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 | 32 | #include <linux/errno.h> |
2ce9738b | 33 | #include <linux/init_task.h> |
ddbcc7e8 PM |
34 | #include <linux/kernel.h> |
35 | #include <linux/list.h> | |
36 | #include <linux/mm.h> | |
37 | #include <linux/mutex.h> | |
38 | #include <linux/mount.h> | |
39 | #include <linux/pagemap.h> | |
a424316c | 40 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
41 | #include <linux/rcupdate.h> |
42 | #include <linux/sched.h> | |
817929ec | 43 | #include <linux/backing-dev.h> |
ddbcc7e8 PM |
44 | #include <linux/seq_file.h> |
45 | #include <linux/slab.h> | |
46 | #include <linux/magic.h> | |
47 | #include <linux/spinlock.h> | |
48 | #include <linux/string.h> | |
bbcb81d0 | 49 | #include <linux/sort.h> |
81a6a5cd | 50 | #include <linux/kmod.h> |
e6a1105b | 51 | #include <linux/module.h> |
846c7bb0 BS |
52 | #include <linux/delayacct.h> |
53 | #include <linux/cgroupstats.h> | |
0ac801fe | 54 | #include <linux/hashtable.h> |
3f8206d4 | 55 | #include <linux/namei.h> |
096b7fe0 | 56 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 57 | #include <linux/idr.h> |
d1d9fd33 | 58 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
0dea1168 KS |
59 | #include <linux/eventfd.h> |
60 | #include <linux/poll.h> | |
081aa458 | 61 | #include <linux/flex_array.h> /* used in cgroup_attach_task */ |
c4c27fbd | 62 | #include <linux/kthread.h> |
4e10f3c9 | 63 | #include <linux/file.h> |
846c7bb0 | 64 | |
60063497 | 65 | #include <linux/atomic.h> |
ddbcc7e8 | 66 | |
e25e2cbb TH |
67 | /* |
68 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
69 | * hierarchy must be performed while holding it. | |
70 | * | |
71 | * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify | |
72 | * cgroupfs_root of any cgroup hierarchy - subsys list, flags, | |
73 | * release_agent_path and so on. Modifying requires both cgroup_mutex and | |
74 | * cgroup_root_mutex. Readers can acquire either of the two. This is to | |
75 | * break the following locking order cycle. | |
76 | * | |
77 | * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem | |
78 | * B. namespace_sem -> cgroup_mutex | |
79 | * | |
80 | * B happens only through cgroup_show_options() and using cgroup_root_mutex | |
81 | * breaks it. | |
82 | */ | |
2219449a TH |
83 | #ifdef CONFIG_PROVE_RCU |
84 | DEFINE_MUTEX(cgroup_mutex); | |
8af01f56 | 85 | EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for lockdep */ |
2219449a | 86 | #else |
81a6a5cd | 87 | static DEFINE_MUTEX(cgroup_mutex); |
2219449a TH |
88 | #endif |
89 | ||
e25e2cbb | 90 | static DEFINE_MUTEX(cgroup_root_mutex); |
81a6a5cd | 91 | |
aae8aab4 BB |
92 | /* |
93 | * Generate an array of cgroup subsystem pointers. At boot time, this is | |
be45c900 | 94 | * populated with the built in subsystems, and modular subsystems are |
aae8aab4 BB |
95 | * registered after that. The mutable section of this array is protected by |
96 | * cgroup_mutex. | |
97 | */ | |
80f4c877 | 98 | #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys, |
5fc0b025 | 99 | #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option) |
9871bf95 | 100 | static struct cgroup_subsys *cgroup_subsys[CGROUP_SUBSYS_COUNT] = { |
ddbcc7e8 PM |
101 | #include <linux/cgroup_subsys.h> |
102 | }; | |
103 | ||
ddbcc7e8 | 104 | /* |
9871bf95 TH |
105 | * The dummy hierarchy, reserved for the subsystems that are otherwise |
106 | * unattached - it never has more than a single cgroup, and all tasks are | |
107 | * part of that cgroup. | |
ddbcc7e8 | 108 | */ |
9871bf95 TH |
109 | static struct cgroupfs_root cgroup_dummy_root; |
110 | ||
111 | /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */ | |
112 | static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup; | |
ddbcc7e8 | 113 | |
05ef1d7c TH |
114 | /* |
115 | * cgroupfs file entry, pointed to from leaf dentry->d_fsdata. | |
116 | */ | |
117 | struct cfent { | |
118 | struct list_head node; | |
119 | struct dentry *dentry; | |
120 | struct cftype *type; | |
105347ba | 121 | struct cgroup_subsys_state *css; |
712317ad LZ |
122 | |
123 | /* file xattrs */ | |
124 | struct simple_xattrs xattrs; | |
05ef1d7c TH |
125 | }; |
126 | ||
0dea1168 | 127 | /* |
25985edc | 128 | * cgroup_event represents events which userspace want to receive. |
0dea1168 KS |
129 | */ |
130 | struct cgroup_event { | |
131 | /* | |
81eeaf04 | 132 | * css which the event belongs to. |
0dea1168 | 133 | */ |
81eeaf04 | 134 | struct cgroup_subsys_state *css; |
0dea1168 KS |
135 | /* |
136 | * Control file which the event associated. | |
137 | */ | |
138 | struct cftype *cft; | |
139 | /* | |
140 | * eventfd to signal userspace about the event. | |
141 | */ | |
142 | struct eventfd_ctx *eventfd; | |
143 | /* | |
144 | * Each of these stored in a list by the cgroup. | |
145 | */ | |
146 | struct list_head list; | |
147 | /* | |
148 | * All fields below needed to unregister event when | |
149 | * userspace closes eventfd. | |
150 | */ | |
151 | poll_table pt; | |
152 | wait_queue_head_t *wqh; | |
153 | wait_queue_t wait; | |
154 | struct work_struct remove; | |
155 | }; | |
38460b48 | 156 | |
ddbcc7e8 PM |
157 | /* The list of hierarchy roots */ |
158 | ||
9871bf95 TH |
159 | static LIST_HEAD(cgroup_roots); |
160 | static int cgroup_root_count; | |
ddbcc7e8 | 161 | |
54e7b4eb TH |
162 | /* |
163 | * Hierarchy ID allocation and mapping. It follows the same exclusion | |
164 | * rules as other root ops - both cgroup_mutex and cgroup_root_mutex for | |
165 | * writes, either for reads. | |
166 | */ | |
1a574231 | 167 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 168 | |
65dff759 LZ |
169 | static struct cgroup_name root_cgroup_name = { .name = "/" }; |
170 | ||
794611a1 LZ |
171 | /* |
172 | * Assign a monotonically increasing serial number to cgroups. It | |
173 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
174 | * numbers. Also, as cgroups are always appended to the parent's | |
175 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
176 | * the ascending serial number order on the list. Protected by |
177 | * cgroup_mutex. | |
794611a1 | 178 | */ |
00356bd5 | 179 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 180 | |
ddbcc7e8 | 181 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
182 | * check for fork/exit handlers to call. This avoids us having to do |
183 | * extra work in the fork/exit path if none of the subsystems need to | |
184 | * be called. | |
ddbcc7e8 | 185 | */ |
8947f9d5 | 186 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 187 | |
628f7cd4 TH |
188 | static struct cftype cgroup_base_files[]; |
189 | ||
f20104de | 190 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 191 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
2bb566cb TH |
192 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
193 | bool is_add); | |
42809dd4 | 194 | |
95109b62 TH |
195 | /** |
196 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
197 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 198 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 199 | * |
ca8bdcaf TH |
200 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
201 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
202 | * the caller is responsible for pinning the returned css if it wants to | |
203 | * keep accessing it outside the said locks. This function may return | |
204 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
205 | */ |
206 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 207 | struct cgroup_subsys *ss) |
95109b62 | 208 | { |
ca8bdcaf TH |
209 | if (ss) |
210 | return rcu_dereference_check(cgrp->subsys[ss->subsys_id], | |
211 | lockdep_is_held(&cgroup_mutex)); | |
212 | else | |
213 | return &cgrp->dummy_css; | |
95109b62 | 214 | } |
42809dd4 | 215 | |
ddbcc7e8 | 216 | /* convenient tests for these bits */ |
54766d4a | 217 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 218 | { |
54766d4a | 219 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
220 | } |
221 | ||
78574cf9 LZ |
222 | /** |
223 | * cgroup_is_descendant - test ancestry | |
224 | * @cgrp: the cgroup to be tested | |
225 | * @ancestor: possible ancestor of @cgrp | |
226 | * | |
227 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
228 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
229 | * and @ancestor are accessible. | |
230 | */ | |
231 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
232 | { | |
233 | while (cgrp) { | |
234 | if (cgrp == ancestor) | |
235 | return true; | |
236 | cgrp = cgrp->parent; | |
237 | } | |
238 | return false; | |
239 | } | |
240 | EXPORT_SYMBOL_GPL(cgroup_is_descendant); | |
ddbcc7e8 | 241 | |
e9685a03 | 242 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
243 | { |
244 | const int bits = | |
bd89aabc PM |
245 | (1 << CGRP_RELEASABLE) | |
246 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
247 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
248 | } |
249 | ||
e9685a03 | 250 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 251 | { |
bd89aabc | 252 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
253 | } |
254 | ||
30159ec7 TH |
255 | /** |
256 | * for_each_subsys - iterate all loaded cgroup subsystems | |
257 | * @ss: the iteration cursor | |
258 | * @i: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
259 | * | |
260 | * Should be called under cgroup_mutex. | |
261 | */ | |
262 | #define for_each_subsys(ss, i) \ | |
263 | for ((i) = 0; (i) < CGROUP_SUBSYS_COUNT; (i)++) \ | |
264 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
265 | !((ss) = cgroup_subsys[i]); })) { } \ | |
266 | else | |
267 | ||
268 | /** | |
269 | * for_each_builtin_subsys - iterate all built-in cgroup subsystems | |
270 | * @ss: the iteration cursor | |
271 | * @i: the index of @ss, CGROUP_BUILTIN_SUBSYS_COUNT after reaching the end | |
272 | * | |
273 | * Bulit-in subsystems are always present and iteration itself doesn't | |
274 | * require any synchronization. | |
275 | */ | |
276 | #define for_each_builtin_subsys(ss, i) \ | |
277 | for ((i) = 0; (i) < CGROUP_BUILTIN_SUBSYS_COUNT && \ | |
278 | (((ss) = cgroup_subsys[i]) || true); (i)++) | |
279 | ||
5549c497 TH |
280 | /* iterate each subsystem attached to a hierarchy */ |
281 | #define for_each_root_subsys(root, ss) \ | |
282 | list_for_each_entry((ss), &(root)->subsys_list, sibling) | |
ddbcc7e8 | 283 | |
5549c497 TH |
284 | /* iterate across the active hierarchies */ |
285 | #define for_each_active_root(root) \ | |
286 | list_for_each_entry((root), &cgroup_roots, root_list) | |
ddbcc7e8 | 287 | |
f6ea9372 TH |
288 | static inline struct cgroup *__d_cgrp(struct dentry *dentry) |
289 | { | |
290 | return dentry->d_fsdata; | |
291 | } | |
292 | ||
05ef1d7c | 293 | static inline struct cfent *__d_cfe(struct dentry *dentry) |
f6ea9372 TH |
294 | { |
295 | return dentry->d_fsdata; | |
296 | } | |
297 | ||
05ef1d7c TH |
298 | static inline struct cftype *__d_cft(struct dentry *dentry) |
299 | { | |
300 | return __d_cfe(dentry)->type; | |
301 | } | |
302 | ||
7ae1bad9 TH |
303 | /** |
304 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
305 | * @cgrp: the cgroup to be checked for liveness | |
306 | * | |
47cfcd09 TH |
307 | * On success, returns true; the mutex should be later unlocked. On |
308 | * failure returns false with no lock held. | |
7ae1bad9 | 309 | */ |
b9777cf8 | 310 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
311 | { |
312 | mutex_lock(&cgroup_mutex); | |
54766d4a | 313 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
314 | mutex_unlock(&cgroup_mutex); |
315 | return false; | |
316 | } | |
317 | return true; | |
318 | } | |
7ae1bad9 | 319 | |
81a6a5cd PM |
320 | /* the list of cgroups eligible for automatic release. Protected by |
321 | * release_list_lock */ | |
322 | static LIST_HEAD(release_list); | |
cdcc136f | 323 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
324 | static void cgroup_release_agent(struct work_struct *work); |
325 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 326 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 327 | |
69d0206c TH |
328 | /* |
329 | * A cgroup can be associated with multiple css_sets as different tasks may | |
330 | * belong to different cgroups on different hierarchies. In the other | |
331 | * direction, a css_set is naturally associated with multiple cgroups. | |
332 | * This M:N relationship is represented by the following link structure | |
333 | * which exists for each association and allows traversing the associations | |
334 | * from both sides. | |
335 | */ | |
336 | struct cgrp_cset_link { | |
337 | /* the cgroup and css_set this link associates */ | |
338 | struct cgroup *cgrp; | |
339 | struct css_set *cset; | |
340 | ||
341 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
342 | struct list_head cset_link; | |
343 | ||
344 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
345 | struct list_head cgrp_link; | |
817929ec PM |
346 | }; |
347 | ||
348 | /* The default css_set - used by init and its children prior to any | |
349 | * hierarchies being mounted. It contains a pointer to the root state | |
350 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
351 | * reference-counted, to improve performance when child cgroups | |
352 | * haven't been created. | |
353 | */ | |
354 | ||
355 | static struct css_set init_css_set; | |
69d0206c | 356 | static struct cgrp_cset_link init_cgrp_cset_link; |
817929ec | 357 | |
0942eeee TH |
358 | /* |
359 | * css_set_lock protects the list of css_set objects, and the chain of | |
360 | * tasks off each css_set. Nests outside task->alloc_lock due to | |
72ec7029 | 361 | * css_task_iter_start(). |
0942eeee | 362 | */ |
817929ec PM |
363 | static DEFINE_RWLOCK(css_set_lock); |
364 | static int css_set_count; | |
365 | ||
7717f7ba PM |
366 | /* |
367 | * hash table for cgroup groups. This improves the performance to find | |
368 | * an existing css_set. This hash doesn't (currently) take into | |
369 | * account cgroups in empty hierarchies. | |
370 | */ | |
472b1053 | 371 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 372 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 373 | |
0ac801fe | 374 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 375 | { |
0ac801fe | 376 | unsigned long key = 0UL; |
30159ec7 TH |
377 | struct cgroup_subsys *ss; |
378 | int i; | |
472b1053 | 379 | |
30159ec7 | 380 | for_each_subsys(ss, i) |
0ac801fe LZ |
381 | key += (unsigned long)css[i]; |
382 | key = (key >> 16) ^ key; | |
472b1053 | 383 | |
0ac801fe | 384 | return key; |
472b1053 LZ |
385 | } |
386 | ||
0942eeee TH |
387 | /* |
388 | * We don't maintain the lists running through each css_set to its task | |
72ec7029 TH |
389 | * until after the first call to css_task_iter_start(). This reduces the |
390 | * fork()/exit() overhead for people who have cgroups compiled into their | |
391 | * kernel but not actually in use. | |
0942eeee | 392 | */ |
8947f9d5 | 393 | static int use_task_css_set_links __read_mostly; |
817929ec | 394 | |
5abb8855 | 395 | static void __put_css_set(struct css_set *cset, int taskexit) |
b4f48b63 | 396 | { |
69d0206c | 397 | struct cgrp_cset_link *link, *tmp_link; |
5abb8855 | 398 | |
146aa1bd LJ |
399 | /* |
400 | * Ensure that the refcount doesn't hit zero while any readers | |
401 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
402 | * rwlock | |
403 | */ | |
5abb8855 | 404 | if (atomic_add_unless(&cset->refcount, -1, 1)) |
146aa1bd LJ |
405 | return; |
406 | write_lock(&css_set_lock); | |
5abb8855 | 407 | if (!atomic_dec_and_test(&cset->refcount)) { |
146aa1bd LJ |
408 | write_unlock(&css_set_lock); |
409 | return; | |
410 | } | |
81a6a5cd | 411 | |
2c6ab6d2 | 412 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
5abb8855 | 413 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
414 | css_set_count--; |
415 | ||
69d0206c | 416 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 417 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 418 | |
69d0206c TH |
419 | list_del(&link->cset_link); |
420 | list_del(&link->cgrp_link); | |
71b5707e | 421 | |
ddd69148 | 422 | /* @cgrp can't go away while we're holding css_set_lock */ |
6f3d828f | 423 | if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { |
81a6a5cd | 424 | if (taskexit) |
bd89aabc PM |
425 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
426 | check_for_release(cgrp); | |
81a6a5cd | 427 | } |
2c6ab6d2 PM |
428 | |
429 | kfree(link); | |
81a6a5cd | 430 | } |
2c6ab6d2 PM |
431 | |
432 | write_unlock(&css_set_lock); | |
5abb8855 | 433 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
434 | } |
435 | ||
817929ec PM |
436 | /* |
437 | * refcounted get/put for css_set objects | |
438 | */ | |
5abb8855 | 439 | static inline void get_css_set(struct css_set *cset) |
817929ec | 440 | { |
5abb8855 | 441 | atomic_inc(&cset->refcount); |
817929ec PM |
442 | } |
443 | ||
5abb8855 | 444 | static inline void put_css_set(struct css_set *cset) |
817929ec | 445 | { |
5abb8855 | 446 | __put_css_set(cset, 0); |
817929ec PM |
447 | } |
448 | ||
5abb8855 | 449 | static inline void put_css_set_taskexit(struct css_set *cset) |
81a6a5cd | 450 | { |
5abb8855 | 451 | __put_css_set(cset, 1); |
81a6a5cd PM |
452 | } |
453 | ||
b326f9d0 | 454 | /** |
7717f7ba | 455 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
456 | * @cset: candidate css_set being tested |
457 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
458 | * @new_cgrp: cgroup that's being entered by the task |
459 | * @template: desired set of css pointers in css_set (pre-calculated) | |
460 | * | |
6f4b7e63 | 461 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
462 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
463 | */ | |
5abb8855 TH |
464 | static bool compare_css_sets(struct css_set *cset, |
465 | struct css_set *old_cset, | |
7717f7ba PM |
466 | struct cgroup *new_cgrp, |
467 | struct cgroup_subsys_state *template[]) | |
468 | { | |
469 | struct list_head *l1, *l2; | |
470 | ||
5abb8855 | 471 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) { |
7717f7ba PM |
472 | /* Not all subsystems matched */ |
473 | return false; | |
474 | } | |
475 | ||
476 | /* | |
477 | * Compare cgroup pointers in order to distinguish between | |
478 | * different cgroups in heirarchies with no subsystems. We | |
479 | * could get by with just this check alone (and skip the | |
480 | * memcmp above) but on most setups the memcmp check will | |
481 | * avoid the need for this more expensive check on almost all | |
482 | * candidates. | |
483 | */ | |
484 | ||
69d0206c TH |
485 | l1 = &cset->cgrp_links; |
486 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 487 | while (1) { |
69d0206c | 488 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 489 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
490 | |
491 | l1 = l1->next; | |
492 | l2 = l2->next; | |
493 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
494 | if (l1 == &cset->cgrp_links) { |
495 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
496 | break; |
497 | } else { | |
69d0206c | 498 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
499 | } |
500 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
501 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
502 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
503 | cgrp1 = link1->cgrp; | |
504 | cgrp2 = link2->cgrp; | |
7717f7ba | 505 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 506 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
507 | |
508 | /* | |
509 | * If this hierarchy is the hierarchy of the cgroup | |
510 | * that's changing, then we need to check that this | |
511 | * css_set points to the new cgroup; if it's any other | |
512 | * hierarchy, then this css_set should point to the | |
513 | * same cgroup as the old css_set. | |
514 | */ | |
5abb8855 TH |
515 | if (cgrp1->root == new_cgrp->root) { |
516 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
517 | return false; |
518 | } else { | |
5abb8855 | 519 | if (cgrp1 != cgrp2) |
7717f7ba PM |
520 | return false; |
521 | } | |
522 | } | |
523 | return true; | |
524 | } | |
525 | ||
b326f9d0 TH |
526 | /** |
527 | * find_existing_css_set - init css array and find the matching css_set | |
528 | * @old_cset: the css_set that we're using before the cgroup transition | |
529 | * @cgrp: the cgroup that we're moving into | |
530 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 531 | */ |
5abb8855 TH |
532 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
533 | struct cgroup *cgrp, | |
534 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 535 | { |
bd89aabc | 536 | struct cgroupfs_root *root = cgrp->root; |
30159ec7 | 537 | struct cgroup_subsys *ss; |
5abb8855 | 538 | struct css_set *cset; |
0ac801fe | 539 | unsigned long key; |
b326f9d0 | 540 | int i; |
817929ec | 541 | |
aae8aab4 BB |
542 | /* |
543 | * Build the set of subsystem state objects that we want to see in the | |
544 | * new css_set. while subsystems can change globally, the entries here | |
545 | * won't change, so no need for locking. | |
546 | */ | |
30159ec7 | 547 | for_each_subsys(ss, i) { |
a1a71b45 | 548 | if (root->subsys_mask & (1UL << i)) { |
817929ec PM |
549 | /* Subsystem is in this hierarchy. So we want |
550 | * the subsystem state from the new | |
551 | * cgroup */ | |
ca8bdcaf | 552 | template[i] = cgroup_css(cgrp, ss); |
817929ec PM |
553 | } else { |
554 | /* Subsystem is not in this hierarchy, so we | |
555 | * don't want to change the subsystem state */ | |
5abb8855 | 556 | template[i] = old_cset->subsys[i]; |
817929ec PM |
557 | } |
558 | } | |
559 | ||
0ac801fe | 560 | key = css_set_hash(template); |
5abb8855 TH |
561 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
562 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
563 | continue; |
564 | ||
565 | /* This css_set matches what we need */ | |
5abb8855 | 566 | return cset; |
472b1053 | 567 | } |
817929ec PM |
568 | |
569 | /* No existing cgroup group matched */ | |
570 | return NULL; | |
571 | } | |
572 | ||
69d0206c | 573 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 574 | { |
69d0206c | 575 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 576 | |
69d0206c TH |
577 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
578 | list_del(&link->cset_link); | |
36553434 LZ |
579 | kfree(link); |
580 | } | |
581 | } | |
582 | ||
69d0206c TH |
583 | /** |
584 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
585 | * @count: the number of links to allocate | |
586 | * @tmp_links: list_head the allocated links are put on | |
587 | * | |
588 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
589 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 590 | */ |
69d0206c | 591 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 592 | { |
69d0206c | 593 | struct cgrp_cset_link *link; |
817929ec | 594 | int i; |
69d0206c TH |
595 | |
596 | INIT_LIST_HEAD(tmp_links); | |
597 | ||
817929ec | 598 | for (i = 0; i < count; i++) { |
f4f4be2b | 599 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 600 | if (!link) { |
69d0206c | 601 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
602 | return -ENOMEM; |
603 | } | |
69d0206c | 604 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
605 | } |
606 | return 0; | |
607 | } | |
608 | ||
c12f65d4 LZ |
609 | /** |
610 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 611 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 612 | * @cset: the css_set to be linked |
c12f65d4 LZ |
613 | * @cgrp: the destination cgroup |
614 | */ | |
69d0206c TH |
615 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
616 | struct cgroup *cgrp) | |
c12f65d4 | 617 | { |
69d0206c | 618 | struct cgrp_cset_link *link; |
c12f65d4 | 619 | |
69d0206c TH |
620 | BUG_ON(list_empty(tmp_links)); |
621 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); | |
622 | link->cset = cset; | |
7717f7ba | 623 | link->cgrp = cgrp; |
69d0206c | 624 | list_move(&link->cset_link, &cgrp->cset_links); |
7717f7ba PM |
625 | /* |
626 | * Always add links to the tail of the list so that the list | |
627 | * is sorted by order of hierarchy creation | |
628 | */ | |
69d0206c | 629 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
630 | } |
631 | ||
b326f9d0 TH |
632 | /** |
633 | * find_css_set - return a new css_set with one cgroup updated | |
634 | * @old_cset: the baseline css_set | |
635 | * @cgrp: the cgroup to be updated | |
636 | * | |
637 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
638 | * substituted into the appropriate hierarchy. | |
817929ec | 639 | */ |
5abb8855 TH |
640 | static struct css_set *find_css_set(struct css_set *old_cset, |
641 | struct cgroup *cgrp) | |
817929ec | 642 | { |
b326f9d0 | 643 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 644 | struct css_set *cset; |
69d0206c TH |
645 | struct list_head tmp_links; |
646 | struct cgrp_cset_link *link; | |
0ac801fe | 647 | unsigned long key; |
472b1053 | 648 | |
b326f9d0 TH |
649 | lockdep_assert_held(&cgroup_mutex); |
650 | ||
817929ec PM |
651 | /* First see if we already have a cgroup group that matches |
652 | * the desired set */ | |
7e9abd89 | 653 | read_lock(&css_set_lock); |
5abb8855 TH |
654 | cset = find_existing_css_set(old_cset, cgrp, template); |
655 | if (cset) | |
656 | get_css_set(cset); | |
7e9abd89 | 657 | read_unlock(&css_set_lock); |
817929ec | 658 | |
5abb8855 TH |
659 | if (cset) |
660 | return cset; | |
817929ec | 661 | |
f4f4be2b | 662 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 663 | if (!cset) |
817929ec PM |
664 | return NULL; |
665 | ||
69d0206c | 666 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 667 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 668 | kfree(cset); |
817929ec PM |
669 | return NULL; |
670 | } | |
671 | ||
5abb8855 | 672 | atomic_set(&cset->refcount, 1); |
69d0206c | 673 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 TH |
674 | INIT_LIST_HEAD(&cset->tasks); |
675 | INIT_HLIST_NODE(&cset->hlist); | |
817929ec PM |
676 | |
677 | /* Copy the set of subsystem state objects generated in | |
678 | * find_existing_css_set() */ | |
5abb8855 | 679 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec PM |
680 | |
681 | write_lock(&css_set_lock); | |
682 | /* Add reference counts and links from the new css_set. */ | |
69d0206c | 683 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 684 | struct cgroup *c = link->cgrp; |
69d0206c | 685 | |
7717f7ba PM |
686 | if (c->root == cgrp->root) |
687 | c = cgrp; | |
69d0206c | 688 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 689 | } |
817929ec | 690 | |
69d0206c | 691 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 692 | |
817929ec | 693 | css_set_count++; |
472b1053 LZ |
694 | |
695 | /* Add this cgroup group to the hash table */ | |
5abb8855 TH |
696 | key = css_set_hash(cset->subsys); |
697 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 698 | |
817929ec PM |
699 | write_unlock(&css_set_lock); |
700 | ||
5abb8855 | 701 | return cset; |
b4f48b63 PM |
702 | } |
703 | ||
7717f7ba PM |
704 | /* |
705 | * Return the cgroup for "task" from the given hierarchy. Must be | |
706 | * called with cgroup_mutex held. | |
707 | */ | |
708 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
709 | struct cgroupfs_root *root) | |
710 | { | |
5abb8855 | 711 | struct css_set *cset; |
7717f7ba PM |
712 | struct cgroup *res = NULL; |
713 | ||
714 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
715 | read_lock(&css_set_lock); | |
716 | /* | |
717 | * No need to lock the task - since we hold cgroup_mutex the | |
718 | * task can't change groups, so the only thing that can happen | |
719 | * is that it exits and its css is set back to init_css_set. | |
720 | */ | |
a8ad805c | 721 | cset = task_css_set(task); |
5abb8855 | 722 | if (cset == &init_css_set) { |
7717f7ba PM |
723 | res = &root->top_cgroup; |
724 | } else { | |
69d0206c TH |
725 | struct cgrp_cset_link *link; |
726 | ||
727 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 728 | struct cgroup *c = link->cgrp; |
69d0206c | 729 | |
7717f7ba PM |
730 | if (c->root == root) { |
731 | res = c; | |
732 | break; | |
733 | } | |
734 | } | |
735 | } | |
736 | read_unlock(&css_set_lock); | |
737 | BUG_ON(!res); | |
738 | return res; | |
739 | } | |
740 | ||
ddbcc7e8 PM |
741 | /* |
742 | * There is one global cgroup mutex. We also require taking | |
743 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
744 | * See "The task_lock() exception", at the end of this comment. | |
745 | * | |
746 | * A task must hold cgroup_mutex to modify cgroups. | |
747 | * | |
748 | * Any task can increment and decrement the count field without lock. | |
749 | * So in general, code holding cgroup_mutex can't rely on the count | |
750 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 751 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
752 | * means that no tasks are currently attached, therefore there is no |
753 | * way a task attached to that cgroup can fork (the other way to | |
754 | * increment the count). So code holding cgroup_mutex can safely | |
755 | * assume that if the count is zero, it will stay zero. Similarly, if | |
756 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
757 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
758 | * needs that mutex. | |
759 | * | |
ddbcc7e8 PM |
760 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
761 | * (usually) take cgroup_mutex. These are the two most performance | |
762 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
763 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
764 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
765 | * to the release agent with the name of the cgroup (path relative to |
766 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
767 | * |
768 | * A cgroup can only be deleted if both its 'count' of using tasks | |
769 | * is zero, and its list of 'children' cgroups is empty. Since all | |
770 | * tasks in the system use _some_ cgroup, and since there is always at | |
771 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
772 | * always has either children cgroups and/or using tasks. So we don't | |
773 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
774 | * | |
775 | * The task_lock() exception | |
776 | * | |
777 | * The need for this exception arises from the action of | |
d0b2fdd2 | 778 | * cgroup_attach_task(), which overwrites one task's cgroup pointer with |
a043e3b2 | 779 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
780 | * several performance critical places that need to reference |
781 | * task->cgroup without the expense of grabbing a system global | |
782 | * mutex. Therefore except as noted below, when dereferencing or, as | |
d0b2fdd2 | 783 | * in cgroup_attach_task(), modifying a task's cgroup pointer we use |
ddbcc7e8 PM |
784 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
785 | * the task_struct routinely used for such matters. | |
786 | * | |
787 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 788 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
789 | */ |
790 | ||
ddbcc7e8 PM |
791 | /* |
792 | * A couple of forward declarations required, due to cyclic reference loop: | |
793 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> | |
794 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations | |
795 | * -> cgroup_mkdir. | |
796 | */ | |
797 | ||
18bb1db3 | 798 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode); |
ddbcc7e8 | 799 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); |
628f7cd4 | 800 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
6e1d5dcc | 801 | static const struct inode_operations cgroup_dir_inode_operations; |
828c0950 | 802 | static const struct file_operations proc_cgroupstats_operations; |
a424316c PM |
803 | |
804 | static struct backing_dev_info cgroup_backing_dev_info = { | |
d993831f | 805 | .name = "cgroup", |
e4ad08fe | 806 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
a424316c | 807 | }; |
ddbcc7e8 | 808 | |
a5e7ed32 | 809 | static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb) |
ddbcc7e8 PM |
810 | { |
811 | struct inode *inode = new_inode(sb); | |
ddbcc7e8 PM |
812 | |
813 | if (inode) { | |
85fe4025 | 814 | inode->i_ino = get_next_ino(); |
ddbcc7e8 | 815 | inode->i_mode = mode; |
76aac0e9 DH |
816 | inode->i_uid = current_fsuid(); |
817 | inode->i_gid = current_fsgid(); | |
ddbcc7e8 PM |
818 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
819 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; | |
820 | } | |
821 | return inode; | |
822 | } | |
823 | ||
65dff759 LZ |
824 | static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry) |
825 | { | |
826 | struct cgroup_name *name; | |
827 | ||
828 | name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL); | |
829 | if (!name) | |
830 | return NULL; | |
831 | strcpy(name->name, dentry->d_name.name); | |
832 | return name; | |
833 | } | |
834 | ||
be445626 LZ |
835 | static void cgroup_free_fn(struct work_struct *work) |
836 | { | |
ea15f8cc | 837 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 LZ |
838 | |
839 | mutex_lock(&cgroup_mutex); | |
be445626 LZ |
840 | cgrp->root->number_of_cgroups--; |
841 | mutex_unlock(&cgroup_mutex); | |
842 | ||
415cf07a LZ |
843 | /* |
844 | * We get a ref to the parent's dentry, and put the ref when | |
845 | * this cgroup is being freed, so it's guaranteed that the | |
846 | * parent won't be destroyed before its children. | |
847 | */ | |
848 | dput(cgrp->parent->dentry); | |
849 | ||
be445626 LZ |
850 | /* |
851 | * Drop the active superblock reference that we took when we | |
cc20e01c LZ |
852 | * created the cgroup. This will free cgrp->root, if we are |
853 | * holding the last reference to @sb. | |
be445626 LZ |
854 | */ |
855 | deactivate_super(cgrp->root->sb); | |
856 | ||
857 | /* | |
858 | * if we're getting rid of the cgroup, refcount should ensure | |
859 | * that there are no pidlists left. | |
860 | */ | |
861 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
862 | ||
863 | simple_xattrs_free(&cgrp->xattrs); | |
864 | ||
65dff759 | 865 | kfree(rcu_dereference_raw(cgrp->name)); |
be445626 LZ |
866 | kfree(cgrp); |
867 | } | |
868 | ||
869 | static void cgroup_free_rcu(struct rcu_head *head) | |
870 | { | |
871 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
872 | ||
ea15f8cc TH |
873 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
874 | schedule_work(&cgrp->destroy_work); | |
be445626 LZ |
875 | } |
876 | ||
ddbcc7e8 PM |
877 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) |
878 | { | |
879 | /* is dentry a directory ? if so, kfree() associated cgroup */ | |
880 | if (S_ISDIR(inode->i_mode)) { | |
bd89aabc | 881 | struct cgroup *cgrp = dentry->d_fsdata; |
be445626 | 882 | |
54766d4a | 883 | BUG_ON(!(cgroup_is_dead(cgrp))); |
be445626 | 884 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
05ef1d7c TH |
885 | } else { |
886 | struct cfent *cfe = __d_cfe(dentry); | |
887 | struct cgroup *cgrp = dentry->d_parent->d_fsdata; | |
888 | ||
889 | WARN_ONCE(!list_empty(&cfe->node) && | |
890 | cgrp != &cgrp->root->top_cgroup, | |
891 | "cfe still linked for %s\n", cfe->type->name); | |
712317ad | 892 | simple_xattrs_free(&cfe->xattrs); |
05ef1d7c | 893 | kfree(cfe); |
ddbcc7e8 PM |
894 | } |
895 | iput(inode); | |
896 | } | |
897 | ||
c72a04e3 AV |
898 | static int cgroup_delete(const struct dentry *d) |
899 | { | |
900 | return 1; | |
901 | } | |
902 | ||
ddbcc7e8 PM |
903 | static void remove_dir(struct dentry *d) |
904 | { | |
905 | struct dentry *parent = dget(d->d_parent); | |
906 | ||
907 | d_delete(d); | |
908 | simple_rmdir(parent->d_inode, d); | |
909 | dput(parent); | |
910 | } | |
911 | ||
2739d3cc | 912 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c TH |
913 | { |
914 | struct cfent *cfe; | |
915 | ||
916 | lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex); | |
917 | lockdep_assert_held(&cgroup_mutex); | |
918 | ||
2739d3cc LZ |
919 | /* |
920 | * If we're doing cleanup due to failure of cgroup_create(), | |
921 | * the corresponding @cfe may not exist. | |
922 | */ | |
05ef1d7c TH |
923 | list_for_each_entry(cfe, &cgrp->files, node) { |
924 | struct dentry *d = cfe->dentry; | |
925 | ||
926 | if (cft && cfe->type != cft) | |
927 | continue; | |
928 | ||
929 | dget(d); | |
930 | d_delete(d); | |
ce27e317 | 931 | simple_unlink(cgrp->dentry->d_inode, d); |
05ef1d7c TH |
932 | list_del_init(&cfe->node); |
933 | dput(d); | |
934 | ||
2739d3cc | 935 | break; |
ddbcc7e8 | 936 | } |
05ef1d7c TH |
937 | } |
938 | ||
13af07df | 939 | /** |
628f7cd4 | 940 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 941 | * @cgrp: target cgroup |
13af07df AR |
942 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
943 | */ | |
628f7cd4 | 944 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 945 | { |
13af07df | 946 | struct cgroup_subsys *ss; |
b420ba7d | 947 | int i; |
05ef1d7c | 948 | |
b420ba7d | 949 | for_each_subsys(ss, i) { |
13af07df | 950 | struct cftype_set *set; |
b420ba7d TH |
951 | |
952 | if (!test_bit(i, &subsys_mask)) | |
13af07df AR |
953 | continue; |
954 | list_for_each_entry(set, &ss->cftsets, node) | |
2bb566cb | 955 | cgroup_addrm_files(cgrp, set->cfts, false); |
13af07df | 956 | } |
ddbcc7e8 PM |
957 | } |
958 | ||
959 | /* | |
960 | * NOTE : the dentry must have been dget()'ed | |
961 | */ | |
962 | static void cgroup_d_remove_dir(struct dentry *dentry) | |
963 | { | |
2fd6b7f5 | 964 | struct dentry *parent; |
ddbcc7e8 | 965 | |
2fd6b7f5 NP |
966 | parent = dentry->d_parent; |
967 | spin_lock(&parent->d_lock); | |
3ec762ad | 968 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
ddbcc7e8 | 969 | list_del_init(&dentry->d_u.d_child); |
2fd6b7f5 NP |
970 | spin_unlock(&dentry->d_lock); |
971 | spin_unlock(&parent->d_lock); | |
ddbcc7e8 PM |
972 | remove_dir(dentry); |
973 | } | |
974 | ||
aae8aab4 | 975 | /* |
cf5d5941 BB |
976 | * Call with cgroup_mutex held. Drops reference counts on modules, including |
977 | * any duplicate ones that parse_cgroupfs_options took. If this function | |
978 | * returns an error, no reference counts are touched. | |
aae8aab4 | 979 | */ |
ddbcc7e8 | 980 | static int rebind_subsystems(struct cgroupfs_root *root, |
a8a648c4 | 981 | unsigned long added_mask, unsigned removed_mask) |
ddbcc7e8 | 982 | { |
bd89aabc | 983 | struct cgroup *cgrp = &root->top_cgroup; |
30159ec7 | 984 | struct cgroup_subsys *ss; |
1d5be6b2 | 985 | unsigned long pinned = 0; |
3126121f | 986 | int i, ret; |
ddbcc7e8 | 987 | |
aae8aab4 | 988 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
e25e2cbb | 989 | BUG_ON(!mutex_is_locked(&cgroup_root_mutex)); |
aae8aab4 | 990 | |
ddbcc7e8 | 991 | /* Check that any added subsystems are currently free */ |
30159ec7 | 992 | for_each_subsys(ss, i) { |
1d5be6b2 | 993 | if (!(added_mask & (1 << i))) |
ddbcc7e8 | 994 | continue; |
30159ec7 | 995 | |
1d5be6b2 | 996 | /* is the subsystem mounted elsewhere? */ |
9871bf95 | 997 | if (ss->root != &cgroup_dummy_root) { |
1d5be6b2 TH |
998 | ret = -EBUSY; |
999 | goto out_put; | |
1000 | } | |
1001 | ||
1002 | /* pin the module */ | |
1003 | if (!try_module_get(ss->module)) { | |
1004 | ret = -ENOENT; | |
1005 | goto out_put; | |
ddbcc7e8 | 1006 | } |
1d5be6b2 | 1007 | pinned |= 1 << i; |
ddbcc7e8 PM |
1008 | } |
1009 | ||
1d5be6b2 TH |
1010 | /* subsys could be missing if unloaded between parsing and here */ |
1011 | if (added_mask != pinned) { | |
1012 | ret = -ENOENT; | |
1013 | goto out_put; | |
ddbcc7e8 PM |
1014 | } |
1015 | ||
3126121f TH |
1016 | ret = cgroup_populate_dir(cgrp, added_mask); |
1017 | if (ret) | |
1d5be6b2 | 1018 | goto out_put; |
3126121f TH |
1019 | |
1020 | /* | |
1021 | * Nothing can fail from this point on. Remove files for the | |
1022 | * removed subsystems and rebind each subsystem. | |
1023 | */ | |
1024 | cgroup_clear_dir(cgrp, removed_mask); | |
ddbcc7e8 | 1025 | |
30159ec7 | 1026 | for_each_subsys(ss, i) { |
ddbcc7e8 | 1027 | unsigned long bit = 1UL << i; |
30159ec7 | 1028 | |
a1a71b45 | 1029 | if (bit & added_mask) { |
ddbcc7e8 | 1030 | /* We're binding this subsystem to this hierarchy */ |
ca8bdcaf TH |
1031 | BUG_ON(cgroup_css(cgrp, ss)); |
1032 | BUG_ON(!cgroup_css(cgroup_dummy_top, ss)); | |
1033 | BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top); | |
a8a648c4 | 1034 | |
73e80ed8 | 1035 | rcu_assign_pointer(cgrp->subsys[i], |
ca8bdcaf TH |
1036 | cgroup_css(cgroup_dummy_top, ss)); |
1037 | cgroup_css(cgrp, ss)->cgroup = cgrp; | |
a8a648c4 | 1038 | |
33a68ac1 | 1039 | list_move(&ss->sibling, &root->subsys_list); |
b2aa30f7 | 1040 | ss->root = root; |
ddbcc7e8 | 1041 | if (ss->bind) |
ca8bdcaf | 1042 | ss->bind(cgroup_css(cgrp, ss)); |
a8a648c4 | 1043 | |
cf5d5941 | 1044 | /* refcount was already taken, and we're keeping it */ |
a8a648c4 | 1045 | root->subsys_mask |= bit; |
a1a71b45 | 1046 | } else if (bit & removed_mask) { |
ddbcc7e8 | 1047 | /* We're removing this subsystem */ |
ca8bdcaf TH |
1048 | BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss)); |
1049 | BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp); | |
a8a648c4 | 1050 | |
ddbcc7e8 | 1051 | if (ss->bind) |
ca8bdcaf | 1052 | ss->bind(cgroup_css(cgroup_dummy_top, ss)); |
73e80ed8 | 1053 | |
ca8bdcaf | 1054 | cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top; |
73e80ed8 TH |
1055 | RCU_INIT_POINTER(cgrp->subsys[i], NULL); |
1056 | ||
9871bf95 TH |
1057 | cgroup_subsys[i]->root = &cgroup_dummy_root; |
1058 | list_move(&ss->sibling, &cgroup_dummy_root.subsys_list); | |
a8a648c4 | 1059 | |
cf5d5941 BB |
1060 | /* subsystem is now free - drop reference on module */ |
1061 | module_put(ss->module); | |
a8a648c4 | 1062 | root->subsys_mask &= ~bit; |
ddbcc7e8 PM |
1063 | } |
1064 | } | |
ddbcc7e8 | 1065 | |
1672d040 TH |
1066 | /* |
1067 | * Mark @root has finished binding subsystems. @root->subsys_mask | |
1068 | * now matches the bound subsystems. | |
1069 | */ | |
1070 | root->flags |= CGRP_ROOT_SUBSYS_BOUND; | |
1071 | ||
ddbcc7e8 | 1072 | return 0; |
1d5be6b2 TH |
1073 | |
1074 | out_put: | |
1075 | for_each_subsys(ss, i) | |
1076 | if (pinned & (1 << i)) | |
1077 | module_put(ss->module); | |
1078 | return ret; | |
ddbcc7e8 PM |
1079 | } |
1080 | ||
34c80b1d | 1081 | static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry) |
ddbcc7e8 | 1082 | { |
34c80b1d | 1083 | struct cgroupfs_root *root = dentry->d_sb->s_fs_info; |
ddbcc7e8 PM |
1084 | struct cgroup_subsys *ss; |
1085 | ||
e25e2cbb | 1086 | mutex_lock(&cgroup_root_mutex); |
5549c497 | 1087 | for_each_root_subsys(root, ss) |
ddbcc7e8 | 1088 | seq_printf(seq, ",%s", ss->name); |
873fe09e TH |
1089 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1090 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1091 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1092 | seq_puts(seq, ",noprefix"); |
93438629 | 1093 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1094 | seq_puts(seq, ",xattr"); |
81a6a5cd PM |
1095 | if (strlen(root->release_agent_path)) |
1096 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
2260e7fc | 1097 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags)) |
97978e6d | 1098 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1099 | if (strlen(root->name)) |
1100 | seq_printf(seq, ",name=%s", root->name); | |
e25e2cbb | 1101 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1102 | return 0; |
1103 | } | |
1104 | ||
1105 | struct cgroup_sb_opts { | |
a1a71b45 | 1106 | unsigned long subsys_mask; |
ddbcc7e8 | 1107 | unsigned long flags; |
81a6a5cd | 1108 | char *release_agent; |
2260e7fc | 1109 | bool cpuset_clone_children; |
c6d57f33 | 1110 | char *name; |
2c6ab6d2 PM |
1111 | /* User explicitly requested empty subsystem */ |
1112 | bool none; | |
c6d57f33 PM |
1113 | |
1114 | struct cgroupfs_root *new_root; | |
2c6ab6d2 | 1115 | |
ddbcc7e8 PM |
1116 | }; |
1117 | ||
aae8aab4 | 1118 | /* |
9871bf95 TH |
1119 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1120 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1121 | * array. This function takes refcounts on subsystems to be used, unless it | |
1122 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1123 | */ |
cf5d5941 | 1124 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1125 | { |
32a8cf23 DL |
1126 | char *token, *o = data; |
1127 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1128 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1129 | struct cgroup_subsys *ss; |
1130 | int i; | |
f9ab5b5b | 1131 | |
aae8aab4 BB |
1132 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1133 | ||
f9ab5b5b LZ |
1134 | #ifdef CONFIG_CPUSETS |
1135 | mask = ~(1UL << cpuset_subsys_id); | |
1136 | #endif | |
ddbcc7e8 | 1137 | |
c6d57f33 | 1138 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1139 | |
1140 | while ((token = strsep(&o, ",")) != NULL) { | |
1141 | if (!*token) | |
1142 | return -EINVAL; | |
32a8cf23 | 1143 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1144 | /* Explicitly have no subsystems */ |
1145 | opts->none = true; | |
32a8cf23 DL |
1146 | continue; |
1147 | } | |
1148 | if (!strcmp(token, "all")) { | |
1149 | /* Mutually exclusive option 'all' + subsystem name */ | |
1150 | if (one_ss) | |
1151 | return -EINVAL; | |
1152 | all_ss = true; | |
1153 | continue; | |
1154 | } | |
873fe09e TH |
1155 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1156 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1157 | continue; | |
1158 | } | |
32a8cf23 | 1159 | if (!strcmp(token, "noprefix")) { |
93438629 | 1160 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1161 | continue; |
1162 | } | |
1163 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1164 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1165 | continue; |
1166 | } | |
03b1cde6 | 1167 | if (!strcmp(token, "xattr")) { |
93438629 | 1168 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1169 | continue; |
1170 | } | |
32a8cf23 | 1171 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1172 | /* Specifying two release agents is forbidden */ |
1173 | if (opts->release_agent) | |
1174 | return -EINVAL; | |
c6d57f33 | 1175 | opts->release_agent = |
e400c285 | 1176 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1177 | if (!opts->release_agent) |
1178 | return -ENOMEM; | |
32a8cf23 DL |
1179 | continue; |
1180 | } | |
1181 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1182 | const char *name = token + 5; |
1183 | /* Can't specify an empty name */ | |
1184 | if (!strlen(name)) | |
1185 | return -EINVAL; | |
1186 | /* Must match [\w.-]+ */ | |
1187 | for (i = 0; i < strlen(name); i++) { | |
1188 | char c = name[i]; | |
1189 | if (isalnum(c)) | |
1190 | continue; | |
1191 | if ((c == '.') || (c == '-') || (c == '_')) | |
1192 | continue; | |
1193 | return -EINVAL; | |
1194 | } | |
1195 | /* Specifying two names is forbidden */ | |
1196 | if (opts->name) | |
1197 | return -EINVAL; | |
1198 | opts->name = kstrndup(name, | |
e400c285 | 1199 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1200 | GFP_KERNEL); |
1201 | if (!opts->name) | |
1202 | return -ENOMEM; | |
32a8cf23 DL |
1203 | |
1204 | continue; | |
1205 | } | |
1206 | ||
30159ec7 | 1207 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1208 | if (strcmp(token, ss->name)) |
1209 | continue; | |
1210 | if (ss->disabled) | |
1211 | continue; | |
1212 | ||
1213 | /* Mutually exclusive option 'all' + subsystem name */ | |
1214 | if (all_ss) | |
1215 | return -EINVAL; | |
a1a71b45 | 1216 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1217 | one_ss = true; |
1218 | ||
1219 | break; | |
1220 | } | |
1221 | if (i == CGROUP_SUBSYS_COUNT) | |
1222 | return -ENOENT; | |
1223 | } | |
1224 | ||
1225 | /* | |
1226 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1227 | * otherwise if 'none', 'name=' and a subsystem name options |
1228 | * were not specified, let's default to 'all' | |
32a8cf23 | 1229 | */ |
30159ec7 TH |
1230 | if (all_ss || (!one_ss && !opts->none && !opts->name)) |
1231 | for_each_subsys(ss, i) | |
1232 | if (!ss->disabled) | |
1233 | set_bit(i, &opts->subsys_mask); | |
ddbcc7e8 | 1234 | |
2c6ab6d2 PM |
1235 | /* Consistency checks */ |
1236 | ||
873fe09e TH |
1237 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1238 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1239 | ||
1240 | if (opts->flags & CGRP_ROOT_NOPREFIX) { | |
1241 | pr_err("cgroup: sane_behavior: noprefix is not allowed\n"); | |
1242 | return -EINVAL; | |
1243 | } | |
1244 | ||
1245 | if (opts->cpuset_clone_children) { | |
1246 | pr_err("cgroup: sane_behavior: clone_children is not allowed\n"); | |
1247 | return -EINVAL; | |
1248 | } | |
1249 | } | |
1250 | ||
f9ab5b5b LZ |
1251 | /* |
1252 | * Option noprefix was introduced just for backward compatibility | |
1253 | * with the old cpuset, so we allow noprefix only if mounting just | |
1254 | * the cpuset subsystem. | |
1255 | */ | |
93438629 | 1256 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1257 | return -EINVAL; |
1258 | ||
2c6ab6d2 PM |
1259 | |
1260 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1261 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1262 | return -EINVAL; |
1263 | ||
1264 | /* | |
1265 | * We either have to specify by name or by subsystems. (So all | |
1266 | * empty hierarchies must have a name). | |
1267 | */ | |
a1a71b45 | 1268 | if (!opts->subsys_mask && !opts->name) |
ddbcc7e8 PM |
1269 | return -EINVAL; |
1270 | ||
1271 | return 0; | |
1272 | } | |
1273 | ||
1274 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) | |
1275 | { | |
1276 | int ret = 0; | |
1277 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1278 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 | 1279 | struct cgroup_sb_opts opts; |
a1a71b45 | 1280 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1281 | |
873fe09e TH |
1282 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1283 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1284 | return -EINVAL; | |
1285 | } | |
1286 | ||
bd89aabc | 1287 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 | 1288 | mutex_lock(&cgroup_mutex); |
e25e2cbb | 1289 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1290 | |
1291 | /* See what subsystems are wanted */ | |
1292 | ret = parse_cgroupfs_options(data, &opts); | |
1293 | if (ret) | |
1294 | goto out_unlock; | |
1295 | ||
a8a648c4 | 1296 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1297 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1298 | task_tgid_nr(current), current->comm); | |
1299 | ||
a1a71b45 AR |
1300 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1301 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1302 | |
cf5d5941 | 1303 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1304 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1305 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1306 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1307 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1308 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1309 | ret = -EINVAL; |
1310 | goto out_unlock; | |
1311 | } | |
1312 | ||
f172e67c TH |
1313 | /* remounting is not allowed for populated hierarchies */ |
1314 | if (root->number_of_cgroups > 1) { | |
1315 | ret = -EBUSY; | |
0670e08b | 1316 | goto out_unlock; |
cf5d5941 | 1317 | } |
ddbcc7e8 | 1318 | |
a8a648c4 | 1319 | ret = rebind_subsystems(root, added_mask, removed_mask); |
3126121f | 1320 | if (ret) |
0670e08b | 1321 | goto out_unlock; |
ddbcc7e8 | 1322 | |
81a6a5cd PM |
1323 | if (opts.release_agent) |
1324 | strcpy(root->release_agent_path, opts.release_agent); | |
ddbcc7e8 | 1325 | out_unlock: |
66bdc9cf | 1326 | kfree(opts.release_agent); |
c6d57f33 | 1327 | kfree(opts.name); |
e25e2cbb | 1328 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 | 1329 | mutex_unlock(&cgroup_mutex); |
bd89aabc | 1330 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
1331 | return ret; |
1332 | } | |
1333 | ||
b87221de | 1334 | static const struct super_operations cgroup_ops = { |
ddbcc7e8 PM |
1335 | .statfs = simple_statfs, |
1336 | .drop_inode = generic_delete_inode, | |
1337 | .show_options = cgroup_show_options, | |
1338 | .remount_fs = cgroup_remount, | |
1339 | }; | |
1340 | ||
cc31edce PM |
1341 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1342 | { | |
1343 | INIT_LIST_HEAD(&cgrp->sibling); | |
1344 | INIT_LIST_HEAD(&cgrp->children); | |
05ef1d7c | 1345 | INIT_LIST_HEAD(&cgrp->files); |
69d0206c | 1346 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1347 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1348 | INIT_LIST_HEAD(&cgrp->pidlists); |
1349 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1350 | cgrp->dummy_css.cgroup = cgrp; |
0dea1168 KS |
1351 | INIT_LIST_HEAD(&cgrp->event_list); |
1352 | spin_lock_init(&cgrp->event_list_lock); | |
03b1cde6 | 1353 | simple_xattrs_init(&cgrp->xattrs); |
cc31edce | 1354 | } |
c6d57f33 | 1355 | |
ddbcc7e8 PM |
1356 | static void init_cgroup_root(struct cgroupfs_root *root) |
1357 | { | |
bd89aabc | 1358 | struct cgroup *cgrp = &root->top_cgroup; |
b0ca5a84 | 1359 | |
ddbcc7e8 PM |
1360 | INIT_LIST_HEAD(&root->subsys_list); |
1361 | INIT_LIST_HEAD(&root->root_list); | |
1362 | root->number_of_cgroups = 1; | |
bd89aabc | 1363 | cgrp->root = root; |
a4ea1cc9 | 1364 | RCU_INIT_POINTER(cgrp->name, &root_cgroup_name); |
cc31edce | 1365 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1366 | idr_init(&root->cgroup_idr); |
ddbcc7e8 PM |
1367 | } |
1368 | ||
fc76df70 | 1369 | static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end) |
2c6ab6d2 | 1370 | { |
1a574231 | 1371 | int id; |
2c6ab6d2 | 1372 | |
54e7b4eb TH |
1373 | lockdep_assert_held(&cgroup_mutex); |
1374 | lockdep_assert_held(&cgroup_root_mutex); | |
1375 | ||
fc76df70 TH |
1376 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end, |
1377 | GFP_KERNEL); | |
1a574231 TH |
1378 | if (id < 0) |
1379 | return id; | |
1380 | ||
1381 | root->hierarchy_id = id; | |
fa3ca07e TH |
1382 | return 0; |
1383 | } | |
1384 | ||
1385 | static void cgroup_exit_root_id(struct cgroupfs_root *root) | |
1386 | { | |
54e7b4eb TH |
1387 | lockdep_assert_held(&cgroup_mutex); |
1388 | lockdep_assert_held(&cgroup_root_mutex); | |
1389 | ||
fa3ca07e | 1390 | if (root->hierarchy_id) { |
1a574231 | 1391 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
fa3ca07e TH |
1392 | root->hierarchy_id = 0; |
1393 | } | |
2c6ab6d2 PM |
1394 | } |
1395 | ||
ddbcc7e8 PM |
1396 | static int cgroup_test_super(struct super_block *sb, void *data) |
1397 | { | |
c6d57f33 | 1398 | struct cgroup_sb_opts *opts = data; |
ddbcc7e8 PM |
1399 | struct cgroupfs_root *root = sb->s_fs_info; |
1400 | ||
c6d57f33 PM |
1401 | /* If we asked for a name then it must match */ |
1402 | if (opts->name && strcmp(opts->name, root->name)) | |
1403 | return 0; | |
ddbcc7e8 | 1404 | |
2c6ab6d2 PM |
1405 | /* |
1406 | * If we asked for subsystems (or explicitly for no | |
1407 | * subsystems) then they must match | |
1408 | */ | |
a1a71b45 AR |
1409 | if ((opts->subsys_mask || opts->none) |
1410 | && (opts->subsys_mask != root->subsys_mask)) | |
ddbcc7e8 PM |
1411 | return 0; |
1412 | ||
1413 | return 1; | |
1414 | } | |
1415 | ||
c6d57f33 PM |
1416 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1417 | { | |
1418 | struct cgroupfs_root *root; | |
1419 | ||
a1a71b45 | 1420 | if (!opts->subsys_mask && !opts->none) |
c6d57f33 PM |
1421 | return NULL; |
1422 | ||
1423 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1424 | if (!root) | |
1425 | return ERR_PTR(-ENOMEM); | |
1426 | ||
1427 | init_cgroup_root(root); | |
2c6ab6d2 | 1428 | |
1672d040 TH |
1429 | /* |
1430 | * We need to set @root->subsys_mask now so that @root can be | |
1431 | * matched by cgroup_test_super() before it finishes | |
1432 | * initialization; otherwise, competing mounts with the same | |
1433 | * options may try to bind the same subsystems instead of waiting | |
1434 | * for the first one leading to unexpected mount errors. | |
1435 | * SUBSYS_BOUND will be set once actual binding is complete. | |
1436 | */ | |
a1a71b45 | 1437 | root->subsys_mask = opts->subsys_mask; |
c6d57f33 PM |
1438 | root->flags = opts->flags; |
1439 | if (opts->release_agent) | |
1440 | strcpy(root->release_agent_path, opts->release_agent); | |
1441 | if (opts->name) | |
1442 | strcpy(root->name, opts->name); | |
2260e7fc TH |
1443 | if (opts->cpuset_clone_children) |
1444 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1445 | return root; |
1446 | } | |
1447 | ||
fa3ca07e | 1448 | static void cgroup_free_root(struct cgroupfs_root *root) |
2c6ab6d2 | 1449 | { |
fa3ca07e TH |
1450 | if (root) { |
1451 | /* hierarhcy ID shoulid already have been released */ | |
1452 | WARN_ON_ONCE(root->hierarchy_id); | |
2c6ab6d2 | 1453 | |
4e96ee8e | 1454 | idr_destroy(&root->cgroup_idr); |
fa3ca07e TH |
1455 | kfree(root); |
1456 | } | |
2c6ab6d2 PM |
1457 | } |
1458 | ||
ddbcc7e8 PM |
1459 | static int cgroup_set_super(struct super_block *sb, void *data) |
1460 | { | |
1461 | int ret; | |
c6d57f33 PM |
1462 | struct cgroup_sb_opts *opts = data; |
1463 | ||
1464 | /* If we don't have a new root, we can't set up a new sb */ | |
1465 | if (!opts->new_root) | |
1466 | return -EINVAL; | |
1467 | ||
a1a71b45 | 1468 | BUG_ON(!opts->subsys_mask && !opts->none); |
ddbcc7e8 PM |
1469 | |
1470 | ret = set_anon_super(sb, NULL); | |
1471 | if (ret) | |
1472 | return ret; | |
1473 | ||
c6d57f33 PM |
1474 | sb->s_fs_info = opts->new_root; |
1475 | opts->new_root->sb = sb; | |
ddbcc7e8 PM |
1476 | |
1477 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
1478 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
1479 | sb->s_magic = CGROUP_SUPER_MAGIC; | |
1480 | sb->s_op = &cgroup_ops; | |
1481 | ||
1482 | return 0; | |
1483 | } | |
1484 | ||
1485 | static int cgroup_get_rootdir(struct super_block *sb) | |
1486 | { | |
0df6a63f AV |
1487 | static const struct dentry_operations cgroup_dops = { |
1488 | .d_iput = cgroup_diput, | |
c72a04e3 | 1489 | .d_delete = cgroup_delete, |
0df6a63f AV |
1490 | }; |
1491 | ||
ddbcc7e8 PM |
1492 | struct inode *inode = |
1493 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); | |
ddbcc7e8 PM |
1494 | |
1495 | if (!inode) | |
1496 | return -ENOMEM; | |
1497 | ||
ddbcc7e8 PM |
1498 | inode->i_fop = &simple_dir_operations; |
1499 | inode->i_op = &cgroup_dir_inode_operations; | |
1500 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
1501 | inc_nlink(inode); | |
48fde701 AV |
1502 | sb->s_root = d_make_root(inode); |
1503 | if (!sb->s_root) | |
ddbcc7e8 | 1504 | return -ENOMEM; |
0df6a63f AV |
1505 | /* for everything else we want ->d_op set */ |
1506 | sb->s_d_op = &cgroup_dops; | |
ddbcc7e8 PM |
1507 | return 0; |
1508 | } | |
1509 | ||
f7e83571 | 1510 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1511 | int flags, const char *unused_dev_name, |
f7e83571 | 1512 | void *data) |
ddbcc7e8 PM |
1513 | { |
1514 | struct cgroup_sb_opts opts; | |
c6d57f33 | 1515 | struct cgroupfs_root *root; |
ddbcc7e8 PM |
1516 | int ret = 0; |
1517 | struct super_block *sb; | |
c6d57f33 | 1518 | struct cgroupfs_root *new_root; |
3126121f | 1519 | struct list_head tmp_links; |
e25e2cbb | 1520 | struct inode *inode; |
3126121f | 1521 | const struct cred *cred; |
ddbcc7e8 PM |
1522 | |
1523 | /* First find the desired set of subsystems */ | |
aae8aab4 | 1524 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1525 | ret = parse_cgroupfs_options(data, &opts); |
aae8aab4 | 1526 | mutex_unlock(&cgroup_mutex); |
c6d57f33 PM |
1527 | if (ret) |
1528 | goto out_err; | |
ddbcc7e8 | 1529 | |
c6d57f33 PM |
1530 | /* |
1531 | * Allocate a new cgroup root. We may not need it if we're | |
1532 | * reusing an existing hierarchy. | |
1533 | */ | |
1534 | new_root = cgroup_root_from_opts(&opts); | |
1535 | if (IS_ERR(new_root)) { | |
1536 | ret = PTR_ERR(new_root); | |
1d5be6b2 | 1537 | goto out_err; |
81a6a5cd | 1538 | } |
c6d57f33 | 1539 | opts.new_root = new_root; |
ddbcc7e8 | 1540 | |
c6d57f33 | 1541 | /* Locate an existing or new sb for this hierarchy */ |
9249e17f | 1542 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts); |
ddbcc7e8 | 1543 | if (IS_ERR(sb)) { |
c6d57f33 | 1544 | ret = PTR_ERR(sb); |
fa3ca07e | 1545 | cgroup_free_root(opts.new_root); |
1d5be6b2 | 1546 | goto out_err; |
ddbcc7e8 PM |
1547 | } |
1548 | ||
c6d57f33 PM |
1549 | root = sb->s_fs_info; |
1550 | BUG_ON(!root); | |
1551 | if (root == opts.new_root) { | |
1552 | /* We used the new root structure, so this is a new hierarchy */ | |
c12f65d4 | 1553 | struct cgroup *root_cgrp = &root->top_cgroup; |
c6d57f33 | 1554 | struct cgroupfs_root *existing_root; |
28fd5dfc | 1555 | int i; |
5abb8855 | 1556 | struct css_set *cset; |
ddbcc7e8 PM |
1557 | |
1558 | BUG_ON(sb->s_root != NULL); | |
1559 | ||
1560 | ret = cgroup_get_rootdir(sb); | |
1561 | if (ret) | |
1562 | goto drop_new_super; | |
817929ec | 1563 | inode = sb->s_root->d_inode; |
ddbcc7e8 | 1564 | |
817929ec | 1565 | mutex_lock(&inode->i_mutex); |
ddbcc7e8 | 1566 | mutex_lock(&cgroup_mutex); |
e25e2cbb | 1567 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 | 1568 | |
4e96ee8e LZ |
1569 | root_cgrp->id = idr_alloc(&root->cgroup_idr, root_cgrp, |
1570 | 0, 1, GFP_KERNEL); | |
1571 | if (root_cgrp->id < 0) | |
1572 | goto unlock_drop; | |
1573 | ||
e25e2cbb TH |
1574 | /* Check for name clashes with existing mounts */ |
1575 | ret = -EBUSY; | |
1576 | if (strlen(root->name)) | |
1577 | for_each_active_root(existing_root) | |
1578 | if (!strcmp(existing_root->name, root->name)) | |
1579 | goto unlock_drop; | |
c6d57f33 | 1580 | |
817929ec PM |
1581 | /* |
1582 | * We're accessing css_set_count without locking | |
1583 | * css_set_lock here, but that's OK - it can only be | |
1584 | * increased by someone holding cgroup_lock, and | |
1585 | * that's us. The worst that can happen is that we | |
1586 | * have some link structures left over | |
1587 | */ | |
69d0206c | 1588 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); |
e25e2cbb TH |
1589 | if (ret) |
1590 | goto unlock_drop; | |
817929ec | 1591 | |
fc76df70 TH |
1592 | /* ID 0 is reserved for dummy root, 1 for unified hierarchy */ |
1593 | ret = cgroup_init_root_id(root, 2, 0); | |
fa3ca07e TH |
1594 | if (ret) |
1595 | goto unlock_drop; | |
1596 | ||
3126121f TH |
1597 | sb->s_root->d_fsdata = root_cgrp; |
1598 | root_cgrp->dentry = sb->s_root; | |
1599 | ||
1600 | /* | |
1601 | * We're inside get_sb() and will call lookup_one_len() to | |
1602 | * create the root files, which doesn't work if SELinux is | |
1603 | * in use. The following cred dancing somehow works around | |
1604 | * it. See 2ce9738ba ("cgroupfs: use init_cred when | |
1605 | * populating new cgroupfs mount") for more details. | |
1606 | */ | |
1607 | cred = override_creds(&init_cred); | |
1608 | ||
2bb566cb | 1609 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); |
3126121f TH |
1610 | if (ret) |
1611 | goto rm_base_files; | |
1612 | ||
a8a648c4 | 1613 | ret = rebind_subsystems(root, root->subsys_mask, 0); |
3126121f TH |
1614 | if (ret) |
1615 | goto rm_base_files; | |
1616 | ||
1617 | revert_creds(cred); | |
1618 | ||
cf5d5941 BB |
1619 | /* |
1620 | * There must be no failure case after here, since rebinding | |
1621 | * takes care of subsystems' refcounts, which are explicitly | |
1622 | * dropped in the failure exit path. | |
1623 | */ | |
ddbcc7e8 | 1624 | |
9871bf95 TH |
1625 | list_add(&root->root_list, &cgroup_roots); |
1626 | cgroup_root_count++; | |
ddbcc7e8 | 1627 | |
817929ec PM |
1628 | /* Link the top cgroup in this hierarchy into all |
1629 | * the css_set objects */ | |
1630 | write_lock(&css_set_lock); | |
5abb8855 | 1631 | hash_for_each(css_set_table, i, cset, hlist) |
69d0206c | 1632 | link_css_set(&tmp_links, cset, root_cgrp); |
817929ec PM |
1633 | write_unlock(&css_set_lock); |
1634 | ||
69d0206c | 1635 | free_cgrp_cset_links(&tmp_links); |
817929ec | 1636 | |
c12f65d4 | 1637 | BUG_ON(!list_empty(&root_cgrp->children)); |
ddbcc7e8 PM |
1638 | BUG_ON(root->number_of_cgroups != 1); |
1639 | ||
e25e2cbb | 1640 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 | 1641 | mutex_unlock(&cgroup_mutex); |
34f77a90 | 1642 | mutex_unlock(&inode->i_mutex); |
c6d57f33 PM |
1643 | } else { |
1644 | /* | |
1645 | * We re-used an existing hierarchy - the new root (if | |
1646 | * any) is not needed | |
1647 | */ | |
fa3ca07e | 1648 | cgroup_free_root(opts.new_root); |
873fe09e | 1649 | |
c7ba8287 | 1650 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1651 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1652 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1653 | ret = -EINVAL; | |
1654 | goto drop_new_super; | |
1655 | } else { | |
1656 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1657 | } | |
873fe09e | 1658 | } |
ddbcc7e8 PM |
1659 | } |
1660 | ||
c6d57f33 PM |
1661 | kfree(opts.release_agent); |
1662 | kfree(opts.name); | |
f7e83571 | 1663 | return dget(sb->s_root); |
ddbcc7e8 | 1664 | |
3126121f TH |
1665 | rm_base_files: |
1666 | free_cgrp_cset_links(&tmp_links); | |
2bb566cb | 1667 | cgroup_addrm_files(&root->top_cgroup, cgroup_base_files, false); |
3126121f | 1668 | revert_creds(cred); |
e25e2cbb | 1669 | unlock_drop: |
fa3ca07e | 1670 | cgroup_exit_root_id(root); |
e25e2cbb TH |
1671 | mutex_unlock(&cgroup_root_mutex); |
1672 | mutex_unlock(&cgroup_mutex); | |
1673 | mutex_unlock(&inode->i_mutex); | |
ddbcc7e8 | 1674 | drop_new_super: |
6f5bbff9 | 1675 | deactivate_locked_super(sb); |
c6d57f33 PM |
1676 | out_err: |
1677 | kfree(opts.release_agent); | |
1678 | kfree(opts.name); | |
f7e83571 | 1679 | return ERR_PTR(ret); |
ddbcc7e8 PM |
1680 | } |
1681 | ||
1682 | static void cgroup_kill_sb(struct super_block *sb) { | |
1683 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1684 | struct cgroup *cgrp = &root->top_cgroup; |
69d0206c | 1685 | struct cgrp_cset_link *link, *tmp_link; |
ddbcc7e8 PM |
1686 | int ret; |
1687 | ||
1688 | BUG_ON(!root); | |
1689 | ||
1690 | BUG_ON(root->number_of_cgroups != 1); | |
bd89aabc | 1691 | BUG_ON(!list_empty(&cgrp->children)); |
ddbcc7e8 | 1692 | |
3126121f | 1693 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 | 1694 | mutex_lock(&cgroup_mutex); |
e25e2cbb | 1695 | mutex_lock(&cgroup_root_mutex); |
ddbcc7e8 PM |
1696 | |
1697 | /* Rebind all subsystems back to the default hierarchy */ | |
1672d040 TH |
1698 | if (root->flags & CGRP_ROOT_SUBSYS_BOUND) { |
1699 | ret = rebind_subsystems(root, 0, root->subsys_mask); | |
1700 | /* Shouldn't be able to fail ... */ | |
1701 | BUG_ON(ret); | |
1702 | } | |
ddbcc7e8 | 1703 | |
817929ec | 1704 | /* |
69d0206c | 1705 | * Release all the links from cset_links to this hierarchy's |
817929ec PM |
1706 | * root cgroup |
1707 | */ | |
1708 | write_lock(&css_set_lock); | |
71cbb949 | 1709 | |
69d0206c TH |
1710 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { |
1711 | list_del(&link->cset_link); | |
1712 | list_del(&link->cgrp_link); | |
817929ec PM |
1713 | kfree(link); |
1714 | } | |
1715 | write_unlock(&css_set_lock); | |
1716 | ||
839ec545 PM |
1717 | if (!list_empty(&root->root_list)) { |
1718 | list_del(&root->root_list); | |
9871bf95 | 1719 | cgroup_root_count--; |
839ec545 | 1720 | } |
e5f6a860 | 1721 | |
fa3ca07e TH |
1722 | cgroup_exit_root_id(root); |
1723 | ||
e25e2cbb | 1724 | mutex_unlock(&cgroup_root_mutex); |
ddbcc7e8 | 1725 | mutex_unlock(&cgroup_mutex); |
3126121f | 1726 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 | 1727 | |
03b1cde6 AR |
1728 | simple_xattrs_free(&cgrp->xattrs); |
1729 | ||
ddbcc7e8 | 1730 | kill_litter_super(sb); |
fa3ca07e | 1731 | cgroup_free_root(root); |
ddbcc7e8 PM |
1732 | } |
1733 | ||
1734 | static struct file_system_type cgroup_fs_type = { | |
1735 | .name = "cgroup", | |
f7e83571 | 1736 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1737 | .kill_sb = cgroup_kill_sb, |
1738 | }; | |
1739 | ||
676db4af GK |
1740 | static struct kobject *cgroup_kobj; |
1741 | ||
a043e3b2 LZ |
1742 | /** |
1743 | * cgroup_path - generate the path of a cgroup | |
1744 | * @cgrp: the cgroup in question | |
1745 | * @buf: the buffer to write the path into | |
1746 | * @buflen: the length of the buffer | |
1747 | * | |
65dff759 LZ |
1748 | * Writes path of cgroup into buf. Returns 0 on success, -errno on error. |
1749 | * | |
1750 | * We can't generate cgroup path using dentry->d_name, as accessing | |
1751 | * dentry->name must be protected by irq-unsafe dentry->d_lock or parent | |
1752 | * inode's i_mutex, while on the other hand cgroup_path() can be called | |
1753 | * with some irq-safe spinlocks held. | |
ddbcc7e8 | 1754 | */ |
bd89aabc | 1755 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
ddbcc7e8 | 1756 | { |
65dff759 | 1757 | int ret = -ENAMETOOLONG; |
ddbcc7e8 | 1758 | char *start; |
febfcef6 | 1759 | |
da1f296f TH |
1760 | if (!cgrp->parent) { |
1761 | if (strlcpy(buf, "/", buflen) >= buflen) | |
1762 | return -ENAMETOOLONG; | |
ddbcc7e8 PM |
1763 | return 0; |
1764 | } | |
1765 | ||
316eb661 | 1766 | start = buf + buflen - 1; |
316eb661 | 1767 | *start = '\0'; |
9a9686b6 | 1768 | |
65dff759 | 1769 | rcu_read_lock(); |
da1f296f | 1770 | do { |
65dff759 LZ |
1771 | const char *name = cgroup_name(cgrp); |
1772 | int len; | |
1773 | ||
1774 | len = strlen(name); | |
ddbcc7e8 | 1775 | if ((start -= len) < buf) |
65dff759 LZ |
1776 | goto out; |
1777 | memcpy(start, name, len); | |
9a9686b6 | 1778 | |
ddbcc7e8 | 1779 | if (--start < buf) |
65dff759 | 1780 | goto out; |
ddbcc7e8 | 1781 | *start = '/'; |
65dff759 LZ |
1782 | |
1783 | cgrp = cgrp->parent; | |
da1f296f | 1784 | } while (cgrp->parent); |
65dff759 | 1785 | ret = 0; |
ddbcc7e8 | 1786 | memmove(buf, start, buf + buflen - start); |
65dff759 LZ |
1787 | out: |
1788 | rcu_read_unlock(); | |
1789 | return ret; | |
ddbcc7e8 | 1790 | } |
67523c48 | 1791 | EXPORT_SYMBOL_GPL(cgroup_path); |
ddbcc7e8 | 1792 | |
857a2beb | 1793 | /** |
913ffdb5 | 1794 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1795 | * @task: target task |
857a2beb TH |
1796 | * @buf: the buffer to write the path into |
1797 | * @buflen: the length of the buffer | |
1798 | * | |
913ffdb5 TH |
1799 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1800 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1801 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1802 | * cgroup controller callbacks. | |
1803 | * | |
1804 | * Returns 0 on success, fails with -%ENAMETOOLONG if @buflen is too short. | |
857a2beb | 1805 | */ |
913ffdb5 | 1806 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb TH |
1807 | { |
1808 | struct cgroupfs_root *root; | |
913ffdb5 TH |
1809 | struct cgroup *cgrp; |
1810 | int hierarchy_id = 1, ret = 0; | |
1811 | ||
1812 | if (buflen < 2) | |
1813 | return -ENAMETOOLONG; | |
857a2beb TH |
1814 | |
1815 | mutex_lock(&cgroup_mutex); | |
1816 | ||
913ffdb5 TH |
1817 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1818 | ||
857a2beb TH |
1819 | if (root) { |
1820 | cgrp = task_cgroup_from_root(task, root); | |
1821 | ret = cgroup_path(cgrp, buf, buflen); | |
913ffdb5 TH |
1822 | } else { |
1823 | /* if no hierarchy exists, everyone is in "/" */ | |
1824 | memcpy(buf, "/", 2); | |
857a2beb TH |
1825 | } |
1826 | ||
1827 | mutex_unlock(&cgroup_mutex); | |
857a2beb TH |
1828 | return ret; |
1829 | } | |
913ffdb5 | 1830 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1831 | |
2f7ee569 TH |
1832 | /* |
1833 | * Control Group taskset | |
1834 | */ | |
134d3373 TH |
1835 | struct task_and_cgroup { |
1836 | struct task_struct *task; | |
1837 | struct cgroup *cgrp; | |
6f4b7e63 | 1838 | struct css_set *cset; |
134d3373 TH |
1839 | }; |
1840 | ||
2f7ee569 TH |
1841 | struct cgroup_taskset { |
1842 | struct task_and_cgroup single; | |
1843 | struct flex_array *tc_array; | |
1844 | int tc_array_len; | |
1845 | int idx; | |
1846 | struct cgroup *cur_cgrp; | |
1847 | }; | |
1848 | ||
1849 | /** | |
1850 | * cgroup_taskset_first - reset taskset and return the first task | |
1851 | * @tset: taskset of interest | |
1852 | * | |
1853 | * @tset iteration is initialized and the first task is returned. | |
1854 | */ | |
1855 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1856 | { | |
1857 | if (tset->tc_array) { | |
1858 | tset->idx = 0; | |
1859 | return cgroup_taskset_next(tset); | |
1860 | } else { | |
1861 | tset->cur_cgrp = tset->single.cgrp; | |
1862 | return tset->single.task; | |
1863 | } | |
1864 | } | |
1865 | EXPORT_SYMBOL_GPL(cgroup_taskset_first); | |
1866 | ||
1867 | /** | |
1868 | * cgroup_taskset_next - iterate to the next task in taskset | |
1869 | * @tset: taskset of interest | |
1870 | * | |
1871 | * Return the next task in @tset. Iteration must have been initialized | |
1872 | * with cgroup_taskset_first(). | |
1873 | */ | |
1874 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1875 | { | |
1876 | struct task_and_cgroup *tc; | |
1877 | ||
1878 | if (!tset->tc_array || tset->idx >= tset->tc_array_len) | |
1879 | return NULL; | |
1880 | ||
1881 | tc = flex_array_get(tset->tc_array, tset->idx++); | |
1882 | tset->cur_cgrp = tc->cgrp; | |
1883 | return tc->task; | |
1884 | } | |
1885 | EXPORT_SYMBOL_GPL(cgroup_taskset_next); | |
1886 | ||
1887 | /** | |
d99c8727 | 1888 | * cgroup_taskset_cur_css - return the matching css for the current task |
2f7ee569 | 1889 | * @tset: taskset of interest |
d99c8727 | 1890 | * @subsys_id: the ID of the target subsystem |
2f7ee569 | 1891 | * |
d99c8727 TH |
1892 | * Return the css for the current (last returned) task of @tset for |
1893 | * subsystem specified by @subsys_id. This function must be preceded by | |
1894 | * either cgroup_taskset_first() or cgroup_taskset_next(). | |
2f7ee569 | 1895 | */ |
d99c8727 TH |
1896 | struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset, |
1897 | int subsys_id) | |
2f7ee569 | 1898 | { |
ca8bdcaf | 1899 | return cgroup_css(tset->cur_cgrp, cgroup_subsys[subsys_id]); |
2f7ee569 | 1900 | } |
d99c8727 | 1901 | EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css); |
2f7ee569 TH |
1902 | |
1903 | /** | |
1904 | * cgroup_taskset_size - return the number of tasks in taskset | |
1905 | * @tset: taskset of interest | |
1906 | */ | |
1907 | int cgroup_taskset_size(struct cgroup_taskset *tset) | |
1908 | { | |
1909 | return tset->tc_array ? tset->tc_array_len : 1; | |
1910 | } | |
1911 | EXPORT_SYMBOL_GPL(cgroup_taskset_size); | |
1912 | ||
1913 | ||
74a1166d BB |
1914 | /* |
1915 | * cgroup_task_migrate - move a task from one cgroup to another. | |
1916 | * | |
d0b2fdd2 | 1917 | * Must be called with cgroup_mutex and threadgroup locked. |
74a1166d | 1918 | */ |
5abb8855 TH |
1919 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1920 | struct task_struct *tsk, | |
1921 | struct css_set *new_cset) | |
74a1166d | 1922 | { |
5abb8855 | 1923 | struct css_set *old_cset; |
74a1166d BB |
1924 | |
1925 | /* | |
026085ef MSB |
1926 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1927 | * setting such that we can't race against cgroup_exit() changing the | |
1928 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1929 | */ |
c84cdf75 | 1930 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1931 | old_cset = task_css_set(tsk); |
74a1166d | 1932 | |
74a1166d | 1933 | task_lock(tsk); |
5abb8855 | 1934 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d BB |
1935 | task_unlock(tsk); |
1936 | ||
1937 | /* Update the css_set linked lists if we're using them */ | |
1938 | write_lock(&css_set_lock); | |
1939 | if (!list_empty(&tsk->cg_list)) | |
5abb8855 | 1940 | list_move(&tsk->cg_list, &new_cset->tasks); |
74a1166d BB |
1941 | write_unlock(&css_set_lock); |
1942 | ||
1943 | /* | |
5abb8855 TH |
1944 | * We just gained a reference on old_cset by taking it from the |
1945 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1946 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1947 | */ |
5abb8855 TH |
1948 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
1949 | put_css_set(old_cset); | |
74a1166d BB |
1950 | } |
1951 | ||
a043e3b2 | 1952 | /** |
081aa458 | 1953 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup |
74a1166d | 1954 | * @cgrp: the cgroup to attach to |
081aa458 LZ |
1955 | * @tsk: the task or the leader of the threadgroup to be attached |
1956 | * @threadgroup: attach the whole threadgroup? | |
74a1166d | 1957 | * |
257058ae | 1958 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
081aa458 | 1959 | * task_lock of @tsk or each thread in the threadgroup individually in turn. |
74a1166d | 1960 | */ |
47cfcd09 TH |
1961 | static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk, |
1962 | bool threadgroup) | |
74a1166d BB |
1963 | { |
1964 | int retval, i, group_size; | |
1965 | struct cgroup_subsys *ss, *failed_ss = NULL; | |
74a1166d BB |
1966 | struct cgroupfs_root *root = cgrp->root; |
1967 | /* threadgroup list cursor and array */ | |
081aa458 | 1968 | struct task_struct *leader = tsk; |
134d3373 | 1969 | struct task_and_cgroup *tc; |
d846687d | 1970 | struct flex_array *group; |
2f7ee569 | 1971 | struct cgroup_taskset tset = { }; |
74a1166d BB |
1972 | |
1973 | /* | |
1974 | * step 0: in order to do expensive, possibly blocking operations for | |
1975 | * every thread, we cannot iterate the thread group list, since it needs | |
1976 | * rcu or tasklist locked. instead, build an array of all threads in the | |
257058ae TH |
1977 | * group - group_rwsem prevents new threads from appearing, and if |
1978 | * threads exit, this will just be an over-estimate. | |
74a1166d | 1979 | */ |
081aa458 LZ |
1980 | if (threadgroup) |
1981 | group_size = get_nr_threads(tsk); | |
1982 | else | |
1983 | group_size = 1; | |
d846687d | 1984 | /* flex_array supports very large thread-groups better than kmalloc. */ |
134d3373 | 1985 | group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL); |
74a1166d BB |
1986 | if (!group) |
1987 | return -ENOMEM; | |
d846687d | 1988 | /* pre-allocate to guarantee space while iterating in rcu read-side. */ |
3ac1707a | 1989 | retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL); |
d846687d BB |
1990 | if (retval) |
1991 | goto out_free_group_list; | |
74a1166d | 1992 | |
74a1166d | 1993 | i = 0; |
fb5d2b4c MSB |
1994 | /* |
1995 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1996 | * already PF_EXITING could be freed from underneath us unless we | |
1997 | * take an rcu_read_lock. | |
1998 | */ | |
1999 | rcu_read_lock(); | |
74a1166d | 2000 | do { |
134d3373 TH |
2001 | struct task_and_cgroup ent; |
2002 | ||
cd3d0952 TH |
2003 | /* @tsk either already exited or can't exit until the end */ |
2004 | if (tsk->flags & PF_EXITING) | |
ea84753c | 2005 | goto next; |
cd3d0952 | 2006 | |
74a1166d BB |
2007 | /* as per above, nr_threads may decrease, but not increase. */ |
2008 | BUG_ON(i >= group_size); | |
134d3373 TH |
2009 | ent.task = tsk; |
2010 | ent.cgrp = task_cgroup_from_root(tsk, root); | |
892a2b90 MSB |
2011 | /* nothing to do if this task is already in the cgroup */ |
2012 | if (ent.cgrp == cgrp) | |
ea84753c | 2013 | goto next; |
61d1d219 MSB |
2014 | /* |
2015 | * saying GFP_ATOMIC has no effect here because we did prealloc | |
2016 | * earlier, but it's good form to communicate our expectations. | |
2017 | */ | |
134d3373 | 2018 | retval = flex_array_put(group, i, &ent, GFP_ATOMIC); |
d846687d | 2019 | BUG_ON(retval != 0); |
74a1166d | 2020 | i++; |
ea84753c | 2021 | next: |
081aa458 LZ |
2022 | if (!threadgroup) |
2023 | break; | |
74a1166d | 2024 | } while_each_thread(leader, tsk); |
fb5d2b4c | 2025 | rcu_read_unlock(); |
74a1166d BB |
2026 | /* remember the number of threads in the array for later. */ |
2027 | group_size = i; | |
2f7ee569 TH |
2028 | tset.tc_array = group; |
2029 | tset.tc_array_len = group_size; | |
74a1166d | 2030 | |
134d3373 TH |
2031 | /* methods shouldn't be called if no task is actually migrating */ |
2032 | retval = 0; | |
892a2b90 | 2033 | if (!group_size) |
b07ef774 | 2034 | goto out_free_group_list; |
134d3373 | 2035 | |
74a1166d BB |
2036 | /* |
2037 | * step 1: check that we can legitimately attach to the cgroup. | |
2038 | */ | |
5549c497 | 2039 | for_each_root_subsys(root, ss) { |
ca8bdcaf | 2040 | struct cgroup_subsys_state *css = cgroup_css(cgrp, ss); |
eb95419b | 2041 | |
74a1166d | 2042 | if (ss->can_attach) { |
eb95419b | 2043 | retval = ss->can_attach(css, &tset); |
74a1166d BB |
2044 | if (retval) { |
2045 | failed_ss = ss; | |
2046 | goto out_cancel_attach; | |
2047 | } | |
2048 | } | |
74a1166d BB |
2049 | } |
2050 | ||
2051 | /* | |
2052 | * step 2: make sure css_sets exist for all threads to be migrated. | |
2053 | * we use find_css_set, which allocates a new one if necessary. | |
2054 | */ | |
74a1166d | 2055 | for (i = 0; i < group_size; i++) { |
a8ad805c TH |
2056 | struct css_set *old_cset; |
2057 | ||
134d3373 | 2058 | tc = flex_array_get(group, i); |
a8ad805c | 2059 | old_cset = task_css_set(tc->task); |
6f4b7e63 LZ |
2060 | tc->cset = find_css_set(old_cset, cgrp); |
2061 | if (!tc->cset) { | |
61d1d219 MSB |
2062 | retval = -ENOMEM; |
2063 | goto out_put_css_set_refs; | |
74a1166d BB |
2064 | } |
2065 | } | |
2066 | ||
2067 | /* | |
494c167c TH |
2068 | * step 3: now that we're guaranteed success wrt the css_sets, |
2069 | * proceed to move all tasks to the new cgroup. There are no | |
2070 | * failure cases after here, so this is the commit point. | |
74a1166d | 2071 | */ |
74a1166d | 2072 | for (i = 0; i < group_size; i++) { |
134d3373 | 2073 | tc = flex_array_get(group, i); |
6f4b7e63 | 2074 | cgroup_task_migrate(tc->cgrp, tc->task, tc->cset); |
74a1166d BB |
2075 | } |
2076 | /* nothing is sensitive to fork() after this point. */ | |
2077 | ||
2078 | /* | |
494c167c | 2079 | * step 4: do subsystem attach callbacks. |
74a1166d | 2080 | */ |
5549c497 | 2081 | for_each_root_subsys(root, ss) { |
ca8bdcaf | 2082 | struct cgroup_subsys_state *css = cgroup_css(cgrp, ss); |
eb95419b | 2083 | |
74a1166d | 2084 | if (ss->attach) |
eb95419b | 2085 | ss->attach(css, &tset); |
74a1166d BB |
2086 | } |
2087 | ||
2088 | /* | |
2089 | * step 5: success! and cleanup | |
2090 | */ | |
74a1166d | 2091 | retval = 0; |
61d1d219 MSB |
2092 | out_put_css_set_refs: |
2093 | if (retval) { | |
2094 | for (i = 0; i < group_size; i++) { | |
2095 | tc = flex_array_get(group, i); | |
6f4b7e63 | 2096 | if (!tc->cset) |
61d1d219 | 2097 | break; |
6f4b7e63 | 2098 | put_css_set(tc->cset); |
61d1d219 | 2099 | } |
74a1166d BB |
2100 | } |
2101 | out_cancel_attach: | |
74a1166d | 2102 | if (retval) { |
5549c497 | 2103 | for_each_root_subsys(root, ss) { |
ca8bdcaf | 2104 | struct cgroup_subsys_state *css = cgroup_css(cgrp, ss); |
eb95419b | 2105 | |
494c167c | 2106 | if (ss == failed_ss) |
74a1166d | 2107 | break; |
74a1166d | 2108 | if (ss->cancel_attach) |
eb95419b | 2109 | ss->cancel_attach(css, &tset); |
74a1166d BB |
2110 | } |
2111 | } | |
74a1166d | 2112 | out_free_group_list: |
d846687d | 2113 | flex_array_free(group); |
74a1166d BB |
2114 | return retval; |
2115 | } | |
2116 | ||
2117 | /* | |
2118 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
2119 | * function to attach either it or all tasks in its threadgroup. Will lock |
2120 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 2121 | */ |
74a1166d | 2122 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2123 | { |
bbcb81d0 | 2124 | struct task_struct *tsk; |
c69e8d9c | 2125 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2126 | int ret; |
2127 | ||
74a1166d BB |
2128 | if (!cgroup_lock_live_group(cgrp)) |
2129 | return -ENODEV; | |
2130 | ||
b78949eb MSB |
2131 | retry_find_task: |
2132 | rcu_read_lock(); | |
bbcb81d0 | 2133 | if (pid) { |
73507f33 | 2134 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2135 | if (!tsk) { |
2136 | rcu_read_unlock(); | |
b78949eb MSB |
2137 | ret= -ESRCH; |
2138 | goto out_unlock_cgroup; | |
bbcb81d0 | 2139 | } |
74a1166d BB |
2140 | /* |
2141 | * even if we're attaching all tasks in the thread group, we | |
2142 | * only need to check permissions on one of them. | |
2143 | */ | |
c69e8d9c | 2144 | tcred = __task_cred(tsk); |
14a590c3 EB |
2145 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2146 | !uid_eq(cred->euid, tcred->uid) && | |
2147 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2148 | rcu_read_unlock(); |
b78949eb MSB |
2149 | ret = -EACCES; |
2150 | goto out_unlock_cgroup; | |
bbcb81d0 | 2151 | } |
b78949eb MSB |
2152 | } else |
2153 | tsk = current; | |
cd3d0952 TH |
2154 | |
2155 | if (threadgroup) | |
b78949eb | 2156 | tsk = tsk->group_leader; |
c4c27fbd MG |
2157 | |
2158 | /* | |
14a40ffc | 2159 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2160 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2161 | * with no rt_runtime allocated. Just say no. | |
2162 | */ | |
14a40ffc | 2163 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2164 | ret = -EINVAL; |
2165 | rcu_read_unlock(); | |
2166 | goto out_unlock_cgroup; | |
2167 | } | |
2168 | ||
b78949eb MSB |
2169 | get_task_struct(tsk); |
2170 | rcu_read_unlock(); | |
2171 | ||
2172 | threadgroup_lock(tsk); | |
2173 | if (threadgroup) { | |
2174 | if (!thread_group_leader(tsk)) { | |
2175 | /* | |
2176 | * a race with de_thread from another thread's exec() | |
2177 | * may strip us of our leadership, if this happens, | |
2178 | * there is no choice but to throw this task away and | |
2179 | * try again; this is | |
2180 | * "double-double-toil-and-trouble-check locking". | |
2181 | */ | |
2182 | threadgroup_unlock(tsk); | |
2183 | put_task_struct(tsk); | |
2184 | goto retry_find_task; | |
2185 | } | |
081aa458 LZ |
2186 | } |
2187 | ||
2188 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2189 | ||
cd3d0952 TH |
2190 | threadgroup_unlock(tsk); |
2191 | ||
bbcb81d0 | 2192 | put_task_struct(tsk); |
b78949eb | 2193 | out_unlock_cgroup: |
47cfcd09 | 2194 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
2195 | return ret; |
2196 | } | |
2197 | ||
7ae1bad9 TH |
2198 | /** |
2199 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2200 | * @from: attach to all cgroups of a given task | |
2201 | * @tsk: the task to be attached | |
2202 | */ | |
2203 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2204 | { | |
2205 | struct cgroupfs_root *root; | |
2206 | int retval = 0; | |
2207 | ||
47cfcd09 | 2208 | mutex_lock(&cgroup_mutex); |
7ae1bad9 | 2209 | for_each_active_root(root) { |
6f4b7e63 | 2210 | struct cgroup *from_cgrp = task_cgroup_from_root(from, root); |
7ae1bad9 | 2211 | |
6f4b7e63 | 2212 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2213 | if (retval) |
2214 | break; | |
2215 | } | |
47cfcd09 | 2216 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2217 | |
2218 | return retval; | |
2219 | } | |
2220 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2221 | ||
182446d0 TH |
2222 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2223 | struct cftype *cft, u64 pid) | |
74a1166d | 2224 | { |
182446d0 | 2225 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2226 | } |
2227 | ||
182446d0 TH |
2228 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2229 | struct cftype *cft, u64 tgid) | |
af351026 | 2230 | { |
182446d0 | 2231 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2232 | } |
2233 | ||
182446d0 TH |
2234 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
2235 | struct cftype *cft, const char *buffer) | |
e788e066 | 2236 | { |
182446d0 | 2237 | BUILD_BUG_ON(sizeof(css->cgroup->root->release_agent_path) < PATH_MAX); |
f4a2589f EK |
2238 | if (strlen(buffer) >= PATH_MAX) |
2239 | return -EINVAL; | |
182446d0 | 2240 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2241 | return -ENODEV; |
e25e2cbb | 2242 | mutex_lock(&cgroup_root_mutex); |
182446d0 | 2243 | strcpy(css->cgroup->root->release_agent_path, buffer); |
e25e2cbb | 2244 | mutex_unlock(&cgroup_root_mutex); |
47cfcd09 | 2245 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2246 | return 0; |
2247 | } | |
2248 | ||
182446d0 TH |
2249 | static int cgroup_release_agent_show(struct cgroup_subsys_state *css, |
2250 | struct cftype *cft, struct seq_file *seq) | |
e788e066 | 2251 | { |
182446d0 TH |
2252 | struct cgroup *cgrp = css->cgroup; |
2253 | ||
e788e066 PM |
2254 | if (!cgroup_lock_live_group(cgrp)) |
2255 | return -ENODEV; | |
2256 | seq_puts(seq, cgrp->root->release_agent_path); | |
2257 | seq_putc(seq, '\n'); | |
47cfcd09 | 2258 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2259 | return 0; |
2260 | } | |
2261 | ||
182446d0 TH |
2262 | static int cgroup_sane_behavior_show(struct cgroup_subsys_state *css, |
2263 | struct cftype *cft, struct seq_file *seq) | |
873fe09e | 2264 | { |
182446d0 | 2265 | seq_printf(seq, "%d\n", cgroup_sane_behavior(css->cgroup)); |
e788e066 PM |
2266 | return 0; |
2267 | } | |
2268 | ||
84eea842 PM |
2269 | /* A buffer size big enough for numbers or short strings */ |
2270 | #define CGROUP_LOCAL_BUFFER_SIZE 64 | |
2271 | ||
182446d0 TH |
2272 | static ssize_t cgroup_write_X64(struct cgroup_subsys_state *css, |
2273 | struct cftype *cft, struct file *file, | |
2274 | const char __user *userbuf, size_t nbytes, | |
2275 | loff_t *unused_ppos) | |
355e0c48 | 2276 | { |
84eea842 | 2277 | char buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
355e0c48 | 2278 | int retval = 0; |
355e0c48 PM |
2279 | char *end; |
2280 | ||
2281 | if (!nbytes) | |
2282 | return -EINVAL; | |
2283 | if (nbytes >= sizeof(buffer)) | |
2284 | return -E2BIG; | |
2285 | if (copy_from_user(buffer, userbuf, nbytes)) | |
2286 | return -EFAULT; | |
2287 | ||
2288 | buffer[nbytes] = 0; /* nul-terminate */ | |
e73d2c61 | 2289 | if (cft->write_u64) { |
478988d3 | 2290 | u64 val = simple_strtoull(strstrip(buffer), &end, 0); |
e73d2c61 PM |
2291 | if (*end) |
2292 | return -EINVAL; | |
182446d0 | 2293 | retval = cft->write_u64(css, cft, val); |
e73d2c61 | 2294 | } else { |
478988d3 | 2295 | s64 val = simple_strtoll(strstrip(buffer), &end, 0); |
e73d2c61 PM |
2296 | if (*end) |
2297 | return -EINVAL; | |
182446d0 | 2298 | retval = cft->write_s64(css, cft, val); |
e73d2c61 | 2299 | } |
355e0c48 PM |
2300 | if (!retval) |
2301 | retval = nbytes; | |
2302 | return retval; | |
2303 | } | |
2304 | ||
182446d0 TH |
2305 | static ssize_t cgroup_write_string(struct cgroup_subsys_state *css, |
2306 | struct cftype *cft, struct file *file, | |
2307 | const char __user *userbuf, size_t nbytes, | |
2308 | loff_t *unused_ppos) | |
db3b1497 | 2309 | { |
84eea842 | 2310 | char local_buffer[CGROUP_LOCAL_BUFFER_SIZE]; |
db3b1497 PM |
2311 | int retval = 0; |
2312 | size_t max_bytes = cft->max_write_len; | |
2313 | char *buffer = local_buffer; | |
2314 | ||
2315 | if (!max_bytes) | |
2316 | max_bytes = sizeof(local_buffer) - 1; | |
2317 | if (nbytes >= max_bytes) | |
2318 | return -E2BIG; | |
2319 | /* Allocate a dynamic buffer if we need one */ | |
2320 | if (nbytes >= sizeof(local_buffer)) { | |
2321 | buffer = kmalloc(nbytes + 1, GFP_KERNEL); | |
2322 | if (buffer == NULL) | |
2323 | return -ENOMEM; | |
2324 | } | |
5a3eb9f6 LZ |
2325 | if (nbytes && copy_from_user(buffer, userbuf, nbytes)) { |
2326 | retval = -EFAULT; | |
2327 | goto out; | |
2328 | } | |
db3b1497 PM |
2329 | |
2330 | buffer[nbytes] = 0; /* nul-terminate */ | |
182446d0 | 2331 | retval = cft->write_string(css, cft, strstrip(buffer)); |
db3b1497 PM |
2332 | if (!retval) |
2333 | retval = nbytes; | |
5a3eb9f6 | 2334 | out: |
db3b1497 PM |
2335 | if (buffer != local_buffer) |
2336 | kfree(buffer); | |
2337 | return retval; | |
2338 | } | |
2339 | ||
ddbcc7e8 | 2340 | static ssize_t cgroup_file_write(struct file *file, const char __user *buf, |
182446d0 | 2341 | size_t nbytes, loff_t *ppos) |
ddbcc7e8 | 2342 | { |
182446d0 | 2343 | struct cfent *cfe = __d_cfe(file->f_dentry); |
ddbcc7e8 | 2344 | struct cftype *cft = __d_cft(file->f_dentry); |
105347ba | 2345 | struct cgroup_subsys_state *css = cfe->css; |
ddbcc7e8 | 2346 | |
355e0c48 | 2347 | if (cft->write) |
182446d0 | 2348 | return cft->write(css, cft, file, buf, nbytes, ppos); |
e73d2c61 | 2349 | if (cft->write_u64 || cft->write_s64) |
182446d0 | 2350 | return cgroup_write_X64(css, cft, file, buf, nbytes, ppos); |
db3b1497 | 2351 | if (cft->write_string) |
182446d0 | 2352 | return cgroup_write_string(css, cft, file, buf, nbytes, ppos); |
d447ea2f | 2353 | if (cft->trigger) { |
182446d0 | 2354 | int ret = cft->trigger(css, (unsigned int)cft->private); |
d447ea2f PE |
2355 | return ret ? ret : nbytes; |
2356 | } | |
355e0c48 | 2357 | return -EINVAL; |
ddbcc7e8 PM |
2358 | } |
2359 | ||
182446d0 TH |
2360 | static ssize_t cgroup_read_u64(struct cgroup_subsys_state *css, |
2361 | struct cftype *cft, struct file *file, | |
2362 | char __user *buf, size_t nbytes, loff_t *ppos) | |
ddbcc7e8 | 2363 | { |
84eea842 | 2364 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
182446d0 | 2365 | u64 val = cft->read_u64(css, cft); |
ddbcc7e8 PM |
2366 | int len = sprintf(tmp, "%llu\n", (unsigned long long) val); |
2367 | ||
2368 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
2369 | } | |
2370 | ||
182446d0 TH |
2371 | static ssize_t cgroup_read_s64(struct cgroup_subsys_state *css, |
2372 | struct cftype *cft, struct file *file, | |
2373 | char __user *buf, size_t nbytes, loff_t *ppos) | |
e73d2c61 | 2374 | { |
84eea842 | 2375 | char tmp[CGROUP_LOCAL_BUFFER_SIZE]; |
182446d0 | 2376 | s64 val = cft->read_s64(css, cft); |
e73d2c61 PM |
2377 | int len = sprintf(tmp, "%lld\n", (long long) val); |
2378 | ||
2379 | return simple_read_from_buffer(buf, nbytes, ppos, tmp, len); | |
2380 | } | |
2381 | ||
ddbcc7e8 | 2382 | static ssize_t cgroup_file_read(struct file *file, char __user *buf, |
182446d0 | 2383 | size_t nbytes, loff_t *ppos) |
ddbcc7e8 | 2384 | { |
182446d0 | 2385 | struct cfent *cfe = __d_cfe(file->f_dentry); |
ddbcc7e8 | 2386 | struct cftype *cft = __d_cft(file->f_dentry); |
105347ba | 2387 | struct cgroup_subsys_state *css = cfe->css; |
ddbcc7e8 PM |
2388 | |
2389 | if (cft->read) | |
182446d0 | 2390 | return cft->read(css, cft, file, buf, nbytes, ppos); |
f4c753b7 | 2391 | if (cft->read_u64) |
182446d0 | 2392 | return cgroup_read_u64(css, cft, file, buf, nbytes, ppos); |
e73d2c61 | 2393 | if (cft->read_s64) |
182446d0 | 2394 | return cgroup_read_s64(css, cft, file, buf, nbytes, ppos); |
ddbcc7e8 PM |
2395 | return -EINVAL; |
2396 | } | |
2397 | ||
91796569 PM |
2398 | /* |
2399 | * seqfile ops/methods for returning structured data. Currently just | |
2400 | * supports string->u64 maps, but can be extended in future. | |
2401 | */ | |
2402 | ||
91796569 PM |
2403 | static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value) |
2404 | { | |
2405 | struct seq_file *sf = cb->state; | |
2406 | return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value); | |
2407 | } | |
2408 | ||
2409 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) | |
2410 | { | |
e0798ce2 LZ |
2411 | struct cfent *cfe = m->private; |
2412 | struct cftype *cft = cfe->type; | |
105347ba | 2413 | struct cgroup_subsys_state *css = cfe->css; |
e0798ce2 | 2414 | |
29486df3 SH |
2415 | if (cft->read_map) { |
2416 | struct cgroup_map_cb cb = { | |
2417 | .fill = cgroup_map_add, | |
2418 | .state = m, | |
2419 | }; | |
182446d0 | 2420 | return cft->read_map(css, cft, &cb); |
29486df3 | 2421 | } |
182446d0 | 2422 | return cft->read_seq_string(css, cft, m); |
91796569 PM |
2423 | } |
2424 | ||
828c0950 | 2425 | static const struct file_operations cgroup_seqfile_operations = { |
91796569 | 2426 | .read = seq_read, |
e788e066 | 2427 | .write = cgroup_file_write, |
91796569 | 2428 | .llseek = seq_lseek, |
e0798ce2 | 2429 | .release = single_release, |
91796569 PM |
2430 | }; |
2431 | ||
ddbcc7e8 PM |
2432 | static int cgroup_file_open(struct inode *inode, struct file *file) |
2433 | { | |
f7d58818 TH |
2434 | struct cfent *cfe = __d_cfe(file->f_dentry); |
2435 | struct cftype *cft = __d_cft(file->f_dentry); | |
105347ba TH |
2436 | struct cgroup *cgrp = __d_cgrp(cfe->dentry->d_parent); |
2437 | struct cgroup_subsys_state *css; | |
ddbcc7e8 | 2438 | int err; |
ddbcc7e8 PM |
2439 | |
2440 | err = generic_file_open(inode, file); | |
2441 | if (err) | |
2442 | return err; | |
75139b82 | 2443 | |
f7d58818 TH |
2444 | /* |
2445 | * If the file belongs to a subsystem, pin the css. Will be | |
2446 | * unpinned either on open failure or release. This ensures that | |
2447 | * @css stays alive for all file operations. | |
2448 | */ | |
105347ba | 2449 | rcu_read_lock(); |
ca8bdcaf TH |
2450 | css = cgroup_css(cgrp, cft->ss); |
2451 | if (cft->ss && !css_tryget(css)) | |
2452 | css = NULL; | |
105347ba | 2453 | rcu_read_unlock(); |
f4f4be2b | 2454 | |
0bfb4aa6 | 2455 | if (!css) |
f7d58818 | 2456 | return -ENODEV; |
75139b82 | 2457 | |
0bfb4aa6 TH |
2458 | /* |
2459 | * @cfe->css is used by read/write/close to determine the | |
2460 | * associated css. @file->private_data would be a better place but | |
2461 | * that's already used by seqfile. Multiple accessors may use it | |
2462 | * simultaneously which is okay as the association never changes. | |
2463 | */ | |
2464 | WARN_ON_ONCE(cfe->css && cfe->css != css); | |
2465 | cfe->css = css; | |
f4f4be2b | 2466 | |
29486df3 | 2467 | if (cft->read_map || cft->read_seq_string) { |
91796569 | 2468 | file->f_op = &cgroup_seqfile_operations; |
e0798ce2 LZ |
2469 | err = single_open(file, cgroup_seqfile_show, cfe); |
2470 | } else if (cft->open) { | |
ddbcc7e8 | 2471 | err = cft->open(inode, file); |
e0798ce2 | 2472 | } |
ddbcc7e8 | 2473 | |
67f4c36f | 2474 | if (css->ss && err) |
f7d58818 | 2475 | css_put(css); |
ddbcc7e8 PM |
2476 | return err; |
2477 | } | |
2478 | ||
2479 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
2480 | { | |
f7d58818 | 2481 | struct cfent *cfe = __d_cfe(file->f_dentry); |
ddbcc7e8 | 2482 | struct cftype *cft = __d_cft(file->f_dentry); |
105347ba | 2483 | struct cgroup_subsys_state *css = cfe->css; |
f7d58818 TH |
2484 | int ret = 0; |
2485 | ||
ddbcc7e8 | 2486 | if (cft->release) |
f7d58818 | 2487 | ret = cft->release(inode, file); |
67f4c36f | 2488 | if (css->ss) |
f7d58818 TH |
2489 | css_put(css); |
2490 | return ret; | |
ddbcc7e8 PM |
2491 | } |
2492 | ||
2493 | /* | |
2494 | * cgroup_rename - Only allow simple rename of directories in place. | |
2495 | */ | |
2496 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
2497 | struct inode *new_dir, struct dentry *new_dentry) | |
2498 | { | |
65dff759 LZ |
2499 | int ret; |
2500 | struct cgroup_name *name, *old_name; | |
2501 | struct cgroup *cgrp; | |
2502 | ||
2503 | /* | |
2504 | * It's convinient to use parent dir's i_mutex to protected | |
2505 | * cgrp->name. | |
2506 | */ | |
2507 | lockdep_assert_held(&old_dir->i_mutex); | |
2508 | ||
ddbcc7e8 PM |
2509 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) |
2510 | return -ENOTDIR; | |
2511 | if (new_dentry->d_inode) | |
2512 | return -EEXIST; | |
2513 | if (old_dir != new_dir) | |
2514 | return -EIO; | |
65dff759 LZ |
2515 | |
2516 | cgrp = __d_cgrp(old_dentry); | |
2517 | ||
6db8e85c TH |
2518 | /* |
2519 | * This isn't a proper migration and its usefulness is very | |
2520 | * limited. Disallow if sane_behavior. | |
2521 | */ | |
2522 | if (cgroup_sane_behavior(cgrp)) | |
2523 | return -EPERM; | |
2524 | ||
65dff759 LZ |
2525 | name = cgroup_alloc_name(new_dentry); |
2526 | if (!name) | |
2527 | return -ENOMEM; | |
2528 | ||
2529 | ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
2530 | if (ret) { | |
2531 | kfree(name); | |
2532 | return ret; | |
2533 | } | |
2534 | ||
a4ea1cc9 | 2535 | old_name = rcu_dereference_protected(cgrp->name, true); |
65dff759 LZ |
2536 | rcu_assign_pointer(cgrp->name, name); |
2537 | ||
2538 | kfree_rcu(old_name, rcu_head); | |
2539 | return 0; | |
ddbcc7e8 PM |
2540 | } |
2541 | ||
03b1cde6 AR |
2542 | static struct simple_xattrs *__d_xattrs(struct dentry *dentry) |
2543 | { | |
2544 | if (S_ISDIR(dentry->d_inode->i_mode)) | |
2545 | return &__d_cgrp(dentry)->xattrs; | |
2546 | else | |
712317ad | 2547 | return &__d_cfe(dentry)->xattrs; |
03b1cde6 AR |
2548 | } |
2549 | ||
2550 | static inline int xattr_enabled(struct dentry *dentry) | |
2551 | { | |
2552 | struct cgroupfs_root *root = dentry->d_sb->s_fs_info; | |
93438629 | 2553 | return root->flags & CGRP_ROOT_XATTR; |
03b1cde6 AR |
2554 | } |
2555 | ||
2556 | static bool is_valid_xattr(const char *name) | |
2557 | { | |
2558 | if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) || | |
2559 | !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN)) | |
2560 | return true; | |
2561 | return false; | |
2562 | } | |
2563 | ||
2564 | static int cgroup_setxattr(struct dentry *dentry, const char *name, | |
2565 | const void *val, size_t size, int flags) | |
2566 | { | |
2567 | if (!xattr_enabled(dentry)) | |
2568 | return -EOPNOTSUPP; | |
2569 | if (!is_valid_xattr(name)) | |
2570 | return -EINVAL; | |
2571 | return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags); | |
2572 | } | |
2573 | ||
2574 | static int cgroup_removexattr(struct dentry *dentry, const char *name) | |
2575 | { | |
2576 | if (!xattr_enabled(dentry)) | |
2577 | return -EOPNOTSUPP; | |
2578 | if (!is_valid_xattr(name)) | |
2579 | return -EINVAL; | |
2580 | return simple_xattr_remove(__d_xattrs(dentry), name); | |
2581 | } | |
2582 | ||
2583 | static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name, | |
2584 | void *buf, size_t size) | |
2585 | { | |
2586 | if (!xattr_enabled(dentry)) | |
2587 | return -EOPNOTSUPP; | |
2588 | if (!is_valid_xattr(name)) | |
2589 | return -EINVAL; | |
2590 | return simple_xattr_get(__d_xattrs(dentry), name, buf, size); | |
2591 | } | |
2592 | ||
2593 | static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size) | |
2594 | { | |
2595 | if (!xattr_enabled(dentry)) | |
2596 | return -EOPNOTSUPP; | |
2597 | return simple_xattr_list(__d_xattrs(dentry), buf, size); | |
2598 | } | |
2599 | ||
828c0950 | 2600 | static const struct file_operations cgroup_file_operations = { |
ddbcc7e8 PM |
2601 | .read = cgroup_file_read, |
2602 | .write = cgroup_file_write, | |
2603 | .llseek = generic_file_llseek, | |
2604 | .open = cgroup_file_open, | |
2605 | .release = cgroup_file_release, | |
2606 | }; | |
2607 | ||
03b1cde6 AR |
2608 | static const struct inode_operations cgroup_file_inode_operations = { |
2609 | .setxattr = cgroup_setxattr, | |
2610 | .getxattr = cgroup_getxattr, | |
2611 | .listxattr = cgroup_listxattr, | |
2612 | .removexattr = cgroup_removexattr, | |
2613 | }; | |
2614 | ||
6e1d5dcc | 2615 | static const struct inode_operations cgroup_dir_inode_operations = { |
786e1448 | 2616 | .lookup = simple_lookup, |
ddbcc7e8 PM |
2617 | .mkdir = cgroup_mkdir, |
2618 | .rmdir = cgroup_rmdir, | |
2619 | .rename = cgroup_rename, | |
03b1cde6 AR |
2620 | .setxattr = cgroup_setxattr, |
2621 | .getxattr = cgroup_getxattr, | |
2622 | .listxattr = cgroup_listxattr, | |
2623 | .removexattr = cgroup_removexattr, | |
ddbcc7e8 PM |
2624 | }; |
2625 | ||
0dea1168 KS |
2626 | /* |
2627 | * Check if a file is a control file | |
2628 | */ | |
2629 | static inline struct cftype *__file_cft(struct file *file) | |
2630 | { | |
496ad9aa | 2631 | if (file_inode(file)->i_fop != &cgroup_file_operations) |
0dea1168 KS |
2632 | return ERR_PTR(-EINVAL); |
2633 | return __d_cft(file->f_dentry); | |
2634 | } | |
2635 | ||
a5e7ed32 | 2636 | static int cgroup_create_file(struct dentry *dentry, umode_t mode, |
5adcee1d NP |
2637 | struct super_block *sb) |
2638 | { | |
ddbcc7e8 PM |
2639 | struct inode *inode; |
2640 | ||
2641 | if (!dentry) | |
2642 | return -ENOENT; | |
2643 | if (dentry->d_inode) | |
2644 | return -EEXIST; | |
2645 | ||
2646 | inode = cgroup_new_inode(mode, sb); | |
2647 | if (!inode) | |
2648 | return -ENOMEM; | |
2649 | ||
2650 | if (S_ISDIR(mode)) { | |
2651 | inode->i_op = &cgroup_dir_inode_operations; | |
2652 | inode->i_fop = &simple_dir_operations; | |
2653 | ||
2654 | /* start off with i_nlink == 2 (for "." entry) */ | |
2655 | inc_nlink(inode); | |
28fd6f30 | 2656 | inc_nlink(dentry->d_parent->d_inode); |
ddbcc7e8 | 2657 | |
b8a2df6a TH |
2658 | /* |
2659 | * Control reaches here with cgroup_mutex held. | |
2660 | * @inode->i_mutex should nest outside cgroup_mutex but we | |
2661 | * want to populate it immediately without releasing | |
2662 | * cgroup_mutex. As @inode isn't visible to anyone else | |
2663 | * yet, trylock will always succeed without affecting | |
2664 | * lockdep checks. | |
2665 | */ | |
2666 | WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex)); | |
ddbcc7e8 PM |
2667 | } else if (S_ISREG(mode)) { |
2668 | inode->i_size = 0; | |
2669 | inode->i_fop = &cgroup_file_operations; | |
03b1cde6 | 2670 | inode->i_op = &cgroup_file_inode_operations; |
ddbcc7e8 | 2671 | } |
ddbcc7e8 PM |
2672 | d_instantiate(dentry, inode); |
2673 | dget(dentry); /* Extra count - pin the dentry in core */ | |
2674 | return 0; | |
2675 | } | |
2676 | ||
099fca32 LZ |
2677 | /** |
2678 | * cgroup_file_mode - deduce file mode of a control file | |
2679 | * @cft: the control file in question | |
2680 | * | |
2681 | * returns cft->mode if ->mode is not 0 | |
2682 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
2683 | * returns S_IRUGO if it has only a read handler | |
2684 | * returns S_IWUSR if it has only a write hander | |
2685 | */ | |
a5e7ed32 | 2686 | static umode_t cgroup_file_mode(const struct cftype *cft) |
099fca32 | 2687 | { |
a5e7ed32 | 2688 | umode_t mode = 0; |
099fca32 LZ |
2689 | |
2690 | if (cft->mode) | |
2691 | return cft->mode; | |
2692 | ||
2693 | if (cft->read || cft->read_u64 || cft->read_s64 || | |
2694 | cft->read_map || cft->read_seq_string) | |
2695 | mode |= S_IRUGO; | |
2696 | ||
2697 | if (cft->write || cft->write_u64 || cft->write_s64 || | |
2698 | cft->write_string || cft->trigger) | |
2699 | mode |= S_IWUSR; | |
2700 | ||
2701 | return mode; | |
2702 | } | |
2703 | ||
2bb566cb | 2704 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2705 | { |
bd89aabc | 2706 | struct dentry *dir = cgrp->dentry; |
05ef1d7c | 2707 | struct cgroup *parent = __d_cgrp(dir); |
ddbcc7e8 | 2708 | struct dentry *dentry; |
05ef1d7c | 2709 | struct cfent *cfe; |
ddbcc7e8 | 2710 | int error; |
a5e7ed32 | 2711 | umode_t mode; |
ddbcc7e8 | 2712 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; |
8e3f6541 | 2713 | |
9fa4db33 TH |
2714 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
2715 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) { | |
2bb566cb | 2716 | strcpy(name, cft->ss->name); |
ddbcc7e8 PM |
2717 | strcat(name, "."); |
2718 | } | |
2719 | strcat(name, cft->name); | |
05ef1d7c | 2720 | |
ddbcc7e8 | 2721 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); |
05ef1d7c TH |
2722 | |
2723 | cfe = kzalloc(sizeof(*cfe), GFP_KERNEL); | |
2724 | if (!cfe) | |
2725 | return -ENOMEM; | |
2726 | ||
ddbcc7e8 | 2727 | dentry = lookup_one_len(name, dir, strlen(name)); |
05ef1d7c | 2728 | if (IS_ERR(dentry)) { |
ddbcc7e8 | 2729 | error = PTR_ERR(dentry); |
05ef1d7c TH |
2730 | goto out; |
2731 | } | |
2732 | ||
d6cbf35d LZ |
2733 | cfe->type = (void *)cft; |
2734 | cfe->dentry = dentry; | |
2735 | dentry->d_fsdata = cfe; | |
2736 | simple_xattrs_init(&cfe->xattrs); | |
2737 | ||
05ef1d7c TH |
2738 | mode = cgroup_file_mode(cft); |
2739 | error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb); | |
2740 | if (!error) { | |
05ef1d7c TH |
2741 | list_add_tail(&cfe->node, &parent->files); |
2742 | cfe = NULL; | |
2743 | } | |
2744 | dput(dentry); | |
2745 | out: | |
2746 | kfree(cfe); | |
ddbcc7e8 PM |
2747 | return error; |
2748 | } | |
2749 | ||
b1f28d31 TH |
2750 | /** |
2751 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2752 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2753 | * @cfts: array of cftypes to be added |
2754 | * @is_add: whether to add or remove | |
2755 | * | |
2756 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2757 | * For removals, this function never fails. If addition fails, this |
2758 | * function doesn't remove files already added. The caller is responsible | |
2759 | * for cleaning up. | |
b1f28d31 | 2760 | */ |
2bb566cb TH |
2761 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2762 | bool is_add) | |
ddbcc7e8 | 2763 | { |
03b1cde6 | 2764 | struct cftype *cft; |
b1f28d31 TH |
2765 | int ret; |
2766 | ||
2767 | lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex); | |
2768 | lockdep_assert_held(&cgroup_mutex); | |
db0416b6 TH |
2769 | |
2770 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2771 | /* does cft->flags tell us to skip this file on @cgrp? */ |
873fe09e TH |
2772 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2773 | continue; | |
f33fddc2 G |
2774 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2775 | continue; | |
2776 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2777 | continue; | |
2778 | ||
2739d3cc | 2779 | if (is_add) { |
2bb566cb | 2780 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2781 | if (ret) { |
2739d3cc | 2782 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2783 | cft->name, ret); |
2784 | return ret; | |
2785 | } | |
2739d3cc LZ |
2786 | } else { |
2787 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2788 | } |
ddbcc7e8 | 2789 | } |
b1f28d31 | 2790 | return 0; |
ddbcc7e8 PM |
2791 | } |
2792 | ||
8e3f6541 | 2793 | static void cgroup_cfts_prepare(void) |
e8c82d20 | 2794 | __acquires(&cgroup_mutex) |
8e3f6541 TH |
2795 | { |
2796 | /* | |
2797 | * Thanks to the entanglement with vfs inode locking, we can't walk | |
2798 | * the existing cgroups under cgroup_mutex and create files. | |
492eb21b TH |
2799 | * Instead, we use css_for_each_descendant_pre() and drop RCU read |
2800 | * lock before calling cgroup_addrm_files(). | |
8e3f6541 | 2801 | */ |
8e3f6541 TH |
2802 | mutex_lock(&cgroup_mutex); |
2803 | } | |
2804 | ||
2bb566cb | 2805 | static int cgroup_cfts_commit(struct cftype *cfts, bool is_add) |
e8c82d20 | 2806 | __releases(&cgroup_mutex) |
8e3f6541 TH |
2807 | { |
2808 | LIST_HEAD(pending); | |
2bb566cb | 2809 | struct cgroup_subsys *ss = cfts[0].ss; |
492eb21b | 2810 | struct cgroup *root = &ss->root->top_cgroup; |
084457f2 | 2811 | struct super_block *sb = ss->root->sb; |
e8c82d20 LZ |
2812 | struct dentry *prev = NULL; |
2813 | struct inode *inode; | |
492eb21b | 2814 | struct cgroup_subsys_state *css; |
00356bd5 | 2815 | u64 update_before; |
9ccece80 | 2816 | int ret = 0; |
8e3f6541 TH |
2817 | |
2818 | /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */ | |
9871bf95 | 2819 | if (!cfts || ss->root == &cgroup_dummy_root || |
e8c82d20 LZ |
2820 | !atomic_inc_not_zero(&sb->s_active)) { |
2821 | mutex_unlock(&cgroup_mutex); | |
9ccece80 | 2822 | return 0; |
8e3f6541 TH |
2823 | } |
2824 | ||
8e3f6541 | 2825 | /* |
e8c82d20 LZ |
2826 | * All cgroups which are created after we drop cgroup_mutex will |
2827 | * have the updated set of files, so we only need to update the | |
00356bd5 | 2828 | * cgroups created before the current @cgroup_serial_nr_next. |
8e3f6541 | 2829 | */ |
00356bd5 | 2830 | update_before = cgroup_serial_nr_next; |
e8c82d20 LZ |
2831 | |
2832 | mutex_unlock(&cgroup_mutex); | |
2833 | ||
e8c82d20 LZ |
2834 | /* add/rm files for all cgroups created before */ |
2835 | rcu_read_lock(); | |
ca8bdcaf | 2836 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2837 | struct cgroup *cgrp = css->cgroup; |
2838 | ||
e8c82d20 LZ |
2839 | if (cgroup_is_dead(cgrp)) |
2840 | continue; | |
2841 | ||
2842 | inode = cgrp->dentry->d_inode; | |
2843 | dget(cgrp->dentry); | |
2844 | rcu_read_unlock(); | |
2845 | ||
2846 | dput(prev); | |
2847 | prev = cgrp->dentry; | |
8e3f6541 TH |
2848 | |
2849 | mutex_lock(&inode->i_mutex); | |
2850 | mutex_lock(&cgroup_mutex); | |
00356bd5 | 2851 | if (cgrp->serial_nr < update_before && !cgroup_is_dead(cgrp)) |
2bb566cb | 2852 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
8e3f6541 TH |
2853 | mutex_unlock(&cgroup_mutex); |
2854 | mutex_unlock(&inode->i_mutex); | |
2855 | ||
e8c82d20 | 2856 | rcu_read_lock(); |
9ccece80 TH |
2857 | if (ret) |
2858 | break; | |
8e3f6541 | 2859 | } |
e8c82d20 LZ |
2860 | rcu_read_unlock(); |
2861 | dput(prev); | |
2862 | deactivate_super(sb); | |
9ccece80 | 2863 | return ret; |
8e3f6541 TH |
2864 | } |
2865 | ||
2866 | /** | |
2867 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2868 | * @ss: target cgroup subsystem | |
2869 | * @cfts: zero-length name terminated array of cftypes | |
2870 | * | |
2871 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2872 | * existing cgroups to which @ss is attached and all future cgroups will | |
2873 | * have them too. This function can be called anytime whether @ss is | |
2874 | * attached or not. | |
2875 | * | |
2876 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2877 | * function currently returns 0 as long as @cfts registration is successful | |
2878 | * even if some file creation attempts on existing cgroups fail. | |
2879 | */ | |
03b1cde6 | 2880 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 TH |
2881 | { |
2882 | struct cftype_set *set; | |
2bb566cb | 2883 | struct cftype *cft; |
9ccece80 | 2884 | int ret; |
8e3f6541 TH |
2885 | |
2886 | set = kzalloc(sizeof(*set), GFP_KERNEL); | |
2887 | if (!set) | |
2888 | return -ENOMEM; | |
2889 | ||
2bb566cb TH |
2890 | for (cft = cfts; cft->name[0] != '\0'; cft++) |
2891 | cft->ss = ss; | |
2892 | ||
8e3f6541 TH |
2893 | cgroup_cfts_prepare(); |
2894 | set->cfts = cfts; | |
2895 | list_add_tail(&set->node, &ss->cftsets); | |
2bb566cb | 2896 | ret = cgroup_cfts_commit(cfts, true); |
9ccece80 | 2897 | if (ret) |
2bb566cb | 2898 | cgroup_rm_cftypes(cfts); |
9ccece80 | 2899 | return ret; |
8e3f6541 TH |
2900 | } |
2901 | EXPORT_SYMBOL_GPL(cgroup_add_cftypes); | |
2902 | ||
79578621 TH |
2903 | /** |
2904 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
2905 | * @cfts: zero-length name terminated array of cftypes |
2906 | * | |
2bb566cb TH |
2907 | * Unregister @cfts. Files described by @cfts are removed from all |
2908 | * existing cgroups and all future cgroups won't have them either. This | |
2909 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
2910 | * |
2911 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 2912 | * registered. |
79578621 | 2913 | */ |
2bb566cb | 2914 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 TH |
2915 | { |
2916 | struct cftype_set *set; | |
2917 | ||
2bb566cb TH |
2918 | if (!cfts || !cfts[0].ss) |
2919 | return -ENOENT; | |
2920 | ||
79578621 TH |
2921 | cgroup_cfts_prepare(); |
2922 | ||
2bb566cb | 2923 | list_for_each_entry(set, &cfts[0].ss->cftsets, node) { |
79578621 | 2924 | if (set->cfts == cfts) { |
f57947d2 LZ |
2925 | list_del(&set->node); |
2926 | kfree(set); | |
2bb566cb | 2927 | cgroup_cfts_commit(cfts, false); |
79578621 TH |
2928 | return 0; |
2929 | } | |
2930 | } | |
2931 | ||
2bb566cb | 2932 | cgroup_cfts_commit(NULL, false); |
79578621 TH |
2933 | return -ENOENT; |
2934 | } | |
2935 | ||
a043e3b2 LZ |
2936 | /** |
2937 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2938 | * @cgrp: the cgroup in question | |
2939 | * | |
2940 | * Return the number of tasks in the cgroup. | |
2941 | */ | |
bd89aabc | 2942 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2943 | { |
2944 | int count = 0; | |
69d0206c | 2945 | struct cgrp_cset_link *link; |
817929ec PM |
2946 | |
2947 | read_lock(&css_set_lock); | |
69d0206c TH |
2948 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2949 | count += atomic_read(&link->cset->refcount); | |
817929ec | 2950 | read_unlock(&css_set_lock); |
bbcb81d0 PM |
2951 | return count; |
2952 | } | |
2953 | ||
817929ec | 2954 | /* |
0942eeee TH |
2955 | * To reduce the fork() overhead for systems that are not actually using |
2956 | * their cgroups capability, we don't maintain the lists running through | |
2957 | * each css_set to its tasks until we see the list actually used - in other | |
72ec7029 | 2958 | * words after the first call to css_task_iter_start(). |
31a7df01 | 2959 | */ |
3df91fe3 | 2960 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2961 | { |
2962 | struct task_struct *p, *g; | |
2963 | write_lock(&css_set_lock); | |
2964 | use_task_css_set_links = 1; | |
3ce3230a FW |
2965 | /* |
2966 | * We need tasklist_lock because RCU is not safe against | |
2967 | * while_each_thread(). Besides, a forking task that has passed | |
2968 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
2969 | * is not guaranteed to have its child immediately visible in the | |
2970 | * tasklist if we walk through it with RCU. | |
2971 | */ | |
2972 | read_lock(&tasklist_lock); | |
31a7df01 CW |
2973 | do_each_thread(g, p) { |
2974 | task_lock(p); | |
0e04388f LZ |
2975 | /* |
2976 | * We should check if the process is exiting, otherwise | |
2977 | * it will race with cgroup_exit() in that the list | |
2978 | * entry won't be deleted though the process has exited. | |
2979 | */ | |
2980 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
a8ad805c | 2981 | list_add(&p->cg_list, &task_css_set(p)->tasks); |
31a7df01 CW |
2982 | task_unlock(p); |
2983 | } while_each_thread(g, p); | |
3ce3230a | 2984 | read_unlock(&tasklist_lock); |
31a7df01 CW |
2985 | write_unlock(&css_set_lock); |
2986 | } | |
2987 | ||
53fa5261 | 2988 | /** |
492eb21b TH |
2989 | * css_next_child - find the next child of a given css |
2990 | * @pos_css: the current position (%NULL to initiate traversal) | |
2991 | * @parent_css: css whose children to walk | |
53fa5261 | 2992 | * |
492eb21b TH |
2993 | * This function returns the next child of @parent_css and should be called |
2994 | * under RCU read lock. The only requirement is that @parent_css and | |
2995 | * @pos_css are accessible. The next sibling is guaranteed to be returned | |
2996 | * regardless of their states. | |
53fa5261 | 2997 | */ |
492eb21b TH |
2998 | struct cgroup_subsys_state * |
2999 | css_next_child(struct cgroup_subsys_state *pos_css, | |
3000 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 3001 | { |
492eb21b TH |
3002 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
3003 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
3004 | struct cgroup *next; |
3005 | ||
3006 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
3007 | ||
3008 | /* | |
3009 | * @pos could already have been removed. Once a cgroup is removed, | |
3010 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
3011 | * changes. As CGRP_DEAD assertion is serialized and happens |
3012 | * before the cgroup is taken off the ->sibling list, if we see it | |
3013 | * unasserted, it's guaranteed that the next sibling hasn't | |
3014 | * finished its grace period even if it's already removed, and thus | |
3015 | * safe to dereference from this RCU critical section. If | |
3016 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
3017 | * to be visible as %true here. | |
3b287a50 TH |
3018 | * |
3019 | * If @pos is dead, its next pointer can't be dereferenced; | |
3020 | * however, as each cgroup is given a monotonically increasing | |
3021 | * unique serial number and always appended to the sibling list, | |
3022 | * the next one can be found by walking the parent's children until | |
3023 | * we see a cgroup with higher serial number than @pos's. While | |
3024 | * this path can be slower, it's taken only when either the current | |
3025 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 3026 | */ |
3b287a50 TH |
3027 | if (!pos) { |
3028 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
3029 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 3030 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
3031 | } else { |
3032 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
3033 | if (next->serial_nr > pos->serial_nr) | |
3034 | break; | |
53fa5261 TH |
3035 | } |
3036 | ||
492eb21b TH |
3037 | if (&next->sibling == &cgrp->children) |
3038 | return NULL; | |
3039 | ||
ca8bdcaf | 3040 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 3041 | } |
492eb21b | 3042 | EXPORT_SYMBOL_GPL(css_next_child); |
53fa5261 | 3043 | |
574bd9f7 | 3044 | /** |
492eb21b | 3045 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3046 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3047 | * @root: css whose descendants to walk |
574bd9f7 | 3048 | * |
492eb21b | 3049 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3050 | * to visit for pre-order traversal of @root's descendants. @root is |
3051 | * included in the iteration and the first node to be visited. | |
75501a6d TH |
3052 | * |
3053 | * While this function requires RCU read locking, it doesn't require the | |
3054 | * whole traversal to be contained in a single RCU critical section. This | |
3055 | * function will return the correct next descendant as long as both @pos | |
492eb21b | 3056 | * and @root are accessible and @pos is a descendant of @root. |
574bd9f7 | 3057 | */ |
492eb21b TH |
3058 | struct cgroup_subsys_state * |
3059 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3060 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3061 | { |
492eb21b | 3062 | struct cgroup_subsys_state *next; |
574bd9f7 TH |
3063 | |
3064 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
3065 | ||
bd8815a6 | 3066 | /* if first iteration, visit @root */ |
7805d000 | 3067 | if (!pos) |
bd8815a6 | 3068 | return root; |
574bd9f7 TH |
3069 | |
3070 | /* visit the first child if exists */ | |
492eb21b | 3071 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3072 | if (next) |
3073 | return next; | |
3074 | ||
3075 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
3076 | while (pos != root) { |
3077 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 3078 | if (next) |
574bd9f7 | 3079 | return next; |
492eb21b | 3080 | pos = css_parent(pos); |
7805d000 | 3081 | } |
574bd9f7 TH |
3082 | |
3083 | return NULL; | |
3084 | } | |
492eb21b | 3085 | EXPORT_SYMBOL_GPL(css_next_descendant_pre); |
574bd9f7 | 3086 | |
12a9d2fe | 3087 | /** |
492eb21b TH |
3088 | * css_rightmost_descendant - return the rightmost descendant of a css |
3089 | * @pos: css of interest | |
12a9d2fe | 3090 | * |
492eb21b TH |
3091 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3092 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3093 | * subtree of @pos. |
75501a6d TH |
3094 | * |
3095 | * While this function requires RCU read locking, it doesn't require the | |
3096 | * whole traversal to be contained in a single RCU critical section. This | |
3097 | * function will return the correct rightmost descendant as long as @pos is | |
3098 | * accessible. | |
12a9d2fe | 3099 | */ |
492eb21b TH |
3100 | struct cgroup_subsys_state * |
3101 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3102 | { |
492eb21b | 3103 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe TH |
3104 | |
3105 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
3106 | ||
3107 | do { | |
3108 | last = pos; | |
3109 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3110 | pos = NULL; | |
492eb21b | 3111 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3112 | pos = tmp; |
3113 | } while (pos); | |
3114 | ||
3115 | return last; | |
3116 | } | |
492eb21b | 3117 | EXPORT_SYMBOL_GPL(css_rightmost_descendant); |
12a9d2fe | 3118 | |
492eb21b TH |
3119 | static struct cgroup_subsys_state * |
3120 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3121 | { |
492eb21b | 3122 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3123 | |
3124 | do { | |
3125 | last = pos; | |
492eb21b | 3126 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3127 | } while (pos); |
3128 | ||
3129 | return last; | |
3130 | } | |
3131 | ||
3132 | /** | |
492eb21b | 3133 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3134 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3135 | * @root: css whose descendants to walk |
574bd9f7 | 3136 | * |
492eb21b | 3137 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3138 | * to visit for post-order traversal of @root's descendants. @root is |
3139 | * included in the iteration and the last node to be visited. | |
75501a6d TH |
3140 | * |
3141 | * While this function requires RCU read locking, it doesn't require the | |
3142 | * whole traversal to be contained in a single RCU critical section. This | |
3143 | * function will return the correct next descendant as long as both @pos | |
3144 | * and @cgroup are accessible and @pos is a descendant of @cgroup. | |
574bd9f7 | 3145 | */ |
492eb21b TH |
3146 | struct cgroup_subsys_state * |
3147 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3148 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3149 | { |
492eb21b | 3150 | struct cgroup_subsys_state *next; |
574bd9f7 TH |
3151 | |
3152 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
3153 | ||
58b79a91 TH |
3154 | /* if first iteration, visit leftmost descendant which may be @root */ |
3155 | if (!pos) | |
3156 | return css_leftmost_descendant(root); | |
574bd9f7 | 3157 | |
bd8815a6 TH |
3158 | /* if we visited @root, we're done */ |
3159 | if (pos == root) | |
3160 | return NULL; | |
3161 | ||
574bd9f7 | 3162 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 3163 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 3164 | if (next) |
492eb21b | 3165 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3166 | |
3167 | /* no sibling left, visit parent */ | |
bd8815a6 | 3168 | return css_parent(pos); |
574bd9f7 | 3169 | } |
492eb21b | 3170 | EXPORT_SYMBOL_GPL(css_next_descendant_post); |
574bd9f7 | 3171 | |
0942eeee | 3172 | /** |
72ec7029 | 3173 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
3174 | * @it: the iterator to advance |
3175 | * | |
3176 | * Advance @it to the next css_set to walk. | |
d515876e | 3177 | */ |
72ec7029 | 3178 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
3179 | { |
3180 | struct list_head *l = it->cset_link; | |
3181 | struct cgrp_cset_link *link; | |
3182 | struct css_set *cset; | |
3183 | ||
3184 | /* Advance to the next non-empty css_set */ | |
3185 | do { | |
3186 | l = l->next; | |
72ec7029 | 3187 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
3188 | it->cset_link = NULL; |
3189 | return; | |
3190 | } | |
3191 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3192 | cset = link->cset; | |
3193 | } while (list_empty(&cset->tasks)); | |
3194 | it->cset_link = l; | |
3195 | it->task = cset->tasks.next; | |
3196 | } | |
3197 | ||
0942eeee | 3198 | /** |
72ec7029 TH |
3199 | * css_task_iter_start - initiate task iteration |
3200 | * @css: the css to walk tasks of | |
0942eeee TH |
3201 | * @it: the task iterator to use |
3202 | * | |
72ec7029 TH |
3203 | * Initiate iteration through the tasks of @css. The caller can call |
3204 | * css_task_iter_next() to walk through the tasks until the function | |
3205 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3206 | * called. | |
0942eeee TH |
3207 | * |
3208 | * Note that this function acquires a lock which is released when the | |
3209 | * iteration finishes. The caller can't sleep while iteration is in | |
3210 | * progress. | |
3211 | */ | |
72ec7029 TH |
3212 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3213 | struct css_task_iter *it) | |
c6ca5750 | 3214 | __acquires(css_set_lock) |
817929ec PM |
3215 | { |
3216 | /* | |
72ec7029 TH |
3217 | * The first time anyone tries to iterate across a css, we need to |
3218 | * enable the list linking each css_set to its tasks, and fix up | |
3219 | * all existing tasks. | |
817929ec | 3220 | */ |
31a7df01 CW |
3221 | if (!use_task_css_set_links) |
3222 | cgroup_enable_task_cg_lists(); | |
3223 | ||
817929ec | 3224 | read_lock(&css_set_lock); |
c59cd3d8 | 3225 | |
72ec7029 TH |
3226 | it->origin_css = css; |
3227 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 3228 | |
72ec7029 | 3229 | css_advance_task_iter(it); |
817929ec PM |
3230 | } |
3231 | ||
0942eeee | 3232 | /** |
72ec7029 | 3233 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3234 | * @it: the task iterator being iterated |
3235 | * | |
3236 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3237 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3238 | * reaches the end. | |
0942eeee | 3239 | */ |
72ec7029 | 3240 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
3241 | { |
3242 | struct task_struct *res; | |
3243 | struct list_head *l = it->task; | |
69d0206c | 3244 | struct cgrp_cset_link *link; |
817929ec PM |
3245 | |
3246 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 3247 | if (!it->cset_link) |
817929ec PM |
3248 | return NULL; |
3249 | res = list_entry(l, struct task_struct, cg_list); | |
3250 | /* Advance iterator to find next entry */ | |
3251 | l = l->next; | |
69d0206c TH |
3252 | link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link); |
3253 | if (l == &link->cset->tasks) { | |
0942eeee TH |
3254 | /* |
3255 | * We reached the end of this task list - move on to the | |
3256 | * next cgrp_cset_link. | |
3257 | */ | |
72ec7029 | 3258 | css_advance_task_iter(it); |
817929ec PM |
3259 | } else { |
3260 | it->task = l; | |
3261 | } | |
3262 | return res; | |
3263 | } | |
3264 | ||
0942eeee | 3265 | /** |
72ec7029 | 3266 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3267 | * @it: the task iterator to finish |
3268 | * | |
72ec7029 | 3269 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3270 | */ |
72ec7029 | 3271 | void css_task_iter_end(struct css_task_iter *it) |
c6ca5750 | 3272 | __releases(css_set_lock) |
817929ec PM |
3273 | { |
3274 | read_unlock(&css_set_lock); | |
3275 | } | |
3276 | ||
31a7df01 CW |
3277 | static inline int started_after_time(struct task_struct *t1, |
3278 | struct timespec *time, | |
3279 | struct task_struct *t2) | |
3280 | { | |
3281 | int start_diff = timespec_compare(&t1->start_time, time); | |
3282 | if (start_diff > 0) { | |
3283 | return 1; | |
3284 | } else if (start_diff < 0) { | |
3285 | return 0; | |
3286 | } else { | |
3287 | /* | |
3288 | * Arbitrarily, if two processes started at the same | |
3289 | * time, we'll say that the lower pointer value | |
3290 | * started first. Note that t2 may have exited by now | |
3291 | * so this may not be a valid pointer any longer, but | |
3292 | * that's fine - it still serves to distinguish | |
3293 | * between two tasks started (effectively) simultaneously. | |
3294 | */ | |
3295 | return t1 > t2; | |
3296 | } | |
3297 | } | |
3298 | ||
3299 | /* | |
3300 | * This function is a callback from heap_insert() and is used to order | |
3301 | * the heap. | |
3302 | * In this case we order the heap in descending task start time. | |
3303 | */ | |
3304 | static inline int started_after(void *p1, void *p2) | |
3305 | { | |
3306 | struct task_struct *t1 = p1; | |
3307 | struct task_struct *t2 = p2; | |
3308 | return started_after_time(t1, &t2->start_time, t2); | |
3309 | } | |
3310 | ||
3311 | /** | |
72ec7029 TH |
3312 | * css_scan_tasks - iterate though all the tasks in a css |
3313 | * @css: the css to iterate tasks of | |
e535837b TH |
3314 | * @test: optional test callback |
3315 | * @process: process callback | |
3316 | * @data: data passed to @test and @process | |
3317 | * @heap: optional pre-allocated heap used for task iteration | |
31a7df01 | 3318 | * |
72ec7029 TH |
3319 | * Iterate through all the tasks in @css, calling @test for each, and if it |
3320 | * returns %true, call @process for it also. | |
31a7df01 | 3321 | * |
e535837b | 3322 | * @test may be NULL, meaning always true (select all tasks), which |
72ec7029 | 3323 | * effectively duplicates css_task_iter_{start,next,end}() but does not |
e535837b TH |
3324 | * lock css_set_lock for the call to @process. |
3325 | * | |
3326 | * It is guaranteed that @process will act on every task that is a member | |
72ec7029 TH |
3327 | * of @css for the duration of this call. This function may or may not |
3328 | * call @process for tasks that exit or move to a different css during the | |
3329 | * call, or are forked or move into the css during the call. | |
31a7df01 | 3330 | * |
e535837b TH |
3331 | * Note that @test may be called with locks held, and may in some |
3332 | * situations be called multiple times for the same task, so it should be | |
3333 | * cheap. | |
31a7df01 | 3334 | * |
e535837b TH |
3335 | * If @heap is non-NULL, a heap has been pre-allocated and will be used for |
3336 | * heap operations (and its "gt" member will be overwritten), else a | |
3337 | * temporary heap will be used (allocation of which may cause this function | |
3338 | * to fail). | |
31a7df01 | 3339 | */ |
72ec7029 TH |
3340 | int css_scan_tasks(struct cgroup_subsys_state *css, |
3341 | bool (*test)(struct task_struct *, void *), | |
3342 | void (*process)(struct task_struct *, void *), | |
3343 | void *data, struct ptr_heap *heap) | |
31a7df01 CW |
3344 | { |
3345 | int retval, i; | |
72ec7029 | 3346 | struct css_task_iter it; |
31a7df01 CW |
3347 | struct task_struct *p, *dropped; |
3348 | /* Never dereference latest_task, since it's not refcounted */ | |
3349 | struct task_struct *latest_task = NULL; | |
3350 | struct ptr_heap tmp_heap; | |
31a7df01 CW |
3351 | struct timespec latest_time = { 0, 0 }; |
3352 | ||
e535837b | 3353 | if (heap) { |
31a7df01 | 3354 | /* The caller supplied our heap and pre-allocated its memory */ |
31a7df01 CW |
3355 | heap->gt = &started_after; |
3356 | } else { | |
3357 | /* We need to allocate our own heap memory */ | |
3358 | heap = &tmp_heap; | |
3359 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
3360 | if (retval) | |
3361 | /* cannot allocate the heap */ | |
3362 | return retval; | |
3363 | } | |
3364 | ||
3365 | again: | |
3366 | /* | |
72ec7029 | 3367 | * Scan tasks in the css, using the @test callback to determine |
e535837b TH |
3368 | * which are of interest, and invoking @process callback on the |
3369 | * ones which need an update. Since we don't want to hold any | |
3370 | * locks during the task updates, gather tasks to be processed in a | |
3371 | * heap structure. The heap is sorted by descending task start | |
3372 | * time. If the statically-sized heap fills up, we overflow tasks | |
3373 | * that started later, and in future iterations only consider tasks | |
3374 | * that started after the latest task in the previous pass. This | |
31a7df01 CW |
3375 | * guarantees forward progress and that we don't miss any tasks. |
3376 | */ | |
3377 | heap->size = 0; | |
72ec7029 TH |
3378 | css_task_iter_start(css, &it); |
3379 | while ((p = css_task_iter_next(&it))) { | |
31a7df01 CW |
3380 | /* |
3381 | * Only affect tasks that qualify per the caller's callback, | |
3382 | * if he provided one | |
3383 | */ | |
e535837b | 3384 | if (test && !test(p, data)) |
31a7df01 CW |
3385 | continue; |
3386 | /* | |
3387 | * Only process tasks that started after the last task | |
3388 | * we processed | |
3389 | */ | |
3390 | if (!started_after_time(p, &latest_time, latest_task)) | |
3391 | continue; | |
3392 | dropped = heap_insert(heap, p); | |
3393 | if (dropped == NULL) { | |
3394 | /* | |
3395 | * The new task was inserted; the heap wasn't | |
3396 | * previously full | |
3397 | */ | |
3398 | get_task_struct(p); | |
3399 | } else if (dropped != p) { | |
3400 | /* | |
3401 | * The new task was inserted, and pushed out a | |
3402 | * different task | |
3403 | */ | |
3404 | get_task_struct(p); | |
3405 | put_task_struct(dropped); | |
3406 | } | |
3407 | /* | |
3408 | * Else the new task was newer than anything already in | |
3409 | * the heap and wasn't inserted | |
3410 | */ | |
3411 | } | |
72ec7029 | 3412 | css_task_iter_end(&it); |
31a7df01 CW |
3413 | |
3414 | if (heap->size) { | |
3415 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 3416 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 3417 | if (i == 0) { |
4fe91d51 PJ |
3418 | latest_time = q->start_time; |
3419 | latest_task = q; | |
31a7df01 CW |
3420 | } |
3421 | /* Process the task per the caller's callback */ | |
e535837b | 3422 | process(q, data); |
4fe91d51 | 3423 | put_task_struct(q); |
31a7df01 CW |
3424 | } |
3425 | /* | |
3426 | * If we had to process any tasks at all, scan again | |
3427 | * in case some of them were in the middle of forking | |
3428 | * children that didn't get processed. | |
3429 | * Not the most efficient way to do it, but it avoids | |
3430 | * having to take callback_mutex in the fork path | |
3431 | */ | |
3432 | goto again; | |
3433 | } | |
3434 | if (heap == &tmp_heap) | |
3435 | heap_free(&tmp_heap); | |
3436 | return 0; | |
3437 | } | |
3438 | ||
e535837b | 3439 | static void cgroup_transfer_one_task(struct task_struct *task, void *data) |
8cc99345 | 3440 | { |
e535837b | 3441 | struct cgroup *new_cgroup = data; |
8cc99345 | 3442 | |
47cfcd09 | 3443 | mutex_lock(&cgroup_mutex); |
8cc99345 | 3444 | cgroup_attach_task(new_cgroup, task, false); |
47cfcd09 | 3445 | mutex_unlock(&cgroup_mutex); |
8cc99345 TH |
3446 | } |
3447 | ||
3448 | /** | |
3449 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
3450 | * @to: cgroup to which the tasks will be moved | |
3451 | * @from: cgroup in which the tasks currently reside | |
3452 | */ | |
3453 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
3454 | { | |
72ec7029 TH |
3455 | return css_scan_tasks(&from->dummy_css, NULL, cgroup_transfer_one_task, |
3456 | to, NULL); | |
8cc99345 TH |
3457 | } |
3458 | ||
bbcb81d0 | 3459 | /* |
102a775e | 3460 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3461 | * |
3462 | * Reading this file can return large amounts of data if a cgroup has | |
3463 | * *lots* of attached tasks. So it may need several calls to read(), | |
3464 | * but we cannot guarantee that the information we produce is correct | |
3465 | * unless we produce it entirely atomically. | |
3466 | * | |
bbcb81d0 | 3467 | */ |
bbcb81d0 | 3468 | |
24528255 LZ |
3469 | /* which pidlist file are we talking about? */ |
3470 | enum cgroup_filetype { | |
3471 | CGROUP_FILE_PROCS, | |
3472 | CGROUP_FILE_TASKS, | |
3473 | }; | |
3474 | ||
3475 | /* | |
3476 | * A pidlist is a list of pids that virtually represents the contents of one | |
3477 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3478 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3479 | * to the cgroup. | |
3480 | */ | |
3481 | struct cgroup_pidlist { | |
3482 | /* | |
3483 | * used to find which pidlist is wanted. doesn't change as long as | |
3484 | * this particular list stays in the list. | |
3485 | */ | |
3486 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3487 | /* array of xids */ | |
3488 | pid_t *list; | |
3489 | /* how many elements the above list has */ | |
3490 | int length; | |
3491 | /* how many files are using the current array */ | |
3492 | int use_count; | |
3493 | /* each of these stored in a list by its cgroup */ | |
3494 | struct list_head links; | |
3495 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3496 | struct cgroup *owner; | |
3497 | /* protects the other fields */ | |
b395890a | 3498 | struct rw_semaphore rwsem; |
24528255 LZ |
3499 | }; |
3500 | ||
d1d9fd33 BB |
3501 | /* |
3502 | * The following two functions "fix" the issue where there are more pids | |
3503 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3504 | * TODO: replace with a kernel-wide solution to this problem | |
3505 | */ | |
3506 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3507 | static void *pidlist_allocate(int count) | |
3508 | { | |
3509 | if (PIDLIST_TOO_LARGE(count)) | |
3510 | return vmalloc(count * sizeof(pid_t)); | |
3511 | else | |
3512 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3513 | } | |
3514 | static void pidlist_free(void *p) | |
3515 | { | |
3516 | if (is_vmalloc_addr(p)) | |
3517 | vfree(p); | |
3518 | else | |
3519 | kfree(p); | |
3520 | } | |
d1d9fd33 | 3521 | |
bbcb81d0 | 3522 | /* |
102a775e | 3523 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3524 | * Returns the number of unique elements. |
bbcb81d0 | 3525 | */ |
6ee211ad | 3526 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3527 | { |
102a775e | 3528 | int src, dest = 1; |
102a775e BB |
3529 | |
3530 | /* | |
3531 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3532 | * edge cases first; no work needs to be done for either | |
3533 | */ | |
3534 | if (length == 0 || length == 1) | |
3535 | return length; | |
3536 | /* src and dest walk down the list; dest counts unique elements */ | |
3537 | for (src = 1; src < length; src++) { | |
3538 | /* find next unique element */ | |
3539 | while (list[src] == list[src-1]) { | |
3540 | src++; | |
3541 | if (src == length) | |
3542 | goto after; | |
3543 | } | |
3544 | /* dest always points to where the next unique element goes */ | |
3545 | list[dest] = list[src]; | |
3546 | dest++; | |
3547 | } | |
3548 | after: | |
102a775e BB |
3549 | return dest; |
3550 | } | |
3551 | ||
3552 | static int cmppid(const void *a, const void *b) | |
3553 | { | |
3554 | return *(pid_t *)a - *(pid_t *)b; | |
3555 | } | |
3556 | ||
72a8cb30 BB |
3557 | /* |
3558 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3559 | * returns with the lock on that pidlist already held, and takes care | |
3560 | * of the use count, or returns NULL with no locks held if we're out of | |
3561 | * memory. | |
3562 | */ | |
3563 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, | |
3564 | enum cgroup_filetype type) | |
3565 | { | |
3566 | struct cgroup_pidlist *l; | |
3567 | /* don't need task_nsproxy() if we're looking at ourself */ | |
17cf22c3 | 3568 | struct pid_namespace *ns = task_active_pid_ns(current); |
b70cc5fd | 3569 | |
72a8cb30 | 3570 | /* |
b395890a | 3571 | * We can't drop the pidlist_mutex before taking the l->rwsem in case |
72a8cb30 BB |
3572 | * the last ref-holder is trying to remove l from the list at the same |
3573 | * time. Holding the pidlist_mutex precludes somebody taking whichever | |
3574 | * list we find out from under us - compare release_pid_array(). | |
3575 | */ | |
3576 | mutex_lock(&cgrp->pidlist_mutex); | |
3577 | list_for_each_entry(l, &cgrp->pidlists, links) { | |
3578 | if (l->key.type == type && l->key.ns == ns) { | |
72a8cb30 | 3579 | /* make sure l doesn't vanish out from under us */ |
b395890a | 3580 | down_write(&l->rwsem); |
72a8cb30 | 3581 | mutex_unlock(&cgrp->pidlist_mutex); |
72a8cb30 BB |
3582 | return l; |
3583 | } | |
3584 | } | |
3585 | /* entry not found; create a new one */ | |
f4f4be2b | 3586 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
72a8cb30 BB |
3587 | if (!l) { |
3588 | mutex_unlock(&cgrp->pidlist_mutex); | |
72a8cb30 BB |
3589 | return l; |
3590 | } | |
b395890a LZ |
3591 | init_rwsem(&l->rwsem); |
3592 | down_write(&l->rwsem); | |
72a8cb30 | 3593 | l->key.type = type; |
b70cc5fd | 3594 | l->key.ns = get_pid_ns(ns); |
72a8cb30 BB |
3595 | l->owner = cgrp; |
3596 | list_add(&l->links, &cgrp->pidlists); | |
3597 | mutex_unlock(&cgrp->pidlist_mutex); | |
3598 | return l; | |
3599 | } | |
3600 | ||
102a775e BB |
3601 | /* |
3602 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3603 | */ | |
72a8cb30 BB |
3604 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3605 | struct cgroup_pidlist **lp) | |
102a775e BB |
3606 | { |
3607 | pid_t *array; | |
3608 | int length; | |
3609 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3610 | struct css_task_iter it; |
817929ec | 3611 | struct task_struct *tsk; |
102a775e BB |
3612 | struct cgroup_pidlist *l; |
3613 | ||
3614 | /* | |
3615 | * If cgroup gets more users after we read count, we won't have | |
3616 | * enough space - tough. This race is indistinguishable to the | |
3617 | * caller from the case that the additional cgroup users didn't | |
3618 | * show up until sometime later on. | |
3619 | */ | |
3620 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3621 | array = pidlist_allocate(length); |
102a775e BB |
3622 | if (!array) |
3623 | return -ENOMEM; | |
3624 | /* now, populate the array */ | |
72ec7029 TH |
3625 | css_task_iter_start(&cgrp->dummy_css, &it); |
3626 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3627 | if (unlikely(n == length)) |
817929ec | 3628 | break; |
102a775e | 3629 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3630 | if (type == CGROUP_FILE_PROCS) |
3631 | pid = task_tgid_vnr(tsk); | |
3632 | else | |
3633 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3634 | if (pid > 0) /* make sure to only use valid results */ |
3635 | array[n++] = pid; | |
817929ec | 3636 | } |
72ec7029 | 3637 | css_task_iter_end(&it); |
102a775e BB |
3638 | length = n; |
3639 | /* now sort & (if procs) strip out duplicates */ | |
3640 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3641 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3642 | length = pidlist_uniq(array, length); |
72a8cb30 BB |
3643 | l = cgroup_pidlist_find(cgrp, type); |
3644 | if (!l) { | |
d1d9fd33 | 3645 | pidlist_free(array); |
72a8cb30 | 3646 | return -ENOMEM; |
102a775e | 3647 | } |
72a8cb30 | 3648 | /* store array, freeing old if necessary - lock already held */ |
d1d9fd33 | 3649 | pidlist_free(l->list); |
102a775e BB |
3650 | l->list = array; |
3651 | l->length = length; | |
3652 | l->use_count++; | |
b395890a | 3653 | up_write(&l->rwsem); |
72a8cb30 | 3654 | *lp = l; |
102a775e | 3655 | return 0; |
bbcb81d0 PM |
3656 | } |
3657 | ||
846c7bb0 | 3658 | /** |
a043e3b2 | 3659 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3660 | * @stats: cgroupstats to fill information into |
3661 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3662 | * been requested. | |
a043e3b2 LZ |
3663 | * |
3664 | * Build and fill cgroupstats so that taskstats can export it to user | |
3665 | * space. | |
846c7bb0 BS |
3666 | */ |
3667 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3668 | { | |
3669 | int ret = -EINVAL; | |
bd89aabc | 3670 | struct cgroup *cgrp; |
72ec7029 | 3671 | struct css_task_iter it; |
846c7bb0 | 3672 | struct task_struct *tsk; |
33d283be | 3673 | |
846c7bb0 | 3674 | /* |
33d283be LZ |
3675 | * Validate dentry by checking the superblock operations, |
3676 | * and make sure it's a directory. | |
846c7bb0 | 3677 | */ |
33d283be LZ |
3678 | if (dentry->d_sb->s_op != &cgroup_ops || |
3679 | !S_ISDIR(dentry->d_inode->i_mode)) | |
846c7bb0 BS |
3680 | goto err; |
3681 | ||
3682 | ret = 0; | |
bd89aabc | 3683 | cgrp = dentry->d_fsdata; |
846c7bb0 | 3684 | |
72ec7029 TH |
3685 | css_task_iter_start(&cgrp->dummy_css, &it); |
3686 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3687 | switch (tsk->state) { |
3688 | case TASK_RUNNING: | |
3689 | stats->nr_running++; | |
3690 | break; | |
3691 | case TASK_INTERRUPTIBLE: | |
3692 | stats->nr_sleeping++; | |
3693 | break; | |
3694 | case TASK_UNINTERRUPTIBLE: | |
3695 | stats->nr_uninterruptible++; | |
3696 | break; | |
3697 | case TASK_STOPPED: | |
3698 | stats->nr_stopped++; | |
3699 | break; | |
3700 | default: | |
3701 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3702 | stats->nr_io_wait++; | |
3703 | break; | |
3704 | } | |
3705 | } | |
72ec7029 | 3706 | css_task_iter_end(&it); |
846c7bb0 | 3707 | |
846c7bb0 BS |
3708 | err: |
3709 | return ret; | |
3710 | } | |
3711 | ||
8f3ff208 | 3712 | |
bbcb81d0 | 3713 | /* |
102a775e | 3714 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3715 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3716 | * in the cgroup->l->list array. |
bbcb81d0 | 3717 | */ |
cc31edce | 3718 | |
102a775e | 3719 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3720 | { |
cc31edce PM |
3721 | /* |
3722 | * Initially we receive a position value that corresponds to | |
3723 | * one more than the last pid shown (or 0 on the first call or | |
3724 | * after a seek to the start). Use a binary-search to find the | |
3725 | * next pid to display, if any | |
3726 | */ | |
102a775e | 3727 | struct cgroup_pidlist *l = s->private; |
cc31edce PM |
3728 | int index = 0, pid = *pos; |
3729 | int *iter; | |
3730 | ||
b395890a | 3731 | down_read(&l->rwsem); |
cc31edce | 3732 | if (pid) { |
102a775e | 3733 | int end = l->length; |
20777766 | 3734 | |
cc31edce PM |
3735 | while (index < end) { |
3736 | int mid = (index + end) / 2; | |
102a775e | 3737 | if (l->list[mid] == pid) { |
cc31edce PM |
3738 | index = mid; |
3739 | break; | |
102a775e | 3740 | } else if (l->list[mid] <= pid) |
cc31edce PM |
3741 | index = mid + 1; |
3742 | else | |
3743 | end = mid; | |
3744 | } | |
3745 | } | |
3746 | /* If we're off the end of the array, we're done */ | |
102a775e | 3747 | if (index >= l->length) |
cc31edce PM |
3748 | return NULL; |
3749 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3750 | iter = l->list + index; |
cc31edce PM |
3751 | *pos = *iter; |
3752 | return iter; | |
3753 | } | |
3754 | ||
102a775e | 3755 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3756 | { |
102a775e | 3757 | struct cgroup_pidlist *l = s->private; |
b395890a | 3758 | up_read(&l->rwsem); |
cc31edce PM |
3759 | } |
3760 | ||
102a775e | 3761 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3762 | { |
102a775e BB |
3763 | struct cgroup_pidlist *l = s->private; |
3764 | pid_t *p = v; | |
3765 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3766 | /* |
3767 | * Advance to the next pid in the array. If this goes off the | |
3768 | * end, we're done | |
3769 | */ | |
3770 | p++; | |
3771 | if (p >= end) { | |
3772 | return NULL; | |
3773 | } else { | |
3774 | *pos = *p; | |
3775 | return p; | |
3776 | } | |
3777 | } | |
3778 | ||
102a775e | 3779 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3780 | { |
3781 | return seq_printf(s, "%d\n", *(int *)v); | |
3782 | } | |
bbcb81d0 | 3783 | |
102a775e BB |
3784 | /* |
3785 | * seq_operations functions for iterating on pidlists through seq_file - | |
3786 | * independent of whether it's tasks or procs | |
3787 | */ | |
3788 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3789 | .start = cgroup_pidlist_start, | |
3790 | .stop = cgroup_pidlist_stop, | |
3791 | .next = cgroup_pidlist_next, | |
3792 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3793 | }; |
3794 | ||
102a775e | 3795 | static void cgroup_release_pid_array(struct cgroup_pidlist *l) |
cc31edce | 3796 | { |
72a8cb30 BB |
3797 | /* |
3798 | * the case where we're the last user of this particular pidlist will | |
3799 | * have us remove it from the cgroup's list, which entails taking the | |
3800 | * mutex. since in pidlist_find the pidlist->lock depends on cgroup-> | |
3801 | * pidlist_mutex, we have to take pidlist_mutex first. | |
3802 | */ | |
3803 | mutex_lock(&l->owner->pidlist_mutex); | |
b395890a | 3804 | down_write(&l->rwsem); |
102a775e BB |
3805 | BUG_ON(!l->use_count); |
3806 | if (!--l->use_count) { | |
72a8cb30 BB |
3807 | /* we're the last user if refcount is 0; remove and free */ |
3808 | list_del(&l->links); | |
3809 | mutex_unlock(&l->owner->pidlist_mutex); | |
d1d9fd33 | 3810 | pidlist_free(l->list); |
72a8cb30 | 3811 | put_pid_ns(l->key.ns); |
b395890a | 3812 | up_write(&l->rwsem); |
72a8cb30 BB |
3813 | kfree(l); |
3814 | return; | |
cc31edce | 3815 | } |
72a8cb30 | 3816 | mutex_unlock(&l->owner->pidlist_mutex); |
b395890a | 3817 | up_write(&l->rwsem); |
bbcb81d0 PM |
3818 | } |
3819 | ||
102a775e | 3820 | static int cgroup_pidlist_release(struct inode *inode, struct file *file) |
cc31edce | 3821 | { |
102a775e | 3822 | struct cgroup_pidlist *l; |
cc31edce PM |
3823 | if (!(file->f_mode & FMODE_READ)) |
3824 | return 0; | |
102a775e BB |
3825 | /* |
3826 | * the seq_file will only be initialized if the file was opened for | |
3827 | * reading; hence we check if it's not null only in that case. | |
3828 | */ | |
3829 | l = ((struct seq_file *)file->private_data)->private; | |
3830 | cgroup_release_pid_array(l); | |
cc31edce PM |
3831 | return seq_release(inode, file); |
3832 | } | |
3833 | ||
102a775e | 3834 | static const struct file_operations cgroup_pidlist_operations = { |
cc31edce PM |
3835 | .read = seq_read, |
3836 | .llseek = seq_lseek, | |
3837 | .write = cgroup_file_write, | |
102a775e | 3838 | .release = cgroup_pidlist_release, |
cc31edce PM |
3839 | }; |
3840 | ||
bbcb81d0 | 3841 | /* |
102a775e BB |
3842 | * The following functions handle opens on a file that displays a pidlist |
3843 | * (tasks or procs). Prepare an array of the process/thread IDs of whoever's | |
3844 | * in the cgroup. | |
bbcb81d0 | 3845 | */ |
102a775e | 3846 | /* helper function for the two below it */ |
72a8cb30 | 3847 | static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type) |
bbcb81d0 | 3848 | { |
bd89aabc | 3849 | struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent); |
72a8cb30 | 3850 | struct cgroup_pidlist *l; |
cc31edce | 3851 | int retval; |
bbcb81d0 | 3852 | |
cc31edce | 3853 | /* Nothing to do for write-only files */ |
bbcb81d0 PM |
3854 | if (!(file->f_mode & FMODE_READ)) |
3855 | return 0; | |
3856 | ||
102a775e | 3857 | /* have the array populated */ |
72a8cb30 | 3858 | retval = pidlist_array_load(cgrp, type, &l); |
102a775e BB |
3859 | if (retval) |
3860 | return retval; | |
3861 | /* configure file information */ | |
3862 | file->f_op = &cgroup_pidlist_operations; | |
cc31edce | 3863 | |
102a775e | 3864 | retval = seq_open(file, &cgroup_pidlist_seq_operations); |
cc31edce | 3865 | if (retval) { |
102a775e | 3866 | cgroup_release_pid_array(l); |
cc31edce | 3867 | return retval; |
bbcb81d0 | 3868 | } |
102a775e | 3869 | ((struct seq_file *)file->private_data)->private = l; |
bbcb81d0 PM |
3870 | return 0; |
3871 | } | |
102a775e BB |
3872 | static int cgroup_tasks_open(struct inode *unused, struct file *file) |
3873 | { | |
72a8cb30 | 3874 | return cgroup_pidlist_open(file, CGROUP_FILE_TASKS); |
102a775e BB |
3875 | } |
3876 | static int cgroup_procs_open(struct inode *unused, struct file *file) | |
3877 | { | |
72a8cb30 | 3878 | return cgroup_pidlist_open(file, CGROUP_FILE_PROCS); |
102a775e | 3879 | } |
bbcb81d0 | 3880 | |
182446d0 TH |
3881 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3882 | struct cftype *cft) | |
81a6a5cd | 3883 | { |
182446d0 | 3884 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3885 | } |
3886 | ||
182446d0 TH |
3887 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3888 | struct cftype *cft, u64 val) | |
6379c106 | 3889 | { |
182446d0 | 3890 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3891 | if (val) |
182446d0 | 3892 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3893 | else |
182446d0 | 3894 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3895 | return 0; |
3896 | } | |
3897 | ||
1c8158ee LZ |
3898 | /* |
3899 | * When dput() is called asynchronously, if umount has been done and | |
3900 | * then deactivate_super() in cgroup_free_fn() kills the superblock, | |
3901 | * there's a small window that vfs will see the root dentry with non-zero | |
3902 | * refcnt and trigger BUG(). | |
3903 | * | |
3904 | * That's why we hold a reference before dput() and drop it right after. | |
3905 | */ | |
3906 | static void cgroup_dput(struct cgroup *cgrp) | |
3907 | { | |
3908 | struct super_block *sb = cgrp->root->sb; | |
3909 | ||
3910 | atomic_inc(&sb->s_active); | |
3911 | dput(cgrp->dentry); | |
3912 | deactivate_super(sb); | |
3913 | } | |
3914 | ||
0dea1168 KS |
3915 | /* |
3916 | * Unregister event and free resources. | |
3917 | * | |
3918 | * Gets called from workqueue. | |
3919 | */ | |
3920 | static void cgroup_event_remove(struct work_struct *work) | |
3921 | { | |
3922 | struct cgroup_event *event = container_of(work, struct cgroup_event, | |
3923 | remove); | |
81eeaf04 | 3924 | struct cgroup_subsys_state *css = event->css; |
0dea1168 | 3925 | |
810cbee4 LZ |
3926 | remove_wait_queue(event->wqh, &event->wait); |
3927 | ||
81eeaf04 | 3928 | event->cft->unregister_event(css, event->cft, event->eventfd); |
0dea1168 | 3929 | |
810cbee4 LZ |
3930 | /* Notify userspace the event is going away. */ |
3931 | eventfd_signal(event->eventfd, 1); | |
3932 | ||
0dea1168 | 3933 | eventfd_ctx_put(event->eventfd); |
0dea1168 | 3934 | kfree(event); |
7941cb02 | 3935 | css_put(css); |
0dea1168 KS |
3936 | } |
3937 | ||
3938 | /* | |
3939 | * Gets called on POLLHUP on eventfd when user closes it. | |
3940 | * | |
3941 | * Called with wqh->lock held and interrupts disabled. | |
3942 | */ | |
3943 | static int cgroup_event_wake(wait_queue_t *wait, unsigned mode, | |
3944 | int sync, void *key) | |
3945 | { | |
3946 | struct cgroup_event *event = container_of(wait, | |
3947 | struct cgroup_event, wait); | |
81eeaf04 | 3948 | struct cgroup *cgrp = event->css->cgroup; |
0dea1168 KS |
3949 | unsigned long flags = (unsigned long)key; |
3950 | ||
3951 | if (flags & POLLHUP) { | |
0dea1168 | 3952 | /* |
810cbee4 LZ |
3953 | * If the event has been detached at cgroup removal, we |
3954 | * can simply return knowing the other side will cleanup | |
3955 | * for us. | |
3956 | * | |
3957 | * We can't race against event freeing since the other | |
3958 | * side will require wqh->lock via remove_wait_queue(), | |
3959 | * which we hold. | |
0dea1168 | 3960 | */ |
810cbee4 LZ |
3961 | spin_lock(&cgrp->event_list_lock); |
3962 | if (!list_empty(&event->list)) { | |
3963 | list_del_init(&event->list); | |
3964 | /* | |
3965 | * We are in atomic context, but cgroup_event_remove() | |
3966 | * may sleep, so we have to call it in workqueue. | |
3967 | */ | |
3968 | schedule_work(&event->remove); | |
3969 | } | |
3970 | spin_unlock(&cgrp->event_list_lock); | |
0dea1168 KS |
3971 | } |
3972 | ||
3973 | return 0; | |
3974 | } | |
3975 | ||
3976 | static void cgroup_event_ptable_queue_proc(struct file *file, | |
3977 | wait_queue_head_t *wqh, poll_table *pt) | |
3978 | { | |
3979 | struct cgroup_event *event = container_of(pt, | |
3980 | struct cgroup_event, pt); | |
3981 | ||
3982 | event->wqh = wqh; | |
3983 | add_wait_queue(wqh, &event->wait); | |
3984 | } | |
3985 | ||
3986 | /* | |
3987 | * Parse input and register new cgroup event handler. | |
3988 | * | |
3989 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3990 | * Interpretation of args is defined by control file implementation. | |
3991 | */ | |
6e6eab0e | 3992 | static int cgroup_write_event_control(struct cgroup_subsys_state *dummy_css, |
182446d0 | 3993 | struct cftype *cft, const char *buffer) |
0dea1168 | 3994 | { |
6e6eab0e | 3995 | struct cgroup *cgrp = dummy_css->cgroup; |
876ede8b | 3996 | struct cgroup_event *event; |
7c918cbb | 3997 | struct cgroup_subsys_state *cfile_css; |
0dea1168 | 3998 | unsigned int efd, cfd; |
4e10f3c9 AV |
3999 | struct fd efile; |
4000 | struct fd cfile; | |
0dea1168 KS |
4001 | char *endp; |
4002 | int ret; | |
4003 | ||
4004 | efd = simple_strtoul(buffer, &endp, 10); | |
4005 | if (*endp != ' ') | |
4006 | return -EINVAL; | |
4007 | buffer = endp + 1; | |
4008 | ||
4009 | cfd = simple_strtoul(buffer, &endp, 10); | |
4010 | if ((*endp != ' ') && (*endp != '\0')) | |
4011 | return -EINVAL; | |
4012 | buffer = endp + 1; | |
4013 | ||
4014 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4015 | if (!event) | |
4016 | return -ENOMEM; | |
6e6eab0e | 4017 | |
0dea1168 KS |
4018 | INIT_LIST_HEAD(&event->list); |
4019 | init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc); | |
4020 | init_waitqueue_func_entry(&event->wait, cgroup_event_wake); | |
4021 | INIT_WORK(&event->remove, cgroup_event_remove); | |
4022 | ||
4e10f3c9 AV |
4023 | efile = fdget(efd); |
4024 | if (!efile.file) { | |
4025 | ret = -EBADF; | |
876ede8b | 4026 | goto out_kfree; |
0dea1168 KS |
4027 | } |
4028 | ||
4e10f3c9 | 4029 | event->eventfd = eventfd_ctx_fileget(efile.file); |
0dea1168 KS |
4030 | if (IS_ERR(event->eventfd)) { |
4031 | ret = PTR_ERR(event->eventfd); | |
876ede8b | 4032 | goto out_put_efile; |
0dea1168 KS |
4033 | } |
4034 | ||
4e10f3c9 AV |
4035 | cfile = fdget(cfd); |
4036 | if (!cfile.file) { | |
0dea1168 | 4037 | ret = -EBADF; |
876ede8b | 4038 | goto out_put_eventfd; |
0dea1168 KS |
4039 | } |
4040 | ||
4041 | /* the process need read permission on control file */ | |
3bfa784a | 4042 | /* AV: shouldn't we check that it's been opened for read instead? */ |
4e10f3c9 | 4043 | ret = inode_permission(file_inode(cfile.file), MAY_READ); |
0dea1168 | 4044 | if (ret < 0) |
876ede8b | 4045 | goto out_put_cfile; |
0dea1168 | 4046 | |
4e10f3c9 | 4047 | event->cft = __file_cft(cfile.file); |
0dea1168 KS |
4048 | if (IS_ERR(event->cft)) { |
4049 | ret = PTR_ERR(event->cft); | |
876ede8b | 4050 | goto out_put_cfile; |
0dea1168 KS |
4051 | } |
4052 | ||
6e6eab0e TH |
4053 | if (!event->cft->ss) { |
4054 | ret = -EBADF; | |
4055 | goto out_put_cfile; | |
0dea1168 KS |
4056 | } |
4057 | ||
f169007b | 4058 | /* |
7c918cbb TH |
4059 | * Determine the css of @cfile, verify it belongs to the same |
4060 | * cgroup as cgroup.event_control, and associate @event with it. | |
7941cb02 TH |
4061 | * Remaining events are automatically removed on cgroup destruction |
4062 | * but the removal is asynchronous, so take an extra ref. | |
f169007b | 4063 | */ |
6e6eab0e TH |
4064 | rcu_read_lock(); |
4065 | ||
4066 | ret = -EINVAL; | |
ca8bdcaf | 4067 | event->css = cgroup_css(cgrp, event->cft->ss); |
4e10f3c9 | 4068 | cfile_css = css_from_dir(cfile.file->f_dentry->d_parent, event->cft->ss); |
7c918cbb | 4069 | if (event->css && event->css == cfile_css && css_tryget(event->css)) |
6e6eab0e TH |
4070 | ret = 0; |
4071 | ||
4072 | rcu_read_unlock(); | |
4073 | if (ret) | |
4074 | goto out_put_cfile; | |
f169007b | 4075 | |
0dea1168 KS |
4076 | if (!event->cft->register_event || !event->cft->unregister_event) { |
4077 | ret = -EINVAL; | |
7941cb02 | 4078 | goto out_put_css; |
0dea1168 KS |
4079 | } |
4080 | ||
6e6eab0e | 4081 | ret = event->cft->register_event(event->css, event->cft, |
0dea1168 KS |
4082 | event->eventfd, buffer); |
4083 | if (ret) | |
7941cb02 | 4084 | goto out_put_css; |
0dea1168 | 4085 | |
4e10f3c9 | 4086 | efile.file->f_op->poll(efile.file, &event->pt); |
0dea1168 KS |
4087 | |
4088 | spin_lock(&cgrp->event_list_lock); | |
4089 | list_add(&event->list, &cgrp->event_list); | |
4090 | spin_unlock(&cgrp->event_list_lock); | |
4091 | ||
4e10f3c9 AV |
4092 | fdput(cfile); |
4093 | fdput(efile); | |
0dea1168 KS |
4094 | |
4095 | return 0; | |
4096 | ||
7941cb02 TH |
4097 | out_put_css: |
4098 | css_put(event->css); | |
876ede8b | 4099 | out_put_cfile: |
4e10f3c9 | 4100 | fdput(cfile); |
876ede8b LZ |
4101 | out_put_eventfd: |
4102 | eventfd_ctx_put(event->eventfd); | |
4103 | out_put_efile: | |
4e10f3c9 | 4104 | fdput(efile); |
876ede8b | 4105 | out_kfree: |
0dea1168 KS |
4106 | kfree(event); |
4107 | ||
4108 | return ret; | |
4109 | } | |
4110 | ||
182446d0 TH |
4111 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4112 | struct cftype *cft) | |
97978e6d | 4113 | { |
182446d0 | 4114 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4115 | } |
4116 | ||
182446d0 TH |
4117 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4118 | struct cftype *cft, u64 val) | |
97978e6d DL |
4119 | { |
4120 | if (val) | |
182446d0 | 4121 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4122 | else |
182446d0 | 4123 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4124 | return 0; |
4125 | } | |
4126 | ||
d5c56ced | 4127 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 4128 | { |
d5c56ced | 4129 | .name = "cgroup.procs", |
102a775e | 4130 | .open = cgroup_procs_open, |
74a1166d | 4131 | .write_u64 = cgroup_procs_write, |
102a775e | 4132 | .release = cgroup_pidlist_release, |
74a1166d | 4133 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 4134 | }, |
81a6a5cd | 4135 | { |
d5c56ced | 4136 | .name = "cgroup.event_control", |
0dea1168 KS |
4137 | .write_string = cgroup_write_event_control, |
4138 | .mode = S_IWUGO, | |
4139 | }, | |
97978e6d DL |
4140 | { |
4141 | .name = "cgroup.clone_children", | |
873fe09e | 4142 | .flags = CFTYPE_INSANE, |
97978e6d DL |
4143 | .read_u64 = cgroup_clone_children_read, |
4144 | .write_u64 = cgroup_clone_children_write, | |
4145 | }, | |
873fe09e TH |
4146 | { |
4147 | .name = "cgroup.sane_behavior", | |
4148 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4149 | .read_seq_string = cgroup_sane_behavior_show, | |
4150 | }, | |
d5c56ced TH |
4151 | |
4152 | /* | |
4153 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
4154 | * don't exist if sane_behavior. If you're depending on these, be | |
4155 | * prepared to be burned. | |
4156 | */ | |
4157 | { | |
4158 | .name = "tasks", | |
4159 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
4160 | .open = cgroup_tasks_open, | |
4161 | .write_u64 = cgroup_tasks_write, | |
4162 | .release = cgroup_pidlist_release, | |
4163 | .mode = S_IRUGO | S_IWUSR, | |
4164 | }, | |
4165 | { | |
4166 | .name = "notify_on_release", | |
4167 | .flags = CFTYPE_INSANE, | |
4168 | .read_u64 = cgroup_read_notify_on_release, | |
4169 | .write_u64 = cgroup_write_notify_on_release, | |
4170 | }, | |
6e6ff25b TH |
4171 | { |
4172 | .name = "release_agent", | |
cc5943a7 | 4173 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
6e6ff25b TH |
4174 | .read_seq_string = cgroup_release_agent_show, |
4175 | .write_string = cgroup_release_agent_write, | |
4176 | .max_write_len = PATH_MAX, | |
4177 | }, | |
db0416b6 | 4178 | { } /* terminate */ |
bbcb81d0 PM |
4179 | }; |
4180 | ||
13af07df | 4181 | /** |
628f7cd4 | 4182 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 4183 | * @cgrp: target cgroup |
13af07df | 4184 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
4185 | * |
4186 | * On failure, no file is added. | |
13af07df | 4187 | */ |
628f7cd4 | 4188 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 4189 | { |
ddbcc7e8 | 4190 | struct cgroup_subsys *ss; |
b420ba7d | 4191 | int i, ret = 0; |
bbcb81d0 | 4192 | |
8e3f6541 | 4193 | /* process cftsets of each subsystem */ |
b420ba7d | 4194 | for_each_subsys(ss, i) { |
8e3f6541 | 4195 | struct cftype_set *set; |
b420ba7d TH |
4196 | |
4197 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 4198 | continue; |
8e3f6541 | 4199 | |
bee55099 | 4200 | list_for_each_entry(set, &ss->cftsets, node) { |
2bb566cb | 4201 | ret = cgroup_addrm_files(cgrp, set->cfts, true); |
bee55099 TH |
4202 | if (ret < 0) |
4203 | goto err; | |
4204 | } | |
ddbcc7e8 | 4205 | } |
ddbcc7e8 | 4206 | return 0; |
bee55099 TH |
4207 | err: |
4208 | cgroup_clear_dir(cgrp, subsys_mask); | |
4209 | return ret; | |
ddbcc7e8 PM |
4210 | } |
4211 | ||
0c21ead1 TH |
4212 | /* |
4213 | * css destruction is four-stage process. | |
4214 | * | |
4215 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4216 | * Implemented in kill_css(). | |
4217 | * | |
4218 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
4219 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
4220 | * by invoking offline_css(). After offlining, the base ref is put. | |
4221 | * Implemented in css_killed_work_fn(). | |
4222 | * | |
4223 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4224 | * accessors are inside RCU read sections. css_release() schedules the | |
4225 | * RCU callback. | |
4226 | * | |
4227 | * 4. After the grace period, the css can be freed. Implemented in | |
4228 | * css_free_work_fn(). | |
4229 | * | |
4230 | * It is actually hairier because both step 2 and 4 require process context | |
4231 | * and thus involve punting to css->destroy_work adding two additional | |
4232 | * steps to the already complex sequence. | |
4233 | */ | |
35ef10da | 4234 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4235 | { |
4236 | struct cgroup_subsys_state *css = | |
35ef10da | 4237 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 4238 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4239 | |
0ae78e0b TH |
4240 | if (css->parent) |
4241 | css_put(css->parent); | |
4242 | ||
0c21ead1 TH |
4243 | css->ss->css_free(css); |
4244 | cgroup_dput(cgrp); | |
48ddbe19 TH |
4245 | } |
4246 | ||
0c21ead1 | 4247 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4248 | { |
4249 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4250 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4251 | |
35ef10da TH |
4252 | /* |
4253 | * css holds an extra ref to @cgrp->dentry which is put on the last | |
0c21ead1 | 4254 | * css_put(). dput() requires process context which we don't have. |
35ef10da TH |
4255 | */ |
4256 | INIT_WORK(&css->destroy_work, css_free_work_fn); | |
4257 | schedule_work(&css->destroy_work); | |
48ddbe19 TH |
4258 | } |
4259 | ||
d3daf28d TH |
4260 | static void css_release(struct percpu_ref *ref) |
4261 | { | |
4262 | struct cgroup_subsys_state *css = | |
4263 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4264 | ||
0c21ead1 | 4265 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4266 | } |
4267 | ||
623f926b TH |
4268 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
4269 | struct cgroup *cgrp) | |
ddbcc7e8 | 4270 | { |
bd89aabc | 4271 | css->cgroup = cgrp; |
72c97e54 | 4272 | css->ss = ss; |
ddbcc7e8 | 4273 | css->flags = 0; |
0ae78e0b TH |
4274 | |
4275 | if (cgrp->parent) | |
ca8bdcaf | 4276 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 4277 | else |
38b53aba | 4278 | css->flags |= CSS_ROOT; |
48ddbe19 | 4279 | |
ca8bdcaf | 4280 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4281 | } |
4282 | ||
2a4ac633 | 4283 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4284 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4285 | { |
623f926b | 4286 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4287 | int ret = 0; |
4288 | ||
a31f2d3f TH |
4289 | lockdep_assert_held(&cgroup_mutex); |
4290 | ||
92fb9748 | 4291 | if (ss->css_online) |
eb95419b | 4292 | ret = ss->css_online(css); |
ae7f164a | 4293 | if (!ret) { |
eb95419b | 4294 | css->flags |= CSS_ONLINE; |
f20104de | 4295 | css->cgroup->nr_css++; |
ae7f164a TH |
4296 | rcu_assign_pointer(css->cgroup->subsys[ss->subsys_id], css); |
4297 | } | |
b1929db4 | 4298 | return ret; |
a31f2d3f TH |
4299 | } |
4300 | ||
2a4ac633 | 4301 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4302 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4303 | { |
623f926b | 4304 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4305 | |
4306 | lockdep_assert_held(&cgroup_mutex); | |
4307 | ||
4308 | if (!(css->flags & CSS_ONLINE)) | |
4309 | return; | |
4310 | ||
d7eeac19 | 4311 | if (ss->css_offline) |
eb95419b | 4312 | ss->css_offline(css); |
a31f2d3f | 4313 | |
eb95419b | 4314 | css->flags &= ~CSS_ONLINE; |
09a503ea | 4315 | css->cgroup->nr_css--; |
0c21ead1 | 4316 | RCU_INIT_POINTER(css->cgroup->subsys[ss->subsys_id], css); |
a31f2d3f TH |
4317 | } |
4318 | ||
ddbcc7e8 | 4319 | /* |
a043e3b2 LZ |
4320 | * cgroup_create - create a cgroup |
4321 | * @parent: cgroup that will be parent of the new cgroup | |
4322 | * @dentry: dentry of the new cgroup | |
4323 | * @mode: mode to set on new inode | |
ddbcc7e8 | 4324 | * |
a043e3b2 | 4325 | * Must be called with the mutex on the parent inode held |
ddbcc7e8 | 4326 | */ |
ddbcc7e8 | 4327 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, |
a5e7ed32 | 4328 | umode_t mode) |
ddbcc7e8 | 4329 | { |
ae7f164a | 4330 | struct cgroup_subsys_state *css_ar[CGROUP_SUBSYS_COUNT] = { }; |
bd89aabc | 4331 | struct cgroup *cgrp; |
65dff759 | 4332 | struct cgroup_name *name; |
ddbcc7e8 PM |
4333 | struct cgroupfs_root *root = parent->root; |
4334 | int err = 0; | |
4335 | struct cgroup_subsys *ss; | |
4336 | struct super_block *sb = root->sb; | |
4337 | ||
0a950f65 | 4338 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
4339 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
4340 | if (!cgrp) | |
ddbcc7e8 PM |
4341 | return -ENOMEM; |
4342 | ||
65dff759 LZ |
4343 | name = cgroup_alloc_name(dentry); |
4344 | if (!name) | |
4345 | goto err_free_cgrp; | |
4346 | rcu_assign_pointer(cgrp->name, name); | |
4347 | ||
4e96ee8e LZ |
4348 | /* |
4349 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4350 | * a half-baked cgroup. | |
4351 | */ | |
4352 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
0a950f65 | 4353 | if (cgrp->id < 0) |
65dff759 | 4354 | goto err_free_name; |
0a950f65 | 4355 | |
976c06bc TH |
4356 | /* |
4357 | * Only live parents can have children. Note that the liveliness | |
4358 | * check isn't strictly necessary because cgroup_mkdir() and | |
4359 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
4360 | * anyway so that locking is contained inside cgroup proper and we | |
4361 | * don't get nasty surprises if we ever grow another caller. | |
4362 | */ | |
4363 | if (!cgroup_lock_live_group(parent)) { | |
4364 | err = -ENODEV; | |
0a950f65 | 4365 | goto err_free_id; |
976c06bc TH |
4366 | } |
4367 | ||
ddbcc7e8 PM |
4368 | /* Grab a reference on the superblock so the hierarchy doesn't |
4369 | * get deleted on unmount if there are child cgroups. This | |
4370 | * can be done outside cgroup_mutex, since the sb can't | |
4371 | * disappear while someone has an open control file on the | |
4372 | * fs */ | |
4373 | atomic_inc(&sb->s_active); | |
4374 | ||
cc31edce | 4375 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4376 | |
fe1c06ca LZ |
4377 | dentry->d_fsdata = cgrp; |
4378 | cgrp->dentry = dentry; | |
4379 | ||
bd89aabc | 4380 | cgrp->parent = parent; |
0ae78e0b | 4381 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 4382 | cgrp->root = parent->root; |
ddbcc7e8 | 4383 | |
b6abdb0e LZ |
4384 | if (notify_on_release(parent)) |
4385 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4386 | ||
2260e7fc TH |
4387 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4388 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4389 | |
5549c497 | 4390 | for_each_root_subsys(root, ss) { |
8c7f6edb | 4391 | struct cgroup_subsys_state *css; |
4528fd05 | 4392 | |
ca8bdcaf | 4393 | css = ss->css_alloc(cgroup_css(parent, ss)); |
ddbcc7e8 PM |
4394 | if (IS_ERR(css)) { |
4395 | err = PTR_ERR(css); | |
4b8b47eb | 4396 | goto err_free_all; |
ddbcc7e8 | 4397 | } |
ae7f164a | 4398 | css_ar[ss->subsys_id] = css; |
d3daf28d TH |
4399 | |
4400 | err = percpu_ref_init(&css->refcnt, css_release); | |
ae7f164a | 4401 | if (err) |
d3daf28d TH |
4402 | goto err_free_all; |
4403 | ||
623f926b | 4404 | init_css(css, ss, cgrp); |
ddbcc7e8 PM |
4405 | } |
4406 | ||
4e139afc TH |
4407 | /* |
4408 | * Create directory. cgroup_create_file() returns with the new | |
4409 | * directory locked on success so that it can be populated without | |
4410 | * dropping cgroup_mutex. | |
4411 | */ | |
28fd6f30 | 4412 | err = cgroup_create_file(dentry, S_IFDIR | mode, sb); |
ddbcc7e8 | 4413 | if (err < 0) |
4b8b47eb | 4414 | goto err_free_all; |
4e139afc | 4415 | lockdep_assert_held(&dentry->d_inode->i_mutex); |
ddbcc7e8 | 4416 | |
00356bd5 | 4417 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 4418 | |
4e139afc | 4419 | /* allocation complete, commit to creation */ |
4e139afc TH |
4420 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
4421 | root->number_of_cgroups++; | |
28fd6f30 | 4422 | |
0ae78e0b TH |
4423 | /* each css holds a ref to the cgroup's dentry and the parent css */ |
4424 | for_each_root_subsys(root, ss) { | |
ae7f164a | 4425 | struct cgroup_subsys_state *css = css_ar[ss->subsys_id]; |
0ae78e0b | 4426 | |
ed957793 | 4427 | dget(dentry); |
930913a3 | 4428 | css_get(css->parent); |
0ae78e0b | 4429 | } |
48ddbe19 | 4430 | |
415cf07a LZ |
4431 | /* hold a ref to the parent's dentry */ |
4432 | dget(parent->dentry); | |
4433 | ||
b1929db4 | 4434 | /* creation succeeded, notify subsystems */ |
5549c497 | 4435 | for_each_root_subsys(root, ss) { |
ae7f164a | 4436 | struct cgroup_subsys_state *css = css_ar[ss->subsys_id]; |
623f926b TH |
4437 | |
4438 | err = online_css(css); | |
b1929db4 TH |
4439 | if (err) |
4440 | goto err_destroy; | |
1f869e87 GC |
4441 | |
4442 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
4443 | parent->parent) { | |
4444 | 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", | |
4445 | current->comm, current->pid, ss->name); | |
4446 | if (!strcmp(ss->name, "memory")) | |
4447 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
4448 | ss->warned_broken_hierarchy = true; | |
4449 | } | |
a8638030 TH |
4450 | } |
4451 | ||
4e96ee8e LZ |
4452 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4453 | ||
2bb566cb | 4454 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
4455 | if (err) |
4456 | goto err_destroy; | |
4457 | ||
4458 | err = cgroup_populate_dir(cgrp, root->subsys_mask); | |
4b8b47eb TH |
4459 | if (err) |
4460 | goto err_destroy; | |
ddbcc7e8 PM |
4461 | |
4462 | mutex_unlock(&cgroup_mutex); | |
bd89aabc | 4463 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
4464 | |
4465 | return 0; | |
4466 | ||
4b8b47eb | 4467 | err_free_all: |
5549c497 | 4468 | for_each_root_subsys(root, ss) { |
ae7f164a | 4469 | struct cgroup_subsys_state *css = css_ar[ss->subsys_id]; |
d3daf28d TH |
4470 | |
4471 | if (css) { | |
4472 | percpu_ref_cancel_init(&css->refcnt); | |
eb95419b | 4473 | ss->css_free(css); |
d3daf28d | 4474 | } |
ddbcc7e8 | 4475 | } |
ddbcc7e8 | 4476 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
4477 | /* Release the reference count that we took on the superblock */ |
4478 | deactivate_super(sb); | |
0a950f65 | 4479 | err_free_id: |
4e96ee8e | 4480 | idr_remove(&root->cgroup_idr, cgrp->id); |
65dff759 LZ |
4481 | err_free_name: |
4482 | kfree(rcu_dereference_raw(cgrp->name)); | |
4b8b47eb | 4483 | err_free_cgrp: |
bd89aabc | 4484 | kfree(cgrp); |
ddbcc7e8 | 4485 | return err; |
4b8b47eb TH |
4486 | |
4487 | err_destroy: | |
4488 | cgroup_destroy_locked(cgrp); | |
4489 | mutex_unlock(&cgroup_mutex); | |
4490 | mutex_unlock(&dentry->d_inode->i_mutex); | |
4491 | return err; | |
ddbcc7e8 PM |
4492 | } |
4493 | ||
18bb1db3 | 4494 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
ddbcc7e8 PM |
4495 | { |
4496 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
4497 | ||
4498 | /* the vfs holds inode->i_mutex already */ | |
4499 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
4500 | } | |
4501 | ||
223dbc38 TH |
4502 | /* |
4503 | * This is called when the refcnt of a css is confirmed to be killed. | |
4504 | * css_tryget() is now guaranteed to fail. | |
4505 | */ | |
4506 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4507 | { |
223dbc38 TH |
4508 | struct cgroup_subsys_state *css = |
4509 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
4510 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 4511 | |
f20104de TH |
4512 | mutex_lock(&cgroup_mutex); |
4513 | ||
09a503ea TH |
4514 | /* |
4515 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
4516 | * initate destruction. | |
4517 | */ | |
4518 | offline_css(css); | |
4519 | ||
f20104de TH |
4520 | /* |
4521 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
4522 | * be disabled before proceeding to the second phase of cgroup | |
4523 | * destruction. If we are the last one, kick it off. | |
4524 | */ | |
09a503ea | 4525 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
4526 | cgroup_destroy_css_killed(cgrp); |
4527 | ||
4528 | mutex_unlock(&cgroup_mutex); | |
09a503ea TH |
4529 | |
4530 | /* | |
4531 | * Put the css refs from kill_css(). Each css holds an extra | |
4532 | * reference to the cgroup's dentry and cgroup removal proceeds | |
4533 | * regardless of css refs. On the last put of each css, whenever | |
4534 | * that may be, the extra dentry ref is put so that dentry | |
4535 | * destruction happens only after all css's are released. | |
4536 | */ | |
4537 | css_put(css); | |
d3daf28d TH |
4538 | } |
4539 | ||
223dbc38 TH |
4540 | /* css kill confirmation processing requires process context, bounce */ |
4541 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4542 | { |
4543 | struct cgroup_subsys_state *css = | |
4544 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4545 | ||
223dbc38 TH |
4546 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
4547 | schedule_work(&css->destroy_work); | |
d3daf28d TH |
4548 | } |
4549 | ||
edae0c33 TH |
4550 | /** |
4551 | * kill_css - destroy a css | |
4552 | * @css: css to destroy | |
4553 | * | |
3c14f8b4 TH |
4554 | * This function initiates destruction of @css by removing cgroup interface |
4555 | * files and putting its base reference. ->css_offline() will be invoked | |
4556 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
4557 | * reference count reaches zero, @css will be released. | |
edae0c33 TH |
4558 | */ |
4559 | static void kill_css(struct cgroup_subsys_state *css) | |
4560 | { | |
3c14f8b4 TH |
4561 | cgroup_clear_dir(css->cgroup, 1 << css->ss->subsys_id); |
4562 | ||
edae0c33 TH |
4563 | /* |
4564 | * Killing would put the base ref, but we need to keep it alive | |
4565 | * until after ->css_offline(). | |
4566 | */ | |
4567 | css_get(css); | |
4568 | ||
4569 | /* | |
4570 | * cgroup core guarantees that, by the time ->css_offline() is | |
4571 | * invoked, no new css reference will be given out via | |
4572 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
4573 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
4574 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4575 | * | |
4576 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4577 | * css is confirmed to be seen as killed on all CPUs. | |
4578 | */ | |
4579 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4580 | } |
4581 | ||
4582 | /** | |
4583 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4584 | * @cgrp: cgroup to be destroyed | |
4585 | * | |
4586 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4587 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
4588 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
4589 | * invoked. To satisfy all the requirements, destruction is implemented in | |
4590 | * the following two steps. | |
4591 | * | |
4592 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4593 | * userland visible parts and start killing the percpu refcnts of | |
4594 | * css's. Set up so that the next stage will be kicked off once all | |
4595 | * the percpu refcnts are confirmed to be killed. | |
4596 | * | |
4597 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4598 | * rest of destruction. Once all cgroup references are gone, the | |
4599 | * cgroup is RCU-freed. | |
4600 | * | |
4601 | * This function implements s1. After this step, @cgrp is gone as far as | |
4602 | * the userland is concerned and a new cgroup with the same name may be | |
4603 | * created. As cgroup doesn't care about the names internally, this | |
4604 | * doesn't cause any problem. | |
4605 | */ | |
42809dd4 TH |
4606 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4607 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4608 | { |
42809dd4 | 4609 | struct dentry *d = cgrp->dentry; |
4ab78683 | 4610 | struct cgroup_event *event, *tmp; |
ed957793 | 4611 | struct cgroup_subsys *ss; |
bb78a92f | 4612 | struct cgroup *child; |
ddd69148 | 4613 | bool empty; |
ddbcc7e8 | 4614 | |
42809dd4 TH |
4615 | lockdep_assert_held(&d->d_inode->i_mutex); |
4616 | lockdep_assert_held(&cgroup_mutex); | |
4617 | ||
ddd69148 | 4618 | /* |
6f3d828f TH |
4619 | * css_set_lock synchronizes access to ->cset_links and prevents |
4620 | * @cgrp from being removed while __put_css_set() is in progress. | |
ddd69148 TH |
4621 | */ |
4622 | read_lock(&css_set_lock); | |
bb78a92f | 4623 | empty = list_empty(&cgrp->cset_links); |
ddd69148 TH |
4624 | read_unlock(&css_set_lock); |
4625 | if (!empty) | |
ddbcc7e8 | 4626 | return -EBUSY; |
a043e3b2 | 4627 | |
bb78a92f HD |
4628 | /* |
4629 | * Make sure there's no live children. We can't test ->children | |
4630 | * emptiness as dead children linger on it while being destroyed; | |
4631 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
4632 | */ | |
4633 | empty = true; | |
4634 | rcu_read_lock(); | |
4635 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
4636 | empty = cgroup_is_dead(child); | |
4637 | if (!empty) | |
4638 | break; | |
4639 | } | |
4640 | rcu_read_unlock(); | |
4641 | if (!empty) | |
4642 | return -EBUSY; | |
4643 | ||
88703267 | 4644 | /* |
edae0c33 TH |
4645 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
4646 | * will be invoked to perform the rest of destruction once the | |
4647 | * percpu refs of all css's are confirmed to be killed. | |
88703267 | 4648 | */ |
edae0c33 | 4649 | for_each_root_subsys(cgrp->root, ss) |
ca8bdcaf | 4650 | kill_css(cgroup_css(cgrp, ss)); |
455050d2 TH |
4651 | |
4652 | /* | |
4653 | * Mark @cgrp dead. This prevents further task migration and child | |
4654 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 4655 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 4656 | * resume iteration after dropping RCU read lock. See |
492eb21b | 4657 | * css_next_child() for details. |
455050d2 | 4658 | */ |
54766d4a | 4659 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 4660 | |
455050d2 TH |
4661 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
4662 | raw_spin_lock(&release_list_lock); | |
4663 | if (!list_empty(&cgrp->release_list)) | |
4664 | list_del_init(&cgrp->release_list); | |
4665 | raw_spin_unlock(&release_list_lock); | |
4666 | ||
4667 | /* | |
f20104de TH |
4668 | * If @cgrp has css's attached, the second stage of cgroup |
4669 | * destruction is kicked off from css_killed_work_fn() after the | |
4670 | * refs of all attached css's are killed. If @cgrp doesn't have | |
4671 | * any css, we kick it off here. | |
4672 | */ | |
4673 | if (!cgrp->nr_css) | |
4674 | cgroup_destroy_css_killed(cgrp); | |
4675 | ||
455050d2 | 4676 | /* |
3c14f8b4 TH |
4677 | * Clear the base files and remove @cgrp directory. The removal |
4678 | * puts the base ref but we aren't quite done with @cgrp yet, so | |
4679 | * hold onto it. | |
455050d2 | 4680 | */ |
2bb566cb | 4681 | cgroup_addrm_files(cgrp, cgroup_base_files, false); |
455050d2 TH |
4682 | dget(d); |
4683 | cgroup_d_remove_dir(d); | |
4684 | ||
4685 | /* | |
4686 | * Unregister events and notify userspace. | |
4687 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4688 | * directory to avoid race between userspace and kernelspace. | |
4689 | */ | |
4690 | spin_lock(&cgrp->event_list_lock); | |
4691 | list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) { | |
4692 | list_del_init(&event->list); | |
4693 | schedule_work(&event->remove); | |
4694 | } | |
4695 | spin_unlock(&cgrp->event_list_lock); | |
4696 | ||
ea15f8cc TH |
4697 | return 0; |
4698 | }; | |
4699 | ||
d3daf28d | 4700 | /** |
f20104de | 4701 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
4702 | * @work: cgroup->destroy_free_work |
4703 | * | |
4704 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
4705 | * destroyed after all css's are offlined and performs the rest of |
4706 | * destruction. This is the second step of destruction described in the | |
4707 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 4708 | */ |
f20104de | 4709 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 4710 | { |
ea15f8cc TH |
4711 | struct cgroup *parent = cgrp->parent; |
4712 | struct dentry *d = cgrp->dentry; | |
ea15f8cc | 4713 | |
f20104de | 4714 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 4715 | |
999cd8a4 | 4716 | /* delete this cgroup from parent->children */ |
eb6fd504 | 4717 | list_del_rcu(&cgrp->sibling); |
ed957793 TH |
4718 | |
4719 | /* | |
4e96ee8e LZ |
4720 | * We should remove the cgroup object from idr before its grace |
4721 | * period starts, so we won't be looking up a cgroup while the | |
4722 | * cgroup is being freed. | |
ed957793 | 4723 | */ |
4e96ee8e LZ |
4724 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4725 | cgrp->id = -1; | |
b0ca5a84 | 4726 | |
ddbcc7e8 | 4727 | dput(d); |
ddbcc7e8 | 4728 | |
bd89aabc | 4729 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 4730 | check_for_release(parent); |
ddbcc7e8 PM |
4731 | } |
4732 | ||
42809dd4 TH |
4733 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) |
4734 | { | |
4735 | int ret; | |
4736 | ||
4737 | mutex_lock(&cgroup_mutex); | |
4738 | ret = cgroup_destroy_locked(dentry->d_fsdata); | |
4739 | mutex_unlock(&cgroup_mutex); | |
4740 | ||
4741 | return ret; | |
4742 | } | |
4743 | ||
8e3f6541 TH |
4744 | static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss) |
4745 | { | |
4746 | INIT_LIST_HEAD(&ss->cftsets); | |
4747 | ||
4748 | /* | |
4749 | * base_cftset is embedded in subsys itself, no need to worry about | |
4750 | * deregistration. | |
4751 | */ | |
4752 | if (ss->base_cftypes) { | |
2bb566cb TH |
4753 | struct cftype *cft; |
4754 | ||
4755 | for (cft = ss->base_cftypes; cft->name[0] != '\0'; cft++) | |
4756 | cft->ss = ss; | |
4757 | ||
8e3f6541 TH |
4758 | ss->base_cftset.cfts = ss->base_cftypes; |
4759 | list_add_tail(&ss->base_cftset.node, &ss->cftsets); | |
4760 | } | |
4761 | } | |
4762 | ||
06a11920 | 4763 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4764 | { |
ddbcc7e8 | 4765 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4766 | |
4767 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4768 | |
648bb56d TH |
4769 | mutex_lock(&cgroup_mutex); |
4770 | ||
8e3f6541 TH |
4771 | /* init base cftset */ |
4772 | cgroup_init_cftsets(ss); | |
4773 | ||
ddbcc7e8 | 4774 | /* Create the top cgroup state for this subsystem */ |
9871bf95 TH |
4775 | list_add(&ss->sibling, &cgroup_dummy_root.subsys_list); |
4776 | ss->root = &cgroup_dummy_root; | |
ca8bdcaf | 4777 | css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss)); |
ddbcc7e8 PM |
4778 | /* We don't handle early failures gracefully */ |
4779 | BUG_ON(IS_ERR(css)); | |
623f926b | 4780 | init_css(css, ss, cgroup_dummy_top); |
ddbcc7e8 | 4781 | |
e8d55fde | 4782 | /* Update the init_css_set to contain a subsys |
817929ec | 4783 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
4784 | * newly registered, all tasks and hence the |
4785 | * init_css_set is in the subsystem's top cgroup. */ | |
b48c6a80 | 4786 | init_css_set.subsys[ss->subsys_id] = css; |
ddbcc7e8 PM |
4787 | |
4788 | need_forkexit_callback |= ss->fork || ss->exit; | |
4789 | ||
e8d55fde LZ |
4790 | /* At system boot, before all subsystems have been |
4791 | * registered, no tasks have been forked, so we don't | |
4792 | * need to invoke fork callbacks here. */ | |
4793 | BUG_ON(!list_empty(&init_task.tasks)); | |
4794 | ||
ae7f164a | 4795 | BUG_ON(online_css(css)); |
a8638030 | 4796 | |
648bb56d TH |
4797 | mutex_unlock(&cgroup_mutex); |
4798 | ||
e6a1105b BB |
4799 | /* this function shouldn't be used with modular subsystems, since they |
4800 | * need to register a subsys_id, among other things */ | |
4801 | BUG_ON(ss->module); | |
4802 | } | |
4803 | ||
4804 | /** | |
4805 | * cgroup_load_subsys: load and register a modular subsystem at runtime | |
4806 | * @ss: the subsystem to load | |
4807 | * | |
4808 | * This function should be called in a modular subsystem's initcall. If the | |
88393161 | 4809 | * subsystem is built as a module, it will be assigned a new subsys_id and set |
e6a1105b BB |
4810 | * up for use. If the subsystem is built-in anyway, work is delegated to the |
4811 | * simpler cgroup_init_subsys. | |
4812 | */ | |
4813 | int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss) | |
4814 | { | |
e6a1105b | 4815 | struct cgroup_subsys_state *css; |
d19e19de | 4816 | int i, ret; |
b67bfe0d | 4817 | struct hlist_node *tmp; |
5abb8855 | 4818 | struct css_set *cset; |
0ac801fe | 4819 | unsigned long key; |
e6a1105b BB |
4820 | |
4821 | /* check name and function validity */ | |
4822 | if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN || | |
92fb9748 | 4823 | ss->css_alloc == NULL || ss->css_free == NULL) |
e6a1105b BB |
4824 | return -EINVAL; |
4825 | ||
4826 | /* | |
4827 | * we don't support callbacks in modular subsystems. this check is | |
4828 | * before the ss->module check for consistency; a subsystem that could | |
4829 | * be a module should still have no callbacks even if the user isn't | |
4830 | * compiling it as one. | |
4831 | */ | |
4832 | if (ss->fork || ss->exit) | |
4833 | return -EINVAL; | |
4834 | ||
4835 | /* | |
4836 | * an optionally modular subsystem is built-in: we want to do nothing, | |
4837 | * since cgroup_init_subsys will have already taken care of it. | |
4838 | */ | |
4839 | if (ss->module == NULL) { | |
be45c900 | 4840 | /* a sanity check */ |
9871bf95 | 4841 | BUG_ON(cgroup_subsys[ss->subsys_id] != ss); |
e6a1105b BB |
4842 | return 0; |
4843 | } | |
4844 | ||
8e3f6541 TH |
4845 | /* init base cftset */ |
4846 | cgroup_init_cftsets(ss); | |
4847 | ||
e6a1105b | 4848 | mutex_lock(&cgroup_mutex); |
9871bf95 | 4849 | cgroup_subsys[ss->subsys_id] = ss; |
e6a1105b BB |
4850 | |
4851 | /* | |
92fb9748 | 4852 | * no ss->css_alloc seems to need anything important in the ss |
9871bf95 | 4853 | * struct, so this can happen first (i.e. before the dummy root |
92fb9748 | 4854 | * attachment). |
e6a1105b | 4855 | */ |
ca8bdcaf | 4856 | css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss)); |
e6a1105b | 4857 | if (IS_ERR(css)) { |
9871bf95 TH |
4858 | /* failure case - need to deassign the cgroup_subsys[] slot. */ |
4859 | cgroup_subsys[ss->subsys_id] = NULL; | |
e6a1105b BB |
4860 | mutex_unlock(&cgroup_mutex); |
4861 | return PTR_ERR(css); | |
4862 | } | |
4863 | ||
9871bf95 TH |
4864 | list_add(&ss->sibling, &cgroup_dummy_root.subsys_list); |
4865 | ss->root = &cgroup_dummy_root; | |
e6a1105b BB |
4866 | |
4867 | /* our new subsystem will be attached to the dummy hierarchy. */ | |
623f926b | 4868 | init_css(css, ss, cgroup_dummy_top); |
e6a1105b BB |
4869 | |
4870 | /* | |
4871 | * Now we need to entangle the css into the existing css_sets. unlike | |
4872 | * in cgroup_init_subsys, there are now multiple css_sets, so each one | |
4873 | * will need a new pointer to it; done by iterating the css_set_table. | |
4874 | * furthermore, modifying the existing css_sets will corrupt the hash | |
4875 | * table state, so each changed css_set will need its hash recomputed. | |
4876 | * this is all done under the css_set_lock. | |
4877 | */ | |
4878 | write_lock(&css_set_lock); | |
5abb8855 | 4879 | hash_for_each_safe(css_set_table, i, tmp, cset, hlist) { |
0ac801fe | 4880 | /* skip entries that we already rehashed */ |
5abb8855 | 4881 | if (cset->subsys[ss->subsys_id]) |
0ac801fe LZ |
4882 | continue; |
4883 | /* remove existing entry */ | |
5abb8855 | 4884 | hash_del(&cset->hlist); |
0ac801fe | 4885 | /* set new value */ |
5abb8855 | 4886 | cset->subsys[ss->subsys_id] = css; |
0ac801fe | 4887 | /* recompute hash and restore entry */ |
5abb8855 TH |
4888 | key = css_set_hash(cset->subsys); |
4889 | hash_add(css_set_table, &cset->hlist, key); | |
e6a1105b BB |
4890 | } |
4891 | write_unlock(&css_set_lock); | |
4892 | ||
ae7f164a | 4893 | ret = online_css(css); |
b1929db4 TH |
4894 | if (ret) |
4895 | goto err_unload; | |
a8638030 | 4896 | |
e6a1105b BB |
4897 | /* success! */ |
4898 | mutex_unlock(&cgroup_mutex); | |
4899 | return 0; | |
d19e19de TH |
4900 | |
4901 | err_unload: | |
4902 | mutex_unlock(&cgroup_mutex); | |
4903 | /* @ss can't be mounted here as try_module_get() would fail */ | |
4904 | cgroup_unload_subsys(ss); | |
4905 | return ret; | |
ddbcc7e8 | 4906 | } |
e6a1105b | 4907 | EXPORT_SYMBOL_GPL(cgroup_load_subsys); |
ddbcc7e8 | 4908 | |
cf5d5941 BB |
4909 | /** |
4910 | * cgroup_unload_subsys: unload a modular subsystem | |
4911 | * @ss: the subsystem to unload | |
4912 | * | |
4913 | * This function should be called in a modular subsystem's exitcall. When this | |
4914 | * function is invoked, the refcount on the subsystem's module will be 0, so | |
4915 | * the subsystem will not be attached to any hierarchy. | |
4916 | */ | |
4917 | void cgroup_unload_subsys(struct cgroup_subsys *ss) | |
4918 | { | |
69d0206c | 4919 | struct cgrp_cset_link *link; |
cf5d5941 BB |
4920 | |
4921 | BUG_ON(ss->module == NULL); | |
4922 | ||
4923 | /* | |
4924 | * we shouldn't be called if the subsystem is in use, and the use of | |
1d5be6b2 | 4925 | * try_module_get() in rebind_subsystems() should ensure that it |
cf5d5941 BB |
4926 | * doesn't start being used while we're killing it off. |
4927 | */ | |
9871bf95 | 4928 | BUG_ON(ss->root != &cgroup_dummy_root); |
cf5d5941 BB |
4929 | |
4930 | mutex_lock(&cgroup_mutex); | |
02ae7486 | 4931 | |
ca8bdcaf | 4932 | offline_css(cgroup_css(cgroup_dummy_top, ss)); |
02ae7486 | 4933 | |
cf5d5941 | 4934 | /* deassign the subsys_id */ |
9871bf95 | 4935 | cgroup_subsys[ss->subsys_id] = NULL; |
cf5d5941 | 4936 | |
9871bf95 | 4937 | /* remove subsystem from the dummy root's list of subsystems */ |
8d258797 | 4938 | list_del_init(&ss->sibling); |
cf5d5941 BB |
4939 | |
4940 | /* | |
9871bf95 TH |
4941 | * disentangle the css from all css_sets attached to the dummy |
4942 | * top. as in loading, we need to pay our respects to the hashtable | |
4943 | * gods. | |
cf5d5941 BB |
4944 | */ |
4945 | write_lock(&css_set_lock); | |
9871bf95 | 4946 | list_for_each_entry(link, &cgroup_dummy_top->cset_links, cset_link) { |
69d0206c | 4947 | struct css_set *cset = link->cset; |
0ac801fe | 4948 | unsigned long key; |
cf5d5941 | 4949 | |
5abb8855 TH |
4950 | hash_del(&cset->hlist); |
4951 | cset->subsys[ss->subsys_id] = NULL; | |
4952 | key = css_set_hash(cset->subsys); | |
4953 | hash_add(css_set_table, &cset->hlist, key); | |
cf5d5941 BB |
4954 | } |
4955 | write_unlock(&css_set_lock); | |
4956 | ||
4957 | /* | |
9871bf95 TH |
4958 | * remove subsystem's css from the cgroup_dummy_top and free it - |
4959 | * need to free before marking as null because ss->css_free needs | |
2ff2a7d0 | 4960 | * the cgrp->subsys pointer to find their state. |
cf5d5941 | 4961 | */ |
ca8bdcaf | 4962 | ss->css_free(cgroup_css(cgroup_dummy_top, ss)); |
73e80ed8 | 4963 | RCU_INIT_POINTER(cgroup_dummy_top->subsys[ss->subsys_id], NULL); |
cf5d5941 BB |
4964 | |
4965 | mutex_unlock(&cgroup_mutex); | |
4966 | } | |
4967 | EXPORT_SYMBOL_GPL(cgroup_unload_subsys); | |
4968 | ||
ddbcc7e8 | 4969 | /** |
a043e3b2 LZ |
4970 | * cgroup_init_early - cgroup initialization at system boot |
4971 | * | |
4972 | * Initialize cgroups at system boot, and initialize any | |
4973 | * subsystems that request early init. | |
ddbcc7e8 PM |
4974 | */ |
4975 | int __init cgroup_init_early(void) | |
4976 | { | |
30159ec7 | 4977 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4978 | int i; |
30159ec7 | 4979 | |
146aa1bd | 4980 | atomic_set(&init_css_set.refcount, 1); |
69d0206c | 4981 | INIT_LIST_HEAD(&init_css_set.cgrp_links); |
817929ec | 4982 | INIT_LIST_HEAD(&init_css_set.tasks); |
472b1053 | 4983 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 4984 | css_set_count = 1; |
9871bf95 TH |
4985 | init_cgroup_root(&cgroup_dummy_root); |
4986 | cgroup_root_count = 1; | |
a4ea1cc9 | 4987 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4988 | |
69d0206c | 4989 | init_cgrp_cset_link.cset = &init_css_set; |
9871bf95 TH |
4990 | init_cgrp_cset_link.cgrp = cgroup_dummy_top; |
4991 | list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links); | |
69d0206c | 4992 | list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links); |
ddbcc7e8 | 4993 | |
30159ec7 TH |
4994 | /* at bootup time, we don't worry about modular subsystems */ |
4995 | for_each_builtin_subsys(ss, i) { | |
ddbcc7e8 PM |
4996 | BUG_ON(!ss->name); |
4997 | BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); | |
92fb9748 TH |
4998 | BUG_ON(!ss->css_alloc); |
4999 | BUG_ON(!ss->css_free); | |
ddbcc7e8 | 5000 | if (ss->subsys_id != i) { |
cfe36bde | 5001 | printk(KERN_ERR "cgroup: Subsys %s id == %d\n", |
ddbcc7e8 PM |
5002 | ss->name, ss->subsys_id); |
5003 | BUG(); | |
5004 | } | |
5005 | ||
5006 | if (ss->early_init) | |
5007 | cgroup_init_subsys(ss); | |
5008 | } | |
5009 | return 0; | |
5010 | } | |
5011 | ||
5012 | /** | |
a043e3b2 LZ |
5013 | * cgroup_init - cgroup initialization |
5014 | * | |
5015 | * Register cgroup filesystem and /proc file, and initialize | |
5016 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5017 | */ |
5018 | int __init cgroup_init(void) | |
5019 | { | |
30159ec7 | 5020 | struct cgroup_subsys *ss; |
0ac801fe | 5021 | unsigned long key; |
30159ec7 | 5022 | int i, err; |
a424316c PM |
5023 | |
5024 | err = bdi_init(&cgroup_backing_dev_info); | |
5025 | if (err) | |
5026 | return err; | |
ddbcc7e8 | 5027 | |
30159ec7 | 5028 | for_each_builtin_subsys(ss, i) { |
ddbcc7e8 PM |
5029 | if (!ss->early_init) |
5030 | cgroup_init_subsys(ss); | |
5031 | } | |
5032 | ||
fa3ca07e | 5033 | /* allocate id for the dummy hierarchy */ |
54e7b4eb TH |
5034 | mutex_lock(&cgroup_mutex); |
5035 | mutex_lock(&cgroup_root_mutex); | |
5036 | ||
82fe9b0d TH |
5037 | /* Add init_css_set to the hash table */ |
5038 | key = css_set_hash(init_css_set.subsys); | |
5039 | hash_add(css_set_table, &init_css_set.hlist, key); | |
5040 | ||
fc76df70 | 5041 | BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1)); |
676db4af | 5042 | |
4e96ee8e LZ |
5043 | err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top, |
5044 | 0, 1, GFP_KERNEL); | |
5045 | BUG_ON(err < 0); | |
5046 | ||
54e7b4eb TH |
5047 | mutex_unlock(&cgroup_root_mutex); |
5048 | mutex_unlock(&cgroup_mutex); | |
5049 | ||
676db4af GK |
5050 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
5051 | if (!cgroup_kobj) { | |
5052 | err = -ENOMEM; | |
5053 | goto out; | |
5054 | } | |
5055 | ||
ddbcc7e8 | 5056 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
5057 | if (err < 0) { |
5058 | kobject_put(cgroup_kobj); | |
ddbcc7e8 | 5059 | goto out; |
676db4af | 5060 | } |
ddbcc7e8 | 5061 | |
46ae220b | 5062 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
a424316c | 5063 | |
ddbcc7e8 | 5064 | out: |
a424316c PM |
5065 | if (err) |
5066 | bdi_destroy(&cgroup_backing_dev_info); | |
5067 | ||
ddbcc7e8 PM |
5068 | return err; |
5069 | } | |
b4f48b63 | 5070 | |
a424316c PM |
5071 | /* |
5072 | * proc_cgroup_show() | |
5073 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5074 | * - Used for /proc/<pid>/cgroup. | |
5075 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
5076 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 5077 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
5078 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
5079 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
5080 | * cgroup to top_cgroup. | |
5081 | */ | |
5082 | ||
5083 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 5084 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
5085 | { |
5086 | struct pid *pid; | |
5087 | struct task_struct *tsk; | |
5088 | char *buf; | |
5089 | int retval; | |
5090 | struct cgroupfs_root *root; | |
5091 | ||
5092 | retval = -ENOMEM; | |
5093 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
5094 | if (!buf) | |
5095 | goto out; | |
5096 | ||
5097 | retval = -ESRCH; | |
5098 | pid = m->private; | |
5099 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
5100 | if (!tsk) | |
5101 | goto out_free; | |
5102 | ||
5103 | retval = 0; | |
5104 | ||
5105 | mutex_lock(&cgroup_mutex); | |
5106 | ||
e5f6a860 | 5107 | for_each_active_root(root) { |
a424316c | 5108 | struct cgroup_subsys *ss; |
bd89aabc | 5109 | struct cgroup *cgrp; |
a424316c PM |
5110 | int count = 0; |
5111 | ||
2c6ab6d2 | 5112 | seq_printf(m, "%d:", root->hierarchy_id); |
5549c497 | 5113 | for_each_root_subsys(root, ss) |
a424316c | 5114 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); |
c6d57f33 PM |
5115 | if (strlen(root->name)) |
5116 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5117 | root->name); | |
a424316c | 5118 | seq_putc(m, ':'); |
7717f7ba | 5119 | cgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 5120 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); |
a424316c PM |
5121 | if (retval < 0) |
5122 | goto out_unlock; | |
5123 | seq_puts(m, buf); | |
5124 | seq_putc(m, '\n'); | |
5125 | } | |
5126 | ||
5127 | out_unlock: | |
5128 | mutex_unlock(&cgroup_mutex); | |
5129 | put_task_struct(tsk); | |
5130 | out_free: | |
5131 | kfree(buf); | |
5132 | out: | |
5133 | return retval; | |
5134 | } | |
5135 | ||
a424316c PM |
5136 | /* Display information about each subsystem and each hierarchy */ |
5137 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5138 | { | |
30159ec7 | 5139 | struct cgroup_subsys *ss; |
a424316c | 5140 | int i; |
a424316c | 5141 | |
8bab8dde | 5142 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5143 | /* |
5144 | * ideally we don't want subsystems moving around while we do this. | |
5145 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5146 | * subsys/hierarchy state. | |
5147 | */ | |
a424316c | 5148 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5149 | |
5150 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
5151 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
5152 | ss->name, ss->root->hierarchy_id, | |
8bab8dde | 5153 | ss->root->number_of_cgroups, !ss->disabled); |
30159ec7 | 5154 | |
a424316c PM |
5155 | mutex_unlock(&cgroup_mutex); |
5156 | return 0; | |
5157 | } | |
5158 | ||
5159 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5160 | { | |
9dce07f1 | 5161 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5162 | } |
5163 | ||
828c0950 | 5164 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5165 | .open = cgroupstats_open, |
5166 | .read = seq_read, | |
5167 | .llseek = seq_lseek, | |
5168 | .release = single_release, | |
5169 | }; | |
5170 | ||
b4f48b63 PM |
5171 | /** |
5172 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 5173 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
5174 | * |
5175 | * Description: A task inherits its parent's cgroup at fork(). | |
5176 | * | |
5177 | * A pointer to the shared css_set was automatically copied in | |
5178 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
9bb71308 TH |
5179 | * it was not made under the protection of RCU or cgroup_mutex, so |
5180 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
5181 | * have already changed current->cgroups, allowing the previously | |
5182 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
5183 | * |
5184 | * At the point that cgroup_fork() is called, 'current' is the parent | |
5185 | * task, and the passed argument 'child' points to the child task. | |
5186 | */ | |
5187 | void cgroup_fork(struct task_struct *child) | |
5188 | { | |
9bb71308 | 5189 | task_lock(current); |
a8ad805c | 5190 | get_css_set(task_css_set(current)); |
817929ec | 5191 | child->cgroups = current->cgroups; |
9bb71308 | 5192 | task_unlock(current); |
817929ec | 5193 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5194 | } |
5195 | ||
817929ec | 5196 | /** |
a043e3b2 LZ |
5197 | * cgroup_post_fork - called on a new task after adding it to the task list |
5198 | * @child: the task in question | |
5199 | * | |
5edee61e TH |
5200 | * Adds the task to the list running through its css_set if necessary and |
5201 | * call the subsystem fork() callbacks. Has to be after the task is | |
5202 | * visible on the task list in case we race with the first call to | |
0942eeee | 5203 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5204 | * list. |
a043e3b2 | 5205 | */ |
817929ec PM |
5206 | void cgroup_post_fork(struct task_struct *child) |
5207 | { | |
30159ec7 | 5208 | struct cgroup_subsys *ss; |
5edee61e TH |
5209 | int i; |
5210 | ||
3ce3230a FW |
5211 | /* |
5212 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
5213 | * under the tasklist_lock and we read it here after we added the child | |
5214 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
5215 | * yet in the tasklist when we walked through it from | |
5216 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
5217 | * should be visible now due to the paired locking and barriers implied | |
5218 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
5219 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
5220 | * lock on fork. | |
5221 | */ | |
817929ec PM |
5222 | if (use_task_css_set_links) { |
5223 | write_lock(&css_set_lock); | |
d8783832 TH |
5224 | task_lock(child); |
5225 | if (list_empty(&child->cg_list)) | |
a8ad805c | 5226 | list_add(&child->cg_list, &task_css_set(child)->tasks); |
d8783832 | 5227 | task_unlock(child); |
817929ec PM |
5228 | write_unlock(&css_set_lock); |
5229 | } | |
5edee61e TH |
5230 | |
5231 | /* | |
5232 | * Call ss->fork(). This must happen after @child is linked on | |
5233 | * css_set; otherwise, @child might change state between ->fork() | |
5234 | * and addition to css_set. | |
5235 | */ | |
5236 | if (need_forkexit_callback) { | |
7d8e0bf5 LZ |
5237 | /* |
5238 | * fork/exit callbacks are supported only for builtin | |
5239 | * subsystems, and the builtin section of the subsys | |
5240 | * array is immutable, so we don't need to lock the | |
5241 | * subsys array here. On the other hand, modular section | |
5242 | * of the array can be freed at module unload, so we | |
5243 | * can't touch that. | |
5244 | */ | |
30159ec7 | 5245 | for_each_builtin_subsys(ss, i) |
5edee61e TH |
5246 | if (ss->fork) |
5247 | ss->fork(child); | |
5edee61e | 5248 | } |
817929ec | 5249 | } |
5edee61e | 5250 | |
b4f48b63 PM |
5251 | /** |
5252 | * cgroup_exit - detach cgroup from exiting task | |
5253 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 5254 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
5255 | * |
5256 | * Description: Detach cgroup from @tsk and release it. | |
5257 | * | |
5258 | * Note that cgroups marked notify_on_release force every task in | |
5259 | * them to take the global cgroup_mutex mutex when exiting. | |
5260 | * This could impact scaling on very large systems. Be reluctant to | |
5261 | * use notify_on_release cgroups where very high task exit scaling | |
5262 | * is required on large systems. | |
5263 | * | |
5264 | * the_top_cgroup_hack: | |
5265 | * | |
5266 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
5267 | * | |
5268 | * We call cgroup_exit() while the task is still competent to | |
5269 | * handle notify_on_release(), then leave the task attached to the | |
5270 | * root cgroup in each hierarchy for the remainder of its exit. | |
5271 | * | |
5272 | * To do this properly, we would increment the reference count on | |
5273 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
5274 | * code we would add a second cgroup function call, to drop that | |
5275 | * reference. This would just create an unnecessary hot spot on | |
5276 | * the top_cgroup reference count, to no avail. | |
5277 | * | |
5278 | * Normally, holding a reference to a cgroup without bumping its | |
5279 | * count is unsafe. The cgroup could go away, or someone could | |
5280 | * attach us to a different cgroup, decrementing the count on | |
5281 | * the first cgroup that we never incremented. But in this case, | |
5282 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
5283 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
5284 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
5285 | */ |
5286 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
5287 | { | |
30159ec7 | 5288 | struct cgroup_subsys *ss; |
5abb8855 | 5289 | struct css_set *cset; |
d41d5a01 | 5290 | int i; |
817929ec PM |
5291 | |
5292 | /* | |
5293 | * Unlink from the css_set task list if necessary. | |
5294 | * Optimistically check cg_list before taking | |
5295 | * css_set_lock | |
5296 | */ | |
5297 | if (!list_empty(&tsk->cg_list)) { | |
5298 | write_lock(&css_set_lock); | |
5299 | if (!list_empty(&tsk->cg_list)) | |
8d258797 | 5300 | list_del_init(&tsk->cg_list); |
817929ec PM |
5301 | write_unlock(&css_set_lock); |
5302 | } | |
5303 | ||
b4f48b63 PM |
5304 | /* Reassign the task to the init_css_set. */ |
5305 | task_lock(tsk); | |
a8ad805c TH |
5306 | cset = task_css_set(tsk); |
5307 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 PZ |
5308 | |
5309 | if (run_callbacks && need_forkexit_callback) { | |
7d8e0bf5 LZ |
5310 | /* |
5311 | * fork/exit callbacks are supported only for builtin | |
5312 | * subsystems, see cgroup_post_fork() for details. | |
5313 | */ | |
30159ec7 | 5314 | for_each_builtin_subsys(ss, i) { |
d41d5a01 | 5315 | if (ss->exit) { |
eb95419b TH |
5316 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
5317 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 5318 | |
eb95419b | 5319 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
5320 | } |
5321 | } | |
5322 | } | |
b4f48b63 | 5323 | task_unlock(tsk); |
d41d5a01 | 5324 | |
5abb8855 | 5325 | put_css_set_taskexit(cset); |
b4f48b63 | 5326 | } |
697f4161 | 5327 | |
bd89aabc | 5328 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5329 | { |
f50daa70 | 5330 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 5331 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
5332 | /* |
5333 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 5334 | * already queued for a userspace notification, queue |
f50daa70 LZ |
5335 | * it now |
5336 | */ | |
81a6a5cd | 5337 | int need_schedule_work = 0; |
f50daa70 | 5338 | |
cdcc136f | 5339 | raw_spin_lock(&release_list_lock); |
54766d4a | 5340 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
5341 | list_empty(&cgrp->release_list)) { |
5342 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
5343 | need_schedule_work = 1; |
5344 | } | |
cdcc136f | 5345 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5346 | if (need_schedule_work) |
5347 | schedule_work(&release_agent_work); | |
5348 | } | |
5349 | } | |
5350 | ||
81a6a5cd PM |
5351 | /* |
5352 | * Notify userspace when a cgroup is released, by running the | |
5353 | * configured release agent with the name of the cgroup (path | |
5354 | * relative to the root of cgroup file system) as the argument. | |
5355 | * | |
5356 | * Most likely, this user command will try to rmdir this cgroup. | |
5357 | * | |
5358 | * This races with the possibility that some other task will be | |
5359 | * attached to this cgroup before it is removed, or that some other | |
5360 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5361 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5362 | * unused, and this cgroup will be reprieved from its death sentence, | |
5363 | * to continue to serve a useful existence. Next time it's released, | |
5364 | * we will get notified again, if it still has 'notify_on_release' set. | |
5365 | * | |
5366 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5367 | * means only wait until the task is successfully execve()'d. The | |
5368 | * separate release agent task is forked by call_usermodehelper(), | |
5369 | * then control in this thread returns here, without waiting for the | |
5370 | * release agent task. We don't bother to wait because the caller of | |
5371 | * this routine has no use for the exit status of the release agent | |
5372 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5373 | */ |
81a6a5cd PM |
5374 | static void cgroup_release_agent(struct work_struct *work) |
5375 | { | |
5376 | BUG_ON(work != &release_agent_work); | |
5377 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 5378 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
5379 | while (!list_empty(&release_list)) { |
5380 | char *argv[3], *envp[3]; | |
5381 | int i; | |
e788e066 | 5382 | char *pathbuf = NULL, *agentbuf = NULL; |
bd89aabc | 5383 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
5384 | struct cgroup, |
5385 | release_list); | |
bd89aabc | 5386 | list_del_init(&cgrp->release_list); |
cdcc136f | 5387 | raw_spin_unlock(&release_list_lock); |
81a6a5cd | 5388 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
e788e066 PM |
5389 | if (!pathbuf) |
5390 | goto continue_free; | |
5391 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
5392 | goto continue_free; | |
5393 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5394 | if (!agentbuf) | |
5395 | goto continue_free; | |
81a6a5cd PM |
5396 | |
5397 | i = 0; | |
e788e066 PM |
5398 | argv[i++] = agentbuf; |
5399 | argv[i++] = pathbuf; | |
81a6a5cd PM |
5400 | argv[i] = NULL; |
5401 | ||
5402 | i = 0; | |
5403 | /* minimal command environment */ | |
5404 | envp[i++] = "HOME=/"; | |
5405 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5406 | envp[i] = NULL; | |
5407 | ||
5408 | /* Drop the lock while we invoke the usermode helper, | |
5409 | * since the exec could involve hitting disk and hence | |
5410 | * be a slow process */ | |
5411 | mutex_unlock(&cgroup_mutex); | |
5412 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 5413 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
5414 | continue_free: |
5415 | kfree(pathbuf); | |
5416 | kfree(agentbuf); | |
cdcc136f | 5417 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 5418 | } |
cdcc136f | 5419 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5420 | mutex_unlock(&cgroup_mutex); |
5421 | } | |
8bab8dde PM |
5422 | |
5423 | static int __init cgroup_disable(char *str) | |
5424 | { | |
30159ec7 | 5425 | struct cgroup_subsys *ss; |
8bab8dde | 5426 | char *token; |
30159ec7 | 5427 | int i; |
8bab8dde PM |
5428 | |
5429 | while ((token = strsep(&str, ",")) != NULL) { | |
5430 | if (!*token) | |
5431 | continue; | |
be45c900 | 5432 | |
30159ec7 TH |
5433 | /* |
5434 | * cgroup_disable, being at boot time, can't know about | |
5435 | * module subsystems, so we don't worry about them. | |
5436 | */ | |
5437 | for_each_builtin_subsys(ss, i) { | |
8bab8dde PM |
5438 | if (!strcmp(token, ss->name)) { |
5439 | ss->disabled = 1; | |
5440 | printk(KERN_INFO "Disabling %s control group" | |
5441 | " subsystem\n", ss->name); | |
5442 | break; | |
5443 | } | |
5444 | } | |
5445 | } | |
5446 | return 1; | |
5447 | } | |
5448 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5449 | |
b77d7b60 | 5450 | /** |
35cf0836 TH |
5451 | * css_from_dir - get corresponding css from the dentry of a cgroup dir |
5452 | * @dentry: directory dentry of interest | |
5453 | * @ss: subsystem of interest | |
b77d7b60 TH |
5454 | * |
5455 | * Must be called under RCU read lock. The caller is responsible for | |
5456 | * pinning the returned css if it needs to be accessed outside the RCU | |
5457 | * critical section. | |
e5d1367f | 5458 | */ |
35cf0836 TH |
5459 | struct cgroup_subsys_state *css_from_dir(struct dentry *dentry, |
5460 | struct cgroup_subsys *ss) | |
e5d1367f SE |
5461 | { |
5462 | struct cgroup *cgrp; | |
e5d1367f | 5463 | |
b77d7b60 TH |
5464 | WARN_ON_ONCE(!rcu_read_lock_held()); |
5465 | ||
35cf0836 TH |
5466 | /* is @dentry a cgroup dir? */ |
5467 | if (!dentry->d_inode || | |
5468 | dentry->d_inode->i_op != &cgroup_dir_inode_operations) | |
e5d1367f SE |
5469 | return ERR_PTR(-EBADF); |
5470 | ||
35cf0836 | 5471 | cgrp = __d_cgrp(dentry); |
ca8bdcaf | 5472 | return cgroup_css(cgrp, ss) ?: ERR_PTR(-ENOENT); |
e5d1367f | 5473 | } |
e5d1367f | 5474 | |
1cb650b9 LZ |
5475 | /** |
5476 | * css_from_id - lookup css by id | |
5477 | * @id: the cgroup id | |
5478 | * @ss: cgroup subsys to be looked into | |
5479 | * | |
5480 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5481 | * Should be called under rcu_read_lock(). | |
5482 | */ | |
5483 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5484 | { | |
5485 | struct cgroup *cgrp; | |
5486 | ||
5487 | rcu_lockdep_assert(rcu_read_lock_held() || | |
5488 | lockdep_is_held(&cgroup_mutex), | |
5489 | "css_from_id() needs proper protection"); | |
5490 | ||
5491 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
5492 | if (cgrp) | |
d1625964 | 5493 | return cgroup_css(cgrp, ss); |
1cb650b9 | 5494 | return NULL; |
e5d1367f SE |
5495 | } |
5496 | ||
fe693435 | 5497 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5498 | static struct cgroup_subsys_state * |
5499 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5500 | { |
5501 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5502 | ||
5503 | if (!css) | |
5504 | return ERR_PTR(-ENOMEM); | |
5505 | ||
5506 | return css; | |
5507 | } | |
5508 | ||
eb95419b | 5509 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5510 | { |
eb95419b | 5511 | kfree(css); |
fe693435 PM |
5512 | } |
5513 | ||
182446d0 TH |
5514 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5515 | struct cftype *cft) | |
fe693435 | 5516 | { |
182446d0 | 5517 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5518 | } |
5519 | ||
182446d0 TH |
5520 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5521 | struct cftype *cft) | |
fe693435 PM |
5522 | { |
5523 | return (u64)(unsigned long)current->cgroups; | |
5524 | } | |
5525 | ||
182446d0 | 5526 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5527 | struct cftype *cft) |
fe693435 PM |
5528 | { |
5529 | u64 count; | |
5530 | ||
5531 | rcu_read_lock(); | |
a8ad805c | 5532 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5533 | rcu_read_unlock(); |
5534 | return count; | |
5535 | } | |
5536 | ||
182446d0 | 5537 | static int current_css_set_cg_links_read(struct cgroup_subsys_state *css, |
7717f7ba PM |
5538 | struct cftype *cft, |
5539 | struct seq_file *seq) | |
5540 | { | |
69d0206c | 5541 | struct cgrp_cset_link *link; |
5abb8855 | 5542 | struct css_set *cset; |
7717f7ba PM |
5543 | |
5544 | read_lock(&css_set_lock); | |
5545 | rcu_read_lock(); | |
5abb8855 | 5546 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5547 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba PM |
5548 | struct cgroup *c = link->cgrp; |
5549 | const char *name; | |
5550 | ||
5551 | if (c->dentry) | |
5552 | name = c->dentry->d_name.name; | |
5553 | else | |
5554 | name = "?"; | |
2c6ab6d2 PM |
5555 | seq_printf(seq, "Root %d group %s\n", |
5556 | c->root->hierarchy_id, name); | |
7717f7ba PM |
5557 | } |
5558 | rcu_read_unlock(); | |
5559 | read_unlock(&css_set_lock); | |
5560 | return 0; | |
5561 | } | |
5562 | ||
5563 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
182446d0 TH |
5564 | static int cgroup_css_links_read(struct cgroup_subsys_state *css, |
5565 | struct cftype *cft, struct seq_file *seq) | |
7717f7ba | 5566 | { |
69d0206c | 5567 | struct cgrp_cset_link *link; |
7717f7ba PM |
5568 | |
5569 | read_lock(&css_set_lock); | |
182446d0 | 5570 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5571 | struct css_set *cset = link->cset; |
7717f7ba PM |
5572 | struct task_struct *task; |
5573 | int count = 0; | |
5abb8855 TH |
5574 | seq_printf(seq, "css_set %p\n", cset); |
5575 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
7717f7ba PM |
5576 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { |
5577 | seq_puts(seq, " ...\n"); | |
5578 | break; | |
5579 | } else { | |
5580 | seq_printf(seq, " task %d\n", | |
5581 | task_pid_vnr(task)); | |
5582 | } | |
5583 | } | |
5584 | } | |
5585 | read_unlock(&css_set_lock); | |
5586 | return 0; | |
5587 | } | |
5588 | ||
182446d0 | 5589 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5590 | { |
182446d0 | 5591 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
5592 | } |
5593 | ||
5594 | static struct cftype debug_files[] = { | |
fe693435 PM |
5595 | { |
5596 | .name = "taskcount", | |
5597 | .read_u64 = debug_taskcount_read, | |
5598 | }, | |
5599 | ||
5600 | { | |
5601 | .name = "current_css_set", | |
5602 | .read_u64 = current_css_set_read, | |
5603 | }, | |
5604 | ||
5605 | { | |
5606 | .name = "current_css_set_refcount", | |
5607 | .read_u64 = current_css_set_refcount_read, | |
5608 | }, | |
5609 | ||
7717f7ba PM |
5610 | { |
5611 | .name = "current_css_set_cg_links", | |
5612 | .read_seq_string = current_css_set_cg_links_read, | |
5613 | }, | |
5614 | ||
5615 | { | |
5616 | .name = "cgroup_css_links", | |
5617 | .read_seq_string = cgroup_css_links_read, | |
5618 | }, | |
5619 | ||
fe693435 PM |
5620 | { |
5621 | .name = "releasable", | |
5622 | .read_u64 = releasable_read, | |
5623 | }, | |
fe693435 | 5624 | |
4baf6e33 TH |
5625 | { } /* terminate */ |
5626 | }; | |
fe693435 PM |
5627 | |
5628 | struct cgroup_subsys debug_subsys = { | |
5629 | .name = "debug", | |
92fb9748 TH |
5630 | .css_alloc = debug_css_alloc, |
5631 | .css_free = debug_css_free, | |
fe693435 | 5632 | .subsys_id = debug_subsys_id, |
4baf6e33 | 5633 | .base_cftypes = debug_files, |
fe693435 PM |
5634 | }; |
5635 | #endif /* CONFIG_CGROUP_DEBUG */ |