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/slab.h> |
45 | #include <linux/magic.h> | |
46 | #include <linux/spinlock.h> | |
47 | #include <linux/string.h> | |
bbcb81d0 | 48 | #include <linux/sort.h> |
81a6a5cd | 49 | #include <linux/kmod.h> |
846c7bb0 BS |
50 | #include <linux/delayacct.h> |
51 | #include <linux/cgroupstats.h> | |
0ac801fe | 52 | #include <linux/hashtable.h> |
3f8206d4 | 53 | #include <linux/namei.h> |
096b7fe0 | 54 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 55 | #include <linux/idr.h> |
d1d9fd33 | 56 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
081aa458 | 57 | #include <linux/flex_array.h> /* used in cgroup_attach_task */ |
c4c27fbd | 58 | #include <linux/kthread.h> |
846c7bb0 | 59 | |
60063497 | 60 | #include <linux/atomic.h> |
ddbcc7e8 | 61 | |
b1a21367 TH |
62 | /* |
63 | * pidlists linger the following amount before being destroyed. The goal | |
64 | * is avoiding frequent destruction in the middle of consecutive read calls | |
65 | * Expiring in the middle is a performance problem not a correctness one. | |
66 | * 1 sec should be enough. | |
67 | */ | |
68 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
69 | ||
8d7e6fb0 TH |
70 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
71 | MAX_CFTYPE_NAME + 2) | |
72 | ||
ace2bee8 TH |
73 | /* |
74 | * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file | |
75 | * creation/removal and hierarchy changing operations including cgroup | |
76 | * creation, removal, css association and controller rebinding. This outer | |
77 | * lock is needed mainly to resolve the circular dependency between kernfs | |
78 | * active ref and cgroup_mutex. cgroup_tree_mutex nests above both. | |
79 | */ | |
80 | static DEFINE_MUTEX(cgroup_tree_mutex); | |
81 | ||
e25e2cbb TH |
82 | /* |
83 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
84 | * hierarchy must be performed while holding it. | |
e25e2cbb | 85 | */ |
2219449a TH |
86 | #ifdef CONFIG_PROVE_RCU |
87 | DEFINE_MUTEX(cgroup_mutex); | |
8af01f56 | 88 | EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for lockdep */ |
2219449a | 89 | #else |
81a6a5cd | 90 | static DEFINE_MUTEX(cgroup_mutex); |
2219449a TH |
91 | #endif |
92 | ||
69e943b7 TH |
93 | /* |
94 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
95 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
96 | */ | |
97 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
98 | ||
ace2bee8 | 99 | #define cgroup_assert_mutexes_or_rcu_locked() \ |
87fb54f1 | 100 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
ace2bee8 | 101 | lockdep_is_held(&cgroup_tree_mutex) || \ |
87fb54f1 | 102 | lockdep_is_held(&cgroup_mutex), \ |
ace2bee8 | 103 | "cgroup_[tree_]mutex or RCU read lock required"); |
87fb54f1 | 104 | |
e5fca243 TH |
105 | /* |
106 | * cgroup destruction makes heavy use of work items and there can be a lot | |
107 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
108 | * destruction work items don't end up filling up max_active of system_wq | |
109 | * which may lead to deadlock. | |
110 | */ | |
111 | static struct workqueue_struct *cgroup_destroy_wq; | |
112 | ||
b1a21367 TH |
113 | /* |
114 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
115 | * separate workqueue as flush domain. | |
116 | */ | |
117 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
118 | ||
3ed80a62 | 119 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 120 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 121 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
122 | #include <linux/cgroup_subsys.h> |
123 | }; | |
073219e9 TH |
124 | #undef SUBSYS |
125 | ||
126 | /* array of cgroup subsystem names */ | |
127 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
128 | static const char *cgroup_subsys_name[] = { | |
129 | #include <linux/cgroup_subsys.h> | |
130 | }; | |
131 | #undef SUBSYS | |
ddbcc7e8 | 132 | |
ddbcc7e8 | 133 | /* |
9871bf95 TH |
134 | * The dummy hierarchy, reserved for the subsystems that are otherwise |
135 | * unattached - it never has more than a single cgroup, and all tasks are | |
136 | * part of that cgroup. | |
ddbcc7e8 | 137 | */ |
9871bf95 TH |
138 | static struct cgroupfs_root cgroup_dummy_root; |
139 | ||
140 | /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */ | |
141 | static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup; | |
ddbcc7e8 PM |
142 | |
143 | /* The list of hierarchy roots */ | |
144 | ||
9871bf95 TH |
145 | static LIST_HEAD(cgroup_roots); |
146 | static int cgroup_root_count; | |
ddbcc7e8 | 147 | |
3417ae1f | 148 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 149 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 150 | |
65dff759 LZ |
151 | static struct cgroup_name root_cgroup_name = { .name = "/" }; |
152 | ||
794611a1 LZ |
153 | /* |
154 | * Assign a monotonically increasing serial number to cgroups. It | |
155 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
156 | * numbers. Also, as cgroups are always appended to the parent's | |
157 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
158 | * the ascending serial number order on the list. Protected by |
159 | * cgroup_mutex. | |
794611a1 | 160 | */ |
00356bd5 | 161 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 162 | |
ddbcc7e8 | 163 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
164 | * check for fork/exit handlers to call. This avoids us having to do |
165 | * extra work in the fork/exit path if none of the subsystems need to | |
166 | * be called. | |
ddbcc7e8 | 167 | */ |
8947f9d5 | 168 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 169 | |
628f7cd4 TH |
170 | static struct cftype cgroup_base_files[]; |
171 | ||
f20104de | 172 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 173 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
2bb566cb TH |
174 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
175 | bool is_add); | |
e605b365 | 176 | static int cgroup_file_release(struct inode *inode, struct file *file); |
b1a21367 | 177 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 178 | |
95109b62 TH |
179 | /** |
180 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
181 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 182 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 183 | * |
ca8bdcaf TH |
184 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
185 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
186 | * the caller is responsible for pinning the returned css if it wants to | |
187 | * keep accessing it outside the said locks. This function may return | |
188 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
189 | */ |
190 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 191 | struct cgroup_subsys *ss) |
95109b62 | 192 | { |
ca8bdcaf | 193 | if (ss) |
aec25020 | 194 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 TH |
195 | lockdep_is_held(&cgroup_tree_mutex) || |
196 | lockdep_is_held(&cgroup_mutex)); | |
ca8bdcaf TH |
197 | else |
198 | return &cgrp->dummy_css; | |
95109b62 | 199 | } |
42809dd4 | 200 | |
ddbcc7e8 | 201 | /* convenient tests for these bits */ |
54766d4a | 202 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 203 | { |
54766d4a | 204 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
205 | } |
206 | ||
78574cf9 LZ |
207 | /** |
208 | * cgroup_is_descendant - test ancestry | |
209 | * @cgrp: the cgroup to be tested | |
210 | * @ancestor: possible ancestor of @cgrp | |
211 | * | |
212 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
213 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
214 | * and @ancestor are accessible. | |
215 | */ | |
216 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
217 | { | |
218 | while (cgrp) { | |
219 | if (cgrp == ancestor) | |
220 | return true; | |
221 | cgrp = cgrp->parent; | |
222 | } | |
223 | return false; | |
224 | } | |
225 | EXPORT_SYMBOL_GPL(cgroup_is_descendant); | |
ddbcc7e8 | 226 | |
e9685a03 | 227 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
228 | { |
229 | const int bits = | |
bd89aabc PM |
230 | (1 << CGRP_RELEASABLE) | |
231 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
232 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
233 | } |
234 | ||
e9685a03 | 235 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 236 | { |
bd89aabc | 237 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
238 | } |
239 | ||
1c6727af TH |
240 | /** |
241 | * for_each_css - iterate all css's of a cgroup | |
242 | * @css: the iteration cursor | |
243 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
244 | * @cgrp: the target cgroup to iterate css's of | |
245 | * | |
246 | * Should be called under cgroup_mutex. | |
247 | */ | |
248 | #define for_each_css(css, ssid, cgrp) \ | |
249 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
250 | if (!((css) = rcu_dereference_check( \ | |
251 | (cgrp)->subsys[(ssid)], \ | |
ace2bee8 | 252 | lockdep_is_held(&cgroup_tree_mutex) || \ |
1c6727af TH |
253 | lockdep_is_held(&cgroup_mutex)))) { } \ |
254 | else | |
255 | ||
30159ec7 | 256 | /** |
3ed80a62 | 257 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 258 | * @ss: the iteration cursor |
780cd8b3 | 259 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 260 | */ |
780cd8b3 | 261 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
262 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
263 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 264 | |
5549c497 TH |
265 | /* iterate across the active hierarchies */ |
266 | #define for_each_active_root(root) \ | |
267 | list_for_each_entry((root), &cgroup_roots, root_list) | |
ddbcc7e8 | 268 | |
f6ea9372 TH |
269 | static inline struct cgroup *__d_cgrp(struct dentry *dentry) |
270 | { | |
271 | return dentry->d_fsdata; | |
272 | } | |
273 | ||
05ef1d7c | 274 | static inline struct cfent *__d_cfe(struct dentry *dentry) |
f6ea9372 TH |
275 | { |
276 | return dentry->d_fsdata; | |
277 | } | |
278 | ||
05ef1d7c TH |
279 | static inline struct cftype *__d_cft(struct dentry *dentry) |
280 | { | |
281 | return __d_cfe(dentry)->type; | |
282 | } | |
283 | ||
7ae1bad9 TH |
284 | /** |
285 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
286 | * @cgrp: the cgroup to be checked for liveness | |
287 | * | |
47cfcd09 TH |
288 | * On success, returns true; the mutex should be later unlocked. On |
289 | * failure returns false with no lock held. | |
7ae1bad9 | 290 | */ |
b9777cf8 | 291 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
292 | { |
293 | mutex_lock(&cgroup_mutex); | |
54766d4a | 294 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
295 | mutex_unlock(&cgroup_mutex); |
296 | return false; | |
297 | } | |
298 | return true; | |
299 | } | |
7ae1bad9 | 300 | |
81a6a5cd PM |
301 | /* the list of cgroups eligible for automatic release. Protected by |
302 | * release_list_lock */ | |
303 | static LIST_HEAD(release_list); | |
cdcc136f | 304 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
305 | static void cgroup_release_agent(struct work_struct *work); |
306 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 307 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 308 | |
69d0206c TH |
309 | /* |
310 | * A cgroup can be associated with multiple css_sets as different tasks may | |
311 | * belong to different cgroups on different hierarchies. In the other | |
312 | * direction, a css_set is naturally associated with multiple cgroups. | |
313 | * This M:N relationship is represented by the following link structure | |
314 | * which exists for each association and allows traversing the associations | |
315 | * from both sides. | |
316 | */ | |
317 | struct cgrp_cset_link { | |
318 | /* the cgroup and css_set this link associates */ | |
319 | struct cgroup *cgrp; | |
320 | struct css_set *cset; | |
321 | ||
322 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
323 | struct list_head cset_link; | |
324 | ||
325 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
326 | struct list_head cgrp_link; | |
817929ec PM |
327 | }; |
328 | ||
329 | /* The default css_set - used by init and its children prior to any | |
330 | * hierarchies being mounted. It contains a pointer to the root state | |
331 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
332 | * reference-counted, to improve performance when child cgroups | |
333 | * haven't been created. | |
334 | */ | |
335 | ||
336 | static struct css_set init_css_set; | |
69d0206c | 337 | static struct cgrp_cset_link init_cgrp_cset_link; |
817929ec | 338 | |
0942eeee TH |
339 | /* |
340 | * css_set_lock protects the list of css_set objects, and the chain of | |
341 | * tasks off each css_set. Nests outside task->alloc_lock due to | |
72ec7029 | 342 | * css_task_iter_start(). |
0942eeee | 343 | */ |
817929ec PM |
344 | static DEFINE_RWLOCK(css_set_lock); |
345 | static int css_set_count; | |
346 | ||
7717f7ba PM |
347 | /* |
348 | * hash table for cgroup groups. This improves the performance to find | |
349 | * an existing css_set. This hash doesn't (currently) take into | |
350 | * account cgroups in empty hierarchies. | |
351 | */ | |
472b1053 | 352 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 353 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 354 | |
0ac801fe | 355 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 356 | { |
0ac801fe | 357 | unsigned long key = 0UL; |
30159ec7 TH |
358 | struct cgroup_subsys *ss; |
359 | int i; | |
472b1053 | 360 | |
30159ec7 | 361 | for_each_subsys(ss, i) |
0ac801fe LZ |
362 | key += (unsigned long)css[i]; |
363 | key = (key >> 16) ^ key; | |
472b1053 | 364 | |
0ac801fe | 365 | return key; |
472b1053 LZ |
366 | } |
367 | ||
0942eeee TH |
368 | /* |
369 | * We don't maintain the lists running through each css_set to its task | |
72ec7029 TH |
370 | * until after the first call to css_task_iter_start(). This reduces the |
371 | * fork()/exit() overhead for people who have cgroups compiled into their | |
372 | * kernel but not actually in use. | |
0942eeee | 373 | */ |
8947f9d5 | 374 | static int use_task_css_set_links __read_mostly; |
817929ec | 375 | |
5abb8855 | 376 | static void __put_css_set(struct css_set *cset, int taskexit) |
b4f48b63 | 377 | { |
69d0206c | 378 | struct cgrp_cset_link *link, *tmp_link; |
5abb8855 | 379 | |
146aa1bd LJ |
380 | /* |
381 | * Ensure that the refcount doesn't hit zero while any readers | |
382 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
383 | * rwlock | |
384 | */ | |
5abb8855 | 385 | if (atomic_add_unless(&cset->refcount, -1, 1)) |
146aa1bd LJ |
386 | return; |
387 | write_lock(&css_set_lock); | |
5abb8855 | 388 | if (!atomic_dec_and_test(&cset->refcount)) { |
146aa1bd LJ |
389 | write_unlock(&css_set_lock); |
390 | return; | |
391 | } | |
81a6a5cd | 392 | |
2c6ab6d2 | 393 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
5abb8855 | 394 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
395 | css_set_count--; |
396 | ||
69d0206c | 397 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 398 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 399 | |
69d0206c TH |
400 | list_del(&link->cset_link); |
401 | list_del(&link->cgrp_link); | |
71b5707e | 402 | |
ddd69148 | 403 | /* @cgrp can't go away while we're holding css_set_lock */ |
6f3d828f | 404 | if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { |
81a6a5cd | 405 | if (taskexit) |
bd89aabc PM |
406 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
407 | check_for_release(cgrp); | |
81a6a5cd | 408 | } |
2c6ab6d2 PM |
409 | |
410 | kfree(link); | |
81a6a5cd | 411 | } |
2c6ab6d2 PM |
412 | |
413 | write_unlock(&css_set_lock); | |
5abb8855 | 414 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
415 | } |
416 | ||
817929ec PM |
417 | /* |
418 | * refcounted get/put for css_set objects | |
419 | */ | |
5abb8855 | 420 | static inline void get_css_set(struct css_set *cset) |
817929ec | 421 | { |
5abb8855 | 422 | atomic_inc(&cset->refcount); |
817929ec PM |
423 | } |
424 | ||
5abb8855 | 425 | static inline void put_css_set(struct css_set *cset) |
817929ec | 426 | { |
5abb8855 | 427 | __put_css_set(cset, 0); |
817929ec PM |
428 | } |
429 | ||
5abb8855 | 430 | static inline void put_css_set_taskexit(struct css_set *cset) |
81a6a5cd | 431 | { |
5abb8855 | 432 | __put_css_set(cset, 1); |
81a6a5cd PM |
433 | } |
434 | ||
b326f9d0 | 435 | /** |
7717f7ba | 436 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
437 | * @cset: candidate css_set being tested |
438 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
439 | * @new_cgrp: cgroup that's being entered by the task |
440 | * @template: desired set of css pointers in css_set (pre-calculated) | |
441 | * | |
6f4b7e63 | 442 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
443 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
444 | */ | |
5abb8855 TH |
445 | static bool compare_css_sets(struct css_set *cset, |
446 | struct css_set *old_cset, | |
7717f7ba PM |
447 | struct cgroup *new_cgrp, |
448 | struct cgroup_subsys_state *template[]) | |
449 | { | |
450 | struct list_head *l1, *l2; | |
451 | ||
5abb8855 | 452 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) { |
7717f7ba PM |
453 | /* Not all subsystems matched */ |
454 | return false; | |
455 | } | |
456 | ||
457 | /* | |
458 | * Compare cgroup pointers in order to distinguish between | |
459 | * different cgroups in heirarchies with no subsystems. We | |
460 | * could get by with just this check alone (and skip the | |
461 | * memcmp above) but on most setups the memcmp check will | |
462 | * avoid the need for this more expensive check on almost all | |
463 | * candidates. | |
464 | */ | |
465 | ||
69d0206c TH |
466 | l1 = &cset->cgrp_links; |
467 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 468 | while (1) { |
69d0206c | 469 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 470 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
471 | |
472 | l1 = l1->next; | |
473 | l2 = l2->next; | |
474 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
475 | if (l1 == &cset->cgrp_links) { |
476 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
477 | break; |
478 | } else { | |
69d0206c | 479 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
480 | } |
481 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
482 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
483 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
484 | cgrp1 = link1->cgrp; | |
485 | cgrp2 = link2->cgrp; | |
7717f7ba | 486 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 487 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
488 | |
489 | /* | |
490 | * If this hierarchy is the hierarchy of the cgroup | |
491 | * that's changing, then we need to check that this | |
492 | * css_set points to the new cgroup; if it's any other | |
493 | * hierarchy, then this css_set should point to the | |
494 | * same cgroup as the old css_set. | |
495 | */ | |
5abb8855 TH |
496 | if (cgrp1->root == new_cgrp->root) { |
497 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
498 | return false; |
499 | } else { | |
5abb8855 | 500 | if (cgrp1 != cgrp2) |
7717f7ba PM |
501 | return false; |
502 | } | |
503 | } | |
504 | return true; | |
505 | } | |
506 | ||
b326f9d0 TH |
507 | /** |
508 | * find_existing_css_set - init css array and find the matching css_set | |
509 | * @old_cset: the css_set that we're using before the cgroup transition | |
510 | * @cgrp: the cgroup that we're moving into | |
511 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 512 | */ |
5abb8855 TH |
513 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
514 | struct cgroup *cgrp, | |
515 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 516 | { |
bd89aabc | 517 | struct cgroupfs_root *root = cgrp->root; |
30159ec7 | 518 | struct cgroup_subsys *ss; |
5abb8855 | 519 | struct css_set *cset; |
0ac801fe | 520 | unsigned long key; |
b326f9d0 | 521 | int i; |
817929ec | 522 | |
aae8aab4 BB |
523 | /* |
524 | * Build the set of subsystem state objects that we want to see in the | |
525 | * new css_set. while subsystems can change globally, the entries here | |
526 | * won't change, so no need for locking. | |
527 | */ | |
30159ec7 | 528 | for_each_subsys(ss, i) { |
a1a71b45 | 529 | if (root->subsys_mask & (1UL << i)) { |
817929ec PM |
530 | /* Subsystem is in this hierarchy. So we want |
531 | * the subsystem state from the new | |
532 | * cgroup */ | |
ca8bdcaf | 533 | template[i] = cgroup_css(cgrp, ss); |
817929ec PM |
534 | } else { |
535 | /* Subsystem is not in this hierarchy, so we | |
536 | * don't want to change the subsystem state */ | |
5abb8855 | 537 | template[i] = old_cset->subsys[i]; |
817929ec PM |
538 | } |
539 | } | |
540 | ||
0ac801fe | 541 | key = css_set_hash(template); |
5abb8855 TH |
542 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
543 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
544 | continue; |
545 | ||
546 | /* This css_set matches what we need */ | |
5abb8855 | 547 | return cset; |
472b1053 | 548 | } |
817929ec PM |
549 | |
550 | /* No existing cgroup group matched */ | |
551 | return NULL; | |
552 | } | |
553 | ||
69d0206c | 554 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 555 | { |
69d0206c | 556 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 557 | |
69d0206c TH |
558 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
559 | list_del(&link->cset_link); | |
36553434 LZ |
560 | kfree(link); |
561 | } | |
562 | } | |
563 | ||
69d0206c TH |
564 | /** |
565 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
566 | * @count: the number of links to allocate | |
567 | * @tmp_links: list_head the allocated links are put on | |
568 | * | |
569 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
570 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 571 | */ |
69d0206c | 572 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 573 | { |
69d0206c | 574 | struct cgrp_cset_link *link; |
817929ec | 575 | int i; |
69d0206c TH |
576 | |
577 | INIT_LIST_HEAD(tmp_links); | |
578 | ||
817929ec | 579 | for (i = 0; i < count; i++) { |
f4f4be2b | 580 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 581 | if (!link) { |
69d0206c | 582 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
583 | return -ENOMEM; |
584 | } | |
69d0206c | 585 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
586 | } |
587 | return 0; | |
588 | } | |
589 | ||
c12f65d4 LZ |
590 | /** |
591 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 592 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 593 | * @cset: the css_set to be linked |
c12f65d4 LZ |
594 | * @cgrp: the destination cgroup |
595 | */ | |
69d0206c TH |
596 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
597 | struct cgroup *cgrp) | |
c12f65d4 | 598 | { |
69d0206c | 599 | struct cgrp_cset_link *link; |
c12f65d4 | 600 | |
69d0206c TH |
601 | BUG_ON(list_empty(tmp_links)); |
602 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); | |
603 | link->cset = cset; | |
7717f7ba | 604 | link->cgrp = cgrp; |
69d0206c | 605 | list_move(&link->cset_link, &cgrp->cset_links); |
7717f7ba PM |
606 | /* |
607 | * Always add links to the tail of the list so that the list | |
608 | * is sorted by order of hierarchy creation | |
609 | */ | |
69d0206c | 610 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
611 | } |
612 | ||
b326f9d0 TH |
613 | /** |
614 | * find_css_set - return a new css_set with one cgroup updated | |
615 | * @old_cset: the baseline css_set | |
616 | * @cgrp: the cgroup to be updated | |
617 | * | |
618 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
619 | * substituted into the appropriate hierarchy. | |
817929ec | 620 | */ |
5abb8855 TH |
621 | static struct css_set *find_css_set(struct css_set *old_cset, |
622 | struct cgroup *cgrp) | |
817929ec | 623 | { |
b326f9d0 | 624 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 625 | struct css_set *cset; |
69d0206c TH |
626 | struct list_head tmp_links; |
627 | struct cgrp_cset_link *link; | |
0ac801fe | 628 | unsigned long key; |
472b1053 | 629 | |
b326f9d0 TH |
630 | lockdep_assert_held(&cgroup_mutex); |
631 | ||
817929ec PM |
632 | /* First see if we already have a cgroup group that matches |
633 | * the desired set */ | |
7e9abd89 | 634 | read_lock(&css_set_lock); |
5abb8855 TH |
635 | cset = find_existing_css_set(old_cset, cgrp, template); |
636 | if (cset) | |
637 | get_css_set(cset); | |
7e9abd89 | 638 | read_unlock(&css_set_lock); |
817929ec | 639 | |
5abb8855 TH |
640 | if (cset) |
641 | return cset; | |
817929ec | 642 | |
f4f4be2b | 643 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 644 | if (!cset) |
817929ec PM |
645 | return NULL; |
646 | ||
69d0206c | 647 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 648 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 649 | kfree(cset); |
817929ec PM |
650 | return NULL; |
651 | } | |
652 | ||
5abb8855 | 653 | atomic_set(&cset->refcount, 1); |
69d0206c | 654 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 TH |
655 | INIT_LIST_HEAD(&cset->tasks); |
656 | INIT_HLIST_NODE(&cset->hlist); | |
817929ec PM |
657 | |
658 | /* Copy the set of subsystem state objects generated in | |
659 | * find_existing_css_set() */ | |
5abb8855 | 660 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec PM |
661 | |
662 | write_lock(&css_set_lock); | |
663 | /* Add reference counts and links from the new css_set. */ | |
69d0206c | 664 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 665 | struct cgroup *c = link->cgrp; |
69d0206c | 666 | |
7717f7ba PM |
667 | if (c->root == cgrp->root) |
668 | c = cgrp; | |
69d0206c | 669 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 670 | } |
817929ec | 671 | |
69d0206c | 672 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 673 | |
817929ec | 674 | css_set_count++; |
472b1053 LZ |
675 | |
676 | /* Add this cgroup group to the hash table */ | |
5abb8855 TH |
677 | key = css_set_hash(cset->subsys); |
678 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 679 | |
817929ec PM |
680 | write_unlock(&css_set_lock); |
681 | ||
5abb8855 | 682 | return cset; |
b4f48b63 PM |
683 | } |
684 | ||
7717f7ba PM |
685 | /* |
686 | * Return the cgroup for "task" from the given hierarchy. Must be | |
687 | * called with cgroup_mutex held. | |
688 | */ | |
689 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
690 | struct cgroupfs_root *root) | |
691 | { | |
5abb8855 | 692 | struct css_set *cset; |
7717f7ba PM |
693 | struct cgroup *res = NULL; |
694 | ||
695 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
696 | read_lock(&css_set_lock); | |
697 | /* | |
698 | * No need to lock the task - since we hold cgroup_mutex the | |
699 | * task can't change groups, so the only thing that can happen | |
700 | * is that it exits and its css is set back to init_css_set. | |
701 | */ | |
a8ad805c | 702 | cset = task_css_set(task); |
5abb8855 | 703 | if (cset == &init_css_set) { |
7717f7ba PM |
704 | res = &root->top_cgroup; |
705 | } else { | |
69d0206c TH |
706 | struct cgrp_cset_link *link; |
707 | ||
708 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 709 | struct cgroup *c = link->cgrp; |
69d0206c | 710 | |
7717f7ba PM |
711 | if (c->root == root) { |
712 | res = c; | |
713 | break; | |
714 | } | |
715 | } | |
716 | } | |
717 | read_unlock(&css_set_lock); | |
718 | BUG_ON(!res); | |
719 | return res; | |
720 | } | |
721 | ||
ddbcc7e8 PM |
722 | /* |
723 | * There is one global cgroup mutex. We also require taking | |
724 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
725 | * See "The task_lock() exception", at the end of this comment. | |
726 | * | |
727 | * A task must hold cgroup_mutex to modify cgroups. | |
728 | * | |
729 | * Any task can increment and decrement the count field without lock. | |
730 | * So in general, code holding cgroup_mutex can't rely on the count | |
731 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 732 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
733 | * means that no tasks are currently attached, therefore there is no |
734 | * way a task attached to that cgroup can fork (the other way to | |
735 | * increment the count). So code holding cgroup_mutex can safely | |
736 | * assume that if the count is zero, it will stay zero. Similarly, if | |
737 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
738 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
739 | * needs that mutex. | |
740 | * | |
ddbcc7e8 PM |
741 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
742 | * (usually) take cgroup_mutex. These are the two most performance | |
743 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
744 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
745 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
746 | * to the release agent with the name of the cgroup (path relative to |
747 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
748 | * |
749 | * A cgroup can only be deleted if both its 'count' of using tasks | |
750 | * is zero, and its list of 'children' cgroups is empty. Since all | |
751 | * tasks in the system use _some_ cgroup, and since there is always at | |
752 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
753 | * always has either children cgroups and/or using tasks. So we don't | |
754 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
755 | * | |
756 | * The task_lock() exception | |
757 | * | |
758 | * The need for this exception arises from the action of | |
d0b2fdd2 | 759 | * cgroup_attach_task(), which overwrites one task's cgroup pointer with |
a043e3b2 | 760 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
761 | * several performance critical places that need to reference |
762 | * task->cgroup without the expense of grabbing a system global | |
763 | * mutex. Therefore except as noted below, when dereferencing or, as | |
d0b2fdd2 | 764 | * in cgroup_attach_task(), modifying a task's cgroup pointer we use |
ddbcc7e8 PM |
765 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
766 | * the task_struct routinely used for such matters. | |
767 | * | |
768 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 769 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
770 | */ |
771 | ||
ddbcc7e8 PM |
772 | /* |
773 | * A couple of forward declarations required, due to cyclic reference loop: | |
774 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> | |
775 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations | |
776 | * -> cgroup_mkdir. | |
777 | */ | |
778 | ||
18bb1db3 | 779 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode); |
ddbcc7e8 | 780 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); |
628f7cd4 | 781 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
6e1d5dcc | 782 | static const struct inode_operations cgroup_dir_inode_operations; |
828c0950 | 783 | static const struct file_operations proc_cgroupstats_operations; |
a424316c PM |
784 | |
785 | static struct backing_dev_info cgroup_backing_dev_info = { | |
d993831f | 786 | .name = "cgroup", |
e4ad08fe | 787 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
a424316c | 788 | }; |
ddbcc7e8 | 789 | |
a5e7ed32 | 790 | static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb) |
ddbcc7e8 PM |
791 | { |
792 | struct inode *inode = new_inode(sb); | |
ddbcc7e8 PM |
793 | |
794 | if (inode) { | |
85fe4025 | 795 | inode->i_ino = get_next_ino(); |
ddbcc7e8 | 796 | inode->i_mode = mode; |
76aac0e9 DH |
797 | inode->i_uid = current_fsuid(); |
798 | inode->i_gid = current_fsgid(); | |
ddbcc7e8 PM |
799 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
800 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; | |
801 | } | |
802 | return inode; | |
803 | } | |
804 | ||
8d7e6fb0 | 805 | static struct cgroup_name *cgroup_alloc_name(const char *name_str) |
65dff759 LZ |
806 | { |
807 | struct cgroup_name *name; | |
808 | ||
8d7e6fb0 | 809 | name = kmalloc(sizeof(*name) + strlen(name_str) + 1, GFP_KERNEL); |
65dff759 LZ |
810 | if (!name) |
811 | return NULL; | |
8d7e6fb0 | 812 | strcpy(name->name, name_str); |
65dff759 LZ |
813 | return name; |
814 | } | |
815 | ||
8d7e6fb0 TH |
816 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
817 | char *buf) | |
818 | { | |
819 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && | |
820 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
821 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
822 | cft->ss->name, cft->name); | |
823 | else | |
824 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
825 | return buf; | |
826 | } | |
827 | ||
be445626 LZ |
828 | static void cgroup_free_fn(struct work_struct *work) |
829 | { | |
ea15f8cc | 830 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 LZ |
831 | |
832 | mutex_lock(&cgroup_mutex); | |
be445626 LZ |
833 | cgrp->root->number_of_cgroups--; |
834 | mutex_unlock(&cgroup_mutex); | |
835 | ||
415cf07a LZ |
836 | /* |
837 | * We get a ref to the parent's dentry, and put the ref when | |
838 | * this cgroup is being freed, so it's guaranteed that the | |
839 | * parent won't be destroyed before its children. | |
840 | */ | |
841 | dput(cgrp->parent->dentry); | |
842 | ||
be445626 LZ |
843 | /* |
844 | * Drop the active superblock reference that we took when we | |
cc20e01c LZ |
845 | * created the cgroup. This will free cgrp->root, if we are |
846 | * holding the last reference to @sb. | |
be445626 LZ |
847 | */ |
848 | deactivate_super(cgrp->root->sb); | |
849 | ||
b1a21367 | 850 | cgroup_pidlist_destroy_all(cgrp); |
be445626 LZ |
851 | |
852 | simple_xattrs_free(&cgrp->xattrs); | |
853 | ||
65dff759 | 854 | kfree(rcu_dereference_raw(cgrp->name)); |
be445626 LZ |
855 | kfree(cgrp); |
856 | } | |
857 | ||
858 | static void cgroup_free_rcu(struct rcu_head *head) | |
859 | { | |
860 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
861 | ||
ea15f8cc | 862 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 863 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
864 | } |
865 | ||
ddbcc7e8 PM |
866 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) |
867 | { | |
868 | /* is dentry a directory ? if so, kfree() associated cgroup */ | |
869 | if (S_ISDIR(inode->i_mode)) { | |
bd89aabc | 870 | struct cgroup *cgrp = dentry->d_fsdata; |
be445626 | 871 | |
54766d4a | 872 | BUG_ON(!(cgroup_is_dead(cgrp))); |
c1a71504 LZ |
873 | |
874 | /* | |
875 | * XXX: cgrp->id is only used to look up css's. As cgroup | |
876 | * and css's lifetimes will be decoupled, it should be made | |
877 | * per-subsystem and moved to css->id so that lookups are | |
878 | * successful until the target css is released. | |
879 | */ | |
0ab02ca8 | 880 | mutex_lock(&cgroup_mutex); |
c1a71504 | 881 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
0ab02ca8 | 882 | mutex_unlock(&cgroup_mutex); |
c1a71504 LZ |
883 | cgrp->id = -1; |
884 | ||
be445626 | 885 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
05ef1d7c TH |
886 | } else { |
887 | struct cfent *cfe = __d_cfe(dentry); | |
888 | struct cgroup *cgrp = dentry->d_parent->d_fsdata; | |
889 | ||
890 | WARN_ONCE(!list_empty(&cfe->node) && | |
891 | cgrp != &cgrp->root->top_cgroup, | |
892 | "cfe still linked for %s\n", cfe->type->name); | |
712317ad | 893 | simple_xattrs_free(&cfe->xattrs); |
05ef1d7c | 894 | kfree(cfe); |
ddbcc7e8 PM |
895 | } |
896 | iput(inode); | |
897 | } | |
898 | ||
899 | static void remove_dir(struct dentry *d) | |
900 | { | |
901 | struct dentry *parent = dget(d->d_parent); | |
902 | ||
903 | d_delete(d); | |
904 | simple_rmdir(parent->d_inode, d); | |
905 | dput(parent); | |
906 | } | |
907 | ||
2739d3cc | 908 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c TH |
909 | { |
910 | struct cfent *cfe; | |
911 | ||
912 | lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex); | |
ace2bee8 | 913 | lockdep_assert_held(&cgroup_tree_mutex); |
05ef1d7c | 914 | |
2739d3cc LZ |
915 | /* |
916 | * If we're doing cleanup due to failure of cgroup_create(), | |
917 | * the corresponding @cfe may not exist. | |
918 | */ | |
05ef1d7c TH |
919 | list_for_each_entry(cfe, &cgrp->files, node) { |
920 | struct dentry *d = cfe->dentry; | |
921 | ||
922 | if (cft && cfe->type != cft) | |
923 | continue; | |
924 | ||
925 | dget(d); | |
926 | d_delete(d); | |
ce27e317 | 927 | simple_unlink(cgrp->dentry->d_inode, d); |
05ef1d7c TH |
928 | list_del_init(&cfe->node); |
929 | dput(d); | |
930 | ||
2739d3cc | 931 | break; |
ddbcc7e8 | 932 | } |
05ef1d7c TH |
933 | } |
934 | ||
13af07df | 935 | /** |
628f7cd4 | 936 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 937 | * @cgrp: target cgroup |
13af07df AR |
938 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
939 | */ | |
628f7cd4 | 940 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 941 | { |
13af07df | 942 | struct cgroup_subsys *ss; |
b420ba7d | 943 | int i; |
05ef1d7c | 944 | |
b420ba7d | 945 | for_each_subsys(ss, i) { |
13af07df | 946 | struct cftype_set *set; |
b420ba7d TH |
947 | |
948 | if (!test_bit(i, &subsys_mask)) | |
13af07df AR |
949 | continue; |
950 | list_for_each_entry(set, &ss->cftsets, node) | |
2bb566cb | 951 | cgroup_addrm_files(cgrp, set->cfts, false); |
13af07df | 952 | } |
ddbcc7e8 PM |
953 | } |
954 | ||
955 | /* | |
956 | * NOTE : the dentry must have been dget()'ed | |
957 | */ | |
958 | static void cgroup_d_remove_dir(struct dentry *dentry) | |
959 | { | |
2fd6b7f5 | 960 | struct dentry *parent; |
ddbcc7e8 | 961 | |
2fd6b7f5 NP |
962 | parent = dentry->d_parent; |
963 | spin_lock(&parent->d_lock); | |
3ec762ad | 964 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
ddbcc7e8 | 965 | list_del_init(&dentry->d_u.d_child); |
2fd6b7f5 NP |
966 | spin_unlock(&dentry->d_lock); |
967 | spin_unlock(&parent->d_lock); | |
ddbcc7e8 PM |
968 | remove_dir(dentry); |
969 | } | |
970 | ||
971 | static int rebind_subsystems(struct cgroupfs_root *root, | |
a8a648c4 | 972 | unsigned long added_mask, unsigned removed_mask) |
ddbcc7e8 | 973 | { |
bd89aabc | 974 | struct cgroup *cgrp = &root->top_cgroup; |
30159ec7 | 975 | struct cgroup_subsys *ss; |
3126121f | 976 | int i, ret; |
ddbcc7e8 | 977 | |
ace2bee8 TH |
978 | lockdep_assert_held(&cgroup_tree_mutex); |
979 | lockdep_assert_held(&cgroup_mutex); | |
aae8aab4 | 980 | |
ddbcc7e8 | 981 | /* Check that any added subsystems are currently free */ |
3ed80a62 TH |
982 | for_each_subsys(ss, i) |
983 | if ((added_mask & (1 << i)) && ss->root != &cgroup_dummy_root) | |
984 | return -EBUSY; | |
ddbcc7e8 | 985 | |
3126121f TH |
986 | ret = cgroup_populate_dir(cgrp, added_mask); |
987 | if (ret) | |
3ed80a62 | 988 | return ret; |
3126121f TH |
989 | |
990 | /* | |
991 | * Nothing can fail from this point on. Remove files for the | |
992 | * removed subsystems and rebind each subsystem. | |
993 | */ | |
4ac06017 | 994 | mutex_unlock(&cgroup_mutex); |
3126121f | 995 | cgroup_clear_dir(cgrp, removed_mask); |
4ac06017 | 996 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 997 | |
30159ec7 | 998 | for_each_subsys(ss, i) { |
ddbcc7e8 | 999 | unsigned long bit = 1UL << i; |
30159ec7 | 1000 | |
a1a71b45 | 1001 | if (bit & added_mask) { |
ddbcc7e8 | 1002 | /* We're binding this subsystem to this hierarchy */ |
ca8bdcaf TH |
1003 | BUG_ON(cgroup_css(cgrp, ss)); |
1004 | BUG_ON(!cgroup_css(cgroup_dummy_top, ss)); | |
1005 | BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top); | |
a8a648c4 | 1006 | |
73e80ed8 | 1007 | rcu_assign_pointer(cgrp->subsys[i], |
ca8bdcaf TH |
1008 | cgroup_css(cgroup_dummy_top, ss)); |
1009 | cgroup_css(cgrp, ss)->cgroup = cgrp; | |
a8a648c4 | 1010 | |
b2aa30f7 | 1011 | ss->root = root; |
ddbcc7e8 | 1012 | if (ss->bind) |
ca8bdcaf | 1013 | ss->bind(cgroup_css(cgrp, ss)); |
a8a648c4 | 1014 | |
cf5d5941 | 1015 | /* refcount was already taken, and we're keeping it */ |
a8a648c4 | 1016 | root->subsys_mask |= bit; |
a1a71b45 | 1017 | } else if (bit & removed_mask) { |
ddbcc7e8 | 1018 | /* We're removing this subsystem */ |
ca8bdcaf TH |
1019 | BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss)); |
1020 | BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp); | |
a8a648c4 | 1021 | |
ddbcc7e8 | 1022 | if (ss->bind) |
ca8bdcaf | 1023 | ss->bind(cgroup_css(cgroup_dummy_top, ss)); |
73e80ed8 | 1024 | |
ca8bdcaf | 1025 | cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top; |
73e80ed8 TH |
1026 | RCU_INIT_POINTER(cgrp->subsys[i], NULL); |
1027 | ||
9871bf95 | 1028 | cgroup_subsys[i]->root = &cgroup_dummy_root; |
a8a648c4 | 1029 | root->subsys_mask &= ~bit; |
ddbcc7e8 PM |
1030 | } |
1031 | } | |
ddbcc7e8 | 1032 | |
1672d040 TH |
1033 | /* |
1034 | * Mark @root has finished binding subsystems. @root->subsys_mask | |
1035 | * now matches the bound subsystems. | |
1036 | */ | |
1037 | root->flags |= CGRP_ROOT_SUBSYS_BOUND; | |
1038 | ||
ddbcc7e8 PM |
1039 | return 0; |
1040 | } | |
1041 | ||
34c80b1d | 1042 | static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry) |
ddbcc7e8 | 1043 | { |
34c80b1d | 1044 | struct cgroupfs_root *root = dentry->d_sb->s_fs_info; |
ddbcc7e8 | 1045 | struct cgroup_subsys *ss; |
b85d2040 | 1046 | int ssid; |
ddbcc7e8 | 1047 | |
b85d2040 TH |
1048 | for_each_subsys(ss, ssid) |
1049 | if (root->subsys_mask & (1 << ssid)) | |
1050 | seq_printf(seq, ",%s", ss->name); | |
873fe09e TH |
1051 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1052 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1053 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1054 | seq_puts(seq, ",noprefix"); |
93438629 | 1055 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1056 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1057 | |
1058 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1059 | if (strlen(root->release_agent_path)) |
1060 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1061 | spin_unlock(&release_agent_path_lock); |
1062 | ||
2260e7fc | 1063 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags)) |
97978e6d | 1064 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1065 | if (strlen(root->name)) |
1066 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1067 | return 0; |
1068 | } | |
1069 | ||
1070 | struct cgroup_sb_opts { | |
a1a71b45 | 1071 | unsigned long subsys_mask; |
ddbcc7e8 | 1072 | unsigned long flags; |
81a6a5cd | 1073 | char *release_agent; |
2260e7fc | 1074 | bool cpuset_clone_children; |
c6d57f33 | 1075 | char *name; |
2c6ab6d2 PM |
1076 | /* User explicitly requested empty subsystem */ |
1077 | bool none; | |
c6d57f33 PM |
1078 | |
1079 | struct cgroupfs_root *new_root; | |
2c6ab6d2 | 1080 | |
ddbcc7e8 PM |
1081 | }; |
1082 | ||
aae8aab4 | 1083 | /* |
9871bf95 TH |
1084 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1085 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1086 | * array. This function takes refcounts on subsystems to be used, unless it | |
1087 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1088 | */ |
cf5d5941 | 1089 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1090 | { |
32a8cf23 DL |
1091 | char *token, *o = data; |
1092 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1093 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1094 | struct cgroup_subsys *ss; |
1095 | int i; | |
f9ab5b5b | 1096 | |
aae8aab4 BB |
1097 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1098 | ||
f9ab5b5b | 1099 | #ifdef CONFIG_CPUSETS |
073219e9 | 1100 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1101 | #endif |
ddbcc7e8 | 1102 | |
c6d57f33 | 1103 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1104 | |
1105 | while ((token = strsep(&o, ",")) != NULL) { | |
1106 | if (!*token) | |
1107 | return -EINVAL; | |
32a8cf23 | 1108 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1109 | /* Explicitly have no subsystems */ |
1110 | opts->none = true; | |
32a8cf23 DL |
1111 | continue; |
1112 | } | |
1113 | if (!strcmp(token, "all")) { | |
1114 | /* Mutually exclusive option 'all' + subsystem name */ | |
1115 | if (one_ss) | |
1116 | return -EINVAL; | |
1117 | all_ss = true; | |
1118 | continue; | |
1119 | } | |
873fe09e TH |
1120 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1121 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1122 | continue; | |
1123 | } | |
32a8cf23 | 1124 | if (!strcmp(token, "noprefix")) { |
93438629 | 1125 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1126 | continue; |
1127 | } | |
1128 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1129 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1130 | continue; |
1131 | } | |
03b1cde6 | 1132 | if (!strcmp(token, "xattr")) { |
93438629 | 1133 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1134 | continue; |
1135 | } | |
32a8cf23 | 1136 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1137 | /* Specifying two release agents is forbidden */ |
1138 | if (opts->release_agent) | |
1139 | return -EINVAL; | |
c6d57f33 | 1140 | opts->release_agent = |
e400c285 | 1141 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1142 | if (!opts->release_agent) |
1143 | return -ENOMEM; | |
32a8cf23 DL |
1144 | continue; |
1145 | } | |
1146 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1147 | const char *name = token + 5; |
1148 | /* Can't specify an empty name */ | |
1149 | if (!strlen(name)) | |
1150 | return -EINVAL; | |
1151 | /* Must match [\w.-]+ */ | |
1152 | for (i = 0; i < strlen(name); i++) { | |
1153 | char c = name[i]; | |
1154 | if (isalnum(c)) | |
1155 | continue; | |
1156 | if ((c == '.') || (c == '-') || (c == '_')) | |
1157 | continue; | |
1158 | return -EINVAL; | |
1159 | } | |
1160 | /* Specifying two names is forbidden */ | |
1161 | if (opts->name) | |
1162 | return -EINVAL; | |
1163 | opts->name = kstrndup(name, | |
e400c285 | 1164 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1165 | GFP_KERNEL); |
1166 | if (!opts->name) | |
1167 | return -ENOMEM; | |
32a8cf23 DL |
1168 | |
1169 | continue; | |
1170 | } | |
1171 | ||
30159ec7 | 1172 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1173 | if (strcmp(token, ss->name)) |
1174 | continue; | |
1175 | if (ss->disabled) | |
1176 | continue; | |
1177 | ||
1178 | /* Mutually exclusive option 'all' + subsystem name */ | |
1179 | if (all_ss) | |
1180 | return -EINVAL; | |
a1a71b45 | 1181 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1182 | one_ss = true; |
1183 | ||
1184 | break; | |
1185 | } | |
1186 | if (i == CGROUP_SUBSYS_COUNT) | |
1187 | return -ENOENT; | |
1188 | } | |
1189 | ||
1190 | /* | |
1191 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1192 | * otherwise if 'none', 'name=' and a subsystem name options |
1193 | * were not specified, let's default to 'all' | |
32a8cf23 | 1194 | */ |
30159ec7 TH |
1195 | if (all_ss || (!one_ss && !opts->none && !opts->name)) |
1196 | for_each_subsys(ss, i) | |
1197 | if (!ss->disabled) | |
1198 | set_bit(i, &opts->subsys_mask); | |
ddbcc7e8 | 1199 | |
2c6ab6d2 PM |
1200 | /* Consistency checks */ |
1201 | ||
873fe09e TH |
1202 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1203 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1204 | ||
1205 | if (opts->flags & CGRP_ROOT_NOPREFIX) { | |
1206 | pr_err("cgroup: sane_behavior: noprefix is not allowed\n"); | |
1207 | return -EINVAL; | |
1208 | } | |
1209 | ||
1210 | if (opts->cpuset_clone_children) { | |
1211 | pr_err("cgroup: sane_behavior: clone_children is not allowed\n"); | |
1212 | return -EINVAL; | |
1213 | } | |
1214 | } | |
1215 | ||
f9ab5b5b LZ |
1216 | /* |
1217 | * Option noprefix was introduced just for backward compatibility | |
1218 | * with the old cpuset, so we allow noprefix only if mounting just | |
1219 | * the cpuset subsystem. | |
1220 | */ | |
93438629 | 1221 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1222 | return -EINVAL; |
1223 | ||
2c6ab6d2 PM |
1224 | |
1225 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1226 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1227 | return -EINVAL; |
1228 | ||
1229 | /* | |
1230 | * We either have to specify by name or by subsystems. (So all | |
1231 | * empty hierarchies must have a name). | |
1232 | */ | |
a1a71b45 | 1233 | if (!opts->subsys_mask && !opts->name) |
ddbcc7e8 PM |
1234 | return -EINVAL; |
1235 | ||
1236 | return 0; | |
1237 | } | |
1238 | ||
1239 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) | |
1240 | { | |
1241 | int ret = 0; | |
1242 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1243 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 | 1244 | struct cgroup_sb_opts opts; |
a1a71b45 | 1245 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1246 | |
873fe09e TH |
1247 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1248 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1249 | return -EINVAL; | |
1250 | } | |
1251 | ||
bd89aabc | 1252 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
ace2bee8 | 1253 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1254 | mutex_lock(&cgroup_mutex); |
1255 | ||
1256 | /* See what subsystems are wanted */ | |
1257 | ret = parse_cgroupfs_options(data, &opts); | |
1258 | if (ret) | |
1259 | goto out_unlock; | |
1260 | ||
a8a648c4 | 1261 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1262 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1263 | task_tgid_nr(current), current->comm); | |
1264 | ||
a1a71b45 AR |
1265 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1266 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1267 | |
cf5d5941 | 1268 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1269 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1270 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1271 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1272 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1273 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1274 | ret = -EINVAL; |
1275 | goto out_unlock; | |
1276 | } | |
1277 | ||
f172e67c TH |
1278 | /* remounting is not allowed for populated hierarchies */ |
1279 | if (root->number_of_cgroups > 1) { | |
1280 | ret = -EBUSY; | |
0670e08b | 1281 | goto out_unlock; |
cf5d5941 | 1282 | } |
ddbcc7e8 | 1283 | |
a8a648c4 | 1284 | ret = rebind_subsystems(root, added_mask, removed_mask); |
3126121f | 1285 | if (ret) |
0670e08b | 1286 | goto out_unlock; |
ddbcc7e8 | 1287 | |
69e943b7 TH |
1288 | if (opts.release_agent) { |
1289 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1290 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1291 | spin_unlock(&release_agent_path_lock); |
1292 | } | |
ddbcc7e8 | 1293 | out_unlock: |
66bdc9cf | 1294 | kfree(opts.release_agent); |
c6d57f33 | 1295 | kfree(opts.name); |
ddbcc7e8 | 1296 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1297 | mutex_unlock(&cgroup_tree_mutex); |
bd89aabc | 1298 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
1299 | return ret; |
1300 | } | |
1301 | ||
b87221de | 1302 | static const struct super_operations cgroup_ops = { |
ddbcc7e8 PM |
1303 | .statfs = simple_statfs, |
1304 | .drop_inode = generic_delete_inode, | |
1305 | .show_options = cgroup_show_options, | |
1306 | .remount_fs = cgroup_remount, | |
1307 | }; | |
1308 | ||
cc31edce PM |
1309 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1310 | { | |
1311 | INIT_LIST_HEAD(&cgrp->sibling); | |
1312 | INIT_LIST_HEAD(&cgrp->children); | |
05ef1d7c | 1313 | INIT_LIST_HEAD(&cgrp->files); |
69d0206c | 1314 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1315 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1316 | INIT_LIST_HEAD(&cgrp->pidlists); |
1317 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1318 | cgrp->dummy_css.cgroup = cgrp; |
03b1cde6 | 1319 | simple_xattrs_init(&cgrp->xattrs); |
cc31edce | 1320 | } |
c6d57f33 | 1321 | |
ddbcc7e8 PM |
1322 | static void init_cgroup_root(struct cgroupfs_root *root) |
1323 | { | |
bd89aabc | 1324 | struct cgroup *cgrp = &root->top_cgroup; |
b0ca5a84 | 1325 | |
ddbcc7e8 PM |
1326 | INIT_LIST_HEAD(&root->root_list); |
1327 | root->number_of_cgroups = 1; | |
bd89aabc | 1328 | cgrp->root = root; |
a4ea1cc9 | 1329 | RCU_INIT_POINTER(cgrp->name, &root_cgroup_name); |
cc31edce | 1330 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1331 | idr_init(&root->cgroup_idr); |
ddbcc7e8 PM |
1332 | } |
1333 | ||
fc76df70 | 1334 | static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end) |
2c6ab6d2 | 1335 | { |
1a574231 | 1336 | int id; |
2c6ab6d2 | 1337 | |
54e7b4eb | 1338 | lockdep_assert_held(&cgroup_mutex); |
54e7b4eb | 1339 | |
fc76df70 TH |
1340 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end, |
1341 | GFP_KERNEL); | |
1a574231 TH |
1342 | if (id < 0) |
1343 | return id; | |
1344 | ||
1345 | root->hierarchy_id = id; | |
fa3ca07e TH |
1346 | return 0; |
1347 | } | |
1348 | ||
1349 | static void cgroup_exit_root_id(struct cgroupfs_root *root) | |
1350 | { | |
54e7b4eb | 1351 | lockdep_assert_held(&cgroup_mutex); |
54e7b4eb | 1352 | |
fa3ca07e | 1353 | if (root->hierarchy_id) { |
1a574231 | 1354 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
fa3ca07e TH |
1355 | root->hierarchy_id = 0; |
1356 | } | |
2c6ab6d2 PM |
1357 | } |
1358 | ||
ddbcc7e8 PM |
1359 | static int cgroup_test_super(struct super_block *sb, void *data) |
1360 | { | |
c6d57f33 | 1361 | struct cgroup_sb_opts *opts = data; |
ddbcc7e8 PM |
1362 | struct cgroupfs_root *root = sb->s_fs_info; |
1363 | ||
c6d57f33 PM |
1364 | /* If we asked for a name then it must match */ |
1365 | if (opts->name && strcmp(opts->name, root->name)) | |
1366 | return 0; | |
ddbcc7e8 | 1367 | |
2c6ab6d2 PM |
1368 | /* |
1369 | * If we asked for subsystems (or explicitly for no | |
1370 | * subsystems) then they must match | |
1371 | */ | |
a1a71b45 AR |
1372 | if ((opts->subsys_mask || opts->none) |
1373 | && (opts->subsys_mask != root->subsys_mask)) | |
ddbcc7e8 PM |
1374 | return 0; |
1375 | ||
1376 | return 1; | |
1377 | } | |
1378 | ||
c6d57f33 PM |
1379 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1380 | { | |
1381 | struct cgroupfs_root *root; | |
1382 | ||
a1a71b45 | 1383 | if (!opts->subsys_mask && !opts->none) |
c6d57f33 PM |
1384 | return NULL; |
1385 | ||
1386 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1387 | if (!root) | |
1388 | return ERR_PTR(-ENOMEM); | |
1389 | ||
1390 | init_cgroup_root(root); | |
2c6ab6d2 | 1391 | |
1672d040 TH |
1392 | /* |
1393 | * We need to set @root->subsys_mask now so that @root can be | |
1394 | * matched by cgroup_test_super() before it finishes | |
1395 | * initialization; otherwise, competing mounts with the same | |
1396 | * options may try to bind the same subsystems instead of waiting | |
1397 | * for the first one leading to unexpected mount errors. | |
1398 | * SUBSYS_BOUND will be set once actual binding is complete. | |
1399 | */ | |
a1a71b45 | 1400 | root->subsys_mask = opts->subsys_mask; |
c6d57f33 PM |
1401 | root->flags = opts->flags; |
1402 | if (opts->release_agent) | |
1403 | strcpy(root->release_agent_path, opts->release_agent); | |
1404 | if (opts->name) | |
1405 | strcpy(root->name, opts->name); | |
2260e7fc TH |
1406 | if (opts->cpuset_clone_children) |
1407 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1408 | return root; |
1409 | } | |
1410 | ||
fa3ca07e | 1411 | static void cgroup_free_root(struct cgroupfs_root *root) |
2c6ab6d2 | 1412 | { |
fa3ca07e TH |
1413 | if (root) { |
1414 | /* hierarhcy ID shoulid already have been released */ | |
1415 | WARN_ON_ONCE(root->hierarchy_id); | |
2c6ab6d2 | 1416 | |
4e96ee8e | 1417 | idr_destroy(&root->cgroup_idr); |
fa3ca07e TH |
1418 | kfree(root); |
1419 | } | |
2c6ab6d2 PM |
1420 | } |
1421 | ||
ddbcc7e8 PM |
1422 | static int cgroup_set_super(struct super_block *sb, void *data) |
1423 | { | |
1424 | int ret; | |
c6d57f33 PM |
1425 | struct cgroup_sb_opts *opts = data; |
1426 | ||
1427 | /* If we don't have a new root, we can't set up a new sb */ | |
1428 | if (!opts->new_root) | |
1429 | return -EINVAL; | |
1430 | ||
a1a71b45 | 1431 | BUG_ON(!opts->subsys_mask && !opts->none); |
ddbcc7e8 PM |
1432 | |
1433 | ret = set_anon_super(sb, NULL); | |
1434 | if (ret) | |
1435 | return ret; | |
1436 | ||
c6d57f33 PM |
1437 | sb->s_fs_info = opts->new_root; |
1438 | opts->new_root->sb = sb; | |
ddbcc7e8 PM |
1439 | |
1440 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
1441 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
1442 | sb->s_magic = CGROUP_SUPER_MAGIC; | |
1443 | sb->s_op = &cgroup_ops; | |
1444 | ||
1445 | return 0; | |
1446 | } | |
1447 | ||
1448 | static int cgroup_get_rootdir(struct super_block *sb) | |
1449 | { | |
0df6a63f AV |
1450 | static const struct dentry_operations cgroup_dops = { |
1451 | .d_iput = cgroup_diput, | |
b26d4cd3 | 1452 | .d_delete = always_delete_dentry, |
0df6a63f AV |
1453 | }; |
1454 | ||
ddbcc7e8 PM |
1455 | struct inode *inode = |
1456 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); | |
ddbcc7e8 PM |
1457 | |
1458 | if (!inode) | |
1459 | return -ENOMEM; | |
1460 | ||
ddbcc7e8 PM |
1461 | inode->i_fop = &simple_dir_operations; |
1462 | inode->i_op = &cgroup_dir_inode_operations; | |
1463 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
1464 | inc_nlink(inode); | |
48fde701 AV |
1465 | sb->s_root = d_make_root(inode); |
1466 | if (!sb->s_root) | |
ddbcc7e8 | 1467 | return -ENOMEM; |
0df6a63f AV |
1468 | /* for everything else we want ->d_op set */ |
1469 | sb->s_d_op = &cgroup_dops; | |
ddbcc7e8 PM |
1470 | return 0; |
1471 | } | |
1472 | ||
d427dfeb TH |
1473 | static int cgroup_setup_root(struct cgroupfs_root *root) |
1474 | { | |
1475 | LIST_HEAD(tmp_links); | |
1476 | struct super_block *sb = root->sb; | |
1477 | struct cgroup *root_cgrp = &root->top_cgroup; | |
1478 | struct cgroupfs_root *existing_root; | |
1479 | struct css_set *cset; | |
1480 | struct inode *inode; | |
1481 | const struct cred *cred; | |
1482 | int i, ret; | |
1483 | ||
1484 | lockdep_assert_held(&cgroup_tree_mutex); | |
1485 | lockdep_assert_held(&cgroup_mutex); | |
1486 | BUG_ON(sb->s_root != NULL); | |
1487 | ||
1488 | mutex_unlock(&cgroup_mutex); | |
1489 | mutex_unlock(&cgroup_tree_mutex); | |
1490 | ||
1491 | ret = cgroup_get_rootdir(sb); | |
1492 | if (ret) { | |
1493 | mutex_lock(&cgroup_tree_mutex); | |
1494 | mutex_lock(&cgroup_mutex); | |
1495 | return ret; | |
1496 | } | |
1497 | inode = sb->s_root->d_inode; | |
1498 | ||
1499 | mutex_lock(&inode->i_mutex); | |
1500 | mutex_lock(&cgroup_tree_mutex); | |
1501 | mutex_lock(&cgroup_mutex); | |
1502 | ||
1503 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); | |
1504 | if (ret < 0) | |
1505 | goto out_unlock; | |
1506 | root_cgrp->id = ret; | |
1507 | ||
1508 | /* check for name clashes with existing mounts */ | |
1509 | ret = -EBUSY; | |
1510 | if (strlen(root->name)) | |
1511 | for_each_active_root(existing_root) | |
1512 | if (!strcmp(existing_root->name, root->name)) | |
1513 | goto out_unlock; | |
1514 | ||
1515 | /* | |
1516 | * We're accessing css_set_count without locking css_set_lock here, | |
1517 | * but that's OK - it can only be increased by someone holding | |
1518 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1519 | * have some link structures left over | |
1520 | */ | |
1521 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1522 | if (ret) | |
1523 | goto out_unlock; | |
1524 | ||
1525 | /* ID 0 is reserved for dummy root, 1 for unified hierarchy */ | |
1526 | ret = cgroup_init_root_id(root, 2, 0); | |
1527 | if (ret) | |
1528 | goto out_unlock; | |
1529 | ||
1530 | sb->s_root->d_fsdata = root_cgrp; | |
1531 | root_cgrp->dentry = sb->s_root; | |
1532 | ||
1533 | /* | |
1534 | * We're inside get_sb() and will call lookup_one_len() to create | |
1535 | * the root files, which doesn't work if SELinux is in use. The | |
1536 | * following cred dancing somehow works around it. See 2ce9738ba | |
1537 | * ("cgroupfs: use init_cred when populating new cgroupfs mount") | |
1538 | * for more details. | |
1539 | */ | |
1540 | cred = override_creds(&init_cred); | |
1541 | ||
1542 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); | |
1543 | if (ret) | |
1544 | goto rm_base_files; | |
1545 | ||
1546 | ret = rebind_subsystems(root, root->subsys_mask, 0); | |
1547 | if (ret) | |
1548 | goto rm_base_files; | |
1549 | ||
1550 | revert_creds(cred); | |
1551 | ||
1552 | /* | |
1553 | * There must be no failure case after here, since rebinding takes | |
1554 | * care of subsystems' refcounts, which are explicitly dropped in | |
1555 | * the failure exit path. | |
1556 | */ | |
1557 | list_add(&root->root_list, &cgroup_roots); | |
1558 | cgroup_root_count++; | |
1559 | ||
1560 | /* | |
1561 | * Link the top cgroup in this hierarchy into all the css_set | |
1562 | * objects. | |
1563 | */ | |
1564 | write_lock(&css_set_lock); | |
1565 | hash_for_each(css_set_table, i, cset, hlist) | |
1566 | link_css_set(&tmp_links, cset, root_cgrp); | |
1567 | write_unlock(&css_set_lock); | |
1568 | ||
1569 | BUG_ON(!list_empty(&root_cgrp->children)); | |
1570 | BUG_ON(root->number_of_cgroups != 1); | |
1571 | ||
1572 | ret = 0; | |
1573 | goto out_unlock; | |
1574 | ||
1575 | rm_base_files: | |
1576 | cgroup_addrm_files(&root->top_cgroup, cgroup_base_files, false); | |
1577 | revert_creds(cred); | |
1578 | cgroup_exit_root_id(root); | |
1579 | out_unlock: | |
1580 | mutex_unlock(&inode->i_mutex); | |
1581 | free_cgrp_cset_links(&tmp_links); | |
1582 | return ret; | |
1583 | } | |
1584 | ||
f7e83571 | 1585 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1586 | int flags, const char *unused_dev_name, |
f7e83571 | 1587 | void *data) |
ddbcc7e8 | 1588 | { |
8e30e2b8 | 1589 | struct super_block *sb = NULL; |
8e30e2b8 | 1590 | struct cgroupfs_root *root = NULL; |
ddbcc7e8 | 1591 | struct cgroup_sb_opts opts; |
c6d57f33 | 1592 | struct cgroupfs_root *new_root; |
8e30e2b8 | 1593 | int ret; |
ddbcc7e8 | 1594 | |
8e30e2b8 | 1595 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1596 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1597 | |
1598 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1599 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1600 | if (ret) |
8e30e2b8 | 1601 | goto out_unlock; |
ddbcc7e8 | 1602 | |
c6d57f33 PM |
1603 | /* |
1604 | * Allocate a new cgroup root. We may not need it if we're | |
1605 | * reusing an existing hierarchy. | |
1606 | */ | |
1607 | new_root = cgroup_root_from_opts(&opts); | |
1608 | if (IS_ERR(new_root)) { | |
1609 | ret = PTR_ERR(new_root); | |
8e30e2b8 | 1610 | goto out_unlock; |
81a6a5cd | 1611 | } |
c6d57f33 | 1612 | opts.new_root = new_root; |
ddbcc7e8 | 1613 | |
c6d57f33 | 1614 | /* Locate an existing or new sb for this hierarchy */ |
8e30e2b8 TH |
1615 | mutex_unlock(&cgroup_mutex); |
1616 | mutex_unlock(&cgroup_tree_mutex); | |
9249e17f | 1617 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts); |
8e30e2b8 TH |
1618 | mutex_lock(&cgroup_tree_mutex); |
1619 | mutex_lock(&cgroup_mutex); | |
ddbcc7e8 | 1620 | if (IS_ERR(sb)) { |
c6d57f33 | 1621 | ret = PTR_ERR(sb); |
fa3ca07e | 1622 | cgroup_free_root(opts.new_root); |
8e30e2b8 | 1623 | goto out_unlock; |
ddbcc7e8 PM |
1624 | } |
1625 | ||
c6d57f33 PM |
1626 | root = sb->s_fs_info; |
1627 | BUG_ON(!root); | |
1628 | if (root == opts.new_root) { | |
d427dfeb | 1629 | ret = cgroup_setup_root(root); |
fa3ca07e | 1630 | if (ret) |
8e30e2b8 | 1631 | goto out_unlock; |
c6d57f33 PM |
1632 | } else { |
1633 | /* | |
1634 | * We re-used an existing hierarchy - the new root (if | |
1635 | * any) is not needed | |
1636 | */ | |
fa3ca07e | 1637 | cgroup_free_root(opts.new_root); |
873fe09e | 1638 | |
c7ba8287 | 1639 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1640 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1641 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1642 | ret = -EINVAL; | |
8e30e2b8 | 1643 | goto out_unlock; |
2a0ff3fb JL |
1644 | } else { |
1645 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1646 | } | |
873fe09e | 1647 | } |
ddbcc7e8 PM |
1648 | } |
1649 | ||
8e30e2b8 | 1650 | ret = 0; |
8e30e2b8 | 1651 | out_unlock: |
e25e2cbb | 1652 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1653 | mutex_unlock(&cgroup_tree_mutex); |
8e30e2b8 TH |
1654 | |
1655 | if (ret && !IS_ERR_OR_NULL(sb)) | |
1656 | deactivate_locked_super(sb); | |
1657 | ||
c6d57f33 PM |
1658 | kfree(opts.release_agent); |
1659 | kfree(opts.name); | |
8e30e2b8 TH |
1660 | |
1661 | if (!ret) | |
1662 | return dget(sb->s_root); | |
1663 | else | |
1664 | return ERR_PTR(ret); | |
ddbcc7e8 PM |
1665 | } |
1666 | ||
dd4b0a46 SP |
1667 | static void cgroup_kill_sb(struct super_block *sb) |
1668 | { | |
ddbcc7e8 | 1669 | struct cgroupfs_root *root = sb->s_fs_info; |
bd89aabc | 1670 | struct cgroup *cgrp = &root->top_cgroup; |
69d0206c | 1671 | struct cgrp_cset_link *link, *tmp_link; |
ddbcc7e8 PM |
1672 | int ret; |
1673 | ||
1674 | BUG_ON(!root); | |
1675 | ||
1676 | BUG_ON(root->number_of_cgroups != 1); | |
bd89aabc | 1677 | BUG_ON(!list_empty(&cgrp->children)); |
ddbcc7e8 | 1678 | |
3126121f | 1679 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
ace2bee8 | 1680 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1681 | mutex_lock(&cgroup_mutex); |
1682 | ||
1683 | /* Rebind all subsystems back to the default hierarchy */ | |
1672d040 TH |
1684 | if (root->flags & CGRP_ROOT_SUBSYS_BOUND) { |
1685 | ret = rebind_subsystems(root, 0, root->subsys_mask); | |
1686 | /* Shouldn't be able to fail ... */ | |
1687 | BUG_ON(ret); | |
1688 | } | |
ddbcc7e8 | 1689 | |
817929ec | 1690 | /* |
69d0206c | 1691 | * Release all the links from cset_links to this hierarchy's |
817929ec PM |
1692 | * root cgroup |
1693 | */ | |
1694 | write_lock(&css_set_lock); | |
71cbb949 | 1695 | |
69d0206c TH |
1696 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { |
1697 | list_del(&link->cset_link); | |
1698 | list_del(&link->cgrp_link); | |
817929ec PM |
1699 | kfree(link); |
1700 | } | |
1701 | write_unlock(&css_set_lock); | |
1702 | ||
839ec545 PM |
1703 | if (!list_empty(&root->root_list)) { |
1704 | list_del(&root->root_list); | |
9871bf95 | 1705 | cgroup_root_count--; |
839ec545 | 1706 | } |
e5f6a860 | 1707 | |
fa3ca07e TH |
1708 | cgroup_exit_root_id(root); |
1709 | ||
ddbcc7e8 | 1710 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1711 | mutex_unlock(&cgroup_tree_mutex); |
3126121f | 1712 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 | 1713 | |
03b1cde6 AR |
1714 | simple_xattrs_free(&cgrp->xattrs); |
1715 | ||
ddbcc7e8 | 1716 | kill_litter_super(sb); |
fa3ca07e | 1717 | cgroup_free_root(root); |
ddbcc7e8 PM |
1718 | } |
1719 | ||
1720 | static struct file_system_type cgroup_fs_type = { | |
1721 | .name = "cgroup", | |
f7e83571 | 1722 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1723 | .kill_sb = cgroup_kill_sb, |
1724 | }; | |
1725 | ||
676db4af GK |
1726 | static struct kobject *cgroup_kobj; |
1727 | ||
a043e3b2 LZ |
1728 | /** |
1729 | * cgroup_path - generate the path of a cgroup | |
1730 | * @cgrp: the cgroup in question | |
1731 | * @buf: the buffer to write the path into | |
1732 | * @buflen: the length of the buffer | |
1733 | * | |
65dff759 LZ |
1734 | * Writes path of cgroup into buf. Returns 0 on success, -errno on error. |
1735 | * | |
1736 | * We can't generate cgroup path using dentry->d_name, as accessing | |
1737 | * dentry->name must be protected by irq-unsafe dentry->d_lock or parent | |
1738 | * inode's i_mutex, while on the other hand cgroup_path() can be called | |
1739 | * with some irq-safe spinlocks held. | |
ddbcc7e8 | 1740 | */ |
bd89aabc | 1741 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
ddbcc7e8 | 1742 | { |
65dff759 | 1743 | int ret = -ENAMETOOLONG; |
ddbcc7e8 | 1744 | char *start; |
febfcef6 | 1745 | |
da1f296f TH |
1746 | if (!cgrp->parent) { |
1747 | if (strlcpy(buf, "/", buflen) >= buflen) | |
1748 | return -ENAMETOOLONG; | |
ddbcc7e8 PM |
1749 | return 0; |
1750 | } | |
1751 | ||
316eb661 | 1752 | start = buf + buflen - 1; |
316eb661 | 1753 | *start = '\0'; |
9a9686b6 | 1754 | |
65dff759 | 1755 | rcu_read_lock(); |
da1f296f | 1756 | do { |
65dff759 LZ |
1757 | const char *name = cgroup_name(cgrp); |
1758 | int len; | |
1759 | ||
1760 | len = strlen(name); | |
ddbcc7e8 | 1761 | if ((start -= len) < buf) |
65dff759 LZ |
1762 | goto out; |
1763 | memcpy(start, name, len); | |
9a9686b6 | 1764 | |
ddbcc7e8 | 1765 | if (--start < buf) |
65dff759 | 1766 | goto out; |
ddbcc7e8 | 1767 | *start = '/'; |
65dff759 LZ |
1768 | |
1769 | cgrp = cgrp->parent; | |
da1f296f | 1770 | } while (cgrp->parent); |
65dff759 | 1771 | ret = 0; |
ddbcc7e8 | 1772 | memmove(buf, start, buf + buflen - start); |
65dff759 LZ |
1773 | out: |
1774 | rcu_read_unlock(); | |
1775 | return ret; | |
ddbcc7e8 | 1776 | } |
67523c48 | 1777 | EXPORT_SYMBOL_GPL(cgroup_path); |
ddbcc7e8 | 1778 | |
857a2beb | 1779 | /** |
913ffdb5 | 1780 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1781 | * @task: target task |
857a2beb TH |
1782 | * @buf: the buffer to write the path into |
1783 | * @buflen: the length of the buffer | |
1784 | * | |
913ffdb5 TH |
1785 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1786 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1787 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1788 | * cgroup controller callbacks. | |
1789 | * | |
1790 | * Returns 0 on success, fails with -%ENAMETOOLONG if @buflen is too short. | |
857a2beb | 1791 | */ |
913ffdb5 | 1792 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb TH |
1793 | { |
1794 | struct cgroupfs_root *root; | |
913ffdb5 TH |
1795 | struct cgroup *cgrp; |
1796 | int hierarchy_id = 1, ret = 0; | |
1797 | ||
1798 | if (buflen < 2) | |
1799 | return -ENAMETOOLONG; | |
857a2beb TH |
1800 | |
1801 | mutex_lock(&cgroup_mutex); | |
1802 | ||
913ffdb5 TH |
1803 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1804 | ||
857a2beb TH |
1805 | if (root) { |
1806 | cgrp = task_cgroup_from_root(task, root); | |
1807 | ret = cgroup_path(cgrp, buf, buflen); | |
913ffdb5 TH |
1808 | } else { |
1809 | /* if no hierarchy exists, everyone is in "/" */ | |
1810 | memcpy(buf, "/", 2); | |
857a2beb TH |
1811 | } |
1812 | ||
1813 | mutex_unlock(&cgroup_mutex); | |
857a2beb TH |
1814 | return ret; |
1815 | } | |
913ffdb5 | 1816 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1817 | |
2f7ee569 TH |
1818 | /* |
1819 | * Control Group taskset | |
1820 | */ | |
134d3373 TH |
1821 | struct task_and_cgroup { |
1822 | struct task_struct *task; | |
1823 | struct cgroup *cgrp; | |
6f4b7e63 | 1824 | struct css_set *cset; |
134d3373 TH |
1825 | }; |
1826 | ||
2f7ee569 TH |
1827 | struct cgroup_taskset { |
1828 | struct task_and_cgroup single; | |
1829 | struct flex_array *tc_array; | |
1830 | int tc_array_len; | |
1831 | int idx; | |
1832 | struct cgroup *cur_cgrp; | |
1833 | }; | |
1834 | ||
1835 | /** | |
1836 | * cgroup_taskset_first - reset taskset and return the first task | |
1837 | * @tset: taskset of interest | |
1838 | * | |
1839 | * @tset iteration is initialized and the first task is returned. | |
1840 | */ | |
1841 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1842 | { | |
1843 | if (tset->tc_array) { | |
1844 | tset->idx = 0; | |
1845 | return cgroup_taskset_next(tset); | |
1846 | } else { | |
1847 | tset->cur_cgrp = tset->single.cgrp; | |
1848 | return tset->single.task; | |
1849 | } | |
1850 | } | |
1851 | EXPORT_SYMBOL_GPL(cgroup_taskset_first); | |
1852 | ||
1853 | /** | |
1854 | * cgroup_taskset_next - iterate to the next task in taskset | |
1855 | * @tset: taskset of interest | |
1856 | * | |
1857 | * Return the next task in @tset. Iteration must have been initialized | |
1858 | * with cgroup_taskset_first(). | |
1859 | */ | |
1860 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1861 | { | |
1862 | struct task_and_cgroup *tc; | |
1863 | ||
1864 | if (!tset->tc_array || tset->idx >= tset->tc_array_len) | |
1865 | return NULL; | |
1866 | ||
1867 | tc = flex_array_get(tset->tc_array, tset->idx++); | |
1868 | tset->cur_cgrp = tc->cgrp; | |
1869 | return tc->task; | |
1870 | } | |
1871 | EXPORT_SYMBOL_GPL(cgroup_taskset_next); | |
1872 | ||
1873 | /** | |
d99c8727 | 1874 | * cgroup_taskset_cur_css - return the matching css for the current task |
2f7ee569 | 1875 | * @tset: taskset of interest |
d99c8727 | 1876 | * @subsys_id: the ID of the target subsystem |
2f7ee569 | 1877 | * |
d99c8727 TH |
1878 | * Return the css for the current (last returned) task of @tset for |
1879 | * subsystem specified by @subsys_id. This function must be preceded by | |
1880 | * either cgroup_taskset_first() or cgroup_taskset_next(). | |
2f7ee569 | 1881 | */ |
d99c8727 TH |
1882 | struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset, |
1883 | int subsys_id) | |
2f7ee569 | 1884 | { |
ca8bdcaf | 1885 | return cgroup_css(tset->cur_cgrp, cgroup_subsys[subsys_id]); |
2f7ee569 | 1886 | } |
d99c8727 | 1887 | EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css); |
2f7ee569 TH |
1888 | |
1889 | /** | |
1890 | * cgroup_taskset_size - return the number of tasks in taskset | |
1891 | * @tset: taskset of interest | |
1892 | */ | |
1893 | int cgroup_taskset_size(struct cgroup_taskset *tset) | |
1894 | { | |
1895 | return tset->tc_array ? tset->tc_array_len : 1; | |
1896 | } | |
1897 | EXPORT_SYMBOL_GPL(cgroup_taskset_size); | |
1898 | ||
1899 | ||
74a1166d BB |
1900 | /* |
1901 | * cgroup_task_migrate - move a task from one cgroup to another. | |
1902 | * | |
d0b2fdd2 | 1903 | * Must be called with cgroup_mutex and threadgroup locked. |
74a1166d | 1904 | */ |
5abb8855 TH |
1905 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1906 | struct task_struct *tsk, | |
1907 | struct css_set *new_cset) | |
74a1166d | 1908 | { |
5abb8855 | 1909 | struct css_set *old_cset; |
74a1166d BB |
1910 | |
1911 | /* | |
026085ef MSB |
1912 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1913 | * setting such that we can't race against cgroup_exit() changing the | |
1914 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1915 | */ |
c84cdf75 | 1916 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1917 | old_cset = task_css_set(tsk); |
74a1166d | 1918 | |
74a1166d | 1919 | task_lock(tsk); |
5abb8855 | 1920 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d BB |
1921 | task_unlock(tsk); |
1922 | ||
1923 | /* Update the css_set linked lists if we're using them */ | |
1924 | write_lock(&css_set_lock); | |
1925 | if (!list_empty(&tsk->cg_list)) | |
5abb8855 | 1926 | list_move(&tsk->cg_list, &new_cset->tasks); |
74a1166d BB |
1927 | write_unlock(&css_set_lock); |
1928 | ||
1929 | /* | |
5abb8855 TH |
1930 | * We just gained a reference on old_cset by taking it from the |
1931 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1932 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1933 | */ |
5abb8855 TH |
1934 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
1935 | put_css_set(old_cset); | |
74a1166d BB |
1936 | } |
1937 | ||
a043e3b2 | 1938 | /** |
081aa458 | 1939 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup |
74a1166d | 1940 | * @cgrp: the cgroup to attach to |
081aa458 LZ |
1941 | * @tsk: the task or the leader of the threadgroup to be attached |
1942 | * @threadgroup: attach the whole threadgroup? | |
74a1166d | 1943 | * |
257058ae | 1944 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
081aa458 | 1945 | * task_lock of @tsk or each thread in the threadgroup individually in turn. |
74a1166d | 1946 | */ |
47cfcd09 TH |
1947 | static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk, |
1948 | bool threadgroup) | |
74a1166d BB |
1949 | { |
1950 | int retval, i, group_size; | |
74a1166d | 1951 | struct cgroupfs_root *root = cgrp->root; |
1c6727af | 1952 | struct cgroup_subsys_state *css, *failed_css = NULL; |
74a1166d | 1953 | /* threadgroup list cursor and array */ |
081aa458 | 1954 | struct task_struct *leader = tsk; |
134d3373 | 1955 | struct task_and_cgroup *tc; |
d846687d | 1956 | struct flex_array *group; |
2f7ee569 | 1957 | struct cgroup_taskset tset = { }; |
74a1166d BB |
1958 | |
1959 | /* | |
1960 | * step 0: in order to do expensive, possibly blocking operations for | |
1961 | * every thread, we cannot iterate the thread group list, since it needs | |
1962 | * rcu or tasklist locked. instead, build an array of all threads in the | |
257058ae TH |
1963 | * group - group_rwsem prevents new threads from appearing, and if |
1964 | * threads exit, this will just be an over-estimate. | |
74a1166d | 1965 | */ |
081aa458 LZ |
1966 | if (threadgroup) |
1967 | group_size = get_nr_threads(tsk); | |
1968 | else | |
1969 | group_size = 1; | |
d846687d | 1970 | /* flex_array supports very large thread-groups better than kmalloc. */ |
134d3373 | 1971 | group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL); |
74a1166d BB |
1972 | if (!group) |
1973 | return -ENOMEM; | |
d846687d | 1974 | /* pre-allocate to guarantee space while iterating in rcu read-side. */ |
3ac1707a | 1975 | retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL); |
d846687d BB |
1976 | if (retval) |
1977 | goto out_free_group_list; | |
74a1166d | 1978 | |
74a1166d | 1979 | i = 0; |
fb5d2b4c MSB |
1980 | /* |
1981 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1982 | * already PF_EXITING could be freed from underneath us unless we | |
1983 | * take an rcu_read_lock. | |
1984 | */ | |
1985 | rcu_read_lock(); | |
74a1166d | 1986 | do { |
134d3373 TH |
1987 | struct task_and_cgroup ent; |
1988 | ||
cd3d0952 TH |
1989 | /* @tsk either already exited or can't exit until the end */ |
1990 | if (tsk->flags & PF_EXITING) | |
ea84753c | 1991 | goto next; |
cd3d0952 | 1992 | |
74a1166d BB |
1993 | /* as per above, nr_threads may decrease, but not increase. */ |
1994 | BUG_ON(i >= group_size); | |
134d3373 TH |
1995 | ent.task = tsk; |
1996 | ent.cgrp = task_cgroup_from_root(tsk, root); | |
892a2b90 MSB |
1997 | /* nothing to do if this task is already in the cgroup */ |
1998 | if (ent.cgrp == cgrp) | |
ea84753c | 1999 | goto next; |
61d1d219 MSB |
2000 | /* |
2001 | * saying GFP_ATOMIC has no effect here because we did prealloc | |
2002 | * earlier, but it's good form to communicate our expectations. | |
2003 | */ | |
134d3373 | 2004 | retval = flex_array_put(group, i, &ent, GFP_ATOMIC); |
d846687d | 2005 | BUG_ON(retval != 0); |
74a1166d | 2006 | i++; |
ea84753c | 2007 | next: |
081aa458 LZ |
2008 | if (!threadgroup) |
2009 | break; | |
74a1166d | 2010 | } while_each_thread(leader, tsk); |
fb5d2b4c | 2011 | rcu_read_unlock(); |
74a1166d BB |
2012 | /* remember the number of threads in the array for later. */ |
2013 | group_size = i; | |
2f7ee569 TH |
2014 | tset.tc_array = group; |
2015 | tset.tc_array_len = group_size; | |
74a1166d | 2016 | |
134d3373 TH |
2017 | /* methods shouldn't be called if no task is actually migrating */ |
2018 | retval = 0; | |
892a2b90 | 2019 | if (!group_size) |
b07ef774 | 2020 | goto out_free_group_list; |
134d3373 | 2021 | |
74a1166d BB |
2022 | /* |
2023 | * step 1: check that we can legitimately attach to the cgroup. | |
2024 | */ | |
1c6727af TH |
2025 | for_each_css(css, i, cgrp) { |
2026 | if (css->ss->can_attach) { | |
2027 | retval = css->ss->can_attach(css, &tset); | |
74a1166d | 2028 | if (retval) { |
1c6727af | 2029 | failed_css = css; |
74a1166d BB |
2030 | goto out_cancel_attach; |
2031 | } | |
2032 | } | |
74a1166d BB |
2033 | } |
2034 | ||
2035 | /* | |
2036 | * step 2: make sure css_sets exist for all threads to be migrated. | |
2037 | * we use find_css_set, which allocates a new one if necessary. | |
2038 | */ | |
74a1166d | 2039 | for (i = 0; i < group_size; i++) { |
a8ad805c TH |
2040 | struct css_set *old_cset; |
2041 | ||
134d3373 | 2042 | tc = flex_array_get(group, i); |
a8ad805c | 2043 | old_cset = task_css_set(tc->task); |
6f4b7e63 LZ |
2044 | tc->cset = find_css_set(old_cset, cgrp); |
2045 | if (!tc->cset) { | |
61d1d219 MSB |
2046 | retval = -ENOMEM; |
2047 | goto out_put_css_set_refs; | |
74a1166d BB |
2048 | } |
2049 | } | |
2050 | ||
2051 | /* | |
494c167c TH |
2052 | * step 3: now that we're guaranteed success wrt the css_sets, |
2053 | * proceed to move all tasks to the new cgroup. There are no | |
2054 | * failure cases after here, so this is the commit point. | |
74a1166d | 2055 | */ |
74a1166d | 2056 | for (i = 0; i < group_size; i++) { |
134d3373 | 2057 | tc = flex_array_get(group, i); |
6f4b7e63 | 2058 | cgroup_task_migrate(tc->cgrp, tc->task, tc->cset); |
74a1166d BB |
2059 | } |
2060 | /* nothing is sensitive to fork() after this point. */ | |
2061 | ||
2062 | /* | |
494c167c | 2063 | * step 4: do subsystem attach callbacks. |
74a1166d | 2064 | */ |
1c6727af TH |
2065 | for_each_css(css, i, cgrp) |
2066 | if (css->ss->attach) | |
2067 | css->ss->attach(css, &tset); | |
74a1166d BB |
2068 | |
2069 | /* | |
2070 | * step 5: success! and cleanup | |
2071 | */ | |
74a1166d | 2072 | retval = 0; |
61d1d219 MSB |
2073 | out_put_css_set_refs: |
2074 | if (retval) { | |
2075 | for (i = 0; i < group_size; i++) { | |
2076 | tc = flex_array_get(group, i); | |
6f4b7e63 | 2077 | if (!tc->cset) |
61d1d219 | 2078 | break; |
6f4b7e63 | 2079 | put_css_set(tc->cset); |
61d1d219 | 2080 | } |
74a1166d BB |
2081 | } |
2082 | out_cancel_attach: | |
74a1166d | 2083 | if (retval) { |
1c6727af TH |
2084 | for_each_css(css, i, cgrp) { |
2085 | if (css == failed_css) | |
74a1166d | 2086 | break; |
1c6727af TH |
2087 | if (css->ss->cancel_attach) |
2088 | css->ss->cancel_attach(css, &tset); | |
74a1166d BB |
2089 | } |
2090 | } | |
74a1166d | 2091 | out_free_group_list: |
d846687d | 2092 | flex_array_free(group); |
74a1166d BB |
2093 | return retval; |
2094 | } | |
2095 | ||
2096 | /* | |
2097 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
2098 | * function to attach either it or all tasks in its threadgroup. Will lock |
2099 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 2100 | */ |
74a1166d | 2101 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2102 | { |
bbcb81d0 | 2103 | struct task_struct *tsk; |
c69e8d9c | 2104 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2105 | int ret; |
2106 | ||
74a1166d BB |
2107 | if (!cgroup_lock_live_group(cgrp)) |
2108 | return -ENODEV; | |
2109 | ||
b78949eb MSB |
2110 | retry_find_task: |
2111 | rcu_read_lock(); | |
bbcb81d0 | 2112 | if (pid) { |
73507f33 | 2113 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2114 | if (!tsk) { |
2115 | rcu_read_unlock(); | |
dd4b0a46 | 2116 | ret = -ESRCH; |
b78949eb | 2117 | goto out_unlock_cgroup; |
bbcb81d0 | 2118 | } |
74a1166d BB |
2119 | /* |
2120 | * even if we're attaching all tasks in the thread group, we | |
2121 | * only need to check permissions on one of them. | |
2122 | */ | |
c69e8d9c | 2123 | tcred = __task_cred(tsk); |
14a590c3 EB |
2124 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2125 | !uid_eq(cred->euid, tcred->uid) && | |
2126 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2127 | rcu_read_unlock(); |
b78949eb MSB |
2128 | ret = -EACCES; |
2129 | goto out_unlock_cgroup; | |
bbcb81d0 | 2130 | } |
b78949eb MSB |
2131 | } else |
2132 | tsk = current; | |
cd3d0952 TH |
2133 | |
2134 | if (threadgroup) | |
b78949eb | 2135 | tsk = tsk->group_leader; |
c4c27fbd MG |
2136 | |
2137 | /* | |
14a40ffc | 2138 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2139 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2140 | * with no rt_runtime allocated. Just say no. | |
2141 | */ | |
14a40ffc | 2142 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2143 | ret = -EINVAL; |
2144 | rcu_read_unlock(); | |
2145 | goto out_unlock_cgroup; | |
2146 | } | |
2147 | ||
b78949eb MSB |
2148 | get_task_struct(tsk); |
2149 | rcu_read_unlock(); | |
2150 | ||
2151 | threadgroup_lock(tsk); | |
2152 | if (threadgroup) { | |
2153 | if (!thread_group_leader(tsk)) { | |
2154 | /* | |
2155 | * a race with de_thread from another thread's exec() | |
2156 | * may strip us of our leadership, if this happens, | |
2157 | * there is no choice but to throw this task away and | |
2158 | * try again; this is | |
2159 | * "double-double-toil-and-trouble-check locking". | |
2160 | */ | |
2161 | threadgroup_unlock(tsk); | |
2162 | put_task_struct(tsk); | |
2163 | goto retry_find_task; | |
2164 | } | |
081aa458 LZ |
2165 | } |
2166 | ||
2167 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2168 | ||
cd3d0952 TH |
2169 | threadgroup_unlock(tsk); |
2170 | ||
bbcb81d0 | 2171 | put_task_struct(tsk); |
b78949eb | 2172 | out_unlock_cgroup: |
47cfcd09 | 2173 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
2174 | return ret; |
2175 | } | |
2176 | ||
7ae1bad9 TH |
2177 | /** |
2178 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2179 | * @from: attach to all cgroups of a given task | |
2180 | * @tsk: the task to be attached | |
2181 | */ | |
2182 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2183 | { | |
2184 | struct cgroupfs_root *root; | |
2185 | int retval = 0; | |
2186 | ||
47cfcd09 | 2187 | mutex_lock(&cgroup_mutex); |
7ae1bad9 | 2188 | for_each_active_root(root) { |
6f4b7e63 | 2189 | struct cgroup *from_cgrp = task_cgroup_from_root(from, root); |
7ae1bad9 | 2190 | |
6f4b7e63 | 2191 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2192 | if (retval) |
2193 | break; | |
2194 | } | |
47cfcd09 | 2195 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2196 | |
2197 | return retval; | |
2198 | } | |
2199 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2200 | ||
182446d0 TH |
2201 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2202 | struct cftype *cft, u64 pid) | |
74a1166d | 2203 | { |
182446d0 | 2204 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2205 | } |
2206 | ||
182446d0 TH |
2207 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2208 | struct cftype *cft, u64 tgid) | |
af351026 | 2209 | { |
182446d0 | 2210 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2211 | } |
2212 | ||
182446d0 TH |
2213 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
2214 | struct cftype *cft, const char *buffer) | |
e788e066 | 2215 | { |
182446d0 | 2216 | BUILD_BUG_ON(sizeof(css->cgroup->root->release_agent_path) < PATH_MAX); |
f4a2589f EK |
2217 | if (strlen(buffer) >= PATH_MAX) |
2218 | return -EINVAL; | |
182446d0 | 2219 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2220 | return -ENODEV; |
69e943b7 | 2221 | spin_lock(&release_agent_path_lock); |
182446d0 | 2222 | strcpy(css->cgroup->root->release_agent_path, buffer); |
69e943b7 | 2223 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2224 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2225 | return 0; |
2226 | } | |
2227 | ||
2da8ca82 | 2228 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2229 | { |
2da8ca82 | 2230 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2231 | |
e788e066 PM |
2232 | if (!cgroup_lock_live_group(cgrp)) |
2233 | return -ENODEV; | |
2234 | seq_puts(seq, cgrp->root->release_agent_path); | |
2235 | seq_putc(seq, '\n'); | |
47cfcd09 | 2236 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2237 | return 0; |
2238 | } | |
2239 | ||
2da8ca82 | 2240 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2241 | { |
2da8ca82 TH |
2242 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2243 | ||
2244 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2245 | return 0; |
2246 | } | |
2247 | ||
84eea842 PM |
2248 | /* A buffer size big enough for numbers or short strings */ |
2249 | #define CGROUP_LOCAL_BUFFER_SIZE 64 | |
2250 | ||
a742c59d TH |
2251 | static ssize_t cgroup_file_write(struct file *file, const char __user *userbuf, |
2252 | size_t nbytes, loff_t *ppos) | |
355e0c48 | 2253 | { |
a742c59d TH |
2254 | struct cfent *cfe = __d_cfe(file->f_dentry); |
2255 | struct cftype *cft = __d_cft(file->f_dentry); | |
2256 | struct cgroup_subsys_state *css = cfe->css; | |
2257 | size_t max_bytes = cft->max_write_len ?: CGROUP_LOCAL_BUFFER_SIZE - 1; | |
2258 | char *buf; | |
2259 | int ret; | |
355e0c48 | 2260 | |
db3b1497 | 2261 | if (nbytes >= max_bytes) |
355e0c48 | 2262 | return -E2BIG; |
355e0c48 | 2263 | |
a742c59d TH |
2264 | buf = kmalloc(nbytes + 1, GFP_KERNEL); |
2265 | if (!buf) | |
2266 | return -ENOMEM; | |
db3b1497 | 2267 | |
a742c59d TH |
2268 | if (copy_from_user(buf, userbuf, nbytes)) { |
2269 | ret = -EFAULT; | |
2270 | goto out_free; | |
2271 | } | |
ddbcc7e8 | 2272 | |
a742c59d TH |
2273 | buf[nbytes] = '\0'; |
2274 | ||
2275 | if (cft->write_string) { | |
2276 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2277 | } else if (cft->write_u64) { | |
2278 | unsigned long long v; | |
2279 | ret = kstrtoull(buf, 0, &v); | |
2280 | if (!ret) | |
2281 | ret = cft->write_u64(css, cft, v); | |
2282 | } else if (cft->write_s64) { | |
2283 | long long v; | |
2284 | ret = kstrtoll(buf, 0, &v); | |
2285 | if (!ret) | |
2286 | ret = cft->write_s64(css, cft, v); | |
2287 | } else if (cft->trigger) { | |
2288 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2289 | } else { |
a742c59d | 2290 | ret = -EINVAL; |
e73d2c61 | 2291 | } |
a742c59d TH |
2292 | out_free: |
2293 | kfree(buf); | |
2294 | return ret ?: nbytes; | |
355e0c48 PM |
2295 | } |
2296 | ||
91796569 PM |
2297 | /* |
2298 | * seqfile ops/methods for returning structured data. Currently just | |
2299 | * supports string->u64 maps, but can be extended in future. | |
2300 | */ | |
2301 | ||
6612f05b | 2302 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2303 | { |
6612f05b | 2304 | struct cftype *cft = seq_cft(seq); |
db3b1497 | 2305 | |
6612f05b TH |
2306 | if (cft->seq_start) { |
2307 | return cft->seq_start(seq, ppos); | |
2308 | } else { | |
2309 | /* | |
2310 | * The same behavior and code as single_open(). Returns | |
2311 | * !NULL if pos is at the beginning; otherwise, NULL. | |
2312 | */ | |
2313 | return NULL + !*ppos; | |
5a3eb9f6 | 2314 | } |
db3b1497 PM |
2315 | } |
2316 | ||
6612f05b | 2317 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2318 | { |
6612f05b | 2319 | struct cftype *cft = seq_cft(seq); |
ddbcc7e8 | 2320 | |
6612f05b TH |
2321 | if (cft->seq_next) { |
2322 | return cft->seq_next(seq, v, ppos); | |
2323 | } else { | |
2324 | /* | |
2325 | * The same behavior and code as single_open(), always | |
2326 | * terminate after the initial read. | |
2327 | */ | |
2328 | ++*ppos; | |
2329 | return NULL; | |
d447ea2f | 2330 | } |
ddbcc7e8 PM |
2331 | } |
2332 | ||
6612f05b | 2333 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2334 | { |
6612f05b | 2335 | struct cftype *cft = seq_cft(seq); |
ddbcc7e8 | 2336 | |
6612f05b TH |
2337 | if (cft->seq_stop) |
2338 | cft->seq_stop(seq, v); | |
ddbcc7e8 PM |
2339 | } |
2340 | ||
91796569 | 2341 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2342 | { |
7da11279 TH |
2343 | struct cftype *cft = seq_cft(m); |
2344 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2345 | |
2da8ca82 TH |
2346 | if (cft->seq_show) |
2347 | return cft->seq_show(m, arg); | |
e73d2c61 | 2348 | |
f4c753b7 | 2349 | if (cft->read_u64) |
896f5199 TH |
2350 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2351 | else if (cft->read_s64) | |
2352 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2353 | else | |
2354 | return -EINVAL; | |
2355 | return 0; | |
91796569 PM |
2356 | } |
2357 | ||
6612f05b TH |
2358 | static struct seq_operations cgroup_seq_operations = { |
2359 | .start = cgroup_seqfile_start, | |
2360 | .next = cgroup_seqfile_next, | |
2361 | .stop = cgroup_seqfile_stop, | |
2362 | .show = cgroup_seqfile_show, | |
91796569 PM |
2363 | }; |
2364 | ||
ddbcc7e8 PM |
2365 | static int cgroup_file_open(struct inode *inode, struct file *file) |
2366 | { | |
f7d58818 TH |
2367 | struct cfent *cfe = __d_cfe(file->f_dentry); |
2368 | struct cftype *cft = __d_cft(file->f_dentry); | |
105347ba TH |
2369 | struct cgroup *cgrp = __d_cgrp(cfe->dentry->d_parent); |
2370 | struct cgroup_subsys_state *css; | |
6612f05b | 2371 | struct cgroup_open_file *of; |
ddbcc7e8 | 2372 | int err; |
ddbcc7e8 PM |
2373 | |
2374 | err = generic_file_open(inode, file); | |
2375 | if (err) | |
2376 | return err; | |
75139b82 | 2377 | |
f7d58818 TH |
2378 | /* |
2379 | * If the file belongs to a subsystem, pin the css. Will be | |
2380 | * unpinned either on open failure or release. This ensures that | |
2381 | * @css stays alive for all file operations. | |
2382 | */ | |
105347ba | 2383 | rcu_read_lock(); |
ca8bdcaf TH |
2384 | css = cgroup_css(cgrp, cft->ss); |
2385 | if (cft->ss && !css_tryget(css)) | |
2386 | css = NULL; | |
105347ba | 2387 | rcu_read_unlock(); |
f4f4be2b | 2388 | |
0bfb4aa6 | 2389 | if (!css) |
f7d58818 | 2390 | return -ENODEV; |
75139b82 | 2391 | |
0bfb4aa6 TH |
2392 | /* |
2393 | * @cfe->css is used by read/write/close to determine the | |
2394 | * associated css. @file->private_data would be a better place but | |
2395 | * that's already used by seqfile. Multiple accessors may use it | |
2396 | * simultaneously which is okay as the association never changes. | |
2397 | */ | |
2398 | WARN_ON_ONCE(cfe->css && cfe->css != css); | |
2399 | cfe->css = css; | |
f4f4be2b | 2400 | |
6612f05b TH |
2401 | of = __seq_open_private(file, &cgroup_seq_operations, |
2402 | sizeof(struct cgroup_open_file)); | |
2403 | if (of) { | |
2404 | of->cfe = cfe; | |
2405 | return 0; | |
e0798ce2 | 2406 | } |
ddbcc7e8 | 2407 | |
6612f05b | 2408 | if (css->ss) |
f7d58818 | 2409 | css_put(css); |
6612f05b | 2410 | return -ENOMEM; |
ddbcc7e8 PM |
2411 | } |
2412 | ||
2413 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
2414 | { | |
f7d58818 | 2415 | struct cfent *cfe = __d_cfe(file->f_dentry); |
105347ba | 2416 | struct cgroup_subsys_state *css = cfe->css; |
f7d58818 | 2417 | |
67f4c36f | 2418 | if (css->ss) |
f7d58818 | 2419 | css_put(css); |
6612f05b | 2420 | return seq_release_private(inode, file); |
ddbcc7e8 PM |
2421 | } |
2422 | ||
2423 | /* | |
2424 | * cgroup_rename - Only allow simple rename of directories in place. | |
2425 | */ | |
2426 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
2427 | struct inode *new_dir, struct dentry *new_dentry) | |
2428 | { | |
65dff759 LZ |
2429 | int ret; |
2430 | struct cgroup_name *name, *old_name; | |
2431 | struct cgroup *cgrp; | |
2432 | ||
2433 | /* | |
2434 | * It's convinient to use parent dir's i_mutex to protected | |
2435 | * cgrp->name. | |
2436 | */ | |
2437 | lockdep_assert_held(&old_dir->i_mutex); | |
2438 | ||
ddbcc7e8 PM |
2439 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) |
2440 | return -ENOTDIR; | |
2441 | if (new_dentry->d_inode) | |
2442 | return -EEXIST; | |
2443 | if (old_dir != new_dir) | |
2444 | return -EIO; | |
65dff759 LZ |
2445 | |
2446 | cgrp = __d_cgrp(old_dentry); | |
2447 | ||
6db8e85c TH |
2448 | /* |
2449 | * This isn't a proper migration and its usefulness is very | |
2450 | * limited. Disallow if sane_behavior. | |
2451 | */ | |
2452 | if (cgroup_sane_behavior(cgrp)) | |
2453 | return -EPERM; | |
2454 | ||
8d7e6fb0 | 2455 | name = cgroup_alloc_name(new_dentry->d_name.name); |
65dff759 LZ |
2456 | if (!name) |
2457 | return -ENOMEM; | |
2458 | ||
2459 | ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
2460 | if (ret) { | |
2461 | kfree(name); | |
2462 | return ret; | |
2463 | } | |
2464 | ||
a4ea1cc9 | 2465 | old_name = rcu_dereference_protected(cgrp->name, true); |
65dff759 LZ |
2466 | rcu_assign_pointer(cgrp->name, name); |
2467 | ||
2468 | kfree_rcu(old_name, rcu_head); | |
2469 | return 0; | |
ddbcc7e8 PM |
2470 | } |
2471 | ||
03b1cde6 AR |
2472 | static struct simple_xattrs *__d_xattrs(struct dentry *dentry) |
2473 | { | |
2474 | if (S_ISDIR(dentry->d_inode->i_mode)) | |
2475 | return &__d_cgrp(dentry)->xattrs; | |
2476 | else | |
712317ad | 2477 | return &__d_cfe(dentry)->xattrs; |
03b1cde6 AR |
2478 | } |
2479 | ||
2480 | static inline int xattr_enabled(struct dentry *dentry) | |
2481 | { | |
2482 | struct cgroupfs_root *root = dentry->d_sb->s_fs_info; | |
93438629 | 2483 | return root->flags & CGRP_ROOT_XATTR; |
03b1cde6 AR |
2484 | } |
2485 | ||
2486 | static bool is_valid_xattr(const char *name) | |
2487 | { | |
2488 | if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) || | |
2489 | !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN)) | |
2490 | return true; | |
2491 | return false; | |
2492 | } | |
2493 | ||
2494 | static int cgroup_setxattr(struct dentry *dentry, const char *name, | |
2495 | const void *val, size_t size, int flags) | |
2496 | { | |
2497 | if (!xattr_enabled(dentry)) | |
2498 | return -EOPNOTSUPP; | |
2499 | if (!is_valid_xattr(name)) | |
2500 | return -EINVAL; | |
2501 | return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags); | |
2502 | } | |
2503 | ||
2504 | static int cgroup_removexattr(struct dentry *dentry, const char *name) | |
2505 | { | |
2506 | if (!xattr_enabled(dentry)) | |
2507 | return -EOPNOTSUPP; | |
2508 | if (!is_valid_xattr(name)) | |
2509 | return -EINVAL; | |
2510 | return simple_xattr_remove(__d_xattrs(dentry), name); | |
2511 | } | |
2512 | ||
2513 | static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name, | |
2514 | void *buf, size_t size) | |
2515 | { | |
2516 | if (!xattr_enabled(dentry)) | |
2517 | return -EOPNOTSUPP; | |
2518 | if (!is_valid_xattr(name)) | |
2519 | return -EINVAL; | |
2520 | return simple_xattr_get(__d_xattrs(dentry), name, buf, size); | |
2521 | } | |
2522 | ||
2523 | static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size) | |
2524 | { | |
2525 | if (!xattr_enabled(dentry)) | |
2526 | return -EOPNOTSUPP; | |
2527 | return simple_xattr_list(__d_xattrs(dentry), buf, size); | |
2528 | } | |
2529 | ||
828c0950 | 2530 | static const struct file_operations cgroup_file_operations = { |
896f5199 | 2531 | .read = seq_read, |
ddbcc7e8 PM |
2532 | .write = cgroup_file_write, |
2533 | .llseek = generic_file_llseek, | |
2534 | .open = cgroup_file_open, | |
2535 | .release = cgroup_file_release, | |
2536 | }; | |
2537 | ||
03b1cde6 AR |
2538 | static const struct inode_operations cgroup_file_inode_operations = { |
2539 | .setxattr = cgroup_setxattr, | |
2540 | .getxattr = cgroup_getxattr, | |
2541 | .listxattr = cgroup_listxattr, | |
2542 | .removexattr = cgroup_removexattr, | |
2543 | }; | |
2544 | ||
6e1d5dcc | 2545 | static const struct inode_operations cgroup_dir_inode_operations = { |
786e1448 | 2546 | .lookup = simple_lookup, |
ddbcc7e8 PM |
2547 | .mkdir = cgroup_mkdir, |
2548 | .rmdir = cgroup_rmdir, | |
2549 | .rename = cgroup_rename, | |
03b1cde6 AR |
2550 | .setxattr = cgroup_setxattr, |
2551 | .getxattr = cgroup_getxattr, | |
2552 | .listxattr = cgroup_listxattr, | |
2553 | .removexattr = cgroup_removexattr, | |
ddbcc7e8 PM |
2554 | }; |
2555 | ||
a5e7ed32 | 2556 | static int cgroup_create_file(struct dentry *dentry, umode_t mode, |
5adcee1d NP |
2557 | struct super_block *sb) |
2558 | { | |
ddbcc7e8 PM |
2559 | struct inode *inode; |
2560 | ||
2561 | if (!dentry) | |
2562 | return -ENOENT; | |
2563 | if (dentry->d_inode) | |
2564 | return -EEXIST; | |
2565 | ||
2566 | inode = cgroup_new_inode(mode, sb); | |
2567 | if (!inode) | |
2568 | return -ENOMEM; | |
2569 | ||
2570 | if (S_ISDIR(mode)) { | |
2571 | inode->i_op = &cgroup_dir_inode_operations; | |
2572 | inode->i_fop = &simple_dir_operations; | |
2573 | ||
2574 | /* start off with i_nlink == 2 (for "." entry) */ | |
2575 | inc_nlink(inode); | |
28fd6f30 | 2576 | inc_nlink(dentry->d_parent->d_inode); |
ddbcc7e8 | 2577 | |
b8a2df6a TH |
2578 | /* |
2579 | * Control reaches here with cgroup_mutex held. | |
2580 | * @inode->i_mutex should nest outside cgroup_mutex but we | |
2581 | * want to populate it immediately without releasing | |
2582 | * cgroup_mutex. As @inode isn't visible to anyone else | |
2583 | * yet, trylock will always succeed without affecting | |
2584 | * lockdep checks. | |
2585 | */ | |
2586 | WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex)); | |
ddbcc7e8 PM |
2587 | } else if (S_ISREG(mode)) { |
2588 | inode->i_size = 0; | |
2589 | inode->i_fop = &cgroup_file_operations; | |
03b1cde6 | 2590 | inode->i_op = &cgroup_file_inode_operations; |
ddbcc7e8 | 2591 | } |
ddbcc7e8 PM |
2592 | d_instantiate(dentry, inode); |
2593 | dget(dentry); /* Extra count - pin the dentry in core */ | |
2594 | return 0; | |
2595 | } | |
2596 | ||
099fca32 LZ |
2597 | /** |
2598 | * cgroup_file_mode - deduce file mode of a control file | |
2599 | * @cft: the control file in question | |
2600 | * | |
2601 | * returns cft->mode if ->mode is not 0 | |
2602 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
2603 | * returns S_IRUGO if it has only a read handler | |
2604 | * returns S_IWUSR if it has only a write hander | |
2605 | */ | |
a5e7ed32 | 2606 | static umode_t cgroup_file_mode(const struct cftype *cft) |
099fca32 | 2607 | { |
a5e7ed32 | 2608 | umode_t mode = 0; |
099fca32 LZ |
2609 | |
2610 | if (cft->mode) | |
2611 | return cft->mode; | |
2612 | ||
2da8ca82 | 2613 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
099fca32 LZ |
2614 | mode |= S_IRUGO; |
2615 | ||
6e0755b0 TH |
2616 | if (cft->write_u64 || cft->write_s64 || cft->write_string || |
2617 | cft->trigger) | |
099fca32 LZ |
2618 | mode |= S_IWUSR; |
2619 | ||
2620 | return mode; | |
2621 | } | |
2622 | ||
2bb566cb | 2623 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2624 | { |
bd89aabc | 2625 | struct dentry *dir = cgrp->dentry; |
05ef1d7c | 2626 | struct cgroup *parent = __d_cgrp(dir); |
ddbcc7e8 | 2627 | struct dentry *dentry; |
05ef1d7c | 2628 | struct cfent *cfe; |
ddbcc7e8 | 2629 | int error; |
a5e7ed32 | 2630 | umode_t mode; |
8d7e6fb0 | 2631 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 2632 | |
ddbcc7e8 | 2633 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); |
05ef1d7c TH |
2634 | |
2635 | cfe = kzalloc(sizeof(*cfe), GFP_KERNEL); | |
2636 | if (!cfe) | |
2637 | return -ENOMEM; | |
2638 | ||
8d7e6fb0 | 2639 | cgroup_file_name(cgrp, cft, name); |
ddbcc7e8 | 2640 | dentry = lookup_one_len(name, dir, strlen(name)); |
05ef1d7c | 2641 | if (IS_ERR(dentry)) { |
ddbcc7e8 | 2642 | error = PTR_ERR(dentry); |
05ef1d7c TH |
2643 | goto out; |
2644 | } | |
2645 | ||
d6cbf35d LZ |
2646 | cfe->type = (void *)cft; |
2647 | cfe->dentry = dentry; | |
2648 | dentry->d_fsdata = cfe; | |
2649 | simple_xattrs_init(&cfe->xattrs); | |
2650 | ||
05ef1d7c TH |
2651 | mode = cgroup_file_mode(cft); |
2652 | error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb); | |
2653 | if (!error) { | |
05ef1d7c TH |
2654 | list_add_tail(&cfe->node, &parent->files); |
2655 | cfe = NULL; | |
2656 | } | |
2657 | dput(dentry); | |
2658 | out: | |
2659 | kfree(cfe); | |
ddbcc7e8 PM |
2660 | return error; |
2661 | } | |
2662 | ||
b1f28d31 TH |
2663 | /** |
2664 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2665 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2666 | * @cfts: array of cftypes to be added |
2667 | * @is_add: whether to add or remove | |
2668 | * | |
2669 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2670 | * For removals, this function never fails. If addition fails, this |
2671 | * function doesn't remove files already added. The caller is responsible | |
2672 | * for cleaning up. | |
b1f28d31 | 2673 | */ |
2bb566cb TH |
2674 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2675 | bool is_add) | |
ddbcc7e8 | 2676 | { |
03b1cde6 | 2677 | struct cftype *cft; |
b1f28d31 TH |
2678 | int ret; |
2679 | ||
2680 | lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex); | |
ace2bee8 | 2681 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2682 | |
2683 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2684 | /* does cft->flags tell us to skip this file on @cgrp? */ |
873fe09e TH |
2685 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2686 | continue; | |
f33fddc2 G |
2687 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2688 | continue; | |
2689 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2690 | continue; | |
2691 | ||
2739d3cc | 2692 | if (is_add) { |
2bb566cb | 2693 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2694 | if (ret) { |
2739d3cc | 2695 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2696 | cft->name, ret); |
2697 | return ret; | |
2698 | } | |
2739d3cc LZ |
2699 | } else { |
2700 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2701 | } |
ddbcc7e8 | 2702 | } |
b1f28d31 | 2703 | return 0; |
ddbcc7e8 PM |
2704 | } |
2705 | ||
8e3f6541 | 2706 | static void cgroup_cfts_prepare(void) |
e8c82d20 | 2707 | __acquires(&cgroup_mutex) |
8e3f6541 TH |
2708 | { |
2709 | /* | |
2710 | * Thanks to the entanglement with vfs inode locking, we can't walk | |
2711 | * the existing cgroups under cgroup_mutex and create files. | |
492eb21b TH |
2712 | * Instead, we use css_for_each_descendant_pre() and drop RCU read |
2713 | * lock before calling cgroup_addrm_files(). | |
8e3f6541 | 2714 | */ |
ace2bee8 | 2715 | mutex_lock(&cgroup_tree_mutex); |
8e3f6541 TH |
2716 | mutex_lock(&cgroup_mutex); |
2717 | } | |
2718 | ||
2bb566cb | 2719 | static int cgroup_cfts_commit(struct cftype *cfts, bool is_add) |
e8c82d20 | 2720 | __releases(&cgroup_mutex) |
8e3f6541 TH |
2721 | { |
2722 | LIST_HEAD(pending); | |
2bb566cb | 2723 | struct cgroup_subsys *ss = cfts[0].ss; |
492eb21b | 2724 | struct cgroup *root = &ss->root->top_cgroup; |
084457f2 | 2725 | struct super_block *sb = ss->root->sb; |
e8c82d20 LZ |
2726 | struct dentry *prev = NULL; |
2727 | struct inode *inode; | |
492eb21b | 2728 | struct cgroup_subsys_state *css; |
00356bd5 | 2729 | u64 update_before; |
9ccece80 | 2730 | int ret = 0; |
8e3f6541 | 2731 | |
4ac06017 TH |
2732 | mutex_unlock(&cgroup_mutex); |
2733 | ||
8e3f6541 | 2734 | /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */ |
9871bf95 | 2735 | if (!cfts || ss->root == &cgroup_dummy_root || |
e8c82d20 | 2736 | !atomic_inc_not_zero(&sb->s_active)) { |
ace2bee8 | 2737 | mutex_unlock(&cgroup_tree_mutex); |
9ccece80 | 2738 | return 0; |
8e3f6541 TH |
2739 | } |
2740 | ||
8e3f6541 | 2741 | /* |
e8c82d20 LZ |
2742 | * All cgroups which are created after we drop cgroup_mutex will |
2743 | * have the updated set of files, so we only need to update the | |
00356bd5 | 2744 | * cgroups created before the current @cgroup_serial_nr_next. |
8e3f6541 | 2745 | */ |
00356bd5 | 2746 | update_before = cgroup_serial_nr_next; |
e8c82d20 | 2747 | |
e8c82d20 | 2748 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2749 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2750 | struct cgroup *cgrp = css->cgroup; |
2751 | ||
e8c82d20 LZ |
2752 | if (cgroup_is_dead(cgrp)) |
2753 | continue; | |
2754 | ||
2755 | inode = cgrp->dentry->d_inode; | |
2756 | dget(cgrp->dentry); | |
e8c82d20 LZ |
2757 | dput(prev); |
2758 | prev = cgrp->dentry; | |
8e3f6541 | 2759 | |
ace2bee8 | 2760 | mutex_unlock(&cgroup_tree_mutex); |
8e3f6541 | 2761 | mutex_lock(&inode->i_mutex); |
ace2bee8 | 2762 | mutex_lock(&cgroup_tree_mutex); |
00356bd5 | 2763 | if (cgrp->serial_nr < update_before && !cgroup_is_dead(cgrp)) |
2bb566cb | 2764 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
8e3f6541 | 2765 | mutex_unlock(&inode->i_mutex); |
9ccece80 TH |
2766 | if (ret) |
2767 | break; | |
8e3f6541 | 2768 | } |
ace2bee8 | 2769 | mutex_unlock(&cgroup_tree_mutex); |
e8c82d20 LZ |
2770 | dput(prev); |
2771 | deactivate_super(sb); | |
9ccece80 | 2772 | return ret; |
8e3f6541 TH |
2773 | } |
2774 | ||
2775 | /** | |
2776 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2777 | * @ss: target cgroup subsystem | |
2778 | * @cfts: zero-length name terminated array of cftypes | |
2779 | * | |
2780 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2781 | * existing cgroups to which @ss is attached and all future cgroups will | |
2782 | * have them too. This function can be called anytime whether @ss is | |
2783 | * attached or not. | |
2784 | * | |
2785 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2786 | * function currently returns 0 as long as @cfts registration is successful | |
2787 | * even if some file creation attempts on existing cgroups fail. | |
2788 | */ | |
03b1cde6 | 2789 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 TH |
2790 | { |
2791 | struct cftype_set *set; | |
2bb566cb | 2792 | struct cftype *cft; |
9ccece80 | 2793 | int ret; |
8e3f6541 TH |
2794 | |
2795 | set = kzalloc(sizeof(*set), GFP_KERNEL); | |
2796 | if (!set) | |
2797 | return -ENOMEM; | |
2798 | ||
2bb566cb TH |
2799 | for (cft = cfts; cft->name[0] != '\0'; cft++) |
2800 | cft->ss = ss; | |
2801 | ||
8e3f6541 TH |
2802 | cgroup_cfts_prepare(); |
2803 | set->cfts = cfts; | |
2804 | list_add_tail(&set->node, &ss->cftsets); | |
2bb566cb | 2805 | ret = cgroup_cfts_commit(cfts, true); |
9ccece80 | 2806 | if (ret) |
2bb566cb | 2807 | cgroup_rm_cftypes(cfts); |
9ccece80 | 2808 | return ret; |
8e3f6541 TH |
2809 | } |
2810 | EXPORT_SYMBOL_GPL(cgroup_add_cftypes); | |
2811 | ||
79578621 TH |
2812 | /** |
2813 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
2814 | * @cfts: zero-length name terminated array of cftypes |
2815 | * | |
2bb566cb TH |
2816 | * Unregister @cfts. Files described by @cfts are removed from all |
2817 | * existing cgroups and all future cgroups won't have them either. This | |
2818 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
2819 | * |
2820 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 2821 | * registered. |
79578621 | 2822 | */ |
2bb566cb | 2823 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 TH |
2824 | { |
2825 | struct cftype_set *set; | |
2826 | ||
2bb566cb TH |
2827 | if (!cfts || !cfts[0].ss) |
2828 | return -ENOENT; | |
2829 | ||
79578621 TH |
2830 | cgroup_cfts_prepare(); |
2831 | ||
2bb566cb | 2832 | list_for_each_entry(set, &cfts[0].ss->cftsets, node) { |
79578621 | 2833 | if (set->cfts == cfts) { |
f57947d2 LZ |
2834 | list_del(&set->node); |
2835 | kfree(set); | |
2bb566cb | 2836 | cgroup_cfts_commit(cfts, false); |
79578621 TH |
2837 | return 0; |
2838 | } | |
2839 | } | |
2840 | ||
2bb566cb | 2841 | cgroup_cfts_commit(NULL, false); |
79578621 TH |
2842 | return -ENOENT; |
2843 | } | |
2844 | ||
a043e3b2 LZ |
2845 | /** |
2846 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2847 | * @cgrp: the cgroup in question | |
2848 | * | |
2849 | * Return the number of tasks in the cgroup. | |
2850 | */ | |
bd89aabc | 2851 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2852 | { |
2853 | int count = 0; | |
69d0206c | 2854 | struct cgrp_cset_link *link; |
817929ec PM |
2855 | |
2856 | read_lock(&css_set_lock); | |
69d0206c TH |
2857 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2858 | count += atomic_read(&link->cset->refcount); | |
817929ec | 2859 | read_unlock(&css_set_lock); |
bbcb81d0 PM |
2860 | return count; |
2861 | } | |
2862 | ||
817929ec | 2863 | /* |
0942eeee TH |
2864 | * To reduce the fork() overhead for systems that are not actually using |
2865 | * their cgroups capability, we don't maintain the lists running through | |
2866 | * each css_set to its tasks until we see the list actually used - in other | |
72ec7029 | 2867 | * words after the first call to css_task_iter_start(). |
31a7df01 | 2868 | */ |
3df91fe3 | 2869 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2870 | { |
2871 | struct task_struct *p, *g; | |
2872 | write_lock(&css_set_lock); | |
2873 | use_task_css_set_links = 1; | |
3ce3230a FW |
2874 | /* |
2875 | * We need tasklist_lock because RCU is not safe against | |
2876 | * while_each_thread(). Besides, a forking task that has passed | |
2877 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
2878 | * is not guaranteed to have its child immediately visible in the | |
2879 | * tasklist if we walk through it with RCU. | |
2880 | */ | |
2881 | read_lock(&tasklist_lock); | |
31a7df01 CW |
2882 | do_each_thread(g, p) { |
2883 | task_lock(p); | |
0e04388f LZ |
2884 | /* |
2885 | * We should check if the process is exiting, otherwise | |
2886 | * it will race with cgroup_exit() in that the list | |
2887 | * entry won't be deleted though the process has exited. | |
2888 | */ | |
2889 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
a8ad805c | 2890 | list_add(&p->cg_list, &task_css_set(p)->tasks); |
31a7df01 CW |
2891 | task_unlock(p); |
2892 | } while_each_thread(g, p); | |
3ce3230a | 2893 | read_unlock(&tasklist_lock); |
31a7df01 CW |
2894 | write_unlock(&css_set_lock); |
2895 | } | |
2896 | ||
53fa5261 | 2897 | /** |
492eb21b TH |
2898 | * css_next_child - find the next child of a given css |
2899 | * @pos_css: the current position (%NULL to initiate traversal) | |
2900 | * @parent_css: css whose children to walk | |
53fa5261 | 2901 | * |
492eb21b | 2902 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
2903 | * under either cgroup_mutex or RCU read lock. The only requirement is |
2904 | * that @parent_css and @pos_css are accessible. The next sibling is | |
2905 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 2906 | */ |
492eb21b TH |
2907 | struct cgroup_subsys_state * |
2908 | css_next_child(struct cgroup_subsys_state *pos_css, | |
2909 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 2910 | { |
492eb21b TH |
2911 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
2912 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
2913 | struct cgroup *next; |
2914 | ||
ace2bee8 | 2915 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
2916 | |
2917 | /* | |
2918 | * @pos could already have been removed. Once a cgroup is removed, | |
2919 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
2920 | * changes. As CGRP_DEAD assertion is serialized and happens |
2921 | * before the cgroup is taken off the ->sibling list, if we see it | |
2922 | * unasserted, it's guaranteed that the next sibling hasn't | |
2923 | * finished its grace period even if it's already removed, and thus | |
2924 | * safe to dereference from this RCU critical section. If | |
2925 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
2926 | * to be visible as %true here. | |
3b287a50 TH |
2927 | * |
2928 | * If @pos is dead, its next pointer can't be dereferenced; | |
2929 | * however, as each cgroup is given a monotonically increasing | |
2930 | * unique serial number and always appended to the sibling list, | |
2931 | * the next one can be found by walking the parent's children until | |
2932 | * we see a cgroup with higher serial number than @pos's. While | |
2933 | * this path can be slower, it's taken only when either the current | |
2934 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 2935 | */ |
3b287a50 TH |
2936 | if (!pos) { |
2937 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
2938 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 2939 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
2940 | } else { |
2941 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
2942 | if (next->serial_nr > pos->serial_nr) | |
2943 | break; | |
53fa5261 TH |
2944 | } |
2945 | ||
492eb21b TH |
2946 | if (&next->sibling == &cgrp->children) |
2947 | return NULL; | |
2948 | ||
ca8bdcaf | 2949 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 2950 | } |
492eb21b | 2951 | EXPORT_SYMBOL_GPL(css_next_child); |
53fa5261 | 2952 | |
574bd9f7 | 2953 | /** |
492eb21b | 2954 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 2955 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2956 | * @root: css whose descendants to walk |
574bd9f7 | 2957 | * |
492eb21b | 2958 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
2959 | * to visit for pre-order traversal of @root's descendants. @root is |
2960 | * included in the iteration and the first node to be visited. | |
75501a6d | 2961 | * |
87fb54f1 TH |
2962 | * While this function requires cgroup_mutex or RCU read locking, it |
2963 | * doesn't require the whole traversal to be contained in a single critical | |
2964 | * section. This function will return the correct next descendant as long | |
2965 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 2966 | */ |
492eb21b TH |
2967 | struct cgroup_subsys_state * |
2968 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
2969 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2970 | { |
492eb21b | 2971 | struct cgroup_subsys_state *next; |
574bd9f7 | 2972 | |
ace2bee8 | 2973 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2974 | |
bd8815a6 | 2975 | /* if first iteration, visit @root */ |
7805d000 | 2976 | if (!pos) |
bd8815a6 | 2977 | return root; |
574bd9f7 TH |
2978 | |
2979 | /* visit the first child if exists */ | |
492eb21b | 2980 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
2981 | if (next) |
2982 | return next; | |
2983 | ||
2984 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
2985 | while (pos != root) { |
2986 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 2987 | if (next) |
574bd9f7 | 2988 | return next; |
492eb21b | 2989 | pos = css_parent(pos); |
7805d000 | 2990 | } |
574bd9f7 TH |
2991 | |
2992 | return NULL; | |
2993 | } | |
492eb21b | 2994 | EXPORT_SYMBOL_GPL(css_next_descendant_pre); |
574bd9f7 | 2995 | |
12a9d2fe | 2996 | /** |
492eb21b TH |
2997 | * css_rightmost_descendant - return the rightmost descendant of a css |
2998 | * @pos: css of interest | |
12a9d2fe | 2999 | * |
492eb21b TH |
3000 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3001 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3002 | * subtree of @pos. |
75501a6d | 3003 | * |
87fb54f1 TH |
3004 | * While this function requires cgroup_mutex or RCU read locking, it |
3005 | * doesn't require the whole traversal to be contained in a single critical | |
3006 | * section. This function will return the correct rightmost descendant as | |
3007 | * long as @pos is accessible. | |
12a9d2fe | 3008 | */ |
492eb21b TH |
3009 | struct cgroup_subsys_state * |
3010 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3011 | { |
492eb21b | 3012 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3013 | |
ace2bee8 | 3014 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
3015 | |
3016 | do { | |
3017 | last = pos; | |
3018 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3019 | pos = NULL; | |
492eb21b | 3020 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3021 | pos = tmp; |
3022 | } while (pos); | |
3023 | ||
3024 | return last; | |
3025 | } | |
492eb21b | 3026 | EXPORT_SYMBOL_GPL(css_rightmost_descendant); |
12a9d2fe | 3027 | |
492eb21b TH |
3028 | static struct cgroup_subsys_state * |
3029 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3030 | { |
492eb21b | 3031 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3032 | |
3033 | do { | |
3034 | last = pos; | |
492eb21b | 3035 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3036 | } while (pos); |
3037 | ||
3038 | return last; | |
3039 | } | |
3040 | ||
3041 | /** | |
492eb21b | 3042 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3043 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3044 | * @root: css whose descendants to walk |
574bd9f7 | 3045 | * |
492eb21b | 3046 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3047 | * to visit for post-order traversal of @root's descendants. @root is |
3048 | * included in the iteration and the last node to be visited. | |
75501a6d | 3049 | * |
87fb54f1 TH |
3050 | * While this function requires cgroup_mutex or RCU read locking, it |
3051 | * doesn't require the whole traversal to be contained in a single critical | |
3052 | * section. This function will return the correct next descendant as long | |
3053 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3054 | * @cgroup. | |
574bd9f7 | 3055 | */ |
492eb21b TH |
3056 | struct cgroup_subsys_state * |
3057 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3058 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3059 | { |
492eb21b | 3060 | struct cgroup_subsys_state *next; |
574bd9f7 | 3061 | |
ace2bee8 | 3062 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 3063 | |
58b79a91 TH |
3064 | /* if first iteration, visit leftmost descendant which may be @root */ |
3065 | if (!pos) | |
3066 | return css_leftmost_descendant(root); | |
574bd9f7 | 3067 | |
bd8815a6 TH |
3068 | /* if we visited @root, we're done */ |
3069 | if (pos == root) | |
3070 | return NULL; | |
3071 | ||
574bd9f7 | 3072 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 3073 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 3074 | if (next) |
492eb21b | 3075 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3076 | |
3077 | /* no sibling left, visit parent */ | |
bd8815a6 | 3078 | return css_parent(pos); |
574bd9f7 | 3079 | } |
492eb21b | 3080 | EXPORT_SYMBOL_GPL(css_next_descendant_post); |
574bd9f7 | 3081 | |
0942eeee | 3082 | /** |
72ec7029 | 3083 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
3084 | * @it: the iterator to advance |
3085 | * | |
3086 | * Advance @it to the next css_set to walk. | |
d515876e | 3087 | */ |
72ec7029 | 3088 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
3089 | { |
3090 | struct list_head *l = it->cset_link; | |
3091 | struct cgrp_cset_link *link; | |
3092 | struct css_set *cset; | |
3093 | ||
3094 | /* Advance to the next non-empty css_set */ | |
3095 | do { | |
3096 | l = l->next; | |
72ec7029 | 3097 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
3098 | it->cset_link = NULL; |
3099 | return; | |
3100 | } | |
3101 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3102 | cset = link->cset; | |
3103 | } while (list_empty(&cset->tasks)); | |
3104 | it->cset_link = l; | |
3105 | it->task = cset->tasks.next; | |
3106 | } | |
3107 | ||
0942eeee | 3108 | /** |
72ec7029 TH |
3109 | * css_task_iter_start - initiate task iteration |
3110 | * @css: the css to walk tasks of | |
0942eeee TH |
3111 | * @it: the task iterator to use |
3112 | * | |
72ec7029 TH |
3113 | * Initiate iteration through the tasks of @css. The caller can call |
3114 | * css_task_iter_next() to walk through the tasks until the function | |
3115 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3116 | * called. | |
0942eeee TH |
3117 | * |
3118 | * Note that this function acquires a lock which is released when the | |
3119 | * iteration finishes. The caller can't sleep while iteration is in | |
3120 | * progress. | |
3121 | */ | |
72ec7029 TH |
3122 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3123 | struct css_task_iter *it) | |
c6ca5750 | 3124 | __acquires(css_set_lock) |
817929ec PM |
3125 | { |
3126 | /* | |
72ec7029 TH |
3127 | * The first time anyone tries to iterate across a css, we need to |
3128 | * enable the list linking each css_set to its tasks, and fix up | |
3129 | * all existing tasks. | |
817929ec | 3130 | */ |
31a7df01 CW |
3131 | if (!use_task_css_set_links) |
3132 | cgroup_enable_task_cg_lists(); | |
3133 | ||
817929ec | 3134 | read_lock(&css_set_lock); |
c59cd3d8 | 3135 | |
72ec7029 TH |
3136 | it->origin_css = css; |
3137 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 3138 | |
72ec7029 | 3139 | css_advance_task_iter(it); |
817929ec PM |
3140 | } |
3141 | ||
0942eeee | 3142 | /** |
72ec7029 | 3143 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3144 | * @it: the task iterator being iterated |
3145 | * | |
3146 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3147 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3148 | * reaches the end. | |
0942eeee | 3149 | */ |
72ec7029 | 3150 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
3151 | { |
3152 | struct task_struct *res; | |
3153 | struct list_head *l = it->task; | |
69d0206c | 3154 | struct cgrp_cset_link *link; |
817929ec PM |
3155 | |
3156 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 3157 | if (!it->cset_link) |
817929ec PM |
3158 | return NULL; |
3159 | res = list_entry(l, struct task_struct, cg_list); | |
3160 | /* Advance iterator to find next entry */ | |
3161 | l = l->next; | |
69d0206c TH |
3162 | link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link); |
3163 | if (l == &link->cset->tasks) { | |
0942eeee TH |
3164 | /* |
3165 | * We reached the end of this task list - move on to the | |
3166 | * next cgrp_cset_link. | |
3167 | */ | |
72ec7029 | 3168 | css_advance_task_iter(it); |
817929ec PM |
3169 | } else { |
3170 | it->task = l; | |
3171 | } | |
3172 | return res; | |
3173 | } | |
3174 | ||
0942eeee | 3175 | /** |
72ec7029 | 3176 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3177 | * @it: the task iterator to finish |
3178 | * | |
72ec7029 | 3179 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3180 | */ |
72ec7029 | 3181 | void css_task_iter_end(struct css_task_iter *it) |
c6ca5750 | 3182 | __releases(css_set_lock) |
817929ec PM |
3183 | { |
3184 | read_unlock(&css_set_lock); | |
3185 | } | |
3186 | ||
31a7df01 CW |
3187 | static inline int started_after_time(struct task_struct *t1, |
3188 | struct timespec *time, | |
3189 | struct task_struct *t2) | |
3190 | { | |
3191 | int start_diff = timespec_compare(&t1->start_time, time); | |
3192 | if (start_diff > 0) { | |
3193 | return 1; | |
3194 | } else if (start_diff < 0) { | |
3195 | return 0; | |
3196 | } else { | |
3197 | /* | |
3198 | * Arbitrarily, if two processes started at the same | |
3199 | * time, we'll say that the lower pointer value | |
3200 | * started first. Note that t2 may have exited by now | |
3201 | * so this may not be a valid pointer any longer, but | |
3202 | * that's fine - it still serves to distinguish | |
3203 | * between two tasks started (effectively) simultaneously. | |
3204 | */ | |
3205 | return t1 > t2; | |
3206 | } | |
3207 | } | |
3208 | ||
3209 | /* | |
3210 | * This function is a callback from heap_insert() and is used to order | |
3211 | * the heap. | |
3212 | * In this case we order the heap in descending task start time. | |
3213 | */ | |
3214 | static inline int started_after(void *p1, void *p2) | |
3215 | { | |
3216 | struct task_struct *t1 = p1; | |
3217 | struct task_struct *t2 = p2; | |
3218 | return started_after_time(t1, &t2->start_time, t2); | |
3219 | } | |
3220 | ||
3221 | /** | |
72ec7029 TH |
3222 | * css_scan_tasks - iterate though all the tasks in a css |
3223 | * @css: the css to iterate tasks of | |
e535837b TH |
3224 | * @test: optional test callback |
3225 | * @process: process callback | |
3226 | * @data: data passed to @test and @process | |
3227 | * @heap: optional pre-allocated heap used for task iteration | |
31a7df01 | 3228 | * |
72ec7029 TH |
3229 | * Iterate through all the tasks in @css, calling @test for each, and if it |
3230 | * returns %true, call @process for it also. | |
31a7df01 | 3231 | * |
e535837b | 3232 | * @test may be NULL, meaning always true (select all tasks), which |
72ec7029 | 3233 | * effectively duplicates css_task_iter_{start,next,end}() but does not |
e535837b TH |
3234 | * lock css_set_lock for the call to @process. |
3235 | * | |
3236 | * It is guaranteed that @process will act on every task that is a member | |
72ec7029 TH |
3237 | * of @css for the duration of this call. This function may or may not |
3238 | * call @process for tasks that exit or move to a different css during the | |
3239 | * call, or are forked or move into the css during the call. | |
31a7df01 | 3240 | * |
e535837b TH |
3241 | * Note that @test may be called with locks held, and may in some |
3242 | * situations be called multiple times for the same task, so it should be | |
3243 | * cheap. | |
31a7df01 | 3244 | * |
e535837b TH |
3245 | * If @heap is non-NULL, a heap has been pre-allocated and will be used for |
3246 | * heap operations (and its "gt" member will be overwritten), else a | |
3247 | * temporary heap will be used (allocation of which may cause this function | |
3248 | * to fail). | |
31a7df01 | 3249 | */ |
72ec7029 TH |
3250 | int css_scan_tasks(struct cgroup_subsys_state *css, |
3251 | bool (*test)(struct task_struct *, void *), | |
3252 | void (*process)(struct task_struct *, void *), | |
3253 | void *data, struct ptr_heap *heap) | |
31a7df01 CW |
3254 | { |
3255 | int retval, i; | |
72ec7029 | 3256 | struct css_task_iter it; |
31a7df01 CW |
3257 | struct task_struct *p, *dropped; |
3258 | /* Never dereference latest_task, since it's not refcounted */ | |
3259 | struct task_struct *latest_task = NULL; | |
3260 | struct ptr_heap tmp_heap; | |
31a7df01 CW |
3261 | struct timespec latest_time = { 0, 0 }; |
3262 | ||
e535837b | 3263 | if (heap) { |
31a7df01 | 3264 | /* The caller supplied our heap and pre-allocated its memory */ |
31a7df01 CW |
3265 | heap->gt = &started_after; |
3266 | } else { | |
3267 | /* We need to allocate our own heap memory */ | |
3268 | heap = &tmp_heap; | |
3269 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
3270 | if (retval) | |
3271 | /* cannot allocate the heap */ | |
3272 | return retval; | |
3273 | } | |
3274 | ||
3275 | again: | |
3276 | /* | |
72ec7029 | 3277 | * Scan tasks in the css, using the @test callback to determine |
e535837b TH |
3278 | * which are of interest, and invoking @process callback on the |
3279 | * ones which need an update. Since we don't want to hold any | |
3280 | * locks during the task updates, gather tasks to be processed in a | |
3281 | * heap structure. The heap is sorted by descending task start | |
3282 | * time. If the statically-sized heap fills up, we overflow tasks | |
3283 | * that started later, and in future iterations only consider tasks | |
3284 | * that started after the latest task in the previous pass. This | |
31a7df01 CW |
3285 | * guarantees forward progress and that we don't miss any tasks. |
3286 | */ | |
3287 | heap->size = 0; | |
72ec7029 TH |
3288 | css_task_iter_start(css, &it); |
3289 | while ((p = css_task_iter_next(&it))) { | |
31a7df01 CW |
3290 | /* |
3291 | * Only affect tasks that qualify per the caller's callback, | |
3292 | * if he provided one | |
3293 | */ | |
e535837b | 3294 | if (test && !test(p, data)) |
31a7df01 CW |
3295 | continue; |
3296 | /* | |
3297 | * Only process tasks that started after the last task | |
3298 | * we processed | |
3299 | */ | |
3300 | if (!started_after_time(p, &latest_time, latest_task)) | |
3301 | continue; | |
3302 | dropped = heap_insert(heap, p); | |
3303 | if (dropped == NULL) { | |
3304 | /* | |
3305 | * The new task was inserted; the heap wasn't | |
3306 | * previously full | |
3307 | */ | |
3308 | get_task_struct(p); | |
3309 | } else if (dropped != p) { | |
3310 | /* | |
3311 | * The new task was inserted, and pushed out a | |
3312 | * different task | |
3313 | */ | |
3314 | get_task_struct(p); | |
3315 | put_task_struct(dropped); | |
3316 | } | |
3317 | /* | |
3318 | * Else the new task was newer than anything already in | |
3319 | * the heap and wasn't inserted | |
3320 | */ | |
3321 | } | |
72ec7029 | 3322 | css_task_iter_end(&it); |
31a7df01 CW |
3323 | |
3324 | if (heap->size) { | |
3325 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 3326 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 3327 | if (i == 0) { |
4fe91d51 PJ |
3328 | latest_time = q->start_time; |
3329 | latest_task = q; | |
31a7df01 CW |
3330 | } |
3331 | /* Process the task per the caller's callback */ | |
e535837b | 3332 | process(q, data); |
4fe91d51 | 3333 | put_task_struct(q); |
31a7df01 CW |
3334 | } |
3335 | /* | |
3336 | * If we had to process any tasks at all, scan again | |
3337 | * in case some of them were in the middle of forking | |
3338 | * children that didn't get processed. | |
3339 | * Not the most efficient way to do it, but it avoids | |
3340 | * having to take callback_mutex in the fork path | |
3341 | */ | |
3342 | goto again; | |
3343 | } | |
3344 | if (heap == &tmp_heap) | |
3345 | heap_free(&tmp_heap); | |
3346 | return 0; | |
3347 | } | |
3348 | ||
e535837b | 3349 | static void cgroup_transfer_one_task(struct task_struct *task, void *data) |
8cc99345 | 3350 | { |
e535837b | 3351 | struct cgroup *new_cgroup = data; |
8cc99345 | 3352 | |
47cfcd09 | 3353 | mutex_lock(&cgroup_mutex); |
8cc99345 | 3354 | cgroup_attach_task(new_cgroup, task, false); |
47cfcd09 | 3355 | mutex_unlock(&cgroup_mutex); |
8cc99345 TH |
3356 | } |
3357 | ||
3358 | /** | |
3359 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
3360 | * @to: cgroup to which the tasks will be moved | |
3361 | * @from: cgroup in which the tasks currently reside | |
3362 | */ | |
3363 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
3364 | { | |
72ec7029 TH |
3365 | return css_scan_tasks(&from->dummy_css, NULL, cgroup_transfer_one_task, |
3366 | to, NULL); | |
8cc99345 TH |
3367 | } |
3368 | ||
bbcb81d0 | 3369 | /* |
102a775e | 3370 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3371 | * |
3372 | * Reading this file can return large amounts of data if a cgroup has | |
3373 | * *lots* of attached tasks. So it may need several calls to read(), | |
3374 | * but we cannot guarantee that the information we produce is correct | |
3375 | * unless we produce it entirely atomically. | |
3376 | * | |
bbcb81d0 | 3377 | */ |
bbcb81d0 | 3378 | |
24528255 LZ |
3379 | /* which pidlist file are we talking about? */ |
3380 | enum cgroup_filetype { | |
3381 | CGROUP_FILE_PROCS, | |
3382 | CGROUP_FILE_TASKS, | |
3383 | }; | |
3384 | ||
3385 | /* | |
3386 | * A pidlist is a list of pids that virtually represents the contents of one | |
3387 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3388 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3389 | * to the cgroup. | |
3390 | */ | |
3391 | struct cgroup_pidlist { | |
3392 | /* | |
3393 | * used to find which pidlist is wanted. doesn't change as long as | |
3394 | * this particular list stays in the list. | |
3395 | */ | |
3396 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3397 | /* array of xids */ | |
3398 | pid_t *list; | |
3399 | /* how many elements the above list has */ | |
3400 | int length; | |
24528255 LZ |
3401 | /* each of these stored in a list by its cgroup */ |
3402 | struct list_head links; | |
3403 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3404 | struct cgroup *owner; | |
b1a21367 TH |
3405 | /* for delayed destruction */ |
3406 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3407 | }; |
3408 | ||
d1d9fd33 BB |
3409 | /* |
3410 | * The following two functions "fix" the issue where there are more pids | |
3411 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3412 | * TODO: replace with a kernel-wide solution to this problem | |
3413 | */ | |
3414 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3415 | static void *pidlist_allocate(int count) | |
3416 | { | |
3417 | if (PIDLIST_TOO_LARGE(count)) | |
3418 | return vmalloc(count * sizeof(pid_t)); | |
3419 | else | |
3420 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3421 | } | |
b1a21367 | 3422 | |
d1d9fd33 BB |
3423 | static void pidlist_free(void *p) |
3424 | { | |
3425 | if (is_vmalloc_addr(p)) | |
3426 | vfree(p); | |
3427 | else | |
3428 | kfree(p); | |
3429 | } | |
d1d9fd33 | 3430 | |
b1a21367 TH |
3431 | /* |
3432 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3433 | * should be left afterwards. | |
3434 | */ | |
3435 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3436 | { | |
3437 | struct cgroup_pidlist *l, *tmp_l; | |
3438 | ||
3439 | mutex_lock(&cgrp->pidlist_mutex); | |
3440 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3441 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3442 | mutex_unlock(&cgrp->pidlist_mutex); | |
3443 | ||
3444 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3445 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3446 | } | |
3447 | ||
3448 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3449 | { | |
3450 | struct delayed_work *dwork = to_delayed_work(work); | |
3451 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3452 | destroy_dwork); | |
3453 | struct cgroup_pidlist *tofree = NULL; | |
3454 | ||
3455 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3456 | |
3457 | /* | |
04502365 TH |
3458 | * Destroy iff we didn't get queued again. The state won't change |
3459 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3460 | */ |
04502365 | 3461 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3462 | list_del(&l->links); |
3463 | pidlist_free(l->list); | |
3464 | put_pid_ns(l->key.ns); | |
3465 | tofree = l; | |
3466 | } | |
3467 | ||
b1a21367 TH |
3468 | mutex_unlock(&l->owner->pidlist_mutex); |
3469 | kfree(tofree); | |
3470 | } | |
3471 | ||
bbcb81d0 | 3472 | /* |
102a775e | 3473 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3474 | * Returns the number of unique elements. |
bbcb81d0 | 3475 | */ |
6ee211ad | 3476 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3477 | { |
102a775e | 3478 | int src, dest = 1; |
102a775e BB |
3479 | |
3480 | /* | |
3481 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3482 | * edge cases first; no work needs to be done for either | |
3483 | */ | |
3484 | if (length == 0 || length == 1) | |
3485 | return length; | |
3486 | /* src and dest walk down the list; dest counts unique elements */ | |
3487 | for (src = 1; src < length; src++) { | |
3488 | /* find next unique element */ | |
3489 | while (list[src] == list[src-1]) { | |
3490 | src++; | |
3491 | if (src == length) | |
3492 | goto after; | |
3493 | } | |
3494 | /* dest always points to where the next unique element goes */ | |
3495 | list[dest] = list[src]; | |
3496 | dest++; | |
3497 | } | |
3498 | after: | |
102a775e BB |
3499 | return dest; |
3500 | } | |
3501 | ||
afb2bc14 TH |
3502 | /* |
3503 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3504 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3505 | * different per namespace, each namespace needs differently sorted list, | |
3506 | * making it impossible to use, for example, single rbtree of member tasks | |
3507 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3508 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3509 | * pidlists keyed by cgroup and namespace. | |
3510 | * | |
3511 | * All this extra complexity was caused by the original implementation | |
3512 | * committing to an entirely unnecessary property. In the long term, we | |
3513 | * want to do away with it. Explicitly scramble sort order if | |
3514 | * sane_behavior so that no such expectation exists in the new interface. | |
3515 | * | |
3516 | * Scrambling is done by swapping every two consecutive bits, which is | |
3517 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3518 | */ | |
3519 | static pid_t pid_fry(pid_t pid) | |
3520 | { | |
3521 | unsigned a = pid & 0x55555555; | |
3522 | unsigned b = pid & 0xAAAAAAAA; | |
3523 | ||
3524 | return (a << 1) | (b >> 1); | |
3525 | } | |
3526 | ||
3527 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3528 | { | |
3529 | if (cgroup_sane_behavior(cgrp)) | |
3530 | return pid_fry(pid); | |
3531 | else | |
3532 | return pid; | |
3533 | } | |
3534 | ||
102a775e BB |
3535 | static int cmppid(const void *a, const void *b) |
3536 | { | |
3537 | return *(pid_t *)a - *(pid_t *)b; | |
3538 | } | |
3539 | ||
afb2bc14 TH |
3540 | static int fried_cmppid(const void *a, const void *b) |
3541 | { | |
3542 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3543 | } | |
3544 | ||
e6b81710 TH |
3545 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3546 | enum cgroup_filetype type) | |
3547 | { | |
3548 | struct cgroup_pidlist *l; | |
3549 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3550 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3551 | ||
3552 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3553 | ||
3554 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3555 | if (l->key.type == type && l->key.ns == ns) | |
3556 | return l; | |
3557 | return NULL; | |
3558 | } | |
3559 | ||
72a8cb30 BB |
3560 | /* |
3561 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3562 | * returns with the lock on that pidlist already held, and takes care | |
3563 | * of the use count, or returns NULL with no locks held if we're out of | |
3564 | * memory. | |
3565 | */ | |
e6b81710 TH |
3566 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3567 | enum cgroup_filetype type) | |
72a8cb30 BB |
3568 | { |
3569 | struct cgroup_pidlist *l; | |
b70cc5fd | 3570 | |
e6b81710 TH |
3571 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3572 | ||
3573 | l = cgroup_pidlist_find(cgrp, type); | |
3574 | if (l) | |
3575 | return l; | |
3576 | ||
72a8cb30 | 3577 | /* entry not found; create a new one */ |
f4f4be2b | 3578 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3579 | if (!l) |
72a8cb30 | 3580 | return l; |
e6b81710 | 3581 | |
b1a21367 | 3582 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3583 | l->key.type = type; |
e6b81710 TH |
3584 | /* don't need task_nsproxy() if we're looking at ourself */ |
3585 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3586 | l->owner = cgrp; |
3587 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3588 | return l; |
3589 | } | |
3590 | ||
102a775e BB |
3591 | /* |
3592 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3593 | */ | |
72a8cb30 BB |
3594 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3595 | struct cgroup_pidlist **lp) | |
102a775e BB |
3596 | { |
3597 | pid_t *array; | |
3598 | int length; | |
3599 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3600 | struct css_task_iter it; |
817929ec | 3601 | struct task_struct *tsk; |
102a775e BB |
3602 | struct cgroup_pidlist *l; |
3603 | ||
4bac00d1 TH |
3604 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3605 | ||
102a775e BB |
3606 | /* |
3607 | * If cgroup gets more users after we read count, we won't have | |
3608 | * enough space - tough. This race is indistinguishable to the | |
3609 | * caller from the case that the additional cgroup users didn't | |
3610 | * show up until sometime later on. | |
3611 | */ | |
3612 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3613 | array = pidlist_allocate(length); |
102a775e BB |
3614 | if (!array) |
3615 | return -ENOMEM; | |
3616 | /* now, populate the array */ | |
72ec7029 TH |
3617 | css_task_iter_start(&cgrp->dummy_css, &it); |
3618 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3619 | if (unlikely(n == length)) |
817929ec | 3620 | break; |
102a775e | 3621 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3622 | if (type == CGROUP_FILE_PROCS) |
3623 | pid = task_tgid_vnr(tsk); | |
3624 | else | |
3625 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3626 | if (pid > 0) /* make sure to only use valid results */ |
3627 | array[n++] = pid; | |
817929ec | 3628 | } |
72ec7029 | 3629 | css_task_iter_end(&it); |
102a775e BB |
3630 | length = n; |
3631 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3632 | if (cgroup_sane_behavior(cgrp)) |
3633 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3634 | else | |
3635 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3636 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3637 | length = pidlist_uniq(array, length); |
e6b81710 | 3638 | |
e6b81710 | 3639 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3640 | if (!l) { |
e6b81710 | 3641 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3642 | pidlist_free(array); |
72a8cb30 | 3643 | return -ENOMEM; |
102a775e | 3644 | } |
e6b81710 TH |
3645 | |
3646 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3647 | pidlist_free(l->list); |
102a775e BB |
3648 | l->list = array; |
3649 | l->length = length; | |
72a8cb30 | 3650 | *lp = l; |
102a775e | 3651 | return 0; |
bbcb81d0 PM |
3652 | } |
3653 | ||
846c7bb0 | 3654 | /** |
a043e3b2 | 3655 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3656 | * @stats: cgroupstats to fill information into |
3657 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3658 | * been requested. | |
a043e3b2 LZ |
3659 | * |
3660 | * Build and fill cgroupstats so that taskstats can export it to user | |
3661 | * space. | |
846c7bb0 BS |
3662 | */ |
3663 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3664 | { | |
3665 | int ret = -EINVAL; | |
bd89aabc | 3666 | struct cgroup *cgrp; |
72ec7029 | 3667 | struct css_task_iter it; |
846c7bb0 | 3668 | struct task_struct *tsk; |
33d283be | 3669 | |
846c7bb0 | 3670 | /* |
33d283be LZ |
3671 | * Validate dentry by checking the superblock operations, |
3672 | * and make sure it's a directory. | |
846c7bb0 | 3673 | */ |
33d283be LZ |
3674 | if (dentry->d_sb->s_op != &cgroup_ops || |
3675 | !S_ISDIR(dentry->d_inode->i_mode)) | |
846c7bb0 BS |
3676 | goto err; |
3677 | ||
3678 | ret = 0; | |
bd89aabc | 3679 | cgrp = dentry->d_fsdata; |
846c7bb0 | 3680 | |
72ec7029 TH |
3681 | css_task_iter_start(&cgrp->dummy_css, &it); |
3682 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3683 | switch (tsk->state) { |
3684 | case TASK_RUNNING: | |
3685 | stats->nr_running++; | |
3686 | break; | |
3687 | case TASK_INTERRUPTIBLE: | |
3688 | stats->nr_sleeping++; | |
3689 | break; | |
3690 | case TASK_UNINTERRUPTIBLE: | |
3691 | stats->nr_uninterruptible++; | |
3692 | break; | |
3693 | case TASK_STOPPED: | |
3694 | stats->nr_stopped++; | |
3695 | break; | |
3696 | default: | |
3697 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3698 | stats->nr_io_wait++; | |
3699 | break; | |
3700 | } | |
3701 | } | |
72ec7029 | 3702 | css_task_iter_end(&it); |
846c7bb0 | 3703 | |
846c7bb0 BS |
3704 | err: |
3705 | return ret; | |
3706 | } | |
3707 | ||
8f3ff208 | 3708 | |
bbcb81d0 | 3709 | /* |
102a775e | 3710 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3711 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3712 | * in the cgroup->l->list array. |
bbcb81d0 | 3713 | */ |
cc31edce | 3714 | |
102a775e | 3715 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3716 | { |
cc31edce PM |
3717 | /* |
3718 | * Initially we receive a position value that corresponds to | |
3719 | * one more than the last pid shown (or 0 on the first call or | |
3720 | * after a seek to the start). Use a binary-search to find the | |
3721 | * next pid to display, if any | |
3722 | */ | |
5d22444f | 3723 | struct cgroup_open_file *of = s->private; |
7da11279 | 3724 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3725 | struct cgroup_pidlist *l; |
7da11279 | 3726 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3727 | int index = 0, pid = *pos; |
4bac00d1 TH |
3728 | int *iter, ret; |
3729 | ||
3730 | mutex_lock(&cgrp->pidlist_mutex); | |
3731 | ||
3732 | /* | |
5d22444f | 3733 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3734 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3735 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3736 | * could already have been destroyed. |
3737 | */ | |
5d22444f TH |
3738 | if (of->priv) |
3739 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3740 | |
3741 | /* | |
3742 | * Either this is the first start() after open or the matching | |
3743 | * pidlist has been destroyed inbetween. Create a new one. | |
3744 | */ | |
5d22444f TH |
3745 | if (!of->priv) { |
3746 | ret = pidlist_array_load(cgrp, type, | |
3747 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3748 | if (ret) |
3749 | return ERR_PTR(ret); | |
3750 | } | |
5d22444f | 3751 | l = of->priv; |
cc31edce | 3752 | |
cc31edce | 3753 | if (pid) { |
102a775e | 3754 | int end = l->length; |
20777766 | 3755 | |
cc31edce PM |
3756 | while (index < end) { |
3757 | int mid = (index + end) / 2; | |
afb2bc14 | 3758 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3759 | index = mid; |
3760 | break; | |
afb2bc14 | 3761 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3762 | index = mid + 1; |
3763 | else | |
3764 | end = mid; | |
3765 | } | |
3766 | } | |
3767 | /* If we're off the end of the array, we're done */ | |
102a775e | 3768 | if (index >= l->length) |
cc31edce PM |
3769 | return NULL; |
3770 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3771 | iter = l->list + index; |
afb2bc14 | 3772 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3773 | return iter; |
3774 | } | |
3775 | ||
102a775e | 3776 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3777 | { |
5d22444f TH |
3778 | struct cgroup_open_file *of = s->private; |
3779 | struct cgroup_pidlist *l = of->priv; | |
62236858 | 3780 | |
5d22444f TH |
3781 | if (l) |
3782 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3783 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3784 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3785 | } |
3786 | ||
102a775e | 3787 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3788 | { |
5d22444f TH |
3789 | struct cgroup_open_file *of = s->private; |
3790 | struct cgroup_pidlist *l = of->priv; | |
102a775e BB |
3791 | pid_t *p = v; |
3792 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3793 | /* |
3794 | * Advance to the next pid in the array. If this goes off the | |
3795 | * end, we're done | |
3796 | */ | |
3797 | p++; | |
3798 | if (p >= end) { | |
3799 | return NULL; | |
3800 | } else { | |
7da11279 | 3801 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3802 | return p; |
3803 | } | |
3804 | } | |
3805 | ||
102a775e | 3806 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3807 | { |
3808 | return seq_printf(s, "%d\n", *(int *)v); | |
3809 | } | |
bbcb81d0 | 3810 | |
102a775e BB |
3811 | /* |
3812 | * seq_operations functions for iterating on pidlists through seq_file - | |
3813 | * independent of whether it's tasks or procs | |
3814 | */ | |
3815 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3816 | .start = cgroup_pidlist_start, | |
3817 | .stop = cgroup_pidlist_stop, | |
3818 | .next = cgroup_pidlist_next, | |
3819 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3820 | }; |
3821 | ||
182446d0 TH |
3822 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3823 | struct cftype *cft) | |
81a6a5cd | 3824 | { |
182446d0 | 3825 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3826 | } |
3827 | ||
182446d0 TH |
3828 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3829 | struct cftype *cft, u64 val) | |
6379c106 | 3830 | { |
182446d0 | 3831 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3832 | if (val) |
182446d0 | 3833 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3834 | else |
182446d0 | 3835 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3836 | return 0; |
3837 | } | |
3838 | ||
1c8158ee LZ |
3839 | /* |
3840 | * When dput() is called asynchronously, if umount has been done and | |
3841 | * then deactivate_super() in cgroup_free_fn() kills the superblock, | |
3842 | * there's a small window that vfs will see the root dentry with non-zero | |
3843 | * refcnt and trigger BUG(). | |
3844 | * | |
3845 | * That's why we hold a reference before dput() and drop it right after. | |
3846 | */ | |
3847 | static void cgroup_dput(struct cgroup *cgrp) | |
3848 | { | |
3849 | struct super_block *sb = cgrp->root->sb; | |
3850 | ||
3851 | atomic_inc(&sb->s_active); | |
3852 | dput(cgrp->dentry); | |
3853 | deactivate_super(sb); | |
3854 | } | |
3855 | ||
182446d0 TH |
3856 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3857 | struct cftype *cft) | |
97978e6d | 3858 | { |
182446d0 | 3859 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3860 | } |
3861 | ||
182446d0 TH |
3862 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3863 | struct cftype *cft, u64 val) | |
97978e6d DL |
3864 | { |
3865 | if (val) | |
182446d0 | 3866 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3867 | else |
182446d0 | 3868 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3869 | return 0; |
3870 | } | |
3871 | ||
d5c56ced | 3872 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3873 | { |
d5c56ced | 3874 | .name = "cgroup.procs", |
6612f05b TH |
3875 | .seq_start = cgroup_pidlist_start, |
3876 | .seq_next = cgroup_pidlist_next, | |
3877 | .seq_stop = cgroup_pidlist_stop, | |
3878 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3879 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3880 | .write_u64 = cgroup_procs_write, |
74a1166d | 3881 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3882 | }, |
97978e6d DL |
3883 | { |
3884 | .name = "cgroup.clone_children", | |
873fe09e | 3885 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3886 | .read_u64 = cgroup_clone_children_read, |
3887 | .write_u64 = cgroup_clone_children_write, | |
3888 | }, | |
873fe09e TH |
3889 | { |
3890 | .name = "cgroup.sane_behavior", | |
3891 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3892 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3893 | }, |
d5c56ced TH |
3894 | |
3895 | /* | |
3896 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3897 | * don't exist if sane_behavior. If you're depending on these, be | |
3898 | * prepared to be burned. | |
3899 | */ | |
3900 | { | |
3901 | .name = "tasks", | |
3902 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3903 | .seq_start = cgroup_pidlist_start, |
3904 | .seq_next = cgroup_pidlist_next, | |
3905 | .seq_stop = cgroup_pidlist_stop, | |
3906 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3907 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3908 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3909 | .mode = S_IRUGO | S_IWUSR, |
3910 | }, | |
3911 | { | |
3912 | .name = "notify_on_release", | |
3913 | .flags = CFTYPE_INSANE, | |
3914 | .read_u64 = cgroup_read_notify_on_release, | |
3915 | .write_u64 = cgroup_write_notify_on_release, | |
3916 | }, | |
6e6ff25b TH |
3917 | { |
3918 | .name = "release_agent", | |
cc5943a7 | 3919 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3920 | .seq_show = cgroup_release_agent_show, |
6e6ff25b TH |
3921 | .write_string = cgroup_release_agent_write, |
3922 | .max_write_len = PATH_MAX, | |
3923 | }, | |
db0416b6 | 3924 | { } /* terminate */ |
bbcb81d0 PM |
3925 | }; |
3926 | ||
13af07df | 3927 | /** |
628f7cd4 | 3928 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3929 | * @cgrp: target cgroup |
13af07df | 3930 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3931 | * |
3932 | * On failure, no file is added. | |
13af07df | 3933 | */ |
628f7cd4 | 3934 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3935 | { |
ddbcc7e8 | 3936 | struct cgroup_subsys *ss; |
b420ba7d | 3937 | int i, ret = 0; |
bbcb81d0 | 3938 | |
8e3f6541 | 3939 | /* process cftsets of each subsystem */ |
b420ba7d | 3940 | for_each_subsys(ss, i) { |
8e3f6541 | 3941 | struct cftype_set *set; |
b420ba7d TH |
3942 | |
3943 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3944 | continue; |
8e3f6541 | 3945 | |
bee55099 | 3946 | list_for_each_entry(set, &ss->cftsets, node) { |
2bb566cb | 3947 | ret = cgroup_addrm_files(cgrp, set->cfts, true); |
bee55099 TH |
3948 | if (ret < 0) |
3949 | goto err; | |
3950 | } | |
ddbcc7e8 | 3951 | } |
ddbcc7e8 | 3952 | return 0; |
bee55099 TH |
3953 | err: |
3954 | cgroup_clear_dir(cgrp, subsys_mask); | |
3955 | return ret; | |
ddbcc7e8 PM |
3956 | } |
3957 | ||
0c21ead1 TH |
3958 | /* |
3959 | * css destruction is four-stage process. | |
3960 | * | |
3961 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3962 | * Implemented in kill_css(). | |
3963 | * | |
3964 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3965 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3966 | * by invoking offline_css(). After offlining, the base ref is put. | |
3967 | * Implemented in css_killed_work_fn(). | |
3968 | * | |
3969 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3970 | * accessors are inside RCU read sections. css_release() schedules the | |
3971 | * RCU callback. | |
3972 | * | |
3973 | * 4. After the grace period, the css can be freed. Implemented in | |
3974 | * css_free_work_fn(). | |
3975 | * | |
3976 | * It is actually hairier because both step 2 and 4 require process context | |
3977 | * and thus involve punting to css->destroy_work adding two additional | |
3978 | * steps to the already complex sequence. | |
3979 | */ | |
35ef10da | 3980 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3981 | { |
3982 | struct cgroup_subsys_state *css = | |
35ef10da | 3983 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 3984 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3985 | |
0ae78e0b TH |
3986 | if (css->parent) |
3987 | css_put(css->parent); | |
3988 | ||
0c21ead1 TH |
3989 | css->ss->css_free(css); |
3990 | cgroup_dput(cgrp); | |
48ddbe19 TH |
3991 | } |
3992 | ||
0c21ead1 | 3993 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3994 | { |
3995 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3996 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3997 | |
35ef10da TH |
3998 | /* |
3999 | * css holds an extra ref to @cgrp->dentry which is put on the last | |
0c21ead1 | 4000 | * css_put(). dput() requires process context which we don't have. |
35ef10da TH |
4001 | */ |
4002 | INIT_WORK(&css->destroy_work, css_free_work_fn); | |
e5fca243 | 4003 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4004 | } |
4005 | ||
d3daf28d TH |
4006 | static void css_release(struct percpu_ref *ref) |
4007 | { | |
4008 | struct cgroup_subsys_state *css = | |
4009 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4010 | ||
aec25020 | 4011 | rcu_assign_pointer(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 4012 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4013 | } |
4014 | ||
623f926b TH |
4015 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
4016 | struct cgroup *cgrp) | |
ddbcc7e8 | 4017 | { |
bd89aabc | 4018 | css->cgroup = cgrp; |
72c97e54 | 4019 | css->ss = ss; |
ddbcc7e8 | 4020 | css->flags = 0; |
0ae78e0b TH |
4021 | |
4022 | if (cgrp->parent) | |
ca8bdcaf | 4023 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 4024 | else |
38b53aba | 4025 | css->flags |= CSS_ROOT; |
48ddbe19 | 4026 | |
ca8bdcaf | 4027 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4028 | } |
4029 | ||
2a4ac633 | 4030 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4031 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4032 | { |
623f926b | 4033 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4034 | int ret = 0; |
4035 | ||
ace2bee8 | 4036 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
4037 | lockdep_assert_held(&cgroup_mutex); |
4038 | ||
92fb9748 | 4039 | if (ss->css_online) |
eb95419b | 4040 | ret = ss->css_online(css); |
ae7f164a | 4041 | if (!ret) { |
eb95419b | 4042 | css->flags |= CSS_ONLINE; |
f20104de | 4043 | css->cgroup->nr_css++; |
aec25020 | 4044 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 4045 | } |
b1929db4 | 4046 | return ret; |
a31f2d3f TH |
4047 | } |
4048 | ||
2a4ac633 | 4049 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4050 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4051 | { |
623f926b | 4052 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 4053 | |
ace2bee8 | 4054 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
4055 | lockdep_assert_held(&cgroup_mutex); |
4056 | ||
4057 | if (!(css->flags & CSS_ONLINE)) | |
4058 | return; | |
4059 | ||
d7eeac19 | 4060 | if (ss->css_offline) |
eb95419b | 4061 | ss->css_offline(css); |
a31f2d3f | 4062 | |
eb95419b | 4063 | css->flags &= ~CSS_ONLINE; |
09a503ea | 4064 | css->cgroup->nr_css--; |
aec25020 | 4065 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); |
a31f2d3f TH |
4066 | } |
4067 | ||
c81c925a TH |
4068 | /** |
4069 | * create_css - create a cgroup_subsys_state | |
4070 | * @cgrp: the cgroup new css will be associated with | |
4071 | * @ss: the subsys of new css | |
4072 | * | |
4073 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
4074 | * css is online and installed in @cgrp with all interface files created. | |
4075 | * Returns 0 on success, -errno on failure. | |
4076 | */ | |
4077 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
4078 | { | |
4079 | struct cgroup *parent = cgrp->parent; | |
4080 | struct cgroup_subsys_state *css; | |
4081 | int err; | |
4082 | ||
4083 | lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex); | |
4084 | lockdep_assert_held(&cgroup_mutex); | |
4085 | ||
4086 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
4087 | if (IS_ERR(css)) | |
4088 | return PTR_ERR(css); | |
4089 | ||
4090 | err = percpu_ref_init(&css->refcnt, css_release); | |
4091 | if (err) | |
4092 | goto err_free; | |
4093 | ||
4094 | init_css(css, ss, cgrp); | |
4095 | ||
aec25020 | 4096 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a TH |
4097 | if (err) |
4098 | goto err_free; | |
4099 | ||
4100 | err = online_css(css); | |
4101 | if (err) | |
4102 | goto err_free; | |
4103 | ||
4104 | dget(cgrp->dentry); | |
4105 | css_get(css->parent); | |
4106 | ||
4107 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
4108 | parent->parent) { | |
4109 | 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", | |
4110 | current->comm, current->pid, ss->name); | |
4111 | if (!strcmp(ss->name, "memory")) | |
4112 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
4113 | ss->warned_broken_hierarchy = true; | |
4114 | } | |
4115 | ||
4116 | return 0; | |
4117 | ||
4118 | err_free: | |
4119 | percpu_ref_cancel_init(&css->refcnt); | |
4120 | ss->css_free(css); | |
4121 | return err; | |
4122 | } | |
4123 | ||
ddbcc7e8 | 4124 | /* |
a043e3b2 LZ |
4125 | * cgroup_create - create a cgroup |
4126 | * @parent: cgroup that will be parent of the new cgroup | |
4127 | * @dentry: dentry of the new cgroup | |
4128 | * @mode: mode to set on new inode | |
ddbcc7e8 | 4129 | * |
a043e3b2 | 4130 | * Must be called with the mutex on the parent inode held |
ddbcc7e8 | 4131 | */ |
ddbcc7e8 | 4132 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, |
a5e7ed32 | 4133 | umode_t mode) |
ddbcc7e8 | 4134 | { |
bd89aabc | 4135 | struct cgroup *cgrp; |
65dff759 | 4136 | struct cgroup_name *name; |
ddbcc7e8 | 4137 | struct cgroupfs_root *root = parent->root; |
b58c8998 | 4138 | int ssid, err; |
ddbcc7e8 PM |
4139 | struct cgroup_subsys *ss; |
4140 | struct super_block *sb = root->sb; | |
4141 | ||
0a950f65 | 4142 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
4143 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
4144 | if (!cgrp) | |
ddbcc7e8 PM |
4145 | return -ENOMEM; |
4146 | ||
8d7e6fb0 | 4147 | name = cgroup_alloc_name(dentry->d_name.name); |
b58c8998 TH |
4148 | if (!name) { |
4149 | err = -ENOMEM; | |
65dff759 | 4150 | goto err_free_cgrp; |
b58c8998 | 4151 | } |
65dff759 LZ |
4152 | rcu_assign_pointer(cgrp->name, name); |
4153 | ||
ace2bee8 TH |
4154 | mutex_lock(&cgroup_tree_mutex); |
4155 | ||
976c06bc TH |
4156 | /* |
4157 | * Only live parents can have children. Note that the liveliness | |
4158 | * check isn't strictly necessary because cgroup_mkdir() and | |
4159 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
4160 | * anyway so that locking is contained inside cgroup proper and we | |
4161 | * don't get nasty surprises if we ever grow another caller. | |
4162 | */ | |
4163 | if (!cgroup_lock_live_group(parent)) { | |
4164 | err = -ENODEV; | |
ace2bee8 | 4165 | goto err_unlock_tree; |
0ab02ca8 LZ |
4166 | } |
4167 | ||
4168 | /* | |
4169 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4170 | * a half-baked cgroup. | |
4171 | */ | |
4172 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
4173 | if (cgrp->id < 0) { | |
4174 | err = -ENOMEM; | |
4175 | goto err_unlock; | |
976c06bc TH |
4176 | } |
4177 | ||
ddbcc7e8 PM |
4178 | /* Grab a reference on the superblock so the hierarchy doesn't |
4179 | * get deleted on unmount if there are child cgroups. This | |
4180 | * can be done outside cgroup_mutex, since the sb can't | |
4181 | * disappear while someone has an open control file on the | |
4182 | * fs */ | |
4183 | atomic_inc(&sb->s_active); | |
4184 | ||
cc31edce | 4185 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4186 | |
fe1c06ca LZ |
4187 | dentry->d_fsdata = cgrp; |
4188 | cgrp->dentry = dentry; | |
4189 | ||
bd89aabc | 4190 | cgrp->parent = parent; |
0ae78e0b | 4191 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 4192 | cgrp->root = parent->root; |
ddbcc7e8 | 4193 | |
b6abdb0e LZ |
4194 | if (notify_on_release(parent)) |
4195 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4196 | ||
2260e7fc TH |
4197 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4198 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4199 | |
4e139afc TH |
4200 | /* |
4201 | * Create directory. cgroup_create_file() returns with the new | |
4202 | * directory locked on success so that it can be populated without | |
4203 | * dropping cgroup_mutex. | |
4204 | */ | |
28fd6f30 | 4205 | err = cgroup_create_file(dentry, S_IFDIR | mode, sb); |
ddbcc7e8 | 4206 | if (err < 0) |
0ab02ca8 | 4207 | goto err_free_id; |
4e139afc | 4208 | lockdep_assert_held(&dentry->d_inode->i_mutex); |
ddbcc7e8 | 4209 | |
00356bd5 | 4210 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 4211 | |
4e139afc | 4212 | /* allocation complete, commit to creation */ |
4e139afc TH |
4213 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
4214 | root->number_of_cgroups++; | |
28fd6f30 | 4215 | |
415cf07a LZ |
4216 | /* hold a ref to the parent's dentry */ |
4217 | dget(parent->dentry); | |
4218 | ||
0d80255e TH |
4219 | /* |
4220 | * @cgrp is now fully operational. If something fails after this | |
4221 | * point, it'll be released via the normal destruction path. | |
4222 | */ | |
4e96ee8e LZ |
4223 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4224 | ||
2bb566cb | 4225 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
4226 | if (err) |
4227 | goto err_destroy; | |
4228 | ||
9d403e99 | 4229 | /* let's create and online css's */ |
b85d2040 TH |
4230 | for_each_subsys(ss, ssid) { |
4231 | if (root->subsys_mask & (1 << ssid)) { | |
4232 | err = create_css(cgrp, ss); | |
4233 | if (err) | |
4234 | goto err_destroy; | |
4235 | } | |
a8638030 | 4236 | } |
ddbcc7e8 PM |
4237 | |
4238 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4239 | mutex_unlock(&cgroup_tree_mutex); |
bd89aabc | 4240 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
4241 | |
4242 | return 0; | |
4243 | ||
0a950f65 | 4244 | err_free_id: |
4e96ee8e | 4245 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
4246 | /* Release the reference count that we took on the superblock */ |
4247 | deactivate_super(sb); | |
4248 | err_unlock: | |
4249 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
4250 | err_unlock_tree: |
4251 | mutex_unlock(&cgroup_tree_mutex); | |
65dff759 | 4252 | kfree(rcu_dereference_raw(cgrp->name)); |
4b8b47eb | 4253 | err_free_cgrp: |
bd89aabc | 4254 | kfree(cgrp); |
ddbcc7e8 | 4255 | return err; |
4b8b47eb TH |
4256 | |
4257 | err_destroy: | |
4258 | cgroup_destroy_locked(cgrp); | |
4259 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4260 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb TH |
4261 | mutex_unlock(&dentry->d_inode->i_mutex); |
4262 | return err; | |
ddbcc7e8 PM |
4263 | } |
4264 | ||
18bb1db3 | 4265 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
ddbcc7e8 PM |
4266 | { |
4267 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
4268 | ||
4269 | /* the vfs holds inode->i_mutex already */ | |
4270 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
4271 | } | |
4272 | ||
223dbc38 TH |
4273 | /* |
4274 | * This is called when the refcnt of a css is confirmed to be killed. | |
4275 | * css_tryget() is now guaranteed to fail. | |
4276 | */ | |
4277 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4278 | { |
223dbc38 TH |
4279 | struct cgroup_subsys_state *css = |
4280 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
4281 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 4282 | |
ace2bee8 | 4283 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
4284 | mutex_lock(&cgroup_mutex); |
4285 | ||
09a503ea TH |
4286 | /* |
4287 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
4288 | * initate destruction. | |
4289 | */ | |
4290 | offline_css(css); | |
4291 | ||
f20104de TH |
4292 | /* |
4293 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
4294 | * be disabled before proceeding to the second phase of cgroup | |
4295 | * destruction. If we are the last one, kick it off. | |
4296 | */ | |
09a503ea | 4297 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
4298 | cgroup_destroy_css_killed(cgrp); |
4299 | ||
4300 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4301 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
4302 | |
4303 | /* | |
4304 | * Put the css refs from kill_css(). Each css holds an extra | |
4305 | * reference to the cgroup's dentry and cgroup removal proceeds | |
4306 | * regardless of css refs. On the last put of each css, whenever | |
4307 | * that may be, the extra dentry ref is put so that dentry | |
4308 | * destruction happens only after all css's are released. | |
4309 | */ | |
4310 | css_put(css); | |
d3daf28d TH |
4311 | } |
4312 | ||
223dbc38 TH |
4313 | /* css kill confirmation processing requires process context, bounce */ |
4314 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4315 | { |
4316 | struct cgroup_subsys_state *css = | |
4317 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4318 | ||
223dbc38 | 4319 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 4320 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
4321 | } |
4322 | ||
edae0c33 TH |
4323 | /** |
4324 | * kill_css - destroy a css | |
4325 | * @css: css to destroy | |
4326 | * | |
3c14f8b4 TH |
4327 | * This function initiates destruction of @css by removing cgroup interface |
4328 | * files and putting its base reference. ->css_offline() will be invoked | |
4329 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
4330 | * reference count reaches zero, @css will be released. | |
edae0c33 TH |
4331 | */ |
4332 | static void kill_css(struct cgroup_subsys_state *css) | |
4333 | { | |
aec25020 | 4334 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 4335 | |
edae0c33 TH |
4336 | /* |
4337 | * Killing would put the base ref, but we need to keep it alive | |
4338 | * until after ->css_offline(). | |
4339 | */ | |
4340 | css_get(css); | |
4341 | ||
4342 | /* | |
4343 | * cgroup core guarantees that, by the time ->css_offline() is | |
4344 | * invoked, no new css reference will be given out via | |
4345 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
4346 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
4347 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4348 | * | |
4349 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4350 | * css is confirmed to be seen as killed on all CPUs. | |
4351 | */ | |
4352 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4353 | } |
4354 | ||
4355 | /** | |
4356 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4357 | * @cgrp: cgroup to be destroyed | |
4358 | * | |
4359 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4360 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
4361 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
4362 | * invoked. To satisfy all the requirements, destruction is implemented in | |
4363 | * the following two steps. | |
4364 | * | |
4365 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4366 | * userland visible parts and start killing the percpu refcnts of | |
4367 | * css's. Set up so that the next stage will be kicked off once all | |
4368 | * the percpu refcnts are confirmed to be killed. | |
4369 | * | |
4370 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4371 | * rest of destruction. Once all cgroup references are gone, the | |
4372 | * cgroup is RCU-freed. | |
4373 | * | |
4374 | * This function implements s1. After this step, @cgrp is gone as far as | |
4375 | * the userland is concerned and a new cgroup with the same name may be | |
4376 | * created. As cgroup doesn't care about the names internally, this | |
4377 | * doesn't cause any problem. | |
4378 | */ | |
42809dd4 TH |
4379 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4380 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4381 | { |
42809dd4 | 4382 | struct dentry *d = cgrp->dentry; |
1c6727af | 4383 | struct cgroup_subsys_state *css; |
bb78a92f | 4384 | struct cgroup *child; |
ddd69148 | 4385 | bool empty; |
1c6727af | 4386 | int ssid; |
ddbcc7e8 | 4387 | |
42809dd4 | 4388 | lockdep_assert_held(&d->d_inode->i_mutex); |
ace2bee8 | 4389 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
4390 | lockdep_assert_held(&cgroup_mutex); |
4391 | ||
ddd69148 | 4392 | /* |
6f3d828f TH |
4393 | * css_set_lock synchronizes access to ->cset_links and prevents |
4394 | * @cgrp from being removed while __put_css_set() is in progress. | |
ddd69148 TH |
4395 | */ |
4396 | read_lock(&css_set_lock); | |
bb78a92f | 4397 | empty = list_empty(&cgrp->cset_links); |
ddd69148 TH |
4398 | read_unlock(&css_set_lock); |
4399 | if (!empty) | |
ddbcc7e8 | 4400 | return -EBUSY; |
a043e3b2 | 4401 | |
bb78a92f HD |
4402 | /* |
4403 | * Make sure there's no live children. We can't test ->children | |
4404 | * emptiness as dead children linger on it while being destroyed; | |
4405 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
4406 | */ | |
4407 | empty = true; | |
4408 | rcu_read_lock(); | |
4409 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
4410 | empty = cgroup_is_dead(child); | |
4411 | if (!empty) | |
4412 | break; | |
4413 | } | |
4414 | rcu_read_unlock(); | |
4415 | if (!empty) | |
4416 | return -EBUSY; | |
4417 | ||
88703267 | 4418 | /* |
edae0c33 TH |
4419 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
4420 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
4421 | * percpu refs of all css's are confirmed to be killed. This |
4422 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 4423 | */ |
4ac06017 | 4424 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
4425 | for_each_css(css, ssid, cgrp) |
4426 | kill_css(css); | |
4ac06017 | 4427 | mutex_lock(&cgroup_mutex); |
455050d2 TH |
4428 | |
4429 | /* | |
4430 | * Mark @cgrp dead. This prevents further task migration and child | |
4431 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 4432 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 4433 | * resume iteration after dropping RCU read lock. See |
492eb21b | 4434 | * css_next_child() for details. |
455050d2 | 4435 | */ |
54766d4a | 4436 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 4437 | |
455050d2 TH |
4438 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
4439 | raw_spin_lock(&release_list_lock); | |
4440 | if (!list_empty(&cgrp->release_list)) | |
4441 | list_del_init(&cgrp->release_list); | |
4442 | raw_spin_unlock(&release_list_lock); | |
4443 | ||
4444 | /* | |
f20104de TH |
4445 | * If @cgrp has css's attached, the second stage of cgroup |
4446 | * destruction is kicked off from css_killed_work_fn() after the | |
4447 | * refs of all attached css's are killed. If @cgrp doesn't have | |
4448 | * any css, we kick it off here. | |
4449 | */ | |
4450 | if (!cgrp->nr_css) | |
4451 | cgroup_destroy_css_killed(cgrp); | |
4452 | ||
455050d2 | 4453 | /* |
3c14f8b4 TH |
4454 | * Clear the base files and remove @cgrp directory. The removal |
4455 | * puts the base ref but we aren't quite done with @cgrp yet, so | |
4456 | * hold onto it. | |
455050d2 | 4457 | */ |
4ac06017 | 4458 | mutex_unlock(&cgroup_mutex); |
2bb566cb | 4459 | cgroup_addrm_files(cgrp, cgroup_base_files, false); |
455050d2 TH |
4460 | dget(d); |
4461 | cgroup_d_remove_dir(d); | |
4ac06017 | 4462 | mutex_lock(&cgroup_mutex); |
455050d2 | 4463 | |
ea15f8cc TH |
4464 | return 0; |
4465 | }; | |
4466 | ||
d3daf28d | 4467 | /** |
f20104de | 4468 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
4469 | * @work: cgroup->destroy_free_work |
4470 | * | |
4471 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
4472 | * destroyed after all css's are offlined and performs the rest of |
4473 | * destruction. This is the second step of destruction described in the | |
4474 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 4475 | */ |
f20104de | 4476 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 4477 | { |
ea15f8cc TH |
4478 | struct cgroup *parent = cgrp->parent; |
4479 | struct dentry *d = cgrp->dentry; | |
ea15f8cc | 4480 | |
ace2bee8 | 4481 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 4482 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 4483 | |
999cd8a4 | 4484 | /* delete this cgroup from parent->children */ |
eb6fd504 | 4485 | list_del_rcu(&cgrp->sibling); |
ed957793 | 4486 | |
ddbcc7e8 | 4487 | dput(d); |
ddbcc7e8 | 4488 | |
bd89aabc | 4489 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 4490 | check_for_release(parent); |
ddbcc7e8 PM |
4491 | } |
4492 | ||
42809dd4 TH |
4493 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) |
4494 | { | |
4495 | int ret; | |
4496 | ||
ace2bee8 | 4497 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 TH |
4498 | mutex_lock(&cgroup_mutex); |
4499 | ret = cgroup_destroy_locked(dentry->d_fsdata); | |
4500 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4501 | mutex_unlock(&cgroup_tree_mutex); |
42809dd4 TH |
4502 | |
4503 | return ret; | |
4504 | } | |
4505 | ||
06a11920 | 4506 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4507 | { |
ddbcc7e8 | 4508 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4509 | |
4510 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4511 | |
ace2bee8 | 4512 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
4513 | mutex_lock(&cgroup_mutex); |
4514 | ||
de00ffa5 | 4515 | INIT_LIST_HEAD(&ss->cftsets); |
8e3f6541 | 4516 | |
ddbcc7e8 | 4517 | /* Create the top cgroup state for this subsystem */ |
9871bf95 | 4518 | ss->root = &cgroup_dummy_root; |
ca8bdcaf | 4519 | css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss)); |
ddbcc7e8 PM |
4520 | /* We don't handle early failures gracefully */ |
4521 | BUG_ON(IS_ERR(css)); | |
623f926b | 4522 | init_css(css, ss, cgroup_dummy_top); |
ddbcc7e8 | 4523 | |
e8d55fde | 4524 | /* Update the init_css_set to contain a subsys |
817929ec | 4525 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
4526 | * newly registered, all tasks and hence the |
4527 | * init_css_set is in the subsystem's top cgroup. */ | |
aec25020 | 4528 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4529 | |
4530 | need_forkexit_callback |= ss->fork || ss->exit; | |
4531 | ||
e8d55fde LZ |
4532 | /* At system boot, before all subsystems have been |
4533 | * registered, no tasks have been forked, so we don't | |
4534 | * need to invoke fork callbacks here. */ | |
4535 | BUG_ON(!list_empty(&init_task.tasks)); | |
4536 | ||
ae7f164a | 4537 | BUG_ON(online_css(css)); |
a8638030 | 4538 | |
648bb56d | 4539 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4540 | mutex_unlock(&cgroup_tree_mutex); |
e6a1105b BB |
4541 | } |
4542 | ||
ddbcc7e8 | 4543 | /** |
a043e3b2 LZ |
4544 | * cgroup_init_early - cgroup initialization at system boot |
4545 | * | |
4546 | * Initialize cgroups at system boot, and initialize any | |
4547 | * subsystems that request early init. | |
ddbcc7e8 PM |
4548 | */ |
4549 | int __init cgroup_init_early(void) | |
4550 | { | |
30159ec7 | 4551 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4552 | int i; |
30159ec7 | 4553 | |
146aa1bd | 4554 | atomic_set(&init_css_set.refcount, 1); |
69d0206c | 4555 | INIT_LIST_HEAD(&init_css_set.cgrp_links); |
817929ec | 4556 | INIT_LIST_HEAD(&init_css_set.tasks); |
472b1053 | 4557 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 4558 | css_set_count = 1; |
9871bf95 TH |
4559 | init_cgroup_root(&cgroup_dummy_root); |
4560 | cgroup_root_count = 1; | |
a4ea1cc9 | 4561 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4562 | |
69d0206c | 4563 | init_cgrp_cset_link.cset = &init_css_set; |
9871bf95 TH |
4564 | init_cgrp_cset_link.cgrp = cgroup_dummy_top; |
4565 | list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links); | |
69d0206c | 4566 | list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links); |
ddbcc7e8 | 4567 | |
3ed80a62 | 4568 | for_each_subsys(ss, i) { |
aec25020 | 4569 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4570 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4571 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4572 | ss->id, ss->name); |
073219e9 TH |
4573 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4574 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4575 | ||
aec25020 | 4576 | ss->id = i; |
073219e9 | 4577 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4578 | |
4579 | if (ss->early_init) | |
4580 | cgroup_init_subsys(ss); | |
4581 | } | |
4582 | return 0; | |
4583 | } | |
4584 | ||
4585 | /** | |
a043e3b2 LZ |
4586 | * cgroup_init - cgroup initialization |
4587 | * | |
4588 | * Register cgroup filesystem and /proc file, and initialize | |
4589 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4590 | */ |
4591 | int __init cgroup_init(void) | |
4592 | { | |
30159ec7 | 4593 | struct cgroup_subsys *ss; |
0ac801fe | 4594 | unsigned long key; |
30159ec7 | 4595 | int i, err; |
a424316c PM |
4596 | |
4597 | err = bdi_init(&cgroup_backing_dev_info); | |
4598 | if (err) | |
4599 | return err; | |
ddbcc7e8 | 4600 | |
3ed80a62 | 4601 | for_each_subsys(ss, i) { |
ddbcc7e8 PM |
4602 | if (!ss->early_init) |
4603 | cgroup_init_subsys(ss); | |
de00ffa5 TH |
4604 | |
4605 | /* | |
4606 | * cftype registration needs kmalloc and can't be done | |
4607 | * during early_init. Register base cftypes separately. | |
4608 | */ | |
4609 | if (ss->base_cftypes) | |
4610 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
ddbcc7e8 PM |
4611 | } |
4612 | ||
fa3ca07e | 4613 | /* allocate id for the dummy hierarchy */ |
54e7b4eb | 4614 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4615 | |
82fe9b0d TH |
4616 | /* Add init_css_set to the hash table */ |
4617 | key = css_set_hash(init_css_set.subsys); | |
4618 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4619 | ||
fc76df70 | 4620 | BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1)); |
676db4af | 4621 | |
4e96ee8e LZ |
4622 | err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top, |
4623 | 0, 1, GFP_KERNEL); | |
4624 | BUG_ON(err < 0); | |
4625 | ||
54e7b4eb TH |
4626 | mutex_unlock(&cgroup_mutex); |
4627 | ||
676db4af GK |
4628 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
4629 | if (!cgroup_kobj) { | |
4630 | err = -ENOMEM; | |
4631 | goto out; | |
4632 | } | |
4633 | ||
ddbcc7e8 | 4634 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
4635 | if (err < 0) { |
4636 | kobject_put(cgroup_kobj); | |
ddbcc7e8 | 4637 | goto out; |
676db4af | 4638 | } |
ddbcc7e8 | 4639 | |
46ae220b | 4640 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
a424316c | 4641 | |
ddbcc7e8 | 4642 | out: |
a424316c PM |
4643 | if (err) |
4644 | bdi_destroy(&cgroup_backing_dev_info); | |
4645 | ||
ddbcc7e8 PM |
4646 | return err; |
4647 | } | |
b4f48b63 | 4648 | |
e5fca243 TH |
4649 | static int __init cgroup_wq_init(void) |
4650 | { | |
4651 | /* | |
4652 | * There isn't much point in executing destruction path in | |
4653 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
ab3f5faa HD |
4654 | * |
4655 | * XXX: Must be ordered to make sure parent is offlined after | |
4656 | * children. The ordering requirement is for memcg where a | |
4657 | * parent's offline may wait for a child's leading to deadlock. In | |
4658 | * the long term, this should be fixed from memcg side. | |
e5fca243 TH |
4659 | * |
4660 | * We would prefer to do this in cgroup_init() above, but that | |
4661 | * is called before init_workqueues(): so leave this until after. | |
4662 | */ | |
ab3f5faa | 4663 | cgroup_destroy_wq = alloc_ordered_workqueue("cgroup_destroy", 0); |
e5fca243 | 4664 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4665 | |
4666 | /* | |
4667 | * Used to destroy pidlists and separate to serve as flush domain. | |
4668 | * Cap @max_active to 1 too. | |
4669 | */ | |
4670 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4671 | 0, 1); | |
4672 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4673 | ||
e5fca243 TH |
4674 | return 0; |
4675 | } | |
4676 | core_initcall(cgroup_wq_init); | |
4677 | ||
a424316c PM |
4678 | /* |
4679 | * proc_cgroup_show() | |
4680 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4681 | * - Used for /proc/<pid>/cgroup. | |
4682 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4683 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4684 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4685 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4686 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4687 | * cgroup to top_cgroup. | |
4688 | */ | |
4689 | ||
4690 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4691 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4692 | { |
4693 | struct pid *pid; | |
4694 | struct task_struct *tsk; | |
4695 | char *buf; | |
4696 | int retval; | |
4697 | struct cgroupfs_root *root; | |
4698 | ||
4699 | retval = -ENOMEM; | |
4700 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4701 | if (!buf) | |
4702 | goto out; | |
4703 | ||
4704 | retval = -ESRCH; | |
4705 | pid = m->private; | |
4706 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4707 | if (!tsk) | |
4708 | goto out_free; | |
4709 | ||
4710 | retval = 0; | |
4711 | ||
4712 | mutex_lock(&cgroup_mutex); | |
4713 | ||
e5f6a860 | 4714 | for_each_active_root(root) { |
a424316c | 4715 | struct cgroup_subsys *ss; |
bd89aabc | 4716 | struct cgroup *cgrp; |
b85d2040 | 4717 | int ssid, count = 0; |
a424316c | 4718 | |
2c6ab6d2 | 4719 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 TH |
4720 | for_each_subsys(ss, ssid) |
4721 | if (root->subsys_mask & (1 << ssid)) | |
4722 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4723 | if (strlen(root->name)) |
4724 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4725 | root->name); | |
a424316c | 4726 | seq_putc(m, ':'); |
7717f7ba | 4727 | cgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 4728 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); |
a424316c PM |
4729 | if (retval < 0) |
4730 | goto out_unlock; | |
4731 | seq_puts(m, buf); | |
4732 | seq_putc(m, '\n'); | |
4733 | } | |
4734 | ||
4735 | out_unlock: | |
4736 | mutex_unlock(&cgroup_mutex); | |
4737 | put_task_struct(tsk); | |
4738 | out_free: | |
4739 | kfree(buf); | |
4740 | out: | |
4741 | return retval; | |
4742 | } | |
4743 | ||
a424316c PM |
4744 | /* Display information about each subsystem and each hierarchy */ |
4745 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4746 | { | |
30159ec7 | 4747 | struct cgroup_subsys *ss; |
a424316c | 4748 | int i; |
a424316c | 4749 | |
8bab8dde | 4750 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4751 | /* |
4752 | * ideally we don't want subsystems moving around while we do this. | |
4753 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4754 | * subsys/hierarchy state. | |
4755 | */ | |
a424316c | 4756 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4757 | |
4758 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4759 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4760 | ss->name, ss->root->hierarchy_id, | |
8bab8dde | 4761 | ss->root->number_of_cgroups, !ss->disabled); |
30159ec7 | 4762 | |
a424316c PM |
4763 | mutex_unlock(&cgroup_mutex); |
4764 | return 0; | |
4765 | } | |
4766 | ||
4767 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4768 | { | |
9dce07f1 | 4769 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4770 | } |
4771 | ||
828c0950 | 4772 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4773 | .open = cgroupstats_open, |
4774 | .read = seq_read, | |
4775 | .llseek = seq_lseek, | |
4776 | .release = single_release, | |
4777 | }; | |
4778 | ||
b4f48b63 PM |
4779 | /** |
4780 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4781 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4782 | * |
4783 | * Description: A task inherits its parent's cgroup at fork(). | |
4784 | * | |
4785 | * A pointer to the shared css_set was automatically copied in | |
4786 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
9bb71308 TH |
4787 | * it was not made under the protection of RCU or cgroup_mutex, so |
4788 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
4789 | * have already changed current->cgroups, allowing the previously | |
4790 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
4791 | * |
4792 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4793 | * task, and the passed argument 'child' points to the child task. | |
4794 | */ | |
4795 | void cgroup_fork(struct task_struct *child) | |
4796 | { | |
9bb71308 | 4797 | task_lock(current); |
a8ad805c | 4798 | get_css_set(task_css_set(current)); |
817929ec | 4799 | child->cgroups = current->cgroups; |
9bb71308 | 4800 | task_unlock(current); |
817929ec | 4801 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4802 | } |
4803 | ||
817929ec | 4804 | /** |
a043e3b2 LZ |
4805 | * cgroup_post_fork - called on a new task after adding it to the task list |
4806 | * @child: the task in question | |
4807 | * | |
5edee61e TH |
4808 | * Adds the task to the list running through its css_set if necessary and |
4809 | * call the subsystem fork() callbacks. Has to be after the task is | |
4810 | * visible on the task list in case we race with the first call to | |
0942eeee | 4811 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4812 | * list. |
a043e3b2 | 4813 | */ |
817929ec PM |
4814 | void cgroup_post_fork(struct task_struct *child) |
4815 | { | |
30159ec7 | 4816 | struct cgroup_subsys *ss; |
5edee61e TH |
4817 | int i; |
4818 | ||
3ce3230a FW |
4819 | /* |
4820 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
4821 | * under the tasklist_lock and we read it here after we added the child | |
4822 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
4823 | * yet in the tasklist when we walked through it from | |
4824 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
4825 | * should be visible now due to the paired locking and barriers implied | |
4826 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
4827 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
4828 | * lock on fork. | |
4829 | */ | |
817929ec PM |
4830 | if (use_task_css_set_links) { |
4831 | write_lock(&css_set_lock); | |
d8783832 TH |
4832 | task_lock(child); |
4833 | if (list_empty(&child->cg_list)) | |
a8ad805c | 4834 | list_add(&child->cg_list, &task_css_set(child)->tasks); |
d8783832 | 4835 | task_unlock(child); |
817929ec PM |
4836 | write_unlock(&css_set_lock); |
4837 | } | |
5edee61e TH |
4838 | |
4839 | /* | |
4840 | * Call ss->fork(). This must happen after @child is linked on | |
4841 | * css_set; otherwise, @child might change state between ->fork() | |
4842 | * and addition to css_set. | |
4843 | */ | |
4844 | if (need_forkexit_callback) { | |
3ed80a62 | 4845 | for_each_subsys(ss, i) |
5edee61e TH |
4846 | if (ss->fork) |
4847 | ss->fork(child); | |
5edee61e | 4848 | } |
817929ec | 4849 | } |
5edee61e | 4850 | |
b4f48b63 PM |
4851 | /** |
4852 | * cgroup_exit - detach cgroup from exiting task | |
4853 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4854 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4855 | * |
4856 | * Description: Detach cgroup from @tsk and release it. | |
4857 | * | |
4858 | * Note that cgroups marked notify_on_release force every task in | |
4859 | * them to take the global cgroup_mutex mutex when exiting. | |
4860 | * This could impact scaling on very large systems. Be reluctant to | |
4861 | * use notify_on_release cgroups where very high task exit scaling | |
4862 | * is required on large systems. | |
4863 | * | |
4864 | * the_top_cgroup_hack: | |
4865 | * | |
4866 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4867 | * | |
4868 | * We call cgroup_exit() while the task is still competent to | |
4869 | * handle notify_on_release(), then leave the task attached to the | |
4870 | * root cgroup in each hierarchy for the remainder of its exit. | |
4871 | * | |
4872 | * To do this properly, we would increment the reference count on | |
4873 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4874 | * code we would add a second cgroup function call, to drop that | |
4875 | * reference. This would just create an unnecessary hot spot on | |
4876 | * the top_cgroup reference count, to no avail. | |
4877 | * | |
4878 | * Normally, holding a reference to a cgroup without bumping its | |
4879 | * count is unsafe. The cgroup could go away, or someone could | |
4880 | * attach us to a different cgroup, decrementing the count on | |
4881 | * the first cgroup that we never incremented. But in this case, | |
4882 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4883 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4884 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4885 | */ |
4886 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4887 | { | |
30159ec7 | 4888 | struct cgroup_subsys *ss; |
5abb8855 | 4889 | struct css_set *cset; |
d41d5a01 | 4890 | int i; |
817929ec PM |
4891 | |
4892 | /* | |
4893 | * Unlink from the css_set task list if necessary. | |
4894 | * Optimistically check cg_list before taking | |
4895 | * css_set_lock | |
4896 | */ | |
4897 | if (!list_empty(&tsk->cg_list)) { | |
4898 | write_lock(&css_set_lock); | |
4899 | if (!list_empty(&tsk->cg_list)) | |
8d258797 | 4900 | list_del_init(&tsk->cg_list); |
817929ec PM |
4901 | write_unlock(&css_set_lock); |
4902 | } | |
4903 | ||
b4f48b63 PM |
4904 | /* Reassign the task to the init_css_set. */ |
4905 | task_lock(tsk); | |
a8ad805c TH |
4906 | cset = task_css_set(tsk); |
4907 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 PZ |
4908 | |
4909 | if (run_callbacks && need_forkexit_callback) { | |
3ed80a62 TH |
4910 | /* see cgroup_post_fork() for details */ |
4911 | for_each_subsys(ss, i) { | |
d41d5a01 | 4912 | if (ss->exit) { |
eb95419b TH |
4913 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4914 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4915 | |
eb95419b | 4916 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4917 | } |
4918 | } | |
4919 | } | |
b4f48b63 | 4920 | task_unlock(tsk); |
d41d5a01 | 4921 | |
5abb8855 | 4922 | put_css_set_taskexit(cset); |
b4f48b63 | 4923 | } |
697f4161 | 4924 | |
bd89aabc | 4925 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4926 | { |
f50daa70 | 4927 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4928 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4929 | /* |
4930 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4931 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4932 | * it now |
4933 | */ | |
81a6a5cd | 4934 | int need_schedule_work = 0; |
f50daa70 | 4935 | |
cdcc136f | 4936 | raw_spin_lock(&release_list_lock); |
54766d4a | 4937 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4938 | list_empty(&cgrp->release_list)) { |
4939 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4940 | need_schedule_work = 1; |
4941 | } | |
cdcc136f | 4942 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4943 | if (need_schedule_work) |
4944 | schedule_work(&release_agent_work); | |
4945 | } | |
4946 | } | |
4947 | ||
81a6a5cd PM |
4948 | /* |
4949 | * Notify userspace when a cgroup is released, by running the | |
4950 | * configured release agent with the name of the cgroup (path | |
4951 | * relative to the root of cgroup file system) as the argument. | |
4952 | * | |
4953 | * Most likely, this user command will try to rmdir this cgroup. | |
4954 | * | |
4955 | * This races with the possibility that some other task will be | |
4956 | * attached to this cgroup before it is removed, or that some other | |
4957 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4958 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4959 | * unused, and this cgroup will be reprieved from its death sentence, | |
4960 | * to continue to serve a useful existence. Next time it's released, | |
4961 | * we will get notified again, if it still has 'notify_on_release' set. | |
4962 | * | |
4963 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4964 | * means only wait until the task is successfully execve()'d. The | |
4965 | * separate release agent task is forked by call_usermodehelper(), | |
4966 | * then control in this thread returns here, without waiting for the | |
4967 | * release agent task. We don't bother to wait because the caller of | |
4968 | * this routine has no use for the exit status of the release agent | |
4969 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4970 | */ |
81a6a5cd PM |
4971 | static void cgroup_release_agent(struct work_struct *work) |
4972 | { | |
4973 | BUG_ON(work != &release_agent_work); | |
4974 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4975 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4976 | while (!list_empty(&release_list)) { |
4977 | char *argv[3], *envp[3]; | |
4978 | int i; | |
e788e066 | 4979 | char *pathbuf = NULL, *agentbuf = NULL; |
bd89aabc | 4980 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4981 | struct cgroup, |
4982 | release_list); | |
bd89aabc | 4983 | list_del_init(&cgrp->release_list); |
cdcc136f | 4984 | raw_spin_unlock(&release_list_lock); |
81a6a5cd | 4985 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
e788e066 PM |
4986 | if (!pathbuf) |
4987 | goto continue_free; | |
4988 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
4989 | goto continue_free; | |
4990 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4991 | if (!agentbuf) | |
4992 | goto continue_free; | |
81a6a5cd PM |
4993 | |
4994 | i = 0; | |
e788e066 PM |
4995 | argv[i++] = agentbuf; |
4996 | argv[i++] = pathbuf; | |
81a6a5cd PM |
4997 | argv[i] = NULL; |
4998 | ||
4999 | i = 0; | |
5000 | /* minimal command environment */ | |
5001 | envp[i++] = "HOME=/"; | |
5002 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5003 | envp[i] = NULL; | |
5004 | ||
5005 | /* Drop the lock while we invoke the usermode helper, | |
5006 | * since the exec could involve hitting disk and hence | |
5007 | * be a slow process */ | |
5008 | mutex_unlock(&cgroup_mutex); | |
5009 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 5010 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
5011 | continue_free: |
5012 | kfree(pathbuf); | |
5013 | kfree(agentbuf); | |
cdcc136f | 5014 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 5015 | } |
cdcc136f | 5016 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5017 | mutex_unlock(&cgroup_mutex); |
5018 | } | |
8bab8dde PM |
5019 | |
5020 | static int __init cgroup_disable(char *str) | |
5021 | { | |
30159ec7 | 5022 | struct cgroup_subsys *ss; |
8bab8dde | 5023 | char *token; |
30159ec7 | 5024 | int i; |
8bab8dde PM |
5025 | |
5026 | while ((token = strsep(&str, ",")) != NULL) { | |
5027 | if (!*token) | |
5028 | continue; | |
be45c900 | 5029 | |
3ed80a62 | 5030 | for_each_subsys(ss, i) { |
8bab8dde PM |
5031 | if (!strcmp(token, ss->name)) { |
5032 | ss->disabled = 1; | |
5033 | printk(KERN_INFO "Disabling %s control group" | |
5034 | " subsystem\n", ss->name); | |
5035 | break; | |
5036 | } | |
5037 | } | |
5038 | } | |
5039 | return 1; | |
5040 | } | |
5041 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5042 | |
b77d7b60 | 5043 | /** |
5a17f543 | 5044 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
5045 | * @dentry: directory dentry of interest |
5046 | * @ss: subsystem of interest | |
b77d7b60 | 5047 | * |
5a17f543 TH |
5048 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5049 | * to get the corresponding css and return it. If such css doesn't exist | |
5050 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5051 | */ |
5a17f543 TH |
5052 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
5053 | struct cgroup_subsys *ss) | |
e5d1367f SE |
5054 | { |
5055 | struct cgroup *cgrp; | |
5a17f543 | 5056 | struct cgroup_subsys_state *css; |
b77d7b60 | 5057 | |
35cf0836 TH |
5058 | /* is @dentry a cgroup dir? */ |
5059 | if (!dentry->d_inode || | |
5060 | dentry->d_inode->i_op != &cgroup_dir_inode_operations) | |
e5d1367f SE |
5061 | return ERR_PTR(-EBADF); |
5062 | ||
5a17f543 TH |
5063 | rcu_read_lock(); |
5064 | ||
35cf0836 | 5065 | cgrp = __d_cgrp(dentry); |
5a17f543 TH |
5066 | css = cgroup_css(cgrp, ss); |
5067 | ||
5068 | if (!css || !css_tryget(css)) | |
5069 | css = ERR_PTR(-ENOENT); | |
5070 | ||
5071 | rcu_read_unlock(); | |
5072 | return css; | |
e5d1367f | 5073 | } |
e5d1367f | 5074 | |
1cb650b9 LZ |
5075 | /** |
5076 | * css_from_id - lookup css by id | |
5077 | * @id: the cgroup id | |
5078 | * @ss: cgroup subsys to be looked into | |
5079 | * | |
5080 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5081 | * Should be called under rcu_read_lock(). | |
5082 | */ | |
5083 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5084 | { | |
5085 | struct cgroup *cgrp; | |
5086 | ||
ace2bee8 | 5087 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
5088 | |
5089 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
5090 | if (cgrp) | |
d1625964 | 5091 | return cgroup_css(cgrp, ss); |
1cb650b9 | 5092 | return NULL; |
e5d1367f SE |
5093 | } |
5094 | ||
fe693435 | 5095 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5096 | static struct cgroup_subsys_state * |
5097 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5098 | { |
5099 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5100 | ||
5101 | if (!css) | |
5102 | return ERR_PTR(-ENOMEM); | |
5103 | ||
5104 | return css; | |
5105 | } | |
5106 | ||
eb95419b | 5107 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5108 | { |
eb95419b | 5109 | kfree(css); |
fe693435 PM |
5110 | } |
5111 | ||
182446d0 TH |
5112 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5113 | struct cftype *cft) | |
fe693435 | 5114 | { |
182446d0 | 5115 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5116 | } |
5117 | ||
182446d0 TH |
5118 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5119 | struct cftype *cft) | |
fe693435 PM |
5120 | { |
5121 | return (u64)(unsigned long)current->cgroups; | |
5122 | } | |
5123 | ||
182446d0 | 5124 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5125 | struct cftype *cft) |
fe693435 PM |
5126 | { |
5127 | u64 count; | |
5128 | ||
5129 | rcu_read_lock(); | |
a8ad805c | 5130 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5131 | rcu_read_unlock(); |
5132 | return count; | |
5133 | } | |
5134 | ||
2da8ca82 | 5135 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5136 | { |
69d0206c | 5137 | struct cgrp_cset_link *link; |
5abb8855 | 5138 | struct css_set *cset; |
7717f7ba PM |
5139 | |
5140 | read_lock(&css_set_lock); | |
5141 | rcu_read_lock(); | |
5abb8855 | 5142 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5143 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba PM |
5144 | struct cgroup *c = link->cgrp; |
5145 | const char *name; | |
5146 | ||
5147 | if (c->dentry) | |
5148 | name = c->dentry->d_name.name; | |
5149 | else | |
5150 | name = "?"; | |
2c6ab6d2 PM |
5151 | seq_printf(seq, "Root %d group %s\n", |
5152 | c->root->hierarchy_id, name); | |
7717f7ba PM |
5153 | } |
5154 | rcu_read_unlock(); | |
5155 | read_unlock(&css_set_lock); | |
5156 | return 0; | |
5157 | } | |
5158 | ||
5159 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 5160 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5161 | { |
2da8ca82 | 5162 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 5163 | struct cgrp_cset_link *link; |
7717f7ba PM |
5164 | |
5165 | read_lock(&css_set_lock); | |
182446d0 | 5166 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5167 | struct css_set *cset = link->cset; |
7717f7ba PM |
5168 | struct task_struct *task; |
5169 | int count = 0; | |
5abb8855 TH |
5170 | seq_printf(seq, "css_set %p\n", cset); |
5171 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
7717f7ba PM |
5172 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { |
5173 | seq_puts(seq, " ...\n"); | |
5174 | break; | |
5175 | } else { | |
5176 | seq_printf(seq, " task %d\n", | |
5177 | task_pid_vnr(task)); | |
5178 | } | |
5179 | } | |
5180 | } | |
5181 | read_unlock(&css_set_lock); | |
5182 | return 0; | |
5183 | } | |
5184 | ||
182446d0 | 5185 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5186 | { |
182446d0 | 5187 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
5188 | } |
5189 | ||
5190 | static struct cftype debug_files[] = { | |
fe693435 PM |
5191 | { |
5192 | .name = "taskcount", | |
5193 | .read_u64 = debug_taskcount_read, | |
5194 | }, | |
5195 | ||
5196 | { | |
5197 | .name = "current_css_set", | |
5198 | .read_u64 = current_css_set_read, | |
5199 | }, | |
5200 | ||
5201 | { | |
5202 | .name = "current_css_set_refcount", | |
5203 | .read_u64 = current_css_set_refcount_read, | |
5204 | }, | |
5205 | ||
7717f7ba PM |
5206 | { |
5207 | .name = "current_css_set_cg_links", | |
2da8ca82 | 5208 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
5209 | }, |
5210 | ||
5211 | { | |
5212 | .name = "cgroup_css_links", | |
2da8ca82 | 5213 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
5214 | }, |
5215 | ||
fe693435 PM |
5216 | { |
5217 | .name = "releasable", | |
5218 | .read_u64 = releasable_read, | |
5219 | }, | |
fe693435 | 5220 | |
4baf6e33 TH |
5221 | { } /* terminate */ |
5222 | }; | |
fe693435 | 5223 | |
073219e9 | 5224 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
5225 | .css_alloc = debug_css_alloc, |
5226 | .css_free = debug_css_free, | |
4baf6e33 | 5227 | .base_cftypes = debug_files, |
fe693435 PM |
5228 | }; |
5229 | #endif /* CONFIG_CGROUP_DEBUG */ |