | |
d7926ee3 SS |
2012 | if (retval) |
2013 | break; | |
2014 | } | |
2015 | cgroup_unlock(); | |
2016 | ||
2017 | return retval; | |
2018 | } | |
31583bb0 | 2019 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); |
d7926ee3 | 2020 | |
74a1166d BB |
2021 | /** |
2022 | * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup | |
2023 | * @cgrp: the cgroup to attach to | |
2024 | * @leader: the threadgroup leader task_struct of the group to be attached | |
2025 | * | |
257058ae TH |
2026 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
2027 | * task_lock of each thread in leader's threadgroup individually in turn. | |
74a1166d | 2028 | */ |
1c6c3fad | 2029 | static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader) |
74a1166d BB |
2030 | { |
2031 | int retval, i, group_size; | |
2032 | struct cgroup_subsys *ss, *failed_ss = NULL; | |
74a1166d | 2033 | /* guaranteed to be initialized later, but the compiler needs this */ |
74a1166d BB |
2034 | struct cgroupfs_root *root = cgrp->root; |
2035 | /* threadgroup list cursor and array */ | |
2036 | struct task_struct *tsk; | |
134d3373 | 2037 | struct task_and_cgroup *tc; |
d846687d | 2038 | struct flex_array *group; |
2f7ee569 | 2039 | struct cgroup_taskset tset = { }; |
74a1166d BB |
2040 | |
2041 | /* | |
2042 | * step 0: in order to do expensive, possibly blocking operations for | |
2043 | * every thread, we cannot iterate the thread group list, since it needs | |
2044 | * rcu or tasklist locked. instead, build an array of all threads in the | |
257058ae TH |
2045 | * group - group_rwsem prevents new threads from appearing, and if |
2046 | * threads exit, this will just be an over-estimate. | |
74a1166d BB |
2047 | */ |
2048 | group_size = get_nr_threads(leader); | |
d846687d | 2049 | /* flex_array supports very large thread-groups better than kmalloc. */ |
134d3373 | 2050 | group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL); |
74a1166d BB |
2051 | if (!group) |
2052 | return -ENOMEM; | |
d846687d BB |
2053 | /* pre-allocate to guarantee space while iterating in rcu read-side. */ |
2054 | retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL); | |
2055 | if (retval) | |
2056 | goto out_free_group_list; | |
74a1166d | 2057 | |
74a1166d BB |
2058 | tsk = leader; |
2059 | i = 0; | |
fb5d2b4c MSB |
2060 | /* |
2061 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2062 | * already PF_EXITING could be freed from underneath us unless we | |
2063 | * take an rcu_read_lock. | |
2064 | */ | |
2065 | rcu_read_lock(); | |
74a1166d | 2066 | do { |
134d3373 TH |
2067 | struct task_and_cgroup ent; |
2068 | ||
cd3d0952 TH |
2069 | /* @tsk either already exited or can't exit until the end */ |
2070 | if (tsk->flags & PF_EXITING) | |
2071 | continue; | |
2072 | ||
74a1166d BB |
2073 | /* as per above, nr_threads may decrease, but not increase. */ |
2074 | BUG_ON(i >= group_size); | |
134d3373 TH |
2075 | ent.task = tsk; |
2076 | ent.cgrp = task_cgroup_from_root(tsk, root); | |
892a2b90 MSB |
2077 | /* nothing to do if this task is already in the cgroup */ |
2078 | if (ent.cgrp == cgrp) | |
2079 | continue; | |
61d1d219 MSB |
2080 | /* |
2081 | * saying GFP_ATOMIC has no effect here because we did prealloc | |
2082 | * earlier, but it's good form to communicate our expectations. | |
2083 | */ | |
134d3373 | 2084 | retval = flex_array_put(group, i, &ent, GFP_ATOMIC); |
d846687d | 2085 | BUG_ON(retval != 0); |
74a1166d BB |
2086 | i++; |
2087 | } while_each_thread(leader, tsk); | |
fb5d2b4c | 2088 | rcu_read_unlock(); |
74a1166d BB |
2089 | /* remember the number of threads in the array for later. */ |
2090 | group_size = i; | |
2f7ee569 TH |
2091 | tset.tc_array = group; |
2092 | tset.tc_array_len = group_size; | |
74a1166d | 2093 | |
134d3373 TH |
2094 | /* methods shouldn't be called if no task is actually migrating */ |
2095 | retval = 0; | |
892a2b90 | 2096 | if (!group_size) |
b07ef774 | 2097 | goto out_free_group_list; |
134d3373 | 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) { | |
761b3ef5 | 2104 | retval = ss->can_attach(cgrp, &tset); |
74a1166d BB |
2105 | if (retval) { |
2106 | failed_ss = ss; | |
2107 | goto out_cancel_attach; | |
2108 | } | |
2109 | } | |
74a1166d BB |
2110 | } |
2111 | ||
2112 | /* | |
2113 | * step 2: make sure css_sets exist for all threads to be migrated. | |
2114 | * we use find_css_set, which allocates a new one if necessary. | |
2115 | */ | |
74a1166d | 2116 | for (i = 0; i < group_size; i++) { |
134d3373 | 2117 | tc = flex_array_get(group, i); |
61d1d219 MSB |
2118 | tc->cg = find_css_set(tc->task->cgroups, cgrp); |
2119 | if (!tc->cg) { | |
2120 | retval = -ENOMEM; | |
2121 | goto out_put_css_set_refs; | |
74a1166d BB |
2122 | } |
2123 | } | |
2124 | ||
2125 | /* | |
494c167c TH |
2126 | * step 3: now that we're guaranteed success wrt the css_sets, |
2127 | * proceed to move all tasks to the new cgroup. There are no | |
2128 | * failure cases after here, so this is the commit point. | |
74a1166d | 2129 | */ |
74a1166d | 2130 | for (i = 0; i < group_size; i++) { |
134d3373 | 2131 | tc = flex_array_get(group, i); |
61d1d219 | 2132 | cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg); |
74a1166d BB |
2133 | } |
2134 | /* nothing is sensitive to fork() after this point. */ | |
2135 | ||
2136 | /* | |
494c167c | 2137 | * step 4: do subsystem attach callbacks. |
74a1166d BB |
2138 | */ |
2139 | for_each_subsys(root, ss) { | |
2140 | if (ss->attach) | |
761b3ef5 | 2141 | ss->attach(cgrp, &tset); |
74a1166d BB |
2142 | } |
2143 | ||
2144 | /* | |
2145 | * step 5: success! and cleanup | |
2146 | */ | |
2147 | synchronize_rcu(); | |
74a1166d | 2148 | retval = 0; |
61d1d219 MSB |
2149 | out_put_css_set_refs: |
2150 | if (retval) { | |
2151 | for (i = 0; i < group_size; i++) { | |
2152 | tc = flex_array_get(group, i); | |
2153 | if (!tc->cg) | |
2154 | break; | |
2155 | put_css_set(tc->cg); | |
2156 | } | |
74a1166d BB |
2157 | } |
2158 | out_cancel_attach: | |
74a1166d BB |
2159 | if (retval) { |
2160 | for_each_subsys(root, ss) { | |
494c167c | 2161 | if (ss == failed_ss) |
74a1166d | 2162 | break; |
74a1166d | 2163 | if (ss->cancel_attach) |
761b3ef5 | 2164 | ss->cancel_attach(cgrp, &tset); |
74a1166d BB |
2165 | } |
2166 | } | |
74a1166d | 2167 | out_free_group_list: |
d846687d | 2168 | flex_array_free(group); |
74a1166d BB |
2169 | return retval; |
2170 | } | |
2171 | ||
2172 | /* | |
2173 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
2174 | * function to attach either it or all tasks in its threadgroup. Will lock |
2175 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 2176 | */ |
74a1166d | 2177 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2178 | { |
bbcb81d0 | 2179 | struct task_struct *tsk; |
c69e8d9c | 2180 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2181 | int ret; |
2182 | ||
74a1166d BB |
2183 | if (!cgroup_lock_live_group(cgrp)) |
2184 | return -ENODEV; | |
2185 | ||
b78949eb MSB |
2186 | retry_find_task: |
2187 | rcu_read_lock(); | |
bbcb81d0 | 2188 | if (pid) { |
73507f33 | 2189 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2190 | if (!tsk) { |
2191 | rcu_read_unlock(); | |
b78949eb MSB |
2192 | ret= -ESRCH; |
2193 | goto out_unlock_cgroup; | |
bbcb81d0 | 2194 | } |
74a1166d BB |
2195 | /* |
2196 | * even if we're attaching all tasks in the thread group, we | |
2197 | * only need to check permissions on one of them. | |
2198 | */ | |
c69e8d9c | 2199 | tcred = __task_cred(tsk); |
14a590c3 EB |
2200 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2201 | !uid_eq(cred->euid, tcred->uid) && | |
2202 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2203 | rcu_read_unlock(); |
b78949eb MSB |
2204 | ret = -EACCES; |
2205 | goto out_unlock_cgroup; | |
bbcb81d0 | 2206 | } |
b78949eb MSB |
2207 | } else |
2208 | tsk = current; | |
cd3d0952 TH |
2209 | |
2210 | if (threadgroup) | |
b78949eb | 2211 | tsk = tsk->group_leader; |
c4c27fbd MG |
2212 | |
2213 | /* | |
2214 | * Workqueue threads may acquire PF_THREAD_BOUND and become | |
2215 | * trapped in a cpuset, or RT worker may be born in a cgroup | |
2216 | * with no rt_runtime allocated. Just say no. | |
2217 | */ | |
2218 | if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) { | |
2219 | ret = -EINVAL; | |
2220 | rcu_read_unlock(); | |
2221 | goto out_unlock_cgroup; | |
2222 | } | |
2223 | ||
b78949eb MSB |
2224 | get_task_struct(tsk); |
2225 | rcu_read_unlock(); | |
2226 | ||
2227 | threadgroup_lock(tsk); | |
2228 | if (threadgroup) { | |
2229 | if (!thread_group_leader(tsk)) { | |
2230 | /* | |
2231 | * a race with de_thread from another thread's exec() | |
2232 | * may strip us of our leadership, if this happens, | |
2233 | * there is no choice but to throw this task away and | |
2234 | * try again; this is | |
2235 | * "double-double-toil-and-trouble-check locking". | |
2236 | */ | |
2237 | threadgroup_unlock(tsk); | |
2238 | put_task_struct(tsk); | |
2239 | goto retry_find_task; | |
2240 | } | |
74a1166d | 2241 | ret = cgroup_attach_proc(cgrp, tsk); |
b78949eb | 2242 | } else |
74a1166d | 2243 | ret = cgroup_attach_task(cgrp, tsk); |
cd3d0952 TH |
2244 | threadgroup_unlock(tsk); |
2245 | ||
bbcb81d0 | 2246 | put_task_struct(tsk); |
b78949eb | 2247 | out_unlock_cgroup: |
74a1166d | 2248 | cgroup_unlock(); |
bbcb81d0 PM |
2249 | return ret; |
2250 | } | |
2251 | ||
af351026 | 2252 | static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid) |
74a1166d BB |
2253 | { |
2254 | return attach_task_by_pid(cgrp, pid, false); | |
2255 | } | |
2256 | ||
2257 | static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid) | |
af351026 | 2258 | { |
b78949eb | 2259 | return attach_task_by_pid(cgrp, tgid, true); |
af351026 PM |
2260 | } |
2261 | ||
e788e066 PM |
2262 | /** |
2263 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
2264 | * @cgrp: the cgroup to be checked for liveness | |
2265 | * | |
84eea842 PM |
2266 | * On success, returns true; the lock should be later released with |
2267 | * cgroup_unlock(). On failure returns false with no lock held. | |
e788e066 | 2268 | */ |
84eea842 | 2269 | bool cgroup_lock_live_group(struct cgroup *cgrp) |
e788e066 PM |
2270 | { |
2271 | mutex_lock(&cgroup_mutex); | |
2272 | if (cgroup_is_removed(cgrp)) { | |
2273 | mutex_unlock(&cgroup_mutex); | |
2274 | return false; | |
2275 | } | |
2276 | return true; | |
2277 | } | |
67523c48 | 2278 | EXPORT_SYMBOL_GPL(cgroup_lock_live_group); |
e788e066 PM |
2279 | |
2280 | static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft, | |
2281 | const char *buffer) | |
2282 | { | |
2283 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); | |
f4a2589f EK |
2284 | if (strlen(buffer) >= PATH_MAX) |
2285 | return -EINVAL; | |
e788e066 PM |
2286 | if (!cgroup_lock_live_group(cgrp)) |
2287 | return -ENODEV; | |
e25e2cbb | 2288 | mutex_lock(&cgroup_root_mutex); |
e788e066 | 2289 | strcpy(cgrp->root->release_agent_path, buffer); |
e25e2cbb | 2290 | mutex_unlock(&cgroup_root_mutex); |
84eea842 | 2291 | cgroup_unlock(); |
e788e066 PM |
2292 | return 0; |
2293 | } | |
2294 | ||
2295 | static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft, | |
2296 | struct seq_file *seq) | |
2297 | { | |
2298 | if (!cgroup_lock_live_group(cgrp)) | |
2299 | return -ENODEV; | |
2300 | seq_puts(seq, cgrp->root->release_agent_path); | |
2301 | seq_putc(seq, '\n'); | |
84eea842 | 2302 | cgroup_unlock(); |
e788e066 PM |
2303 | return 0; |
2304 | } | |
2305 | ||
84eea842 PM |
2306 | /* A buffer size big enough for numbers or short strings */ |
2307 | #define CGROUP_LOCAL_BUFFER_SIZE 64 | |
2308 | ||
e73d2c61 | 2309 | static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft, |
f4c753b7 PM |
2310 | struct file *file, |
2311 | const char __user *userbuf, | |
2312 | size_t nbytes, loff_t *unused_ppos) | |
355e0c48 | 2313 | { |
84eea842 | 2314 | char buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
355e0c48 | 2315 | int retval = 0; |
355e0c48 PM |
2316 | char *end; |
2317 | ||
2318 | if (!nbytes) | |
2319 | return -EINVAL; | |
2320 | if (nbytes >= sizeof(buffer)) | |
2321 | return -E2BIG; | |
2322 | if (copy_from_user(buffer, userbuf, nbytes)) | |
2323 | return -EFAULT; | |
2324 | ||
2325 | buffer[nbytes] = 0; /* nul-terminate */ | |
e73d2c61 | 2326 | if (cft->write_u64) { |
478988d3 | 2327 | u64 val = simple_strtoull(strstrip(buffer), &end, 0); |
e73d2c61 PM |
2328 | if (*end) |
2329 | return -EINVAL; | |
2330 | retval = cft->write_u64(cgrp, cft, val); | |
2331 | } else { | |
478988d3 | 2332 | s64 val = simple_strtoll(strstrip(buffer), &end, 0); |
e73d2c61 PM |
2333 | if (*end) |
2334 | return -EINVAL; | |
2335 | retval = cft->write_s64(cgrp, cft, val); | |
2336 | } | |
355e0c48 PM |
2337 | if (!retval) |
2338 | retval = nbytes; | |
2339 | return retval; | |
2340 | } | |
2341 | ||
db3b1497 PM |
2342 | static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft, |
2343 | struct file *file, | |
2344 | const char __user *userbuf, | |
2345 | size_t nbytes, loff_t *unused_ppos) | |
2346 | { | |
84eea842 | 2347 | char local_buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
db3b1497 PM |
2348 | int retval = 0; |
2349 | size_t max_bytes = cft->max_write_len; | |
2350 | char *buffer = local_buffer; | |
2351 | ||
2352 | if (!max_bytes) | |
2353 | max_bytes = sizeof(local_buffer) - 1; | |
2354 | if (nbytes >= max_bytes) | |
2355 | return -E2BIG; | |
2356 | /* Allocate a dynamic buffer if we need one */ | |
2357 | if (nbytes >= sizeof(local_buffer)) { | |
2358 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); | |
2359 | if (buffer == NULL) | |
2360 | return -ENOMEM; | |
2361 | } | |
5a3eb9f6 LZ |
2362 | if (nbytes && copy_from_user(buffer, userbuf, nbytes)) { |
2363 | retval = -EFAULT; | |
2364 | goto out; | |
2365 | } | |
db3b1497 PM |
2366 | |
2367 | buffer[nbytes] = 0; /* nul-terminate */ | |
478988d3 | 2368 | retval = cft->write_string(cgrp, cft, strstrip(buffer)); |
db3b1497 PM |
2369 | if (!retval) |
2370 | retval = nbytes; | |
5a3eb9f6 | 2371 | out: |
db3b1497 PM |
2372 | if (buffer != local_buffer) |
2373 | kfree(buffer); | |
2374 | return retval; | |
2375 | } | |
2376 | ||
ddbcc7e8 PM |
2377 | static ssize_t cgroup_file_write(struct file *file, const char __user *buf, |
2378 | size_t nbytes, loff_t *ppos) | |
2379 | { | |
2380 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 2381 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 2382 | |
75139b82 | 2383 | if (cgroup_is_removed(cgrp)) |
ddbcc7e8 | 2384 | return -ENODEV; |
355e0c48 | 2385 | if (cft->write) |
bd89aabc | 2386 | return cft->write(cgrp, cft, file, buf, nbytes, ppos); |
e73d2c61 PM |
2387 | if (cft->write_u64 || cft->write_s64) |
2388 | return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos); | |
db3b1497 PM |
2389 | if (cft->write_string) |
2390 | return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos); | |
d447ea2f PE |
2391 | if (cft->trigger) { |
2392 | int ret = cft->trigger(cgrp, (unsigned int)cft->private); | |
2393 | return ret ? ret : nbytes; | |
2394 | } | |
355e0c48 | 2395 | return -EINVAL; |
ddbcc7e8 PM |
2396 | } |
2397 | ||
f4c753b7 PM |
2398 | static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft, |
2399 | struct file *file, | |
2400 | char __user *buf, size_t nbytes, | |
2401 | loff_t *ppos) | |
ddbcc7e8 | 2402 | { |
84eea842 | 2403 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
f4c753b7 | 2404 | u64 val = cft->read_u64(cgrp, cft); |
ddbcc7e8 PM |
2405 | int len = sprintf(tmp, "%llu\n", (unsigned long long) val); |
2406 | ||
2407 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
2408 | } | |
2409 | ||
e73d2c61 PM |
2410 | static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft, |
2411 | struct file *file, | |
2412 | char __user *buf, size_t nbytes, | |
2413 | loff_t *ppos) | |
2414 | { | |
84eea842 | 2415 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
e73d2c61 PM |
2416 | s64 val = cft->read_s64(cgrp, cft); |
2417 | int len = sprintf(tmp, "%lld\n", (long long) val); | |
2418 | ||
2419 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
2420 | } | |
2421 | ||
ddbcc7e8 PM |
2422 | static ssize_t cgroup_file_read(struct file *file, char __user *buf, |
2423 | size_t nbytes, loff_t *ppos) | |
2424 | { | |
2425 | struct cftype *cft = __d_cft(file->f_dentry); | |
bd89aabc | 2426 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
ddbcc7e8 | 2427 | |
75139b82 | 2428 | if (cgroup_is_removed(cgrp)) |
ddbcc7e8 PM |
2429 | return -ENODEV; |
2430 | ||
2431 | if (cft->read) | |
bd89aabc | 2432 | return cft->read(cgrp, cft, file, buf, nbytes, ppos); |
f4c753b7 PM |
2433 | if (cft->read_u64) |
2434 | return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos); | |
e73d2c61 PM |
2435 | if (cft->read_s64) |
2436 | return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos); | |
ddbcc7e8 PM |
2437 | return -EINVAL; |
2438 | } | |
2439 | ||
91796569 PM |
2440 | /* |
2441 | * seqfile ops/methods for returning structured data. Currently just | |
2442 | * supports string->u64 maps, but can be extended in future. | |
2443 | */ | |
2444 | ||
2445 | struct cgroup_seqfile_state { | |
2446 | struct cftype *cft; | |
2447 | struct cgroup *cgroup; | |
2448 | }; | |
2449 | ||
2450 | static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value) | |
2451 | { | |
2452 | struct seq_file *sf = cb->state; | |
2453 | return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value); | |
2454 | } | |
2455 | ||
2456 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) | |
2457 | { | |
2458 | struct cgroup_seqfile_state *state = m->private; | |
2459 | struct cftype *cft = state->cft; | |
29486df3 SH |
2460 | if (cft->read_map) { |
2461 | struct cgroup_map_cb cb = { | |
2462 | .fill = cgroup_map_add, | |
2463 | .state = m, | |
2464 | }; | |
2465 | return cft->read_map(state->cgroup, cft, &cb); | |
2466 | } | |
2467 | return cft->read_seq_string(state->cgroup, cft, m); | |
91796569 PM |
2468 | } |
2469 | ||
96930a63 | 2470 | static int cgroup_seqfile_release(struct inode *inode, struct file *file) |
91796569 PM |
2471 | { |
2472 | struct seq_file *seq = file->private_data; | |
2473 | kfree(seq->private); | |
2474 | return single_release(inode, file); | |
2475 | } | |
2476 | ||
828c0950 | 2477 | static const struct file_operations cgroup_seqfile_operations = { |
91796569 | 2478 | .read = seq_read, |
e788e066 | 2479 | .write = cgroup_file_write, |
91796569 PM |
2480 | .llseek = seq_lseek, |
2481 | .release = cgroup_seqfile_release, | |
2482 | }; | |
2483 | ||
ddbcc7e8 PM |
2484 | static int cgroup_file_open(struct inode *inode, struct file *file) |
2485 | { | |
2486 | int err; | |
2487 | struct cftype *cft; | |
2488 | ||
2489 | err = generic_file_open(inode, file); | |
2490 | if (err) | |
2491 | return err; | |
ddbcc7e8 | 2492 | cft = __d_cft(file->f_dentry); |
75139b82 | 2493 | |
29486df3 | 2494 | if (cft->read_map || cft->read_seq_string) { |
91796569 PM |
2495 | struct cgroup_seqfile_state *state = |
2496 | kzalloc(sizeof(*state), GFP_USER); | |
2497 | if (!state) | |
2498 | return -ENOMEM; | |
2499 | state->cft = cft; | |
2500 | state->cgroup = __d_cgrp(file->f_dentry->d_parent); | |
2501 | file->f_op = &cgroup_seqfile_operations; | |
2502 | err = single_open(file, cgroup_seqfile_show, state); | |
2503 | if (err < 0) | |
2504 | kfree(state); | |
2505 | } else if (cft->open) | |
ddbcc7e8 PM |
2506 | err = cft->open(inode, file); |
2507 | else | |
2508 | err = 0; | |
2509 | ||
2510 | return err; | |
2511 | } | |
2512 | ||
2513 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
2514 | { | |
2515 | struct cftype *cft = __d_cft(file->f_dentry); | |
2516 | if (cft->release) | |
2517 | return cft->release(inode, file); | |
2518 | return 0; | |
2519 | } | |
2520 | ||
2521 | /* | |
2522 | * cgroup_rename - Only allow simple rename of directories in place. | |
2523 | */ | |
2524 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
2525 | struct inode *new_dir, struct dentry *new_dentry) | |
2526 | { | |
2527 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) | |
2528 | return -ENOTDIR; | |
2529 | if (new_dentry->d_inode) | |
2530 | return -EEXIST; | |
2531 | if (old_dir != new_dir) | |
2532 | return -EIO; | |
2533 | return simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
2534 | } | |
2535 | ||
03b1cde6 AR |
2536 | static struct simple_xattrs *__d_xattrs(struct dentry *dentry) |
2537 | { | |
2538 | if (S_ISDIR(dentry->d_inode->i_mode)) | |
2539 | return &__d_cgrp(dentry)->xattrs; | |
2540 | else | |
2541 | return &__d_cft(dentry)->xattrs; | |
2542 | } | |
2543 | ||
2544 | static inline int xattr_enabled(struct dentry *dentry) | |
2545 | { | |
2546 | struct cgroupfs_root *root = dentry->d_sb->s_fs_info; | |
2547 | return test_bit(ROOT_XATTR, &root->flags); | |
2548 | } | |
2549 | ||
2550 | static bool is_valid_xattr(const char *name) | |
2551 | { | |
2552 | if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) || | |
2553 | !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN)) | |
2554 | return true; | |
2555 | return false; | |
2556 | } | |
2557 | ||
2558 | static int cgroup_setxattr(struct dentry *dentry, const char *name, | |
2559 | const void *val, size_t size, int flags) | |
2560 | { | |
2561 | if (!xattr_enabled(dentry)) | |
2562 | return -EOPNOTSUPP; | |
2563 | if (!is_valid_xattr(name)) | |
2564 | return -EINVAL; | |
2565 | return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags); | |
2566 | } | |
2567 | ||
2568 | static int cgroup_removexattr(struct dentry *dentry, const char *name) | |
2569 | { | |
2570 | if (!xattr_enabled(dentry)) | |
2571 | return -EOPNOTSUPP; | |
2572 | if (!is_valid_xattr(name)) | |
2573 | return -EINVAL; | |
2574 | return simple_xattr_remove(__d_xattrs(dentry), name); | |
2575 | } | |
2576 | ||
2577 | static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name, | |
2578 | void *buf, size_t size) | |
2579 | { | |
2580 | if (!xattr_enabled(dentry)) | |
2581 | return -EOPNOTSUPP; | |
2582 | if (!is_valid_xattr(name)) | |
2583 | return -EINVAL; | |
2584 | return simple_xattr_get(__d_xattrs(dentry), name, buf, size); | |
2585 | } | |
2586 | ||
2587 | static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size) | |
2588 | { | |
2589 | if (!xattr_enabled(dentry)) | |
2590 | return -EOPNOTSUPP; | |
2591 | return simple_xattr_list(__d_xattrs(dentry), buf, size); | |
2592 | } | |
2593 | ||
828c0950 | 2594 | static const struct file_operations cgroup_file_operations = { |
ddbcc7e8 PM |
2595 | .read = cgroup_file_read, |
2596 | .write = cgroup_file_write, | |
2597 | .llseek = generic_file_llseek, | |
2598 | .open = cgroup_file_open, | |
2599 | .release = cgroup_file_release, | |
2600 | }; | |
2601 | ||
03b1cde6 AR |
2602 | static const struct inode_operations cgroup_file_inode_operations = { |
2603 | .setxattr = cgroup_setxattr, | |
2604 | .getxattr = cgroup_getxattr, | |
2605 | .listxattr = cgroup_listxattr, | |
2606 | .removexattr = cgroup_removexattr, | |
2607 | }; | |
2608 | ||
6e1d5dcc | 2609 | static const struct inode_operations cgroup_dir_inode_operations = { |
c72a04e3 | 2610 | .lookup = cgroup_lookup, |
ddbcc7e8 PM |
2611 | .mkdir = cgroup_mkdir, |
2612 | .rmdir = cgroup_rmdir, | |
2613 | .rename = cgroup_rename, | |
03b1cde6 AR |
2614 | .setxattr = cgroup_setxattr, |
2615 | .getxattr = cgroup_getxattr, | |
2616 | .listxattr = cgroup_listxattr, | |
2617 | .removexattr = cgroup_removexattr, | |
ddbcc7e8 PM |
2618 | }; |
2619 | ||
00cd8dd3 | 2620 | static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) |
c72a04e3 AV |
2621 | { |
2622 | if (dentry->d_name.len > NAME_MAX) | |
2623 | return ERR_PTR(-ENAMETOOLONG); | |
2624 | d_add(dentry, NULL); | |
2625 | return NULL; | |
2626 | } | |
2627 | ||
0dea1168 KS |
2628 | /* |
2629 | * Check if a file is a control file | |
2630 | */ | |
2631 | static inline struct cftype *__file_cft(struct file *file) | |
2632 | { | |
2633 | if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations) | |
2634 | return ERR_PTR(-EINVAL); | |
2635 | return __d_cft(file->f_dentry); | |
2636 | } | |
2637 | ||
a5e7ed32 | 2638 | static int cgroup_create_file(struct dentry *dentry, umode_t mode, |
5adcee1d NP |
2639 | struct super_block *sb) |
2640 | { | |
ddbcc7e8 PM |
2641 | struct inode *inode; |
2642 | ||
2643 | if (!dentry) | |
2644 | return -ENOENT; | |
2645 | if (dentry->d_inode) | |
2646 | return -EEXIST; | |
2647 | ||
2648 | inode = cgroup_new_inode(mode, sb); | |
2649 | if (!inode) | |
2650 | return -ENOMEM; | |
2651 | ||
2652 | if (S_ISDIR(mode)) { | |
2653 | inode->i_op = &cgroup_dir_inode_operations; | |
2654 | inode->i_fop = &simple_dir_operations; | |
2655 | ||
2656 | /* start off with i_nlink == 2 (for "." entry) */ | |
2657 | inc_nlink(inode); | |
28fd6f30 | 2658 | inc_nlink(dentry->d_parent->d_inode); |
ddbcc7e8 | 2659 | |
b8a2df6a TH |
2660 | /* |
2661 | * Control reaches here with cgroup_mutex held. | |
2662 | * @inode->i_mutex should nest outside cgroup_mutex but we | |
2663 | * want to populate it immediately without releasing | |
2664 | * cgroup_mutex. As @inode isn't visible to anyone else | |
2665 | * yet, trylock will always succeed without affecting | |
2666 | * lockdep checks. | |
2667 | */ | |
2668 | WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex)); | |
ddbcc7e8 PM |
2669 | } else if (S_ISREG(mode)) { |
2670 | inode->i_size = 0; | |
2671 | inode->i_fop = &cgroup_file_operations; | |
03b1cde6 | 2672 | inode->i_op = &cgroup_file_inode_operations; |
ddbcc7e8 | 2673 | } |
ddbcc7e8 PM |
2674 | d_instantiate(dentry, inode); |
2675 | dget(dentry); /* Extra count - pin the dentry in core */ | |
2676 | return 0; | |
2677 | } | |
2678 | ||
099fca32 LZ |
2679 | /** |
2680 | * cgroup_file_mode - deduce file mode of a control file | |
2681 | * @cft: the control file in question | |
2682 | * | |
2683 | * returns cft->mode if ->mode is not 0 | |
2684 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
2685 | * returns S_IRUGO if it has only a read handler | |
2686 | * returns S_IWUSR if it has only a write hander | |
2687 | */ | |
a5e7ed32 | 2688 | static umode_t cgroup_file_mode(const struct cftype *cft) |
099fca32 | 2689 | { |
a5e7ed32 | 2690 | umode_t mode = 0; |
099fca32 LZ |
2691 | |
2692 | if (cft->mode) | |
2693 | return cft->mode; | |
2694 | ||
2695 | if (cft->read || cft->read_u64 || cft->read_s64 || | |
2696 | cft->read_map || cft->read_seq_string) | |
2697 | mode |= S_IRUGO; | |
2698 | ||
2699 | if (cft->write || cft->write_u64 || cft->write_s64 || | |
2700 | cft->write_string || cft->trigger) | |
2701 | mode |= S_IWUSR; | |
2702 | ||
2703 | return mode; | |
2704 | } | |
2705 | ||
db0416b6 | 2706 | static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys, |
03b1cde6 | 2707 | struct cftype *cft) |
ddbcc7e8 | 2708 | { |
bd89aabc | 2709 | struct dentry *dir = cgrp->dentry; |
05ef1d7c | 2710 | struct cgroup *parent = __d_cgrp(dir); |
ddbcc7e8 | 2711 | struct dentry *dentry; |
05ef1d7c | 2712 | struct cfent *cfe; |
ddbcc7e8 | 2713 | int error; |
a5e7ed32 | 2714 | umode_t mode; |
ddbcc7e8 | 2715 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; |
8e3f6541 | 2716 | |
03b1cde6 AR |
2717 | simple_xattrs_init(&cft->xattrs); |
2718 | ||
8e3f6541 TH |
2719 | /* does @cft->flags tell us to skip creation on @cgrp? */ |
2720 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) | |
2721 | return 0; | |
2722 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2723 | return 0; | |
2724 | ||
bd89aabc | 2725 | if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) { |
ddbcc7e8 PM |
2726 | strcpy(name, subsys->name); |
2727 | strcat(name, "."); | |
2728 | } | |
2729 | strcat(name, cft->name); | |
05ef1d7c | 2730 | |
ddbcc7e8 | 2731 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); |
05ef1d7c TH |
2732 | |
2733 | cfe = kzalloc(sizeof(*cfe), GFP_KERNEL); | |
2734 | if (!cfe) | |
2735 | return -ENOMEM; | |
2736 | ||
ddbcc7e8 | 2737 | dentry = lookup_one_len(name, dir, strlen(name)); |
05ef1d7c | 2738 | if (IS_ERR(dentry)) { |
ddbcc7e8 | 2739 | error = PTR_ERR(dentry); |
05ef1d7c TH |
2740 | goto out; |
2741 | } | |
2742 | ||
2743 | mode = cgroup_file_mode(cft); | |
2744 | error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb); | |
2745 | if (!error) { | |
2746 | cfe->type = (void *)cft; | |
2747 | cfe->dentry = dentry; | |
2748 | dentry->d_fsdata = cfe; | |
2749 | list_add_tail(&cfe->node, &parent->files); | |
2750 | cfe = NULL; | |
2751 | } | |
2752 | dput(dentry); | |
2753 | out: | |
2754 | kfree(cfe); | |
ddbcc7e8 PM |
2755 | return error; |
2756 | } | |
2757 | ||
79578621 | 2758 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys, |
03b1cde6 | 2759 | struct cftype cfts[], bool is_add) |
ddbcc7e8 | 2760 | { |
03b1cde6 | 2761 | struct cftype *cft; |
db0416b6 TH |
2762 | int err, ret = 0; |
2763 | ||
2764 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
79578621 TH |
2765 | if (is_add) |
2766 | err = cgroup_add_file(cgrp, subsys, cft); | |
2767 | else | |
2768 | err = cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2769 | if (err) { |
79578621 TH |
2770 | pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n", |
2771 | is_add ? "add" : "remove", cft->name, err); | |
db0416b6 TH |
2772 | ret = err; |
2773 | } | |
ddbcc7e8 | 2774 | } |
db0416b6 | 2775 | return ret; |
ddbcc7e8 PM |
2776 | } |
2777 | ||
8e3f6541 TH |
2778 | static DEFINE_MUTEX(cgroup_cft_mutex); |
2779 | ||
2780 | static void cgroup_cfts_prepare(void) | |
2781 | __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex) | |
2782 | { | |
2783 | /* | |
2784 | * Thanks to the entanglement with vfs inode locking, we can't walk | |
2785 | * the existing cgroups under cgroup_mutex and create files. | |
2786 | * Instead, we increment reference on all cgroups and build list of | |
2787 | * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure | |
2788 | * exclusive access to the field. | |
2789 | */ | |
2790 | mutex_lock(&cgroup_cft_mutex); | |
2791 | mutex_lock(&cgroup_mutex); | |
2792 | } | |
2793 | ||
2794 | static void cgroup_cfts_commit(struct cgroup_subsys *ss, | |
03b1cde6 | 2795 | struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2796 | __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex) |
2797 | { | |
2798 | LIST_HEAD(pending); | |
2799 | struct cgroup *cgrp, *n; | |
8e3f6541 TH |
2800 | |
2801 | /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */ | |
2802 | if (cfts && ss->root != &rootnode) { | |
2803 | list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) { | |
2804 | dget(cgrp->dentry); | |
2805 | list_add_tail(&cgrp->cft_q_node, &pending); | |
2806 | } | |
2807 | } | |
2808 | ||
2809 | mutex_unlock(&cgroup_mutex); | |
2810 | ||
2811 | /* | |
2812 | * All new cgroups will see @cfts update on @ss->cftsets. Add/rm | |
2813 | * files for all cgroups which were created before. | |
2814 | */ | |
2815 | list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) { | |
2816 | struct inode *inode = cgrp->dentry->d_inode; | |
2817 | ||
2818 | mutex_lock(&inode->i_mutex); | |
2819 | mutex_lock(&cgroup_mutex); | |
2820 | if (!cgroup_is_removed(cgrp)) | |
79578621 | 2821 | cgroup_addrm_files(cgrp, ss, cfts, is_add); |
8e3f6541 TH |
2822 | mutex_unlock(&cgroup_mutex); |
2823 | mutex_unlock(&inode->i_mutex); | |
2824 | ||
2825 | list_del_init(&cgrp->cft_q_node); | |
2826 | dput(cgrp->dentry); | |
2827 | } | |
2828 | ||
2829 | mutex_unlock(&cgroup_cft_mutex); | |
2830 | } | |
2831 | ||
2832 | /** | |
2833 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2834 | * @ss: target cgroup subsystem | |
2835 | * @cfts: zero-length name terminated array of cftypes | |
2836 | * | |
2837 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2838 | * existing cgroups to which @ss is attached and all future cgroups will | |
2839 | * have them too. This function can be called anytime whether @ss is | |
2840 | * attached or not. | |
2841 | * | |
2842 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2843 | * function currently returns 0 as long as @cfts registration is successful | |
2844 | * even if some file creation attempts on existing cgroups fail. | |
2845 | */ | |
03b1cde6 | 2846 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 TH |
2847 | { |
2848 | struct cftype_set *set; | |
2849 | ||
2850 | set = kzalloc(sizeof(*set), GFP_KERNEL); | |
2851 | if (!set) | |
2852 | return -ENOMEM; | |
2853 | ||
2854 | cgroup_cfts_prepare(); | |
2855 | set->cfts = cfts; | |
2856 | list_add_tail(&set->node, &ss->cftsets); | |
79578621 | 2857 | cgroup_cfts_commit(ss, cfts, true); |
8e3f6541 TH |
2858 | |
2859 | return 0; | |
2860 | } | |
2861 | EXPORT_SYMBOL_GPL(cgroup_add_cftypes); | |
2862 | ||
79578621 TH |
2863 | /** |
2864 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
2865 | * @ss: target cgroup subsystem | |
2866 | * @cfts: zero-length name terminated array of cftypes | |
2867 | * | |
2868 | * Unregister @cfts from @ss. Files described by @cfts are removed from | |
2869 | * all existing cgroups to which @ss is attached and all future cgroups | |
2870 | * won't have them either. This function can be called anytime whether @ss | |
2871 | * is attached or not. | |
2872 | * | |
2873 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2874 | * registered with @ss. | |
2875 | */ | |
03b1cde6 | 2876 | int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
79578621 TH |
2877 | { |
2878 | struct cftype_set *set; | |
2879 | ||
2880 | cgroup_cfts_prepare(); | |
2881 | ||
2882 | list_for_each_entry(set, &ss->cftsets, node) { | |
2883 | if (set->cfts == cfts) { | |
2884 | list_del_init(&set->node); | |
2885 | cgroup_cfts_commit(ss, cfts, false); | |
2886 | return 0; | |
2887 | } | |
2888 | } | |
2889 | ||
2890 | cgroup_cfts_commit(ss, NULL, false); | |
2891 | return -ENOENT; | |
2892 | } | |
2893 | ||
a043e3b2 LZ |
2894 | /** |
2895 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2896 | * @cgrp: the cgroup in question | |
2897 | * | |
2898 | * Return the number of tasks in the cgroup. | |
2899 | */ | |
bd89aabc | 2900 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2901 | { |
2902 | int count = 0; | |
71cbb949 | 2903 | struct cg_cgroup_link *link; |
817929ec PM |
2904 | |
2905 | read_lock(&css_set_lock); | |
71cbb949 | 2906 | list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) { |
146aa1bd | 2907 | count += atomic_read(&link->cg->refcount); |
817929ec PM |
2908 | } |
2909 | read_unlock(&css_set_lock); | |
bbcb81d0 PM |
2910 | return count; |
2911 | } | |
2912 | ||
817929ec PM |
2913 | /* |
2914 | * Advance a list_head iterator. The iterator should be positioned at | |
2915 | * the start of a css_set | |
2916 | */ | |
bd89aabc | 2917 | static void cgroup_advance_iter(struct cgroup *cgrp, |
7717f7ba | 2918 | struct cgroup_iter *it) |
817929ec PM |
2919 | { |
2920 | struct list_head *l = it->cg_link; | |
2921 | struct cg_cgroup_link *link; | |
2922 | struct css_set *cg; | |
2923 | ||
2924 | /* Advance to the next non-empty css_set */ | |
2925 | do { | |
2926 | l = l->next; | |
bd89aabc | 2927 | if (l == &cgrp->css_sets) { |
817929ec PM |
2928 | it->cg_link = NULL; |
2929 | return; | |
2930 | } | |
bd89aabc | 2931 | link = list_entry(l, struct cg_cgroup_link, cgrp_link_list); |
817929ec PM |
2932 | cg = link->cg; |
2933 | } while (list_empty(&cg->tasks)); | |
2934 | it->cg_link = l; | |
2935 | it->task = cg->tasks.next; | |
2936 | } | |
2937 | ||
31a7df01 CW |
2938 | /* |
2939 | * To reduce the fork() overhead for systems that are not actually | |
2940 | * using their cgroups capability, we don't maintain the lists running | |
2941 | * through each css_set to its tasks until we see the list actually | |
2942 | * used - in other words after the first call to cgroup_iter_start(). | |
31a7df01 | 2943 | */ |
3df91fe3 | 2944 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2945 | { |
2946 | struct task_struct *p, *g; | |
2947 | write_lock(&css_set_lock); | |
2948 | use_task_css_set_links = 1; | |
3ce3230a FW |
2949 | /* |
2950 | * We need tasklist_lock because RCU is not safe against | |
2951 | * while_each_thread(). Besides, a forking task that has passed | |
2952 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
2953 | * is not guaranteed to have its child immediately visible in the | |
2954 | * tasklist if we walk through it with RCU. | |
2955 | */ | |
2956 | read_lock(&tasklist_lock); | |
31a7df01 CW |
2957 | do_each_thread(g, p) { |
2958 | task_lock(p); | |
0e04388f LZ |
2959 | /* |
2960 | * We should check if the process is exiting, otherwise | |
2961 | * it will race with cgroup_exit() in that the list | |
2962 | * entry won't be deleted though the process has exited. | |
2963 | */ | |
2964 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
31a7df01 CW |
2965 | list_add(&p->cg_list, &p->cgroups->tasks); |
2966 | task_unlock(p); | |
2967 | } while_each_thread(g, p); | |
3ce3230a | 2968 | read_unlock(&tasklist_lock); |
31a7df01 CW |
2969 | write_unlock(&css_set_lock); |
2970 | } | |
2971 | ||
574bd9f7 TH |
2972 | /** |
2973 | * cgroup_next_descendant_pre - find the next descendant for pre-order walk | |
2974 | * @pos: the current position (%NULL to initiate traversal) | |
2975 | * @cgroup: cgroup whose descendants to walk | |
2976 | * | |
2977 | * To be used by cgroup_for_each_descendant_pre(). Find the next | |
2978 | * descendant to visit for pre-order traversal of @cgroup's descendants. | |
2979 | */ | |
2980 | struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos, | |
2981 | struct cgroup *cgroup) | |
2982 | { | |
2983 | struct cgroup *next; | |
2984 | ||
2985 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
2986 | ||
2987 | /* if first iteration, pretend we just visited @cgroup */ | |
2988 | if (!pos) { | |
2989 | if (list_empty(&cgroup->children)) | |
2990 | return NULL; | |
2991 | pos = cgroup; | |
2992 | } | |
2993 | ||
2994 | /* visit the first child if exists */ | |
2995 | next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling); | |
2996 | if (next) | |
2997 | return next; | |
2998 | ||
2999 | /* no child, visit my or the closest ancestor's next sibling */ | |
3000 | do { | |
3001 | next = list_entry_rcu(pos->sibling.next, struct cgroup, | |
3002 | sibling); | |
3003 | if (&next->sibling != &pos->parent->children) | |
3004 | return next; | |
3005 | ||
3006 | pos = pos->parent; | |
3007 | } while (pos != cgroup); | |
3008 | ||
3009 | return NULL; | |
3010 | } | |
3011 | EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre); | |
3012 | ||
3013 | static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos) | |
3014 | { | |
3015 | struct cgroup *last; | |
3016 | ||
3017 | do { | |
3018 | last = pos; | |
3019 | pos = list_first_or_null_rcu(&pos->children, struct cgroup, | |
3020 | sibling); | |
3021 | } while (pos); | |
3022 | ||
3023 | return last; | |
3024 | } | |
3025 | ||
3026 | /** | |
3027 | * cgroup_next_descendant_post - find the next descendant for post-order walk | |
3028 | * @pos: the current position (%NULL to initiate traversal) | |
3029 | * @cgroup: cgroup whose descendants to walk | |
3030 | * | |
3031 | * To be used by cgroup_for_each_descendant_post(). Find the next | |
3032 | * descendant to visit for post-order traversal of @cgroup's descendants. | |
3033 | */ | |
3034 | struct cgroup *cgroup_next_descendant_post(struct cgroup *pos, | |
3035 | struct cgroup *cgroup) | |
3036 | { | |
3037 | struct cgroup *next; | |
3038 | ||
3039 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
3040 | ||
3041 | /* if first iteration, visit the leftmost descendant */ | |
3042 | if (!pos) { | |
3043 | next = cgroup_leftmost_descendant(cgroup); | |
3044 | return next != cgroup ? next : NULL; | |
3045 | } | |
3046 | ||
3047 | /* if there's an unvisited sibling, visit its leftmost descendant */ | |
3048 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); | |
3049 | if (&next->sibling != &pos->parent->children) | |
3050 | return cgroup_leftmost_descendant(next); | |
3051 | ||
3052 | /* no sibling left, visit parent */ | |
3053 | next = pos->parent; | |
3054 | return next != cgroup ? next : NULL; | |
3055 | } | |
3056 | EXPORT_SYMBOL_GPL(cgroup_next_descendant_post); | |
3057 | ||
bd89aabc | 3058 | void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it) |
c6ca5750 | 3059 | __acquires(css_set_lock) |
817929ec PM |
3060 | { |
3061 | /* | |
3062 | * The first time anyone tries to iterate across a cgroup, | |
3063 | * we need to enable the list linking each css_set to its | |
3064 | * tasks, and fix up all existing tasks. | |
3065 | */ | |
31a7df01 CW |
3066 | if (!use_task_css_set_links) |
3067 | cgroup_enable_task_cg_lists(); | |
3068 | ||
817929ec | 3069 | read_lock(&css_set_lock); |
bd89aabc PM |
3070 | it->cg_link = &cgrp->css_sets; |
3071 | cgroup_advance_iter(cgrp, it); | |
817929ec PM |
3072 | } |
3073 | ||
bd89aabc | 3074 | struct task_struct *cgroup_iter_next(struct cgroup *cgrp, |
817929ec PM |
3075 | struct cgroup_iter *it) |
3076 | { | |
3077 | struct task_struct *res; | |
3078 | struct list_head *l = it->task; | |
2019f634 | 3079 | struct cg_cgroup_link *link; |
817929ec PM |
3080 | |
3081 | /* If the iterator cg is NULL, we have no tasks */ | |
3082 | if (!it->cg_link) | |
3083 | return NULL; | |
3084 | res = list_entry(l, struct task_struct, cg_list); | |
3085 | /* Advance iterator to find next entry */ | |
3086 | l = l->next; | |
2019f634 LJ |
3087 | link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list); |
3088 | if (l == &link->cg->tasks) { | |
817929ec PM |
3089 | /* We reached the end of this task list - move on to |
3090 | * the next cg_cgroup_link */ | |
bd89aabc | 3091 | cgroup_advance_iter(cgrp, it); |
817929ec PM |
3092 | } else { |
3093 | it->task = l; | |
3094 | } | |
3095 | return res; | |
3096 | } | |
3097 | ||
bd89aabc | 3098 | void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it) |
c6ca5750 | 3099 | __releases(css_set_lock) |
817929ec PM |
3100 | { |
3101 | read_unlock(&css_set_lock); | |
3102 | } | |
3103 | ||
31a7df01 CW |
3104 | static inline int started_after_time(struct task_struct *t1, |
3105 | struct timespec *time, | |
3106 | struct task_struct *t2) | |
3107 | { | |
3108 | int start_diff = timespec_compare(&t1->start_time, time); | |
3109 | if (start_diff > 0) { | |
3110 | return 1; | |
3111 | } else if (start_diff < 0) { | |
3112 | return 0; | |
3113 | } else { | |
3114 | /* | |
3115 | * Arbitrarily, if two processes started at the same | |
3116 | * time, we'll say that the lower pointer value | |
3117 | * started first. Note that t2 may have exited by now | |
3118 | * so this may not be a valid pointer any longer, but | |
3119 | * that's fine - it still serves to distinguish | |
3120 | * between two tasks started (effectively) simultaneously. | |
3121 | */ | |
3122 | return t1 > t2; | |
3123 | } | |
3124 | } | |
3125 | ||
3126 | /* | |
3127 | * This function is a callback from heap_insert() and is used to order | |
3128 | * the heap. | |
3129 | * In this case we order the heap in descending task start time. | |
3130 | */ | |
3131 | static inline int started_after(void *p1, void *p2) | |
3132 | { | |
3133 | struct task_struct *t1 = p1; | |
3134 | struct task_struct *t2 = p2; | |
3135 | return started_after_time(t1, &t2->start_time, t2); | |
3136 | } | |
3137 | ||
3138 | /** | |
3139 | * cgroup_scan_tasks - iterate though all the tasks in a cgroup | |
3140 | * @scan: struct cgroup_scanner containing arguments for the scan | |
3141 | * | |
3142 | * Arguments include pointers to callback functions test_task() and | |
3143 | * process_task(). | |
3144 | * Iterate through all the tasks in a cgroup, calling test_task() for each, | |
3145 | * and if it returns true, call process_task() for it also. | |
3146 | * The test_task pointer may be NULL, meaning always true (select all tasks). | |
3147 | * Effectively duplicates cgroup_iter_{start,next,end}() | |
3148 | * but does not lock css_set_lock for the call to process_task(). | |
3149 | * The struct cgroup_scanner may be embedded in any structure of the caller's | |
3150 | * creation. | |
3151 | * It is guaranteed that process_task() will act on every task that | |
3152 | * is a member of the cgroup for the duration of this call. This | |
3153 | * function may or may not call process_task() for tasks that exit | |
3154 | * or move to a different cgroup during the call, or are forked or | |
3155 | * move into the cgroup during the call. | |
3156 | * | |
3157 | * Note that test_task() may be called with locks held, and may in some | |
3158 | * situations be called multiple times for the same task, so it should | |
3159 | * be cheap. | |
3160 | * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been | |
3161 | * pre-allocated and will be used for heap operations (and its "gt" member will | |
3162 | * be overwritten), else a temporary heap will be used (allocation of which | |
3163 | * may cause this function to fail). | |
3164 | */ | |
3165 | int cgroup_scan_tasks(struct cgroup_scanner *scan) | |
3166 | { | |
3167 | int retval, i; | |
3168 | struct cgroup_iter it; | |
3169 | struct task_struct *p, *dropped; | |
3170 | /* Never dereference latest_task, since it's not refcounted */ | |
3171 | struct task_struct *latest_task = NULL; | |
3172 | struct ptr_heap tmp_heap; | |
3173 | struct ptr_heap *heap; | |
3174 | struct timespec latest_time = { 0, 0 }; | |
3175 | ||
3176 | if (scan->heap) { | |
3177 | /* The caller supplied our heap and pre-allocated its memory */ | |
3178 | heap = scan->heap; | |
3179 | heap->gt = &started_after; | |
3180 | } else { | |
3181 | /* We need to allocate our own heap memory */ | |
3182 | heap = &tmp_heap; | |
3183 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
3184 | if (retval) | |
3185 | /* cannot allocate the heap */ | |
3186 | return retval; | |
3187 | } | |
3188 | ||
3189 | again: | |
3190 | /* | |
3191 | * Scan tasks in the cgroup, using the scanner's "test_task" callback | |
3192 | * to determine which are of interest, and using the scanner's | |
3193 | * "process_task" callback to process any of them that need an update. | |
3194 | * Since we don't want to hold any locks during the task updates, | |
3195 | * gather tasks to be processed in a heap structure. | |
3196 | * The heap is sorted by descending task start time. | |
3197 | * If the statically-sized heap fills up, we overflow tasks that | |
3198 | * started later, and in future iterations only consider tasks that | |
3199 | * started after the latest task in the previous pass. This | |
3200 | * guarantees forward progress and that we don't miss any tasks. | |
3201 | */ | |
3202 | heap->size = 0; | |
3203 | cgroup_iter_start(scan->cg, &it); | |
3204 | while ((p = cgroup_iter_next(scan->cg, &it))) { | |
3205 | /* | |
3206 | * Only affect tasks that qualify per the caller's callback, | |
3207 | * if he provided one | |
3208 | */ | |
3209 | if (scan->test_task && !scan->test_task(p, scan)) | |
3210 | continue; | |
3211 | /* | |
3212 | * Only process tasks that started after the last task | |
3213 | * we processed | |
3214 | */ | |
3215 | if (!started_after_time(p, &latest_time, latest_task)) | |
3216 | continue; | |
3217 | dropped = heap_insert(heap, p); | |
3218 | if (dropped == NULL) { | |
3219 | /* | |
3220 | * The new task was inserted; the heap wasn't | |
3221 | * previously full | |
3222 | */ | |
3223 | get_task_struct(p); | |
3224 | } else if (dropped != p) { | |
3225 | /* | |
3226 | * The new task was inserted, and pushed out a | |
3227 | * different task | |
3228 | */ | |
3229 | get_task_struct(p); | |
3230 | put_task_struct(dropped); | |
3231 | } | |
3232 | /* | |
3233 | * Else the new task was newer than anything already in | |
3234 | * the heap and wasn't inserted | |
3235 | */ | |
3236 | } | |
3237 | cgroup_iter_end(scan->cg, &it); | |
3238 | ||
3239 | if (heap->size) { | |
3240 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 3241 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 3242 | if (i == 0) { |
4fe91d51 PJ |
3243 | latest_time = q->start_time; |
3244 | latest_task = q; | |
31a7df01 CW |
3245 | } |
3246 | /* Process the task per the caller's callback */ | |
4fe91d51 PJ |
3247 | scan->process_task(q, scan); |
3248 | put_task_struct(q); | |
31a7df01 CW |
3249 | } |
3250 | /* | |
3251 | * If we had to process any tasks at all, scan again | |
3252 | * in case some of them were in the middle of forking | |
3253 | * children that didn't get processed. | |
3254 | * Not the most efficient way to do it, but it avoids | |
3255 | * having to take callback_mutex in the fork path | |
3256 | */ | |
3257 | goto again; | |
3258 | } | |
3259 | if (heap == &tmp_heap) | |
3260 | heap_free(&tmp_heap); | |
3261 | return 0; | |
3262 | } | |
3263 | ||
bbcb81d0 | 3264 | /* |
102a775e | 3265 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3266 | * |
3267 | * Reading this file can return large amounts of data if a cgroup has | |
3268 | * *lots* of attached tasks. So it may need several calls to read(), | |
3269 | * but we cannot guarantee that the information we produce is correct | |
3270 | * unless we produce it entirely atomically. | |
3271 | * | |
bbcb81d0 | 3272 | */ |
bbcb81d0 | 3273 | |
24528255 LZ |
3274 | /* which pidlist file are we talking about? */ |
3275 | enum cgroup_filetype { | |
3276 | CGROUP_FILE_PROCS, | |
3277 | CGROUP_FILE_TASKS, | |
3278 | }; | |
3279 | ||
3280 | /* | |
3281 | * A pidlist is a list of pids that virtually represents the contents of one | |
3282 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3283 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3284 | * to the cgroup. | |
3285 | */ | |
3286 | struct cgroup_pidlist { | |
3287 | /* | |
3288 | * used to find which pidlist is wanted. doesn't change as long as | |
3289 | * this particular list stays in the list. | |
3290 | */ | |
3291 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3292 | /* array of xids */ | |
3293 | pid_t *list; | |
3294 | /* how many elements the above list has */ | |
3295 | int length; | |
3296 | /* how many files are using the current array */ | |
3297 | int use_count; | |
3298 | /* each of these stored in a list by its cgroup */ | |
3299 | struct list_head links; | |
3300 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3301 | struct cgroup *owner; | |
3302 | /* protects the other fields */ | |
3303 | struct rw_semaphore mutex; | |
3304 | }; | |
3305 | ||
d1d9fd33 BB |
3306 | /* |
3307 | * The following two functions "fix" the issue where there are more pids | |
3308 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3309 | * TODO: replace with a kernel-wide solution to this problem | |
3310 | */ | |
3311 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3312 | static void *pidlist_allocate(int count) | |
3313 | { | |
3314 | if (PIDLIST_TOO_LARGE(count)) | |
3315 | return vmalloc(count * sizeof(pid_t)); | |
3316 | else | |
3317 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3318 | } | |
3319 | static void pidlist_free(void *p) | |
3320 | { | |
3321 | if (is_vmalloc_addr(p)) | |
3322 | vfree(p); | |
3323 | else | |
3324 | kfree(p); | |
3325 | } | |
3326 | static void *pidlist_resize(void *p, int newcount) | |
3327 | { | |
3328 | void *newlist; | |
3329 | /* note: if new alloc fails, old p will still be valid either way */ | |
3330 | if (is_vmalloc_addr(p)) { | |
3331 | newlist = vmalloc(newcount * sizeof(pid_t)); | |
3332 | if (!newlist) | |
3333 | return NULL; | |
3334 | memcpy(newlist, p, newcount * sizeof(pid_t)); | |
3335 | vfree(p); | |
3336 | } else { | |
3337 | newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL); | |
3338 | } | |
3339 | return newlist; | |
3340 | } | |
3341 | ||
bbcb81d0 | 3342 | /* |
102a775e BB |
3343 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
3344 | * If the new stripped list is sufficiently smaller and there's enough memory | |
3345 | * to allocate a new buffer, will let go of the unneeded memory. Returns the | |
3346 | * number of unique elements. | |
bbcb81d0 | 3347 | */ |
102a775e BB |
3348 | /* is the size difference enough that we should re-allocate the array? */ |
3349 | #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new)) | |
3350 | static int pidlist_uniq(pid_t **p, int length) | |
bbcb81d0 | 3351 | { |
102a775e BB |
3352 | int src, dest = 1; |
3353 | pid_t *list = *p; | |
3354 | pid_t *newlist; | |
3355 | ||
3356 | /* | |
3357 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3358 | * edge cases first; no work needs to be done for either | |
3359 | */ | |
3360 | if (length == 0 || length == 1) | |
3361 | return length; | |
3362 | /* src and dest walk down the list; dest counts unique elements */ | |
3363 | for (src = 1; src < length; src++) { | |
3364 | /* find next unique element */ | |
3365 | while (list[src] == list[src-1]) { | |
3366 | src++; | |
3367 | if (src == length) | |
3368 | goto after; | |
3369 | } | |
3370 | /* dest always points to where the next unique element goes */ | |
3371 | list[dest] = list[src]; | |
3372 | dest++; | |
3373 | } | |
3374 | after: | |
3375 | /* | |
3376 | * if the length difference is large enough, we want to allocate a | |
3377 | * smaller buffer to save memory. if this fails due to out of memory, | |
3378 | * we'll just stay with what we've got. | |
3379 | */ | |
3380 | if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) { | |
d1d9fd33 | 3381 | newlist = pidlist_resize(list, dest); |
102a775e BB |
3382 | if (newlist) |
3383 | *p = newlist; | |
3384 | } | |
3385 | return dest; | |
3386 | } | |
3387 | ||
3388 | static int cmppid(const void *a, const void *b) | |
3389 | { | |
3390 | return *(pid_t *)a - *(pid_t *)b; | |
3391 | } | |
3392 | ||
72a8cb30 BB |
3393 | /* |
3394 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3395 | * returns with the lock on that pidlist already held, and takes care | |
3396 | * of the use count, or returns NULL with no locks held if we're out of | |
3397 | * memory. | |
3398 | */ | |
3399 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
3400 | enum cgroup_filetype type) | |
3401 | { | |
3402 | struct cgroup_pidlist *l; | |
3403 | /* don't need task_nsproxy() if we're looking at ourself */ | |
b70cc5fd LZ |
3404 | struct pid_namespace *ns = current->nsproxy->pid_ns; |
3405 | ||
72a8cb30 BB |
3406 | /* |
3407 | * We can't drop the pidlist_mutex before taking the l->mutex in case | |
3408 | * the last ref-holder is trying to remove l from the list at the same | |
3409 | * time. Holding the pidlist_mutex precludes somebody taking whichever | |
3410 | * list we find out from under us - compare release_pid_array(). | |
3411 | */ | |
3412 | mutex_lock(&cgrp->pidlist_mutex); | |
3413 | list_for_each_entry(l, &cgrp->pidlists, links) { | |
3414 | if (l->key.type == type && l->key.ns == ns) { | |
72a8cb30 BB |
3415 | /* make sure l doesn't vanish out from under us */ |
3416 | down_write(&l->mutex); | |
3417 | mutex_unlock(&cgrp->pidlist_mutex); | |
72a8cb30 BB |
3418 | return l; |
3419 | } | |
3420 | } | |
3421 | /* entry not found; create a new one */ | |
3422 | l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); | |
3423 | if (!l) { | |
3424 | mutex_unlock(&cgrp->pidlist_mutex); | |
72a8cb30 BB |
3425 | return l; |
3426 | } | |
3427 | init_rwsem(&l->mutex); | |
3428 | down_write(&l->mutex); | |
3429 | l->key.type = type; | |
b70cc5fd | 3430 | l->key.ns = get_pid_ns(ns); |
72a8cb30 BB |
3431 | l->use_count = 0; /* don't increment here */ |
3432 | l->list = NULL; | |
3433 | l->owner = cgrp; | |
3434 | list_add(&l->links, &cgrp->pidlists); | |
3435 | mutex_unlock(&cgrp->pidlist_mutex); | |
3436 | return l; | |
3437 | } | |
3438 | ||
102a775e BB |
3439 | /* |
3440 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3441 | */ | |
72a8cb30 BB |
3442 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3443 | struct cgroup_pidlist **lp) | |
102a775e BB |
3444 | { |
3445 | pid_t *array; | |
3446 | int length; | |
3447 | int pid, n = 0; /* used for populating the array */ | |
817929ec PM |
3448 | struct cgroup_iter it; |
3449 | struct task_struct *tsk; | |
102a775e BB |
3450 | struct cgroup_pidlist *l; |
3451 | ||
3452 | /* | |
3453 | * If cgroup gets more users after we read count, we won't have | |
3454 | * enough space - tough. This race is indistinguishable to the | |
3455 | * caller from the case that the additional cgroup users didn't | |
3456 | * show up until sometime later on. | |
3457 | */ | |
3458 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3459 | array = pidlist_allocate(length); |
102a775e BB |
3460 | if (!array) |
3461 | return -ENOMEM; | |
3462 | /* now, populate the array */ | |
bd89aabc PM |
3463 | cgroup_iter_start(cgrp, &it); |
3464 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
102a775e | 3465 | if (unlikely(n == length)) |
817929ec | 3466 | break; |
102a775e | 3467 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3468 | if (type == CGROUP_FILE_PROCS) |
3469 | pid = task_tgid_vnr(tsk); | |
3470 | else | |
3471 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3472 | if (pid > 0) /* make sure to only use valid results */ |
3473 | array[n++] = pid; | |
817929ec | 3474 | } |
bd89aabc | 3475 | cgroup_iter_end(cgrp, &it); |
102a775e BB |
3476 | length = n; |
3477 | /* now sort & (if procs) strip out duplicates */ | |
3478 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3479 | if (type == CGROUP_FILE_PROCS) |
102a775e | 3480 | length = pidlist_uniq(&array, length); |
72a8cb30 BB |
3481 | l = cgroup_pidlist_find(cgrp, type); |
3482 | if (!l) { | |
d1d9fd33 | 3483 | pidlist_free(array); |
72a8cb30 | 3484 | return -ENOMEM; |
102a775e | 3485 | } |
72a8cb30 | 3486 | /* store array, freeing old if necessary - lock already held */ |
d1d9fd33 | 3487 | pidlist_free(l->list); |
102a775e BB |
3488 | l->list = array; |
3489 | l->length = length; | |
3490 | l->use_count++; | |
3491 | up_write(&l->mutex); | |
72a8cb30 | 3492 | *lp = l; |
102a775e | 3493 | return 0; |
bbcb81d0 PM |
3494 | } |
3495 | ||
846c7bb0 | 3496 | /** |
a043e3b2 | 3497 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3498 | * @stats: cgroupstats to fill information into |
3499 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3500 | * been requested. | |
a043e3b2 LZ |
3501 | * |
3502 | * Build and fill cgroupstats so that taskstats can export it to user | |
3503 | * space. | |
846c7bb0 BS |
3504 | */ |
3505 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3506 | { | |
3507 | int ret = -EINVAL; | |
bd89aabc | 3508 | struct cgroup *cgrp; |
846c7bb0 BS |
3509 | struct cgroup_iter it; |
3510 | struct task_struct *tsk; | |
33d283be | 3511 | |
846c7bb0 | 3512 | /* |
33d283be LZ |
3513 | * Validate dentry by checking the superblock operations, |
3514 | * and make sure it's a directory. | |
846c7bb0 | 3515 | */ |
33d283be LZ |
3516 | if (dentry->d_sb->s_op != &cgroup_ops || |
3517 | !S_ISDIR(dentry->d_inode->i_mode)) | |
846c7bb0 BS |
3518 | goto err; |
3519 | ||
3520 | ret = 0; | |
bd89aabc | 3521 | cgrp = dentry->d_fsdata; |
846c7bb0 | 3522 | |
bd89aabc PM |
3523 | cgroup_iter_start(cgrp, &it); |
3524 | while ((tsk = cgroup_iter_next(cgrp, &it))) { | |
846c7bb0 BS |
3525 | switch (tsk->state) { |
3526 | case TASK_RUNNING: | |
3527 | stats->nr_running++; | |
3528 | break; | |
3529 | case TASK_INTERRUPTIBLE: | |
3530 | stats->nr_sleeping++; | |
3531 | break; | |
3532 | case TASK_UNINTERRUPTIBLE: | |
3533 | stats->nr_uninterruptible++; | |
3534 | break; | |
3535 | case TASK_STOPPED: | |
3536 | stats->nr_stopped++; | |
3537 | break; | |
3538 | default: | |
3539 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3540 | stats->nr_io_wait++; | |
3541 | break; | |
3542 | } | |
3543 | } | |
bd89aabc | 3544 | cgroup_iter_end(cgrp, &it); |
846c7bb0 | 3545 | |
846c7bb0 BS |
3546 | err: |
3547 | return ret; | |
3548 | } | |
3549 | ||
8f3ff208 | 3550 | |
bbcb81d0 | 3551 | /* |
102a775e | 3552 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3553 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3554 | * in the cgroup->l->list array. |
bbcb81d0 | 3555 | */ |
cc31edce | 3556 | |
102a775e | 3557 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3558 | { |
cc31edce PM |
3559 | /* |
3560 | * Initially we receive a position value that corresponds to | |
3561 | * one more than the last pid shown (or 0 on the first call or | |
3562 | * after a seek to the start). Use a binary-search to find the | |
3563 | * next pid to display, if any | |
3564 | */ | |
102a775e | 3565 | struct cgroup_pidlist *l = s->private; |
cc31edce PM |
3566 | int index = 0, pid = *pos; |
3567 | int *iter; | |
3568 | ||
102a775e | 3569 | down_read(&l->mutex); |
cc31edce | 3570 | if (pid) { |
102a775e | 3571 | int end = l->length; |
20777766 | 3572 | |
cc31edce PM |
3573 | while (index < end) { |
3574 | int mid = (index + end) / 2; | |
102a775e | 3575 | if (l->list[mid] == pid) { |
cc31edce PM |
3576 | index = mid; |
3577 | break; | |
102a775e | 3578 | } else if (l->list[mid] <= pid) |
cc31edce PM |
3579 | index = mid + 1; |
3580 | else | |
3581 | end = mid; | |
3582 | } | |
3583 | } | |
3584 | /* If we're off the end of the array, we're done */ | |
102a775e | 3585 | if (index >= l->length) |
cc31edce PM |
3586 | return NULL; |
3587 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3588 | iter = l->list + index; |
cc31edce PM |
3589 | *pos = *iter; |
3590 | return iter; | |
3591 | } | |
3592 | ||
102a775e | 3593 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3594 | { |
102a775e BB |
3595 | struct cgroup_pidlist *l = s->private; |
3596 | up_read(&l->mutex); | |
cc31edce PM |
3597 | } |
3598 | ||
102a775e | 3599 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3600 | { |
102a775e BB |
3601 | struct cgroup_pidlist *l = s->private; |
3602 | pid_t *p = v; | |
3603 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3604 | /* |
3605 | * Advance to the next pid in the array. If this goes off the | |
3606 | * end, we're done | |
3607 | */ | |
3608 | p++; | |
3609 | if (p >= end) { | |
3610 | return NULL; | |
3611 | } else { | |
3612 | *pos = *p; | |
3613 | return p; | |
3614 | } | |
3615 | } | |
3616 | ||
102a775e | 3617 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3618 | { |
3619 | return seq_printf(s, "%d\n", *(int *)v); | |
3620 | } | |
bbcb81d0 | 3621 | |
102a775e BB |
3622 | /* |
3623 | * seq_operations functions for iterating on pidlists through seq_file - | |
3624 | * independent of whether it's tasks or procs | |
3625 | */ | |
3626 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3627 | .start = cgroup_pidlist_start, | |
3628 | .stop = cgroup_pidlist_stop, | |
3629 | .next = cgroup_pidlist_next, | |
3630 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3631 | }; |
3632 | ||
102a775e | 3633 | static void cgroup_release_pid_array(struct cgroup_pidlist *l) |
cc31edce | 3634 | { |
72a8cb30 BB |
3635 | /* |
3636 | * the case where we're the last user of this particular pidlist will | |
3637 | * have us remove it from the cgroup's list, which entails taking the | |
3638 | * mutex. since in pidlist_find the pidlist->lock depends on cgroup-> | |
3639 | * pidlist_mutex, we have to take pidlist_mutex first. | |
3640 | */ | |
3641 | mutex_lock(&l->owner->pidlist_mutex); | |
102a775e BB |
3642 | down_write(&l->mutex); |
3643 | BUG_ON(!l->use_count); | |
3644 | if (!--l->use_count) { | |
72a8cb30 BB |
3645 | /* we're the last user if refcount is 0; remove and free */ |
3646 | list_del(&l->links); | |
3647 | mutex_unlock(&l->owner->pidlist_mutex); | |
d1d9fd33 | 3648 | pidlist_free(l->list); |
72a8cb30 BB |
3649 | put_pid_ns(l->key.ns); |
3650 | up_write(&l->mutex); | |
3651 | kfree(l); | |
3652 | return; | |
cc31edce | 3653 | } |
72a8cb30 | 3654 | mutex_unlock(&l->owner->pidlist_mutex); |
102a775e | 3655 | up_write(&l->mutex); |
bbcb81d0 PM |
3656 | } |
3657 | ||
102a775e | 3658 | static int cgroup_pidlist_release(struct inode *inode, struct file *file) |
cc31edce | 3659 | { |
102a775e | 3660 | struct cgroup_pidlist *l; |
cc31edce PM |
3661 | if (!(file->f_mode & FMODE_READ)) |
3662 | return 0; | |
102a775e BB |
3663 | /* |
3664 | * the seq_file will only be initialized if the file was opened for | |
3665 | * reading; hence we check if it's not null only in that case. | |
3666 | */ | |
3667 | l = ((struct seq_file *)file->private_data)->private; | |
3668 | cgroup_release_pid_array(l); | |
cc31edce PM |
3669 | return seq_release(inode, file); |
3670 | } | |
3671 | ||
102a775e | 3672 | static const struct file_operations cgroup_pidlist_operations = { |
cc31edce PM |
3673 | .read = seq_read, |
3674 | .llseek = seq_lseek, | |
3675 | .write = cgroup_file_write, | |
102a775e | 3676 | .release = cgroup_pidlist_release, |
cc31edce PM |
3677 | }; |
3678 | ||
bbcb81d0 | 3679 | /* |
102a775e BB |
3680 | * The following functions handle opens on a file that displays a pidlist |
3681 | * (tasks or procs). Prepare an array of the process/thread IDs of whoever's | |
3682 | * in the cgroup. | |
bbcb81d0 | 3683 | */ |
102a775e | 3684 | /* helper function for the two below it */ |
72a8cb30 | 3685 | static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type) |
bbcb81d0 | 3686 | { |
bd89aabc | 3687 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
72a8cb30 | 3688 | struct cgroup_pidlist *l; |
cc31edce | 3689 | int retval; |
bbcb81d0 | 3690 | |
cc31edce | 3691 | /* Nothing to do for write-only files */ |
bbcb81d0 PM |
3692 | if (!(file->f_mode & FMODE_READ)) |
3693 | return 0; | |
3694 | ||
102a775e | 3695 | /* have the array populated */ |
72a8cb30 | 3696 | retval = pidlist_array_load(cgrp, type, &l); |
102a775e BB |
3697 | if (retval) |
3698 | return retval; | |
3699 | /* configure file information */ | |
3700 | file->f_op = &cgroup_pidlist_operations; | |
cc31edce | 3701 | |
102a775e | 3702 | retval = seq_open(file, &cgroup_pidlist_seq_operations); |
cc31edce | 3703 | if (retval) { |
102a775e | 3704 | cgroup_release_pid_array(l); |
cc31edce | 3705 | return retval; |
bbcb81d0 | 3706 | } |
102a775e | 3707 | ((struct seq_file *)file->private_data)->private = l; |
bbcb81d0 PM |
3708 | return 0; |
3709 | } | |
102a775e BB |
3710 | static int cgroup_tasks_open(struct inode *unused, struct file *file) |
3711 | { | |
72a8cb30 | 3712 | return cgroup_pidlist_open(file, CGROUP_FILE_TASKS); |
102a775e BB |
3713 | } |
3714 | static int cgroup_procs_open(struct inode *unused, struct file *file) | |
3715 | { | |
72a8cb30 | 3716 | return cgroup_pidlist_open(file, CGROUP_FILE_PROCS); |
102a775e | 3717 | } |
bbcb81d0 | 3718 | |
bd89aabc | 3719 | static u64 cgroup_read_notify_on_release(struct cgroup *cgrp, |
81a6a5cd PM |
3720 | struct cftype *cft) |
3721 | { | |
bd89aabc | 3722 | return notify_on_release(cgrp); |
81a6a5cd PM |
3723 | } |
3724 | ||
6379c106 PM |
3725 | static int cgroup_write_notify_on_release(struct cgroup *cgrp, |
3726 | struct cftype *cft, | |
3727 | u64 val) | |
3728 | { | |
3729 | clear_bit(CGRP_RELEASABLE, &cgrp->flags); | |
3730 | if (val) | |
3731 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3732 | else | |
3733 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3734 | return 0; | |
3735 | } | |
3736 | ||
0dea1168 KS |
3737 | /* |
3738 | * Unregister event and free resources. | |
3739 | * | |
3740 | * Gets called from workqueue. | |
3741 | */ | |
3742 | static void cgroup_event_remove(struct work_struct *work) | |
3743 | { | |
3744 | struct cgroup_event *event = container_of(work, struct cgroup_event, | |
3745 | remove); | |
3746 | struct cgroup *cgrp = event->cgrp; | |
3747 | ||
0dea1168 KS |
3748 | event->cft->unregister_event(cgrp, event->cft, event->eventfd); |
3749 | ||
3750 | eventfd_ctx_put(event->eventfd); | |
0dea1168 | 3751 | kfree(event); |
a0a4db54 | 3752 | dput(cgrp->dentry); |
0dea1168 KS |
3753 | } |
3754 | ||
3755 | /* | |
3756 | * Gets called on POLLHUP on eventfd when user closes it. | |
3757 | * | |
3758 | * Called with wqh->lock held and interrupts disabled. | |
3759 | */ | |
3760 | static int cgroup_event_wake(wait_queue_t *wait, unsigned mode, | |
3761 | int sync, void *key) | |
3762 | { | |
3763 | struct cgroup_event *event = container_of(wait, | |
3764 | struct cgroup_event, wait); | |
3765 | struct cgroup *cgrp = event->cgrp; | |
3766 | unsigned long flags = (unsigned long)key; | |
3767 | ||
3768 | if (flags & POLLHUP) { | |
a93d2f17 | 3769 | __remove_wait_queue(event->wqh, &event->wait); |
0dea1168 KS |
3770 | spin_lock(&cgrp->event_list_lock); |
3771 | list_del(&event->list); | |
3772 | spin_unlock(&cgrp->event_list_lock); | |
3773 | /* | |
3774 | * We are in atomic context, but cgroup_event_remove() may | |
3775 | * sleep, so we have to call it in workqueue. | |
3776 | */ | |
3777 | schedule_work(&event->remove); | |
3778 | } | |
3779 | ||
3780 | return 0; | |
3781 | } | |
3782 | ||
3783 | static void cgroup_event_ptable_queue_proc(struct file *file, | |
3784 | wait_queue_head_t *wqh, poll_table *pt) | |
3785 | { | |
3786 | struct cgroup_event *event = container_of(pt, | |
3787 | struct cgroup_event, pt); | |
3788 | ||
3789 | event->wqh = wqh; | |
3790 | add_wait_queue(wqh, &event->wait); | |
3791 | } | |
3792 | ||
3793 | /* | |
3794 | * Parse input and register new cgroup event handler. | |
3795 | * | |
3796 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3797 | * Interpretation of args is defined by control file implementation. | |
3798 | */ | |
3799 | static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft, | |
3800 | const char *buffer) | |
3801 | { | |
3802 | struct cgroup_event *event = NULL; | |
3803 | unsigned int efd, cfd; | |
3804 | struct file *efile = NULL; | |
3805 | struct file *cfile = NULL; | |
3806 | char *endp; | |
3807 | int ret; | |
3808 | ||
3809 | efd = simple_strtoul(buffer, &endp, 10); | |
3810 | if (*endp != ' ') | |
3811 | return -EINVAL; | |
3812 | buffer = endp + 1; | |
3813 | ||
3814 | cfd = simple_strtoul(buffer, &endp, 10); | |
3815 | if ((*endp != ' ') && (*endp != '\0')) | |
3816 | return -EINVAL; | |
3817 | buffer = endp + 1; | |
3818 | ||
3819 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3820 | if (!event) | |
3821 | return -ENOMEM; | |
3822 | event->cgrp = cgrp; | |
3823 | INIT_LIST_HEAD(&event->list); | |
3824 | init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc); | |
3825 | init_waitqueue_func_entry(&event->wait, cgroup_event_wake); | |
3826 | INIT_WORK(&event->remove, cgroup_event_remove); | |
3827 | ||
3828 | efile = eventfd_fget(efd); | |
3829 | if (IS_ERR(efile)) { | |
3830 | ret = PTR_ERR(efile); | |
3831 | goto fail; | |
3832 | } | |
3833 | ||
3834 | event->eventfd = eventfd_ctx_fileget(efile); | |
3835 | if (IS_ERR(event->eventfd)) { | |
3836 | ret = PTR_ERR(event->eventfd); | |
3837 | goto fail; | |
3838 | } | |
3839 | ||
3840 | cfile = fget(cfd); | |
3841 | if (!cfile) { | |
3842 | ret = -EBADF; | |
3843 | goto fail; | |
3844 | } | |
3845 | ||
3846 | /* the process need read permission on control file */ | |
3bfa784a AV |
3847 | /* AV: shouldn't we check that it's been opened for read instead? */ |
3848 | ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ); | |
0dea1168 KS |
3849 | if (ret < 0) |
3850 | goto fail; | |
3851 | ||
3852 | event->cft = __file_cft(cfile); | |
3853 | if (IS_ERR(event->cft)) { | |
3854 | ret = PTR_ERR(event->cft); | |
3855 | goto fail; | |
3856 | } | |
3857 | ||
3858 | if (!event->cft->register_event || !event->cft->unregister_event) { | |
3859 | ret = -EINVAL; | |
3860 | goto fail; | |
3861 | } | |
3862 | ||
3863 | ret = event->cft->register_event(cgrp, event->cft, | |
3864 | event->eventfd, buffer); | |
3865 | if (ret) | |
3866 | goto fail; | |
3867 | ||
3868 | if (efile->f_op->poll(efile, &event->pt) & POLLHUP) { | |
3869 | event->cft->unregister_event(cgrp, event->cft, event->eventfd); | |
3870 | ret = 0; | |
3871 | goto fail; | |
3872 | } | |
3873 | ||
a0a4db54 KS |
3874 | /* |
3875 | * Events should be removed after rmdir of cgroup directory, but before | |
3876 | * destroying subsystem state objects. Let's take reference to cgroup | |
3877 | * directory dentry to do that. | |
3878 | */ | |
3879 | dget(cgrp->dentry); | |
3880 | ||
0dea1168 KS |
3881 | spin_lock(&cgrp->event_list_lock); |
3882 | list_add(&event->list, &cgrp->event_list); | |
3883 | spin_unlock(&cgrp->event_list_lock); | |
3884 | ||
3885 | fput(cfile); | |
3886 | fput(efile); | |
3887 | ||
3888 | return 0; | |
3889 | ||
3890 | fail: | |
3891 | if (cfile) | |
3892 | fput(cfile); | |
3893 | ||
3894 | if (event && event->eventfd && !IS_ERR(event->eventfd)) | |
3895 | eventfd_ctx_put(event->eventfd); | |
3896 | ||
3897 | if (!IS_ERR_OR_NULL(efile)) | |
3898 | fput(efile); | |
3899 | ||
3900 | kfree(event); | |
3901 | ||
3902 | return ret; | |
3903 | } | |
3904 | ||
97978e6d DL |
3905 | static u64 cgroup_clone_children_read(struct cgroup *cgrp, |
3906 | struct cftype *cft) | |
3907 | { | |
3908 | return clone_children(cgrp); | |
3909 | } | |
3910 | ||
3911 | static int cgroup_clone_children_write(struct cgroup *cgrp, | |
3912 | struct cftype *cft, | |
3913 | u64 val) | |
3914 | { | |
3915 | if (val) | |
3916 | set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); | |
3917 | else | |
3918 | clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); | |
3919 | return 0; | |
3920 | } | |
3921 | ||
bbcb81d0 PM |
3922 | /* |
3923 | * for the common functions, 'private' gives the type of file | |
3924 | */ | |
102a775e BB |
3925 | /* for hysterical raisins, we can't put this on the older files */ |
3926 | #define CGROUP_FILE_GENERIC_PREFIX "cgroup." | |
81a6a5cd PM |
3927 | static struct cftype files[] = { |
3928 | { | |
3929 | .name = "tasks", | |
3930 | .open = cgroup_tasks_open, | |
af351026 | 3931 | .write_u64 = cgroup_tasks_write, |
102a775e | 3932 | .release = cgroup_pidlist_release, |
099fca32 | 3933 | .mode = S_IRUGO | S_IWUSR, |
81a6a5cd | 3934 | }, |
102a775e BB |
3935 | { |
3936 | .name = CGROUP_FILE_GENERIC_PREFIX "procs", | |
3937 | .open = cgroup_procs_open, | |
74a1166d | 3938 | .write_u64 = cgroup_procs_write, |
102a775e | 3939 | .release = cgroup_pidlist_release, |
74a1166d | 3940 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3941 | }, |
81a6a5cd PM |
3942 | { |
3943 | .name = "notify_on_release", | |
f4c753b7 | 3944 | .read_u64 = cgroup_read_notify_on_release, |
6379c106 | 3945 | .write_u64 = cgroup_write_notify_on_release, |
81a6a5cd | 3946 | }, |
0dea1168 KS |
3947 | { |
3948 | .name = CGROUP_FILE_GENERIC_PREFIX "event_control", | |
3949 | .write_string = cgroup_write_event_control, | |
3950 | .mode = S_IWUGO, | |
3951 | }, | |
97978e6d DL |
3952 | { |
3953 | .name = "cgroup.clone_children", | |
3954 | .read_u64 = cgroup_clone_children_read, | |
3955 | .write_u64 = cgroup_clone_children_write, | |
3956 | }, | |
6e6ff25b TH |
3957 | { |
3958 | .name = "release_agent", | |
3959 | .flags = CFTYPE_ONLY_ON_ROOT, | |
3960 | .read_seq_string = cgroup_release_agent_show, | |
3961 | .write_string = cgroup_release_agent_write, | |
3962 | .max_write_len = PATH_MAX, | |
3963 | }, | |
db0416b6 | 3964 | { } /* terminate */ |
bbcb81d0 PM |
3965 | }; |
3966 | ||
13af07df AR |
3967 | /** |
3968 | * cgroup_populate_dir - selectively creation of files in a directory | |
3969 | * @cgrp: target cgroup | |
3970 | * @base_files: true if the base files should be added | |
3971 | * @subsys_mask: mask of the subsystem ids whose files should be added | |
3972 | */ | |
3973 | static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files, | |
3974 | unsigned long subsys_mask) | |
ddbcc7e8 PM |
3975 | { |
3976 | int err; | |
3977 | struct cgroup_subsys *ss; | |
3978 | ||
13af07df AR |
3979 | if (base_files) { |
3980 | err = cgroup_addrm_files(cgrp, NULL, files, true); | |
3981 | if (err < 0) | |
3982 | return err; | |
3983 | } | |
bbcb81d0 | 3984 | |
8e3f6541 | 3985 | /* process cftsets of each subsystem */ |
bd89aabc | 3986 | for_each_subsys(cgrp->root, ss) { |
8e3f6541 | 3987 | struct cftype_set *set; |
13af07df AR |
3988 | if (!test_bit(ss->subsys_id, &subsys_mask)) |
3989 | continue; | |
8e3f6541 | 3990 | |
db0416b6 | 3991 | list_for_each_entry(set, &ss->cftsets, node) |
79578621 | 3992 | cgroup_addrm_files(cgrp, ss, set->cfts, true); |
ddbcc7e8 | 3993 | } |
8e3f6541 | 3994 | |
38460b48 KH |
3995 | /* This cgroup is ready now */ |
3996 | for_each_subsys(cgrp->root, ss) { | |
3997 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
3998 | /* | |
3999 | * Update id->css pointer and make this css visible from | |
4000 | * CSS ID functions. This pointer will be dereferened | |
4001 | * from RCU-read-side without locks. | |
4002 | */ | |
4003 | if (css->id) | |
4004 | rcu_assign_pointer(css->id->css, css); | |
4005 | } | |
ddbcc7e8 PM |
4006 | |
4007 | return 0; | |
4008 | } | |
4009 | ||
48ddbe19 TH |
4010 | static void css_dput_fn(struct work_struct *work) |
4011 | { | |
4012 | struct cgroup_subsys_state *css = | |
4013 | container_of(work, struct cgroup_subsys_state, dput_work); | |
5db9a4d9 TH |
4014 | struct dentry *dentry = css->cgroup->dentry; |
4015 | struct super_block *sb = dentry->d_sb; | |
48ddbe19 | 4016 | |
5db9a4d9 TH |
4017 | atomic_inc(&sb->s_active); |
4018 | dput(dentry); | |
4019 | deactivate_super(sb); | |
48ddbe19 TH |
4020 | } |
4021 | ||
ddbcc7e8 PM |
4022 | static void init_cgroup_css(struct cgroup_subsys_state *css, |
4023 | struct cgroup_subsys *ss, | |
bd89aabc | 4024 | struct cgroup *cgrp) |
ddbcc7e8 | 4025 | { |
bd89aabc | 4026 | css->cgroup = cgrp; |
e7c5ec91 | 4027 | atomic_set(&css->refcnt, 1); |
ddbcc7e8 | 4028 | css->flags = 0; |
38460b48 | 4029 | css->id = NULL; |
bd89aabc | 4030 | if (cgrp == dummytop) |
38b53aba | 4031 | css->flags |= CSS_ROOT; |
bd89aabc PM |
4032 | BUG_ON(cgrp->subsys[ss->subsys_id]); |
4033 | cgrp->subsys[ss->subsys_id] = css; | |
48ddbe19 TH |
4034 | |
4035 | /* | |
ed957793 TH |
4036 | * css holds an extra ref to @cgrp->dentry which is put on the last |
4037 | * css_put(). dput() requires process context, which css_put() may | |
4038 | * be called without. @css->dput_work will be used to invoke | |
4039 | * dput() asynchronously from css_put(). | |
48ddbe19 TH |
4040 | */ |
4041 | INIT_WORK(&css->dput_work, css_dput_fn); | |
ddbcc7e8 PM |
4042 | } |
4043 | ||
a31f2d3f TH |
4044 | /* invoke ->post_create() on a new CSS and mark it online */ |
4045 | static void online_css(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
4046 | { | |
4047 | lockdep_assert_held(&cgroup_mutex); | |
4048 | ||
4049 | if (ss->post_create) | |
4050 | ss->post_create(cgrp); | |
4051 | cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE; | |
4052 | } | |
4053 | ||
4054 | /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */ | |
4055 | static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
4056 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
4057 | { | |
4058 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
4059 | ||
4060 | lockdep_assert_held(&cgroup_mutex); | |
4061 | ||
4062 | if (!(css->flags & CSS_ONLINE)) | |
4063 | return; | |
4064 | ||
4065 | /* | |
4066 | * pre_destroy() should be called with cgroup_mutex unlocked. See | |
4067 | * 3fa59dfbc3 ("cgroup: fix potential deadlock in pre_destroy") for | |
4068 | * details. This temporary unlocking should go away once | |
4069 | * cgroup_mutex is unexported from controllers. | |
4070 | */ | |
4071 | if (ss->pre_destroy) { | |
4072 | mutex_unlock(&cgroup_mutex); | |
4073 | ss->pre_destroy(cgrp); | |
4074 | mutex_lock(&cgroup_mutex); | |
4075 | } | |
4076 | ||
4077 | cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE; | |
4078 | } | |
4079 | ||
ddbcc7e8 | 4080 | /* |
a043e3b2 LZ |
4081 | * cgroup_create - create a cgroup |
4082 | * @parent: cgroup that will be parent of the new cgroup | |
4083 | * @dentry: dentry of the new cgroup | |
4084 | * @mode: mode to set on new inode | |
ddbcc7e8 | 4085 | * |
a043e3b2 | 4086 | * Must be called with the mutex on the parent inode held |
ddbcc7e8 | 4087 | */ |
ddbcc7e8 | 4088 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, |
a5e7ed32 | 4089 | umode_t mode) |
ddbcc7e8 | 4090 | { |
bd89aabc | 4091 | struct cgroup *cgrp; |
ddbcc7e8 PM |
4092 | struct cgroupfs_root *root = parent->root; |
4093 | int err = 0; | |
4094 | struct cgroup_subsys *ss; | |
4095 | struct super_block *sb = root->sb; | |
4096 | ||
bd89aabc PM |
4097 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
4098 | if (!cgrp) | |
ddbcc7e8 PM |
4099 | return -ENOMEM; |
4100 | ||
976c06bc TH |
4101 | /* |
4102 | * Only live parents can have children. Note that the liveliness | |
4103 | * check isn't strictly necessary because cgroup_mkdir() and | |
4104 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
4105 | * anyway so that locking is contained inside cgroup proper and we | |
4106 | * don't get nasty surprises if we ever grow another caller. | |
4107 | */ | |
4108 | if (!cgroup_lock_live_group(parent)) { | |
4109 | err = -ENODEV; | |
4b8b47eb | 4110 | goto err_free_cgrp; |
976c06bc TH |
4111 | } |
4112 | ||
ddbcc7e8 PM |
4113 | /* Grab a reference on the superblock so the hierarchy doesn't |
4114 | * get deleted on unmount if there are child cgroups. This | |
4115 | * can be done outside cgroup_mutex, since the sb can't | |
4116 | * disappear while someone has an open control file on the | |
4117 | * fs */ | |
4118 | atomic_inc(&sb->s_active); | |
4119 | ||
cc31edce | 4120 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4121 | |
bd89aabc PM |
4122 | cgrp->parent = parent; |
4123 | cgrp->root = parent->root; | |
4124 | cgrp->top_cgroup = parent->top_cgroup; | |
ddbcc7e8 | 4125 | |
b6abdb0e LZ |
4126 | if (notify_on_release(parent)) |
4127 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4128 | ||
97978e6d DL |
4129 | if (clone_children(parent)) |
4130 | set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); | |
4131 | ||
ddbcc7e8 | 4132 | for_each_subsys(root, ss) { |
8c7f6edb | 4133 | struct cgroup_subsys_state *css; |
4528fd05 | 4134 | |
8c7f6edb | 4135 | css = ss->create(cgrp); |
ddbcc7e8 PM |
4136 | if (IS_ERR(css)) { |
4137 | err = PTR_ERR(css); | |
4b8b47eb | 4138 | goto err_free_all; |
ddbcc7e8 | 4139 | } |
bd89aabc | 4140 | init_cgroup_css(css, ss, cgrp); |
4528fd05 LZ |
4141 | if (ss->use_id) { |
4142 | err = alloc_css_id(ss, parent, cgrp); | |
4143 | if (err) | |
4b8b47eb | 4144 | goto err_free_all; |
4528fd05 | 4145 | } |
38460b48 | 4146 | /* At error, ->destroy() callback has to free assigned ID. */ |
97978e6d | 4147 | if (clone_children(parent) && ss->post_clone) |
761b3ef5 | 4148 | ss->post_clone(cgrp); |
8c7f6edb TH |
4149 | |
4150 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
4151 | parent->parent) { | |
4152 | pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n", | |
4153 | current->comm, current->pid, ss->name); | |
4154 | if (!strcmp(ss->name, "memory")) | |
4155 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
4156 | ss->warned_broken_hierarchy = true; | |
4157 | } | |
ddbcc7e8 PM |
4158 | } |
4159 | ||
4e139afc TH |
4160 | /* |
4161 | * Create directory. cgroup_create_file() returns with the new | |
4162 | * directory locked on success so that it can be populated without | |
4163 | * dropping cgroup_mutex. | |
4164 | */ | |
28fd6f30 | 4165 | err = cgroup_create_file(dentry, S_IFDIR | mode, sb); |
ddbcc7e8 | 4166 | if (err < 0) |
4b8b47eb | 4167 | goto err_free_all; |
4e139afc | 4168 | lockdep_assert_held(&dentry->d_inode->i_mutex); |
ddbcc7e8 | 4169 | |
4e139afc | 4170 | /* allocation complete, commit to creation */ |
28fd6f30 | 4171 | dentry->d_fsdata = cgrp; |
febfcef6 | 4172 | cgrp->dentry = dentry; |
4e139afc TH |
4173 | list_add_tail(&cgrp->allcg_node, &root->allcg_list); |
4174 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); | |
4175 | root->number_of_cgroups++; | |
28fd6f30 | 4176 | |
a8638030 TH |
4177 | for_each_subsys(root, ss) { |
4178 | /* each css holds a ref to the cgroup's dentry */ | |
ed957793 | 4179 | dget(dentry); |
48ddbe19 | 4180 | |
a8638030 | 4181 | /* creation succeeded, notify subsystems */ |
a31f2d3f | 4182 | online_css(ss, cgrp); |
a8638030 TH |
4183 | } |
4184 | ||
a1a71b45 | 4185 | err = cgroup_populate_dir(cgrp, true, root->subsys_mask); |
4b8b47eb TH |
4186 | if (err) |
4187 | goto err_destroy; | |
ddbcc7e8 PM |
4188 | |
4189 | mutex_unlock(&cgroup_mutex); | |
bd89aabc | 4190 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
4191 | |
4192 | return 0; | |
4193 | ||
4b8b47eb | 4194 | err_free_all: |
ddbcc7e8 | 4195 | for_each_subsys(root, ss) { |
bd89aabc | 4196 | if (cgrp->subsys[ss->subsys_id]) |
761b3ef5 | 4197 | ss->destroy(cgrp); |
ddbcc7e8 | 4198 | } |
ddbcc7e8 | 4199 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
4200 | /* Release the reference count that we took on the superblock */ |
4201 | deactivate_super(sb); | |
4b8b47eb | 4202 | err_free_cgrp: |
bd89aabc | 4203 | kfree(cgrp); |
ddbcc7e8 | 4204 | return err; |
4b8b47eb TH |
4205 | |
4206 | err_destroy: | |
4207 | cgroup_destroy_locked(cgrp); | |
4208 | mutex_unlock(&cgroup_mutex); | |
4209 | mutex_unlock(&dentry->d_inode->i_mutex); | |
4210 | return err; | |
ddbcc7e8 PM |
4211 | } |
4212 | ||
18bb1db3 | 4213 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
ddbcc7e8 PM |
4214 | { |
4215 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
4216 | ||
4217 | /* the vfs holds inode->i_mutex already */ | |
4218 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
4219 | } | |
4220 | ||
28b4c27b TH |
4221 | /* |
4222 | * Check the reference count on each subsystem. Since we already | |
4223 | * established that there are no tasks in the cgroup, if the css refcount | |
4224 | * is also 1, then there should be no outstanding references, so the | |
4225 | * subsystem is safe to destroy. We scan across all subsystems rather than | |
4226 | * using the per-hierarchy linked list of mounted subsystems since we can | |
4227 | * be called via check_for_release() with no synchronization other than | |
4228 | * RCU, and the subsystem linked list isn't RCU-safe. | |
4229 | */ | |
55b6fd01 | 4230 | static int cgroup_has_css_refs(struct cgroup *cgrp) |
81a6a5cd | 4231 | { |
81a6a5cd | 4232 | int i; |
28b4c27b | 4233 | |
aae8aab4 BB |
4234 | /* |
4235 | * We won't need to lock the subsys array, because the subsystems | |
4236 | * we're concerned about aren't going anywhere since our cgroup root | |
4237 | * has a reference on them. | |
4238 | */ | |
81a6a5cd PM |
4239 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
4240 | struct cgroup_subsys *ss = subsys[i]; | |
4241 | struct cgroup_subsys_state *css; | |
28b4c27b | 4242 | |
aae8aab4 BB |
4243 | /* Skip subsystems not present or not in this hierarchy */ |
4244 | if (ss == NULL || ss->root != cgrp->root) | |
81a6a5cd | 4245 | continue; |
28b4c27b | 4246 | |
bd89aabc | 4247 | css = cgrp->subsys[ss->subsys_id]; |
28b4c27b TH |
4248 | /* |
4249 | * When called from check_for_release() it's possible | |
81a6a5cd PM |
4250 | * that by this point the cgroup has been removed |
4251 | * and the css deleted. But a false-positive doesn't | |
4252 | * matter, since it can only happen if the cgroup | |
4253 | * has been deleted and hence no longer needs the | |
28b4c27b TH |
4254 | * release agent to be called anyway. |
4255 | */ | |
4256 | if (css && css_refcnt(css) > 1) | |
81a6a5cd | 4257 | return 1; |
81a6a5cd PM |
4258 | } |
4259 | return 0; | |
4260 | } | |
4261 | ||
42809dd4 TH |
4262 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4263 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4264 | { |
42809dd4 TH |
4265 | struct dentry *d = cgrp->dentry; |
4266 | struct cgroup *parent = cgrp->parent; | |
ec64f515 | 4267 | DEFINE_WAIT(wait); |
4ab78683 | 4268 | struct cgroup_event *event, *tmp; |
ed957793 | 4269 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4270 | |
42809dd4 TH |
4271 | lockdep_assert_held(&d->d_inode->i_mutex); |
4272 | lockdep_assert_held(&cgroup_mutex); | |
4273 | ||
4274 | if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) | |
ddbcc7e8 | 4275 | return -EBUSY; |
a043e3b2 | 4276 | |
88703267 | 4277 | /* |
1a90dd50 TH |
4278 | * Block new css_tryget() by deactivating refcnt and mark @cgrp |
4279 | * removed. This makes future css_tryget() and child creation | |
4280 | * attempts fail thus maintaining the removal conditions verified | |
4281 | * above. | |
88703267 | 4282 | */ |
ed957793 TH |
4283 | for_each_subsys(cgrp->root, ss) { |
4284 | struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id]; | |
88703267 | 4285 | |
ed957793 TH |
4286 | WARN_ON(atomic_read(&css->refcnt) < 0); |
4287 | atomic_add(CSS_DEACT_BIAS, &css->refcnt); | |
88703267 | 4288 | } |
1a90dd50 | 4289 | set_bit(CGRP_REMOVED, &cgrp->flags); |
ddbcc7e8 | 4290 | |
a31f2d3f | 4291 | /* tell subsystems to initate destruction */ |
1a90dd50 | 4292 | for_each_subsys(cgrp->root, ss) |
a31f2d3f | 4293 | offline_css(ss, cgrp); |
ed957793 TH |
4294 | |
4295 | /* | |
ed957793 TH |
4296 | * Put all the base refs. Each css holds an extra reference to the |
4297 | * cgroup's dentry and cgroup removal proceeds regardless of css | |
4298 | * refs. On the last put of each css, whenever that may be, the | |
4299 | * extra dentry ref is put so that dentry destruction happens only | |
4300 | * after all css's are released. | |
4301 | */ | |
e9316080 TH |
4302 | for_each_subsys(cgrp->root, ss) |
4303 | css_put(cgrp->subsys[ss->subsys_id]); | |
ddbcc7e8 | 4304 | |
cdcc136f | 4305 | raw_spin_lock(&release_list_lock); |
bd89aabc | 4306 | if (!list_empty(&cgrp->release_list)) |
8d258797 | 4307 | list_del_init(&cgrp->release_list); |
cdcc136f | 4308 | raw_spin_unlock(&release_list_lock); |
999cd8a4 | 4309 | |
999cd8a4 | 4310 | /* delete this cgroup from parent->children */ |
eb6fd504 | 4311 | list_del_rcu(&cgrp->sibling); |
b0ca5a84 TH |
4312 | list_del_init(&cgrp->allcg_node); |
4313 | ||
42809dd4 | 4314 | dget(d); |
ddbcc7e8 PM |
4315 | cgroup_d_remove_dir(d); |
4316 | dput(d); | |
ddbcc7e8 | 4317 | |
bd89aabc | 4318 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd PM |
4319 | check_for_release(parent); |
4320 | ||
4ab78683 KS |
4321 | /* |
4322 | * Unregister events and notify userspace. | |
4323 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4324 | * directory to avoid race between userspace and kernelspace | |
4325 | */ | |
4326 | spin_lock(&cgrp->event_list_lock); | |
4327 | list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) { | |
4328 | list_del(&event->list); | |
4329 | remove_wait_queue(event->wqh, &event->wait); | |
4330 | eventfd_signal(event->eventfd, 1); | |
4331 | schedule_work(&event->remove); | |
4332 | } | |
4333 | spin_unlock(&cgrp->event_list_lock); | |
4334 | ||
ddbcc7e8 PM |
4335 | return 0; |
4336 | } | |
4337 | ||
42809dd4 TH |
4338 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) |
4339 | { | |
4340 | int ret; | |
4341 | ||
4342 | mutex_lock(&cgroup_mutex); | |
4343 | ret = cgroup_destroy_locked(dentry->d_fsdata); | |
4344 | mutex_unlock(&cgroup_mutex); | |
4345 | ||
4346 | return ret; | |
4347 | } | |
4348 | ||
8e3f6541 TH |
4349 | static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss) |
4350 | { | |
4351 | INIT_LIST_HEAD(&ss->cftsets); | |
4352 | ||
4353 | /* | |
4354 | * base_cftset is embedded in subsys itself, no need to worry about | |
4355 | * deregistration. | |
4356 | */ | |
4357 | if (ss->base_cftypes) { | |
4358 | ss->base_cftset.cfts = ss->base_cftypes; | |
4359 | list_add_tail(&ss->base_cftset.node, &ss->cftsets); | |
4360 | } | |
4361 | } | |
4362 | ||
06a11920 | 4363 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4364 | { |
ddbcc7e8 | 4365 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4366 | |
4367 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4368 | |
648bb56d TH |
4369 | mutex_lock(&cgroup_mutex); |
4370 | ||
8e3f6541 TH |
4371 | /* init base cftset */ |
4372 | cgroup_init_cftsets(ss); | |
4373 | ||
ddbcc7e8 | 4374 | /* Create the top cgroup state for this subsystem */ |
33a68ac1 | 4375 | list_add(&ss->sibling, &rootnode.subsys_list); |
ddbcc7e8 | 4376 | ss->root = &rootnode; |
761b3ef5 | 4377 | css = ss->create(dummytop); |
ddbcc7e8 PM |
4378 | /* We don't handle early failures gracefully */ |
4379 | BUG_ON(IS_ERR(css)); | |
4380 | init_cgroup_css(css, ss, dummytop); | |
4381 | ||
e8d55fde | 4382 | /* Update the init_css_set to contain a subsys |
817929ec | 4383 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
4384 | * newly registered, all tasks and hence the |
4385 | * init_css_set is in the subsystem's top cgroup. */ | |
b48c6a80 | 4386 | init_css_set.subsys[ss->subsys_id] = css; |
ddbcc7e8 PM |
4387 | |
4388 | need_forkexit_callback |= ss->fork || ss->exit; | |
4389 | ||
e8d55fde LZ |
4390 | /* At system boot, before all subsystems have been |
4391 | * registered, no tasks have been forked, so we don't | |
4392 | * need to invoke fork callbacks here. */ | |
4393 | BUG_ON(!list_empty(&init_task.tasks)); | |
4394 | ||
ddbcc7e8 | 4395 | ss->active = 1; |
a31f2d3f | 4396 | online_css(ss, dummytop); |
a8638030 | 4397 | |
648bb56d TH |
4398 | mutex_unlock(&cgroup_mutex); |
4399 | ||
e6a1105b BB |
4400 | /* this function shouldn't be used with modular subsystems, since they |
4401 | * need to register a subsys_id, among other things */ | |
4402 | BUG_ON(ss->module); | |
4403 | } | |
4404 | ||
4405 | /** | |
4406 | * cgroup_load_subsys: load and register a modular subsystem at runtime | |
4407 | * @ss: the subsystem to load | |
4408 | * | |
4409 | * This function should be called in a modular subsystem's initcall. If the | |
88393161 | 4410 | * subsystem is built as a module, it will be assigned a new subsys_id and set |
e6a1105b BB |
4411 | * up for use. If the subsystem is built-in anyway, work is delegated to the |
4412 | * simpler cgroup_init_subsys. | |
4413 | */ | |
4414 | int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss) | |
4415 | { | |
e6a1105b | 4416 | struct cgroup_subsys_state *css; |
d19e19de | 4417 | int i, ret; |
e6a1105b BB |
4418 | |
4419 | /* check name and function validity */ | |
4420 | if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN || | |
4421 | ss->create == NULL || ss->destroy == NULL) | |
4422 | return -EINVAL; | |
4423 | ||
4424 | /* | |
4425 | * we don't support callbacks in modular subsystems. this check is | |
4426 | * before the ss->module check for consistency; a subsystem that could | |
4427 | * be a module should still have no callbacks even if the user isn't | |
4428 | * compiling it as one. | |
4429 | */ | |
4430 | if (ss->fork || ss->exit) | |
4431 | return -EINVAL; | |
4432 | ||
4433 | /* | |
4434 | * an optionally modular subsystem is built-in: we want to do nothing, | |
4435 | * since cgroup_init_subsys will have already taken care of it. | |
4436 | */ | |
4437 | if (ss->module == NULL) { | |
be45c900 | 4438 | /* a sanity check */ |
e6a1105b BB |
4439 | BUG_ON(subsys[ss->subsys_id] != ss); |
4440 | return 0; | |
4441 | } | |
4442 | ||
8e3f6541 TH |
4443 | /* init base cftset */ |
4444 | cgroup_init_cftsets(ss); | |
4445 | ||
e6a1105b | 4446 | mutex_lock(&cgroup_mutex); |
8a8e04df | 4447 | subsys[ss->subsys_id] = ss; |
e6a1105b BB |
4448 | |
4449 | /* | |
4450 | * no ss->create seems to need anything important in the ss struct, so | |
4451 | * this can happen first (i.e. before the rootnode attachment). | |
4452 | */ | |
761b3ef5 | 4453 | css = ss->create(dummytop); |
e6a1105b BB |
4454 | if (IS_ERR(css)) { |
4455 | /* failure case - need to deassign the subsys[] slot. */ | |
8a8e04df | 4456 | subsys[ss->subsys_id] = NULL; |
e6a1105b BB |
4457 | mutex_unlock(&cgroup_mutex); |
4458 | return PTR_ERR(css); | |
4459 | } | |
4460 | ||
4461 | list_add(&ss->sibling, &rootnode.subsys_list); | |
4462 | ss->root = &rootnode; | |
4463 | ||
4464 | /* our new subsystem will be attached to the dummy hierarchy. */ | |
4465 | init_cgroup_css(css, ss, dummytop); | |
4466 | /* init_idr must be after init_cgroup_css because it sets css->id. */ | |
4467 | if (ss->use_id) { | |
d19e19de TH |
4468 | ret = cgroup_init_idr(ss, css); |
4469 | if (ret) | |
4470 | goto err_unload; | |
e6a1105b BB |
4471 | } |
4472 | ||
4473 | /* | |
4474 | * Now we need to entangle the css into the existing css_sets. unlike | |
4475 | * in cgroup_init_subsys, there are now multiple css_sets, so each one | |
4476 | * will need a new pointer to it; done by iterating the css_set_table. | |
4477 | * furthermore, modifying the existing css_sets will corrupt the hash | |
4478 | * table state, so each changed css_set will need its hash recomputed. | |
4479 | * this is all done under the css_set_lock. | |
4480 | */ | |
4481 | write_lock(&css_set_lock); | |
4482 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) { | |
4483 | struct css_set *cg; | |
4484 | struct hlist_node *node, *tmp; | |
4485 | struct hlist_head *bucket = &css_set_table[i], *new_bucket; | |
4486 | ||
4487 | hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) { | |
4488 | /* skip entries that we already rehashed */ | |
4489 | if (cg->subsys[ss->subsys_id]) | |
4490 | continue; | |
4491 | /* remove existing entry */ | |
4492 | hlist_del(&cg->hlist); | |
4493 | /* set new value */ | |
4494 | cg->subsys[ss->subsys_id] = css; | |
4495 | /* recompute hash and restore entry */ | |
4496 | new_bucket = css_set_hash(cg->subsys); | |
4497 | hlist_add_head(&cg->hlist, new_bucket); | |
4498 | } | |
4499 | } | |
4500 | write_unlock(&css_set_lock); | |
4501 | ||
e6a1105b | 4502 | ss->active = 1; |
a31f2d3f | 4503 | online_css(ss, dummytop); |
a8638030 | 4504 | |
e6a1105b BB |
4505 | /* success! */ |
4506 | mutex_unlock(&cgroup_mutex); | |
4507 | return 0; | |
d19e19de TH |
4508 | |
4509 | err_unload: | |
4510 | mutex_unlock(&cgroup_mutex); | |
4511 | /* @ss can't be mounted here as try_module_get() would fail */ | |
4512 | cgroup_unload_subsys(ss); | |
4513 | return ret; | |
ddbcc7e8 | 4514 | } |
e6a1105b | 4515 | EXPORT_SYMBOL_GPL(cgroup_load_subsys); |
ddbcc7e8 | 4516 | |
cf5d5941 BB |
4517 | /** |
4518 | * cgroup_unload_subsys: unload a modular subsystem | |
4519 | * @ss: the subsystem to unload | |
4520 | * | |
4521 | * This function should be called in a modular subsystem's exitcall. When this | |
4522 | * function is invoked, the refcount on the subsystem's module will be 0, so | |
4523 | * the subsystem will not be attached to any hierarchy. | |
4524 | */ | |
4525 | void cgroup_unload_subsys(struct cgroup_subsys *ss) | |
4526 | { | |
4527 | struct cg_cgroup_link *link; | |
4528 | struct hlist_head *hhead; | |
4529 | ||
4530 | BUG_ON(ss->module == NULL); | |
4531 | ||
4532 | /* | |
4533 | * we shouldn't be called if the subsystem is in use, and the use of | |
4534 | * try_module_get in parse_cgroupfs_options should ensure that it | |
4535 | * doesn't start being used while we're killing it off. | |
4536 | */ | |
4537 | BUG_ON(ss->root != &rootnode); | |
4538 | ||
4539 | mutex_lock(&cgroup_mutex); | |
02ae7486 | 4540 | |
a31f2d3f | 4541 | offline_css(ss, dummytop); |
02ae7486 TH |
4542 | ss->active = 0; |
4543 | ||
4544 | if (ss->use_id) { | |
4545 | idr_remove_all(&ss->idr); | |
4546 | idr_destroy(&ss->idr); | |
4547 | } | |
4548 | ||
cf5d5941 | 4549 | /* deassign the subsys_id */ |
cf5d5941 BB |
4550 | subsys[ss->subsys_id] = NULL; |
4551 | ||
4552 | /* remove subsystem from rootnode's list of subsystems */ | |
8d258797 | 4553 | list_del_init(&ss->sibling); |
cf5d5941 BB |
4554 | |
4555 | /* | |
4556 | * disentangle the css from all css_sets attached to the dummytop. as | |
4557 | * in loading, we need to pay our respects to the hashtable gods. | |
4558 | */ | |
4559 | write_lock(&css_set_lock); | |
4560 | list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) { | |
4561 | struct css_set *cg = link->cg; | |
4562 | ||
4563 | hlist_del(&cg->hlist); | |
cf5d5941 BB |
4564 | cg->subsys[ss->subsys_id] = NULL; |
4565 | hhead = css_set_hash(cg->subsys); | |
4566 | hlist_add_head(&cg->hlist, hhead); | |
4567 | } | |
4568 | write_unlock(&css_set_lock); | |
4569 | ||
4570 | /* | |
4571 | * remove subsystem's css from the dummytop and free it - need to free | |
4572 | * before marking as null because ss->destroy needs the cgrp->subsys | |
4573 | * pointer to find their state. note that this also takes care of | |
4574 | * freeing the css_id. | |
4575 | */ | |
761b3ef5 | 4576 | ss->destroy(dummytop); |
cf5d5941 BB |
4577 | dummytop->subsys[ss->subsys_id] = NULL; |
4578 | ||
4579 | mutex_unlock(&cgroup_mutex); | |
4580 | } | |
4581 | EXPORT_SYMBOL_GPL(cgroup_unload_subsys); | |
4582 | ||
ddbcc7e8 | 4583 | /** |
a043e3b2 LZ |
4584 | * cgroup_init_early - cgroup initialization at system boot |
4585 | * | |
4586 | * Initialize cgroups at system boot, and initialize any | |
4587 | * subsystems that request early init. | |
ddbcc7e8 PM |
4588 | */ |
4589 | int __init cgroup_init_early(void) | |
4590 | { | |
4591 | int i; | |
146aa1bd | 4592 | atomic_set(&init_css_set.refcount, 1); |
817929ec PM |
4593 | INIT_LIST_HEAD(&init_css_set.cg_links); |
4594 | INIT_LIST_HEAD(&init_css_set.tasks); | |
472b1053 | 4595 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 4596 | css_set_count = 1; |
ddbcc7e8 | 4597 | init_cgroup_root(&rootnode); |
817929ec PM |
4598 | root_count = 1; |
4599 | init_task.cgroups = &init_css_set; | |
4600 | ||
4601 | init_css_set_link.cg = &init_css_set; | |
7717f7ba | 4602 | init_css_set_link.cgrp = dummytop; |
bd89aabc | 4603 | list_add(&init_css_set_link.cgrp_link_list, |
817929ec PM |
4604 | &rootnode.top_cgroup.css_sets); |
4605 | list_add(&init_css_set_link.cg_link_list, | |
4606 | &init_css_set.cg_links); | |
ddbcc7e8 | 4607 | |
472b1053 LZ |
4608 | for (i = 0; i < CSS_SET_TABLE_SIZE; i++) |
4609 | INIT_HLIST_HEAD(&css_set_table[i]); | |
4610 | ||
be45c900 | 4611 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
ddbcc7e8 PM |
4612 | struct cgroup_subsys *ss = subsys[i]; |
4613 | ||
be45c900 DW |
4614 | /* at bootup time, we don't worry about modular subsystems */ |
4615 | if (!ss || ss->module) | |
4616 | continue; | |
4617 | ||
ddbcc7e8 PM |
4618 | BUG_ON(!ss->name); |
4619 | BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); | |
4620 | BUG_ON(!ss->create); | |
4621 | BUG_ON(!ss->destroy); | |
4622 | if (ss->subsys_id != i) { | |
cfe36bde | 4623 | printk(KERN_ERR "cgroup: Subsys %s id == %d\n", |
ddbcc7e8 PM |
4624 | ss->name, ss->subsys_id); |
4625 | BUG(); | |
4626 | } | |
4627 | ||
4628 | if (ss->early_init) | |
4629 | cgroup_init_subsys(ss); | |
4630 | } | |
4631 | return 0; | |
4632 | } | |
4633 | ||
4634 | /** | |
a043e3b2 LZ |
4635 | * cgroup_init - cgroup initialization |
4636 | * | |
4637 | * Register cgroup filesystem and /proc file, and initialize | |
4638 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4639 | */ |
4640 | int __init cgroup_init(void) | |
4641 | { | |
4642 | int err; | |
4643 | int i; | |
472b1053 | 4644 | struct hlist_head *hhead; |
a424316c PM |
4645 | |
4646 | err = bdi_init(&cgroup_backing_dev_info); | |
4647 | if (err) | |
4648 | return err; | |
ddbcc7e8 | 4649 | |
be45c900 | 4650 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
ddbcc7e8 | 4651 | struct cgroup_subsys *ss = subsys[i]; |
be45c900 DW |
4652 | |
4653 | /* at bootup time, we don't worry about modular subsystems */ | |
4654 | if (!ss || ss->module) | |
4655 | continue; | |
ddbcc7e8 PM |
4656 | if (!ss->early_init) |
4657 | cgroup_init_subsys(ss); | |
38460b48 | 4658 | if (ss->use_id) |
e6a1105b | 4659 | cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]); |
ddbcc7e8 PM |
4660 | } |
4661 | ||
472b1053 LZ |
4662 | /* Add init_css_set to the hash table */ |
4663 | hhead = css_set_hash(init_css_set.subsys); | |
4664 | hlist_add_head(&init_css_set.hlist, hhead); | |
2c6ab6d2 | 4665 | BUG_ON(!init_root_id(&rootnode)); |
676db4af GKH |
4666 | |
4667 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); | |
4668 | if (!cgroup_kobj) { | |
4669 | err = -ENOMEM; | |
4670 | goto out; | |
4671 | } | |
4672 | ||
ddbcc7e8 | 4673 | err = register_filesystem(&cgroup_fs_type); |
676db4af GKH |
4674 | if (err < 0) { |
4675 | kobject_put(cgroup_kobj); | |
ddbcc7e8 | 4676 | goto out; |
676db4af | 4677 | } |
ddbcc7e8 | 4678 | |
46ae220b | 4679 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
a424316c | 4680 | |
ddbcc7e8 | 4681 | out: |
a424316c PM |
4682 | if (err) |
4683 | bdi_destroy(&cgroup_backing_dev_info); | |
4684 | ||
ddbcc7e8 PM |
4685 | return err; |
4686 | } | |
b4f48b63 | 4687 | |
a424316c PM |
4688 | /* |
4689 | * proc_cgroup_show() | |
4690 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4691 | * - Used for /proc/<pid>/cgroup. | |
4692 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4693 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4694 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4695 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4696 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4697 | * cgroup to top_cgroup. | |
4698 | */ | |
4699 | ||
4700 | /* TODO: Use a proper seq_file iterator */ | |
4701 | static int proc_cgroup_show(struct seq_file *m, void *v) | |
4702 | { | |
4703 | struct pid *pid; | |
4704 | struct task_struct *tsk; | |
4705 | char *buf; | |
4706 | int retval; | |
4707 | struct cgroupfs_root *root; | |
4708 | ||
4709 | retval = -ENOMEM; | |
4710 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4711 | if (!buf) | |
4712 | goto out; | |
4713 | ||
4714 | retval = -ESRCH; | |
4715 | pid = m->private; | |
4716 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4717 | if (!tsk) | |
4718 | goto out_free; | |
4719 | ||
4720 | retval = 0; | |
4721 | ||
4722 | mutex_lock(&cgroup_mutex); | |
4723 | ||
e5f6a860 | 4724 | for_each_active_root(root) { |
a424316c | 4725 | struct cgroup_subsys *ss; |
bd89aabc | 4726 | struct cgroup *cgrp; |
a424316c PM |
4727 | int count = 0; |
4728 | ||
2c6ab6d2 | 4729 | seq_printf(m, "%d:", root->hierarchy_id); |
a424316c PM |
4730 | for_each_subsys(root, ss) |
4731 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4732 | if (strlen(root->name)) |
4733 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4734 | root->name); | |
a424316c | 4735 | seq_putc(m, ':'); |
7717f7ba | 4736 | cgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 4737 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); |
a424316c PM |
4738 | if (retval < 0) |
4739 | goto out_unlock; | |
4740 | seq_puts(m, buf); | |
4741 | seq_putc(m, '\n'); | |
4742 | } | |
4743 | ||
4744 | out_unlock: | |
4745 | mutex_unlock(&cgroup_mutex); | |
4746 | put_task_struct(tsk); | |
4747 | out_free: | |
4748 | kfree(buf); | |
4749 | out: | |
4750 | return retval; | |
4751 | } | |
4752 | ||
4753 | static int cgroup_open(struct inode *inode, struct file *file) | |
4754 | { | |
4755 | struct pid *pid = PROC_I(inode)->pid; | |
4756 | return single_open(file, proc_cgroup_show, pid); | |
4757 | } | |
4758 | ||
828c0950 | 4759 | const struct file_operations proc_cgroup_operations = { |
a424316c PM |
4760 | .open = cgroup_open, |
4761 | .read = seq_read, | |
4762 | .llseek = seq_lseek, | |
4763 | .release = single_release, | |
4764 | }; | |
4765 | ||
4766 | /* Display information about each subsystem and each hierarchy */ | |
4767 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4768 | { | |
4769 | int i; | |
a424316c | 4770 | |
8bab8dde | 4771 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4772 | /* |
4773 | * ideally we don't want subsystems moving around while we do this. | |
4774 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4775 | * subsys/hierarchy state. | |
4776 | */ | |
a424316c | 4777 | mutex_lock(&cgroup_mutex); |
a424316c PM |
4778 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
4779 | struct cgroup_subsys *ss = subsys[i]; | |
aae8aab4 BB |
4780 | if (ss == NULL) |
4781 | continue; | |
2c6ab6d2 PM |
4782 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4783 | ss->name, ss->root->hierarchy_id, | |
8bab8dde | 4784 | ss->root->number_of_cgroups, !ss->disabled); |
a424316c PM |
4785 | } |
4786 | mutex_unlock(&cgroup_mutex); | |
4787 | return 0; | |
4788 | } | |
4789 | ||
4790 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4791 | { | |
9dce07f1 | 4792 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4793 | } |
4794 | ||
828c0950 | 4795 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4796 | .open = cgroupstats_open, |
4797 | .read = seq_read, | |
4798 | .llseek = seq_lseek, | |
4799 | .release = single_release, | |
4800 | }; | |
4801 | ||
b4f48b63 PM |
4802 | /** |
4803 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4804 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4805 | * |
4806 | * Description: A task inherits its parent's cgroup at fork(). | |
4807 | * | |
4808 | * A pointer to the shared css_set was automatically copied in | |
4809 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
9bb71308 TH |
4810 | * it was not made under the protection of RCU or cgroup_mutex, so |
4811 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
4812 | * have already changed current->cgroups, allowing the previously | |
4813 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
4814 | * |
4815 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4816 | * task, and the passed argument 'child' points to the child task. | |
4817 | */ | |
4818 | void cgroup_fork(struct task_struct *child) | |
4819 | { | |
9bb71308 | 4820 | task_lock(current); |
817929ec PM |
4821 | child->cgroups = current->cgroups; |
4822 | get_css_set(child->cgroups); | |
9bb71308 | 4823 | task_unlock(current); |
817929ec | 4824 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4825 | } |
4826 | ||
817929ec | 4827 | /** |
a043e3b2 LZ |
4828 | * cgroup_post_fork - called on a new task after adding it to the task list |
4829 | * @child: the task in question | |
4830 | * | |
5edee61e TH |
4831 | * Adds the task to the list running through its css_set if necessary and |
4832 | * call the subsystem fork() callbacks. Has to be after the task is | |
4833 | * visible on the task list in case we race with the first call to | |
4834 | * cgroup_iter_start() - to guarantee that the new task ends up on its | |
4835 | * list. | |
a043e3b2 | 4836 | */ |
817929ec PM |
4837 | void cgroup_post_fork(struct task_struct *child) |
4838 | { | |
5edee61e TH |
4839 | int i; |
4840 | ||
3ce3230a FW |
4841 | /* |
4842 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
4843 | * under the tasklist_lock and we read it here after we added the child | |
4844 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
4845 | * yet in the tasklist when we walked through it from | |
4846 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
4847 | * should be visible now due to the paired locking and barriers implied | |
4848 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
4849 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
4850 | * lock on fork. | |
4851 | */ | |
817929ec PM |
4852 | if (use_task_css_set_links) { |
4853 | write_lock(&css_set_lock); | |
d8783832 TH |
4854 | task_lock(child); |
4855 | if (list_empty(&child->cg_list)) | |
817929ec | 4856 | list_add(&child->cg_list, &child->cgroups->tasks); |
d8783832 | 4857 | task_unlock(child); |
817929ec PM |
4858 | write_unlock(&css_set_lock); |
4859 | } | |
5edee61e TH |
4860 | |
4861 | /* | |
4862 | * Call ss->fork(). This must happen after @child is linked on | |
4863 | * css_set; otherwise, @child might change state between ->fork() | |
4864 | * and addition to css_set. | |
4865 | */ | |
4866 | if (need_forkexit_callback) { | |
4867 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { | |
4868 | struct cgroup_subsys *ss = subsys[i]; | |
4869 | ||
4870 | /* | |
4871 | * fork/exit callbacks are supported only for | |
4872 | * builtin subsystems and we don't need further | |
4873 | * synchronization as they never go away. | |
4874 | */ | |
4875 | if (!ss || ss->module) | |
4876 | continue; | |
4877 | ||
4878 | if (ss->fork) | |
4879 | ss->fork(child); | |
4880 | } | |
4881 | } | |
817929ec | 4882 | } |
5edee61e | 4883 | |
b4f48b63 PM |
4884 | /** |
4885 | * cgroup_exit - detach cgroup from exiting task | |
4886 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4887 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4888 | * |
4889 | * Description: Detach cgroup from @tsk and release it. | |
4890 | * | |
4891 | * Note that cgroups marked notify_on_release force every task in | |
4892 | * them to take the global cgroup_mutex mutex when exiting. | |
4893 | * This could impact scaling on very large systems. Be reluctant to | |
4894 | * use notify_on_release cgroups where very high task exit scaling | |
4895 | * is required on large systems. | |
4896 | * | |
4897 | * the_top_cgroup_hack: | |
4898 | * | |
4899 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4900 | * | |
4901 | * We call cgroup_exit() while the task is still competent to | |
4902 | * handle notify_on_release(), then leave the task attached to the | |
4903 | * root cgroup in each hierarchy for the remainder of its exit. | |
4904 | * | |
4905 | * To do this properly, we would increment the reference count on | |
4906 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4907 | * code we would add a second cgroup function call, to drop that | |
4908 | * reference. This would just create an unnecessary hot spot on | |
4909 | * the top_cgroup reference count, to no avail. | |
4910 | * | |
4911 | * Normally, holding a reference to a cgroup without bumping its | |
4912 | * count is unsafe. The cgroup could go away, or someone could | |
4913 | * attach us to a different cgroup, decrementing the count on | |
4914 | * the first cgroup that we never incremented. But in this case, | |
4915 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4916 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4917 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4918 | */ |
4919 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4920 | { | |
817929ec | 4921 | struct css_set *cg; |
d41d5a01 | 4922 | int i; |
817929ec PM |
4923 | |
4924 | /* | |
4925 | * Unlink from the css_set task list if necessary. | |
4926 | * Optimistically check cg_list before taking | |
4927 | * css_set_lock | |
4928 | */ | |
4929 | if (!list_empty(&tsk->cg_list)) { | |
4930 | write_lock(&css_set_lock); | |
4931 | if (!list_empty(&tsk->cg_list)) | |
8d258797 | 4932 | list_del_init(&tsk->cg_list); |
817929ec PM |
4933 | write_unlock(&css_set_lock); |
4934 | } | |
4935 | ||
b4f48b63 PM |
4936 | /* Reassign the task to the init_css_set. */ |
4937 | task_lock(tsk); | |
817929ec PM |
4938 | cg = tsk->cgroups; |
4939 | tsk->cgroups = &init_css_set; | |
d41d5a01 PZ |
4940 | |
4941 | if (run_callbacks && need_forkexit_callback) { | |
be45c900 | 4942 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
d41d5a01 | 4943 | struct cgroup_subsys *ss = subsys[i]; |
be45c900 DW |
4944 | |
4945 | /* modular subsystems can't use callbacks */ | |
4946 | if (!ss || ss->module) | |
4947 | continue; | |
4948 | ||
d41d5a01 PZ |
4949 | if (ss->exit) { |
4950 | struct cgroup *old_cgrp = | |
4951 | rcu_dereference_raw(cg->subsys[i])->cgroup; | |
4952 | struct cgroup *cgrp = task_cgroup(tsk, i); | |
761b3ef5 | 4953 | ss->exit(cgrp, old_cgrp, tsk); |
d41d5a01 PZ |
4954 | } |
4955 | } | |
4956 | } | |
b4f48b63 | 4957 | task_unlock(tsk); |
d41d5a01 | 4958 | |
817929ec | 4959 | if (cg) |
81a6a5cd | 4960 | put_css_set_taskexit(cg); |
b4f48b63 | 4961 | } |
697f4161 | 4962 | |
a043e3b2 | 4963 | /** |
313e924c | 4964 | * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp |
a043e3b2 | 4965 | * @cgrp: the cgroup in question |
313e924c | 4966 | * @task: the task in question |
a043e3b2 | 4967 | * |
313e924c GN |
4968 | * See if @cgrp is a descendant of @task's cgroup in the appropriate |
4969 | * hierarchy. | |
697f4161 PM |
4970 | * |
4971 | * If we are sending in dummytop, then presumably we are creating | |
4972 | * the top cgroup in the subsystem. | |
4973 | * | |
4974 | * Called only by the ns (nsproxy) cgroup. | |
4975 | */ | |
313e924c | 4976 | int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task) |
697f4161 PM |
4977 | { |
4978 | int ret; | |
4979 | struct cgroup *target; | |
697f4161 | 4980 | |
bd89aabc | 4981 | if (cgrp == dummytop) |
697f4161 PM |
4982 | return 1; |
4983 | ||
7717f7ba | 4984 | target = task_cgroup_from_root(task, cgrp->root); |
bd89aabc PM |
4985 | while (cgrp != target && cgrp!= cgrp->top_cgroup) |
4986 | cgrp = cgrp->parent; | |
4987 | ret = (cgrp == target); | |
697f4161 PM |
4988 | return ret; |
4989 | } | |
81a6a5cd | 4990 | |
bd89aabc | 4991 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd PM |
4992 | { |
4993 | /* All of these checks rely on RCU to keep the cgroup | |
4994 | * structure alive */ | |
bd89aabc PM |
4995 | if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count) |
4996 | && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) { | |
81a6a5cd PM |
4997 | /* Control Group is currently removeable. If it's not |
4998 | * already queued for a userspace notification, queue | |
4999 | * it now */ | |
5000 | int need_schedule_work = 0; | |
cdcc136f | 5001 | raw_spin_lock(&release_list_lock); |
bd89aabc PM |
5002 | if (!cgroup_is_removed(cgrp) && |
5003 | list_empty(&cgrp->release_list)) { | |
5004 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
5005 | need_schedule_work = 1; |
5006 | } | |
cdcc136f | 5007 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5008 | if (need_schedule_work) |
5009 | schedule_work(&release_agent_work); | |
5010 | } | |
5011 | } | |
5012 | ||
d7b9fff7 | 5013 | /* Caller must verify that the css is not for root cgroup */ |
28b4c27b TH |
5014 | bool __css_tryget(struct cgroup_subsys_state *css) |
5015 | { | |
e9316080 TH |
5016 | while (true) { |
5017 | int t, v; | |
28b4c27b | 5018 | |
e9316080 TH |
5019 | v = css_refcnt(css); |
5020 | t = atomic_cmpxchg(&css->refcnt, v, v + 1); | |
5021 | if (likely(t == v)) | |
28b4c27b | 5022 | return true; |
e9316080 TH |
5023 | else if (t < 0) |
5024 | return false; | |
28b4c27b | 5025 | cpu_relax(); |
e9316080 | 5026 | } |
28b4c27b TH |
5027 | } |
5028 | EXPORT_SYMBOL_GPL(__css_tryget); | |
5029 | ||
5030 | /* Caller must verify that the css is not for root cgroup */ | |
5031 | void __css_put(struct cgroup_subsys_state *css) | |
81a6a5cd | 5032 | { |
bd89aabc | 5033 | struct cgroup *cgrp = css->cgroup; |
8e3bbf42 | 5034 | int v; |
28b4c27b | 5035 | |
81a6a5cd | 5036 | rcu_read_lock(); |
8e3bbf42 SQ |
5037 | v = css_unbias_refcnt(atomic_dec_return(&css->refcnt)); |
5038 | ||
5039 | switch (v) { | |
48ddbe19 | 5040 | case 1: |
ec64f515 KH |
5041 | if (notify_on_release(cgrp)) { |
5042 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
5043 | check_for_release(cgrp); | |
5044 | } | |
48ddbe19 TH |
5045 | break; |
5046 | case 0: | |
ed957793 | 5047 | schedule_work(&css->dput_work); |
48ddbe19 | 5048 | break; |
81a6a5cd PM |
5049 | } |
5050 | rcu_read_unlock(); | |
5051 | } | |
67523c48 | 5052 | EXPORT_SYMBOL_GPL(__css_put); |
81a6a5cd PM |
5053 | |
5054 | /* | |
5055 | * Notify userspace when a cgroup is released, by running the | |
5056 | * configured release agent with the name of the cgroup (path | |
5057 | * relative to the root of cgroup file system) as the argument. | |
5058 | * | |
5059 | * Most likely, this user command will try to rmdir this cgroup. | |
5060 | * | |
5061 | * This races with the possibility that some other task will be | |
5062 | * attached to this cgroup before it is removed, or that some other | |
5063 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5064 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5065 | * unused, and this cgroup will be reprieved from its death sentence, | |
5066 | * to continue to serve a useful existence. Next time it's released, | |
5067 | * we will get notified again, if it still has 'notify_on_release' set. | |
5068 | * | |
5069 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5070 | * means only wait until the task is successfully execve()'d. The | |
5071 | * separate release agent task is forked by call_usermodehelper(), | |
5072 | * then control in this thread returns here, without waiting for the | |
5073 | * release agent task. We don't bother to wait because the caller of | |
5074 | * this routine has no use for the exit status of the release agent | |
5075 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5076 | */ |
81a6a5cd PM |
5077 | static void cgroup_release_agent(struct work_struct *work) |
5078 | { | |
5079 | BUG_ON(work != &release_agent_work); | |
5080 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 5081 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
5082 | while (!list_empty(&release_list)) { |
5083 | char *argv[3], *envp[3]; | |
5084 | int i; | |
e788e066 | 5085 | char *pathbuf = NULL, *agentbuf = NULL; |
bd89aabc | 5086 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
5087 | struct cgroup, |
5088 | release_list); | |
bd89aabc | 5089 | list_del_init(&cgrp->release_list); |
cdcc136f | 5090 | raw_spin_unlock(&release_list_lock); |
81a6a5cd | 5091 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
e788e066 PM |
5092 | if (!pathbuf) |
5093 | goto continue_free; | |
5094 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
5095 | goto continue_free; | |
5096 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5097 | if (!agentbuf) | |
5098 | goto continue_free; | |
81a6a5cd PM |
5099 | |
5100 | i = 0; | |
e788e066 PM |
5101 | argv[i++] = agentbuf; |
5102 | argv[i++] = pathbuf; | |
81a6a5cd PM |
5103 | argv[i] = NULL; |
5104 | ||
5105 | i = 0; | |
5106 | /* minimal command environment */ | |
5107 | envp[i++] = "HOME=/"; | |
5108 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5109 | envp[i] = NULL; | |
5110 | ||
5111 | /* Drop the lock while we invoke the usermode helper, | |
5112 | * since the exec could involve hitting disk and hence | |
5113 | * be a slow process */ | |
5114 | mutex_unlock(&cgroup_mutex); | |
5115 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 5116 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
5117 | continue_free: |
5118 | kfree(pathbuf); | |
5119 | kfree(agentbuf); | |
cdcc136f | 5120 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 5121 | } |
cdcc136f | 5122 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5123 | mutex_unlock(&cgroup_mutex); |
5124 | } | |
8bab8dde PM |
5125 | |
5126 | static int __init cgroup_disable(char *str) | |
5127 | { | |
5128 | int i; | |
5129 | char *token; | |
5130 | ||
5131 | while ((token = strsep(&str, ",")) != NULL) { | |
5132 | if (!*token) | |
5133 | continue; | |
be45c900 | 5134 | for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { |
8bab8dde PM |
5135 | struct cgroup_subsys *ss = subsys[i]; |
5136 | ||
be45c900 DW |
5137 | /* |
5138 | * cgroup_disable, being at boot time, can't | |
5139 | * know about module subsystems, so we don't | |
5140 | * worry about them. | |
5141 | */ | |
5142 | if (!ss || ss->module) | |
5143 | continue; | |
5144 | ||
8bab8dde PM |
5145 | if (!strcmp(token, ss->name)) { |
5146 | ss->disabled = 1; | |
5147 | printk(KERN_INFO "Disabling %s control group" | |
5148 | " subsystem\n", ss->name); | |
5149 | break; | |
5150 | } | |
5151 | } | |
5152 | } | |
5153 | return 1; | |
5154 | } | |
5155 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 KH |
5156 | |
5157 | /* | |
5158 | * Functons for CSS ID. | |
5159 | */ | |
5160 | ||
5161 | /* | |
5162 | *To get ID other than 0, this should be called when !cgroup_is_removed(). | |
5163 | */ | |
5164 | unsigned short css_id(struct cgroup_subsys_state *css) | |
5165 | { | |
7f0f1546 KH |
5166 | struct css_id *cssid; |
5167 | ||
5168 | /* | |
5169 | * This css_id() can return correct value when somone has refcnt | |
5170 | * on this or this is under rcu_read_lock(). Once css->id is allocated, | |
5171 | * it's unchanged until freed. | |
5172 | */ | |
28b4c27b | 5173 | cssid = rcu_dereference_check(css->id, css_refcnt(css)); |
38460b48 KH |
5174 | |
5175 | if (cssid) | |
5176 | return cssid->id; | |
5177 | return 0; | |
5178 | } | |
67523c48 | 5179 | EXPORT_SYMBOL_GPL(css_id); |
38460b48 KH |
5180 | |
5181 | unsigned short css_depth(struct cgroup_subsys_state *css) | |
5182 | { | |
7f0f1546 KH |
5183 | struct css_id *cssid; |
5184 | ||
28b4c27b | 5185 | cssid = rcu_dereference_check(css->id, css_refcnt(css)); |
38460b48 KH |
5186 | |
5187 | if (cssid) | |
5188 | return cssid->depth; | |
5189 | return 0; | |
5190 | } | |
67523c48 | 5191 | EXPORT_SYMBOL_GPL(css_depth); |
38460b48 | 5192 | |
747388d7 KH |
5193 | /** |
5194 | * css_is_ancestor - test "root" css is an ancestor of "child" | |
5195 | * @child: the css to be tested. | |
5196 | * @root: the css supporsed to be an ancestor of the child. | |
5197 | * | |
5198 | * Returns true if "root" is an ancestor of "child" in its hierarchy. Because | |
91c63734 | 5199 | * this function reads css->id, the caller must hold rcu_read_lock(). |
747388d7 KH |
5200 | * But, considering usual usage, the csses should be valid objects after test. |
5201 | * Assuming that the caller will do some action to the child if this returns | |
5202 | * returns true, the caller must take "child";s reference count. | |
5203 | * If "child" is valid object and this returns true, "root" is valid, too. | |
5204 | */ | |
5205 | ||
38460b48 | 5206 | bool css_is_ancestor(struct cgroup_subsys_state *child, |
0b7f569e | 5207 | const struct cgroup_subsys_state *root) |
38460b48 | 5208 | { |
747388d7 KH |
5209 | struct css_id *child_id; |
5210 | struct css_id *root_id; | |
38460b48 | 5211 | |
747388d7 | 5212 | child_id = rcu_dereference(child->id); |
91c63734 JW |
5213 | if (!child_id) |
5214 | return false; | |
747388d7 | 5215 | root_id = rcu_dereference(root->id); |
91c63734 JW |
5216 | if (!root_id) |
5217 | return false; | |
5218 | if (child_id->depth < root_id->depth) | |
5219 | return false; | |
5220 | if (child_id->stack[root_id->depth] != root_id->id) | |
5221 | return false; | |
5222 | return true; | |
38460b48 KH |
5223 | } |
5224 | ||
38460b48 KH |
5225 | void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css) |
5226 | { | |
5227 | struct css_id *id = css->id; | |
5228 | /* When this is called before css_id initialization, id can be NULL */ | |
5229 | if (!id) | |
5230 | return; | |
5231 | ||
5232 | BUG_ON(!ss->use_id); | |
5233 | ||
5234 | rcu_assign_pointer(id->css, NULL); | |
5235 | rcu_assign_pointer(css->id, NULL); | |
42aee6c4 | 5236 | spin_lock(&ss->id_lock); |
38460b48 | 5237 | idr_remove(&ss->idr, id->id); |
42aee6c4 | 5238 | spin_unlock(&ss->id_lock); |
025cea99 | 5239 | kfree_rcu(id, rcu_head); |
38460b48 | 5240 | } |
67523c48 | 5241 | EXPORT_SYMBOL_GPL(free_css_id); |
38460b48 KH |
5242 | |
5243 | /* | |
5244 | * This is called by init or create(). Then, calls to this function are | |
5245 | * always serialized (By cgroup_mutex() at create()). | |
5246 | */ | |
5247 | ||
5248 | static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth) | |
5249 | { | |
5250 | struct css_id *newid; | |
5251 | int myid, error, size; | |
5252 | ||
5253 | BUG_ON(!ss->use_id); | |
5254 | ||
5255 | size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1); | |
5256 | newid = kzalloc(size, GFP_KERNEL); | |
5257 | if (!newid) | |
5258 | return ERR_PTR(-ENOMEM); | |
5259 | /* get id */ | |
5260 | if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) { | |
5261 | error = -ENOMEM; | |
5262 | goto err_out; | |
5263 | } | |
42aee6c4 | 5264 | spin_lock(&ss->id_lock); |
38460b48 KH |
5265 | /* Don't use 0. allocates an ID of 1-65535 */ |
5266 | error = idr_get_new_above(&ss->idr, newid, 1, &myid); | |
42aee6c4 | 5267 | spin_unlock(&ss->id_lock); |
38460b48 KH |
5268 | |
5269 | /* Returns error when there are no free spaces for new ID.*/ | |
5270 | if (error) { | |
5271 | error = -ENOSPC; | |
5272 | goto err_out; | |
5273 | } | |
5274 | if (myid > CSS_ID_MAX) | |
5275 | goto remove_idr; | |
5276 | ||
5277 | newid->id = myid; | |
5278 | newid->depth = depth; | |
5279 | return newid; | |
5280 | remove_idr: | |
5281 | error = -ENOSPC; | |
42aee6c4 | 5282 | spin_lock(&ss->id_lock); |
38460b48 | 5283 | idr_remove(&ss->idr, myid); |
42aee6c4 | 5284 | spin_unlock(&ss->id_lock); |
38460b48 KH |
5285 | err_out: |
5286 | kfree(newid); | |
5287 | return ERR_PTR(error); | |
5288 | ||
5289 | } | |
5290 | ||
e6a1105b BB |
5291 | static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss, |
5292 | struct cgroup_subsys_state *rootcss) | |
38460b48 KH |
5293 | { |
5294 | struct css_id *newid; | |
38460b48 | 5295 | |
42aee6c4 | 5296 | spin_lock_init(&ss->id_lock); |
38460b48 KH |
5297 | idr_init(&ss->idr); |
5298 | ||
38460b48 KH |
5299 | newid = get_new_cssid(ss, 0); |
5300 | if (IS_ERR(newid)) | |
5301 | return PTR_ERR(newid); | |
5302 | ||
5303 | newid->stack[0] = newid->id; | |
5304 | newid->css = rootcss; | |
5305 | rootcss->id = newid; | |
5306 | return 0; | |
5307 | } | |
5308 | ||
5309 | static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent, | |
5310 | struct cgroup *child) | |
5311 | { | |
5312 | int subsys_id, i, depth = 0; | |
5313 | struct cgroup_subsys_state *parent_css, *child_css; | |
fae9c791 | 5314 | struct css_id *child_id, *parent_id; |
38460b48 KH |
5315 | |
5316 | subsys_id = ss->subsys_id; | |
5317 | parent_css = parent->subsys[subsys_id]; | |
5318 | child_css = child->subsys[subsys_id]; | |
38460b48 | 5319 | parent_id = parent_css->id; |
94b3dd0f | 5320 | depth = parent_id->depth + 1; |
38460b48 KH |
5321 | |
5322 | child_id = get_new_cssid(ss, depth); | |
5323 | if (IS_ERR(child_id)) | |
5324 | return PTR_ERR(child_id); | |
5325 | ||
5326 | for (i = 0; i < depth; i++) | |
5327 | child_id->stack[i] = parent_id->stack[i]; | |
5328 | child_id->stack[depth] = child_id->id; | |
5329 | /* | |
5330 | * child_id->css pointer will be set after this cgroup is available | |
5331 | * see cgroup_populate_dir() | |
5332 | */ | |
5333 | rcu_assign_pointer(child_css->id, child_id); | |
5334 | ||
5335 | return 0; | |
5336 | } | |
5337 | ||
5338 | /** | |
5339 | * css_lookup - lookup css by id | |
5340 | * @ss: cgroup subsys to be looked into. | |
5341 | * @id: the id | |
5342 | * | |
5343 | * Returns pointer to cgroup_subsys_state if there is valid one with id. | |
5344 | * NULL if not. Should be called under rcu_read_lock() | |
5345 | */ | |
5346 | struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id) | |
5347 | { | |
5348 | struct css_id *cssid = NULL; | |
5349 | ||
5350 | BUG_ON(!ss->use_id); | |
5351 | cssid = idr_find(&ss->idr, id); | |
5352 | ||
5353 | if (unlikely(!cssid)) | |
5354 | return NULL; | |
5355 | ||
5356 | return rcu_dereference(cssid->css); | |
5357 | } | |
67523c48 | 5358 | EXPORT_SYMBOL_GPL(css_lookup); |
38460b48 KH |
5359 | |
5360 | /** | |
5361 | * css_get_next - lookup next cgroup under specified hierarchy. | |
5362 | * @ss: pointer to subsystem | |
5363 | * @id: current position of iteration. | |
5364 | * @root: pointer to css. search tree under this. | |
5365 | * @foundid: position of found object. | |
5366 | * | |
5367 | * Search next css under the specified hierarchy of rootid. Calling under | |
5368 | * rcu_read_lock() is necessary. Returns NULL if it reaches the end. | |
5369 | */ | |
5370 | struct cgroup_subsys_state * | |
5371 | css_get_next(struct cgroup_subsys *ss, int id, | |
5372 | struct cgroup_subsys_state *root, int *foundid) | |
5373 | { | |
5374 | struct cgroup_subsys_state *ret = NULL; | |
5375 | struct css_id *tmp; | |
5376 | int tmpid; | |
5377 | int rootid = css_id(root); | |
5378 | int depth = css_depth(root); | |
5379 | ||
5380 | if (!rootid) | |
5381 | return NULL; | |
5382 | ||
5383 | BUG_ON(!ss->use_id); | |
ca464d69 HD |
5384 | WARN_ON_ONCE(!rcu_read_lock_held()); |
5385 | ||
38460b48 KH |
5386 | /* fill start point for scan */ |
5387 | tmpid = id; | |
5388 | while (1) { | |
5389 | /* | |
5390 | * scan next entry from bitmap(tree), tmpid is updated after | |
5391 | * idr_get_next(). | |
5392 | */ | |
38460b48 | 5393 | tmp = idr_get_next(&ss->idr, &tmpid); |
38460b48 KH |
5394 | if (!tmp) |
5395 | break; | |
5396 | if (tmp->depth >= depth && tmp->stack[depth] == rootid) { | |
5397 | ret = rcu_dereference(tmp->css); | |
5398 | if (ret) { | |
5399 | *foundid = tmpid; | |
5400 | break; | |
5401 | } | |
5402 | } | |
5403 | /* continue to scan from next id */ | |
5404 | tmpid = tmpid + 1; | |
5405 | } | |
5406 | return ret; | |
5407 | } | |
5408 | ||
e5d1367f SE |
5409 | /* |
5410 | * get corresponding css from file open on cgroupfs directory | |
5411 | */ | |
5412 | struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id) | |
5413 | { | |
5414 | struct cgroup *cgrp; | |
5415 | struct inode *inode; | |
5416 | struct cgroup_subsys_state *css; | |
5417 | ||
5418 | inode = f->f_dentry->d_inode; | |
5419 | /* check in cgroup filesystem dir */ | |
5420 | if (inode->i_op != &cgroup_dir_inode_operations) | |
5421 | return ERR_PTR(-EBADF); | |
5422 | ||
5423 | if (id < 0 || id >= CGROUP_SUBSYS_COUNT) | |
5424 | return ERR_PTR(-EINVAL); | |
5425 | ||
5426 | /* get cgroup */ | |
5427 | cgrp = __d_cgrp(f->f_dentry); | |
5428 | css = cgrp->subsys[id]; | |
5429 | return css ? css : ERR_PTR(-ENOENT); | |
5430 | } | |
5431 | ||
fe693435 | 5432 | #ifdef CONFIG_CGROUP_DEBUG |
761b3ef5 | 5433 | static struct cgroup_subsys_state *debug_create(struct cgroup *cont) |
fe693435 PM |
5434 | { |
5435 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5436 | ||
5437 | if (!css) | |
5438 | return ERR_PTR(-ENOMEM); | |
5439 | ||
5440 | return css; | |
5441 | } | |
5442 | ||
761b3ef5 | 5443 | static void debug_destroy(struct cgroup *cont) |
fe693435 PM |
5444 | { |
5445 | kfree(cont->subsys[debug_subsys_id]); | |
5446 | } | |
5447 | ||
5448 | static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft) | |
5449 | { | |
5450 | return atomic_read(&cont->count); | |
5451 | } | |
5452 | ||
5453 | static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft) | |
5454 | { | |
5455 | return cgroup_task_count(cont); | |
5456 | } | |
5457 | ||
5458 | static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft) | |
5459 | { | |
5460 | return (u64)(unsigned long)current->cgroups; | |
5461 | } | |
5462 | ||
5463 | static u64 current_css_set_refcount_read(struct cgroup *cont, | |
5464 | struct cftype *cft) | |
5465 | { | |
5466 | u64 count; | |
5467 | ||
5468 | rcu_read_lock(); | |
5469 | count = atomic_read(¤t->cgroups->refcount); | |
5470 | rcu_read_unlock(); | |
5471 | return count; | |
5472 | } | |
5473 | ||
7717f7ba PM |
5474 | static int current_css_set_cg_links_read(struct cgroup *cont, |
5475 | struct cftype *cft, | |
5476 | struct seq_file *seq) | |
5477 | { | |
5478 | struct cg_cgroup_link *link; | |
5479 | struct css_set *cg; | |
5480 | ||
5481 | read_lock(&css_set_lock); | |
5482 | rcu_read_lock(); | |
5483 | cg = rcu_dereference(current->cgroups); | |
5484 | list_for_each_entry(link, &cg->cg_links, cg_link_list) { | |
5485 | struct cgroup *c = link->cgrp; | |
5486 | const char *name; | |
5487 | ||
5488 | if (c->dentry) | |
5489 | name = c->dentry->d_name.name; | |
5490 | else | |
5491 | name = "?"; | |
2c6ab6d2 PM |
5492 | seq_printf(seq, "Root %d group %s\n", |
5493 | c->root->hierarchy_id, name); | |
7717f7ba PM |
5494 | } |
5495 | rcu_read_unlock(); | |
5496 | read_unlock(&css_set_lock); | |
5497 | return 0; | |
5498 | } | |
5499 | ||
5500 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
5501 | static int cgroup_css_links_read(struct cgroup *cont, | |
5502 | struct cftype *cft, | |
5503 | struct seq_file *seq) | |
5504 | { | |
5505 | struct cg_cgroup_link *link; | |
5506 | ||
5507 | read_lock(&css_set_lock); | |
5508 | list_for_each_entry(link, &cont->css_sets, cgrp_link_list) { | |
5509 | struct css_set *cg = link->cg; | |
5510 | struct task_struct *task; | |
5511 | int count = 0; | |
5512 | seq_printf(seq, "css_set %p\n", cg); | |
5513 | list_for_each_entry(task, &cg->tasks, cg_list) { | |
5514 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { | |
5515 | seq_puts(seq, " ...\n"); | |
5516 | break; | |
5517 | } else { | |
5518 | seq_printf(seq, " task %d\n", | |
5519 | task_pid_vnr(task)); | |
5520 | } | |
5521 | } | |
5522 | } | |
5523 | read_unlock(&css_set_lock); | |
5524 | return 0; | |
5525 | } | |
5526 | ||
fe693435 PM |
5527 | static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft) |
5528 | { | |
5529 | return test_bit(CGRP_RELEASABLE, &cgrp->flags); | |
5530 | } | |
5531 | ||
5532 | static struct cftype debug_files[] = { | |
5533 | { | |
5534 | .name = "cgroup_refcount", | |
5535 | .read_u64 = cgroup_refcount_read, | |
5536 | }, | |
5537 | { | |
5538 | .name = "taskcount", | |
5539 | .read_u64 = debug_taskcount_read, | |
5540 | }, | |
5541 | ||
5542 | { | |
5543 | .name = "current_css_set", | |
5544 | .read_u64 = current_css_set_read, | |
5545 | }, | |
5546 | ||
5547 | { | |
5548 | .name = "current_css_set_refcount", | |
5549 | .read_u64 = current_css_set_refcount_read, | |
5550 | }, | |
5551 | ||
7717f7ba PM |
5552 | { |
5553 | .name = "current_css_set_cg_links", | |
5554 | .read_seq_string = current_css_set_cg_links_read, | |
5555 | }, | |
5556 | ||
5557 | { | |
5558 | .name = "cgroup_css_links", | |
5559 | .read_seq_string = cgroup_css_links_read, | |
5560 | }, | |
5561 | ||
fe693435 PM |
5562 | { |
5563 | .name = "releasable", | |
5564 | .read_u64 = releasable_read, | |
5565 | }, | |
fe693435 | 5566 | |
4baf6e33 TH |
5567 | { } /* terminate */ |
5568 | }; | |
fe693435 PM |
5569 | |
5570 | struct cgroup_subsys debug_subsys = { | |
5571 | .name = "debug", | |
5572 | .create = debug_create, | |
5573 | .destroy = debug_destroy, | |
fe693435 | 5574 | .subsys_id = debug_subsys_id, |
4baf6e33 | 5575 | .base_cftypes = debug_files, |
fe693435 PM |
5576 | }; |
5577 | #endif /* CONFIG_CGROUP_DEBUG */ |