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 | ||
e25e2cbb TH |
70 | /* |
71 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
72 | * hierarchy must be performed while holding it. | |
e25e2cbb | 73 | */ |
2219449a TH |
74 | #ifdef CONFIG_PROVE_RCU |
75 | DEFINE_MUTEX(cgroup_mutex); | |
8af01f56 | 76 | EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for lockdep */ |
2219449a | 77 | #else |
81a6a5cd | 78 | static DEFINE_MUTEX(cgroup_mutex); |
2219449a TH |
79 | #endif |
80 | ||
69e943b7 TH |
81 | /* |
82 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
83 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
84 | */ | |
85 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
86 | ||
87fb54f1 TH |
87 | #define cgroup_assert_mutex_or_rcu_locked() \ |
88 | rcu_lockdep_assert(rcu_read_lock_held() || \ | |
89 | lockdep_is_held(&cgroup_mutex), \ | |
90 | "cgroup_mutex or RCU read lock required"); | |
91 | ||
e5fca243 TH |
92 | /* |
93 | * cgroup destruction makes heavy use of work items and there can be a lot | |
94 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
95 | * destruction work items don't end up filling up max_active of system_wq | |
96 | * which may lead to deadlock. | |
97 | */ | |
98 | static struct workqueue_struct *cgroup_destroy_wq; | |
99 | ||
b1a21367 TH |
100 | /* |
101 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
102 | * separate workqueue as flush domain. | |
103 | */ | |
104 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
105 | ||
3ed80a62 | 106 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 107 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 108 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
109 | #include <linux/cgroup_subsys.h> |
110 | }; | |
073219e9 TH |
111 | #undef SUBSYS |
112 | ||
113 | /* array of cgroup subsystem names */ | |
114 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
115 | static const char *cgroup_subsys_name[] = { | |
116 | #include <linux/cgroup_subsys.h> | |
117 | }; | |
118 | #undef SUBSYS | |
ddbcc7e8 | 119 | |
ddbcc7e8 | 120 | /* |
9871bf95 TH |
121 | * The dummy hierarchy, reserved for the subsystems that are otherwise |
122 | * unattached - it never has more than a single cgroup, and all tasks are | |
123 | * part of that cgroup. | |
ddbcc7e8 | 124 | */ |
9871bf95 TH |
125 | static struct cgroupfs_root cgroup_dummy_root; |
126 | ||
127 | /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */ | |
128 | static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup; | |
ddbcc7e8 PM |
129 | |
130 | /* The list of hierarchy roots */ | |
131 | ||
9871bf95 TH |
132 | static LIST_HEAD(cgroup_roots); |
133 | static int cgroup_root_count; | |
ddbcc7e8 | 134 | |
3417ae1f | 135 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 136 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 137 | |
65dff759 LZ |
138 | static struct cgroup_name root_cgroup_name = { .name = "/" }; |
139 | ||
794611a1 LZ |
140 | /* |
141 | * Assign a monotonically increasing serial number to cgroups. It | |
142 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
143 | * numbers. Also, as cgroups are always appended to the parent's | |
144 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
145 | * the ascending serial number order on the list. Protected by |
146 | * cgroup_mutex. | |
794611a1 | 147 | */ |
00356bd5 | 148 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 149 | |
ddbcc7e8 | 150 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
151 | * check for fork/exit handlers to call. This avoids us having to do |
152 | * extra work in the fork/exit path if none of the subsystems need to | |
153 | * be called. | |
ddbcc7e8 | 154 | */ |
8947f9d5 | 155 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 156 | |
628f7cd4 TH |
157 | static struct cftype cgroup_base_files[]; |
158 | ||
f20104de | 159 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 160 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
2bb566cb TH |
161 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
162 | bool is_add); | |
e605b365 | 163 | static int cgroup_file_release(struct inode *inode, struct file *file); |
b1a21367 | 164 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 165 | |
95109b62 TH |
166 | /** |
167 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
168 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 169 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 170 | * |
ca8bdcaf TH |
171 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
172 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
173 | * the caller is responsible for pinning the returned css if it wants to | |
174 | * keep accessing it outside the said locks. This function may return | |
175 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
176 | */ |
177 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 178 | struct cgroup_subsys *ss) |
95109b62 | 179 | { |
ca8bdcaf | 180 | if (ss) |
aec25020 | 181 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ca8bdcaf TH |
182 | lockdep_is_held(&cgroup_mutex)); |
183 | else | |
184 | return &cgrp->dummy_css; | |
95109b62 | 185 | } |
42809dd4 | 186 | |
ddbcc7e8 | 187 | /* convenient tests for these bits */ |
54766d4a | 188 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 189 | { |
54766d4a | 190 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
191 | } |
192 | ||
78574cf9 LZ |
193 | /** |
194 | * cgroup_is_descendant - test ancestry | |
195 | * @cgrp: the cgroup to be tested | |
196 | * @ancestor: possible ancestor of @cgrp | |
197 | * | |
198 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
199 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
200 | * and @ancestor are accessible. | |
201 | */ | |
202 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
203 | { | |
204 | while (cgrp) { | |
205 | if (cgrp == ancestor) | |
206 | return true; | |
207 | cgrp = cgrp->parent; | |
208 | } | |
209 | return false; | |
210 | } | |
211 | EXPORT_SYMBOL_GPL(cgroup_is_descendant); | |
ddbcc7e8 | 212 | |
e9685a03 | 213 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
214 | { |
215 | const int bits = | |
bd89aabc PM |
216 | (1 << CGRP_RELEASABLE) | |
217 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
218 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
219 | } |
220 | ||
e9685a03 | 221 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 222 | { |
bd89aabc | 223 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
224 | } |
225 | ||
1c6727af TH |
226 | /** |
227 | * for_each_css - iterate all css's of a cgroup | |
228 | * @css: the iteration cursor | |
229 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
230 | * @cgrp: the target cgroup to iterate css's of | |
231 | * | |
232 | * Should be called under cgroup_mutex. | |
233 | */ | |
234 | #define for_each_css(css, ssid, cgrp) \ | |
235 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
236 | if (!((css) = rcu_dereference_check( \ | |
237 | (cgrp)->subsys[(ssid)], \ | |
238 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
239 | else | |
240 | ||
30159ec7 | 241 | /** |
3ed80a62 | 242 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 243 | * @ss: the iteration cursor |
780cd8b3 | 244 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 245 | */ |
780cd8b3 | 246 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
247 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
248 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 249 | |
5549c497 TH |
250 | /* iterate across the active hierarchies */ |
251 | #define for_each_active_root(root) \ | |
252 | list_for_each_entry((root), &cgroup_roots, root_list) | |
ddbcc7e8 | 253 | |
f6ea9372 TH |
254 | static inline struct cgroup *__d_cgrp(struct dentry *dentry) |
255 | { | |
256 | return dentry->d_fsdata; | |
257 | } | |
258 | ||
05ef1d7c | 259 | static inline struct cfent *__d_cfe(struct dentry *dentry) |
f6ea9372 TH |
260 | { |
261 | return dentry->d_fsdata; | |
262 | } | |
263 | ||
05ef1d7c TH |
264 | static inline struct cftype *__d_cft(struct dentry *dentry) |
265 | { | |
266 | return __d_cfe(dentry)->type; | |
267 | } | |
268 | ||
7ae1bad9 TH |
269 | /** |
270 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
271 | * @cgrp: the cgroup to be checked for liveness | |
272 | * | |
47cfcd09 TH |
273 | * On success, returns true; the mutex should be later unlocked. On |
274 | * failure returns false with no lock held. | |
7ae1bad9 | 275 | */ |
b9777cf8 | 276 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
277 | { |
278 | mutex_lock(&cgroup_mutex); | |
54766d4a | 279 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
280 | mutex_unlock(&cgroup_mutex); |
281 | return false; | |
282 | } | |
283 | return true; | |
284 | } | |
7ae1bad9 | 285 | |
81a6a5cd PM |
286 | /* the list of cgroups eligible for automatic release. Protected by |
287 | * release_list_lock */ | |
288 | static LIST_HEAD(release_list); | |
cdcc136f | 289 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
290 | static void cgroup_release_agent(struct work_struct *work); |
291 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 292 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 293 | |
69d0206c TH |
294 | /* |
295 | * A cgroup can be associated with multiple css_sets as different tasks may | |
296 | * belong to different cgroups on different hierarchies. In the other | |
297 | * direction, a css_set is naturally associated with multiple cgroups. | |
298 | * This M:N relationship is represented by the following link structure | |
299 | * which exists for each association and allows traversing the associations | |
300 | * from both sides. | |
301 | */ | |
302 | struct cgrp_cset_link { | |
303 | /* the cgroup and css_set this link associates */ | |
304 | struct cgroup *cgrp; | |
305 | struct css_set *cset; | |
306 | ||
307 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
308 | struct list_head cset_link; | |
309 | ||
310 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
311 | struct list_head cgrp_link; | |
817929ec PM |
312 | }; |
313 | ||
314 | /* The default css_set - used by init and its children prior to any | |
315 | * hierarchies being mounted. It contains a pointer to the root state | |
316 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
317 | * reference-counted, to improve performance when child cgroups | |
318 | * haven't been created. | |
319 | */ | |
320 | ||
321 | static struct css_set init_css_set; | |
69d0206c | 322 | static struct cgrp_cset_link init_cgrp_cset_link; |
817929ec | 323 | |
0942eeee TH |
324 | /* |
325 | * css_set_lock protects the list of css_set objects, and the chain of | |
326 | * tasks off each css_set. Nests outside task->alloc_lock due to | |
72ec7029 | 327 | * css_task_iter_start(). |
0942eeee | 328 | */ |
817929ec PM |
329 | static DEFINE_RWLOCK(css_set_lock); |
330 | static int css_set_count; | |
331 | ||
7717f7ba PM |
332 | /* |
333 | * hash table for cgroup groups. This improves the performance to find | |
334 | * an existing css_set. This hash doesn't (currently) take into | |
335 | * account cgroups in empty hierarchies. | |
336 | */ | |
472b1053 | 337 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 338 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 339 | |
0ac801fe | 340 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 341 | { |
0ac801fe | 342 | unsigned long key = 0UL; |
30159ec7 TH |
343 | struct cgroup_subsys *ss; |
344 | int i; | |
472b1053 | 345 | |
30159ec7 | 346 | for_each_subsys(ss, i) |
0ac801fe LZ |
347 | key += (unsigned long)css[i]; |
348 | key = (key >> 16) ^ key; | |
472b1053 | 349 | |
0ac801fe | 350 | return key; |
472b1053 LZ |
351 | } |
352 | ||
0942eeee TH |
353 | /* |
354 | * We don't maintain the lists running through each css_set to its task | |
72ec7029 TH |
355 | * until after the first call to css_task_iter_start(). This reduces the |
356 | * fork()/exit() overhead for people who have cgroups compiled into their | |
357 | * kernel but not actually in use. | |
0942eeee | 358 | */ |
8947f9d5 | 359 | static int use_task_css_set_links __read_mostly; |
817929ec | 360 | |
5abb8855 | 361 | static void __put_css_set(struct css_set *cset, int taskexit) |
b4f48b63 | 362 | { |
69d0206c | 363 | struct cgrp_cset_link *link, *tmp_link; |
5abb8855 | 364 | |
146aa1bd LJ |
365 | /* |
366 | * Ensure that the refcount doesn't hit zero while any readers | |
367 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
368 | * rwlock | |
369 | */ | |
5abb8855 | 370 | if (atomic_add_unless(&cset->refcount, -1, 1)) |
146aa1bd LJ |
371 | return; |
372 | write_lock(&css_set_lock); | |
5abb8855 | 373 | if (!atomic_dec_and_test(&cset->refcount)) { |
146aa1bd LJ |
374 | write_unlock(&css_set_lock); |
375 | return; | |
376 | } | |
81a6a5cd | 377 | |
2c6ab6d2 | 378 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
5abb8855 | 379 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
380 | css_set_count--; |
381 | ||
69d0206c | 382 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 383 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 384 | |
69d0206c TH |
385 | list_del(&link->cset_link); |
386 | list_del(&link->cgrp_link); | |
71b5707e | 387 | |
ddd69148 | 388 | /* @cgrp can't go away while we're holding css_set_lock */ |
6f3d828f | 389 | if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { |
81a6a5cd | 390 | if (taskexit) |
bd89aabc PM |
391 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
392 | check_for_release(cgrp); | |
81a6a5cd | 393 | } |
2c6ab6d2 PM |
394 | |
395 | kfree(link); | |
81a6a5cd | 396 | } |
2c6ab6d2 PM |
397 | |
398 | write_unlock(&css_set_lock); | |
5abb8855 | 399 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
400 | } |
401 | ||
817929ec PM |
402 | /* |
403 | * refcounted get/put for css_set objects | |
404 | */ | |
5abb8855 | 405 | static inline void get_css_set(struct css_set *cset) |
817929ec | 406 | { |
5abb8855 | 407 | atomic_inc(&cset->refcount); |
817929ec PM |
408 | } |
409 | ||
5abb8855 | 410 | static inline void put_css_set(struct css_set *cset) |
817929ec | 411 | { |
5abb8855 | 412 | __put_css_set(cset, 0); |
817929ec PM |
413 | } |
414 | ||
5abb8855 | 415 | static inline void put_css_set_taskexit(struct css_set *cset) |
81a6a5cd | 416 | { |
5abb8855 | 417 | __put_css_set(cset, 1); |
81a6a5cd PM |
418 | } |
419 | ||
b326f9d0 | 420 | /** |
7717f7ba | 421 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
422 | * @cset: candidate css_set being tested |
423 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
424 | * @new_cgrp: cgroup that's being entered by the task |
425 | * @template: desired set of css pointers in css_set (pre-calculated) | |
426 | * | |
6f4b7e63 | 427 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
428 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
429 | */ | |
5abb8855 TH |
430 | static bool compare_css_sets(struct css_set *cset, |
431 | struct css_set *old_cset, | |
7717f7ba PM |
432 | struct cgroup *new_cgrp, |
433 | struct cgroup_subsys_state *template[]) | |
434 | { | |
435 | struct list_head *l1, *l2; | |
436 | ||
5abb8855 | 437 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) { |
7717f7ba PM |
438 | /* Not all subsystems matched */ |
439 | return false; | |
440 | } | |
441 | ||
442 | /* | |
443 | * Compare cgroup pointers in order to distinguish between | |
444 | * different cgroups in heirarchies with no subsystems. We | |
445 | * could get by with just this check alone (and skip the | |
446 | * memcmp above) but on most setups the memcmp check will | |
447 | * avoid the need for this more expensive check on almost all | |
448 | * candidates. | |
449 | */ | |
450 | ||
69d0206c TH |
451 | l1 = &cset->cgrp_links; |
452 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 453 | while (1) { |
69d0206c | 454 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 455 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
456 | |
457 | l1 = l1->next; | |
458 | l2 = l2->next; | |
459 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
460 | if (l1 == &cset->cgrp_links) { |
461 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
462 | break; |
463 | } else { | |
69d0206c | 464 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
465 | } |
466 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
467 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
468 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
469 | cgrp1 = link1->cgrp; | |
470 | cgrp2 = link2->cgrp; | |
7717f7ba | 471 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 472 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
473 | |
474 | /* | |
475 | * If this hierarchy is the hierarchy of the cgroup | |
476 | * that's changing, then we need to check that this | |
477 | * css_set points to the new cgroup; if it's any other | |
478 | * hierarchy, then this css_set should point to the | |
479 | * same cgroup as the old css_set. | |
480 | */ | |
5abb8855 TH |
481 | if (cgrp1->root == new_cgrp->root) { |
482 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
483 | return false; |
484 | } else { | |
5abb8855 | 485 | if (cgrp1 != cgrp2) |
7717f7ba PM |
486 | return false; |
487 | } | |
488 | } | |
489 | return true; | |
490 | } | |
491 | ||
b326f9d0 TH |
492 | /** |
493 | * find_existing_css_set - init css array and find the matching css_set | |
494 | * @old_cset: the css_set that we're using before the cgroup transition | |
495 | * @cgrp: the cgroup that we're moving into | |
496 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 497 | */ |
5abb8855 TH |
498 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
499 | struct cgroup *cgrp, | |
500 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 501 | { |
bd89aabc | 502 | struct cgroupfs_root *root = cgrp->root; |
30159ec7 | 503 | struct cgroup_subsys *ss; |
5abb8855 | 504 | struct css_set *cset; |
0ac801fe | 505 | unsigned long key; |
b326f9d0 | 506 | int i; |
817929ec | 507 | |
aae8aab4 BB |
508 | /* |
509 | * Build the set of subsystem state objects that we want to see in the | |
510 | * new css_set. while subsystems can change globally, the entries here | |
511 | * won't change, so no need for locking. | |
512 | */ | |
30159ec7 | 513 | for_each_subsys(ss, i) { |
a1a71b45 | 514 | if (root->subsys_mask & (1UL << i)) { |
817929ec PM |
515 | /* Subsystem is in this hierarchy. So we want |
516 | * the subsystem state from the new | |
517 | * cgroup */ | |
ca8bdcaf | 518 | template[i] = cgroup_css(cgrp, ss); |
817929ec PM |
519 | } else { |
520 | /* Subsystem is not in this hierarchy, so we | |
521 | * don't want to change the subsystem state */ | |
5abb8855 | 522 | template[i] = old_cset->subsys[i]; |
817929ec PM |
523 | } |
524 | } | |
525 | ||
0ac801fe | 526 | key = css_set_hash(template); |
5abb8855 TH |
527 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
528 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
529 | continue; |
530 | ||
531 | /* This css_set matches what we need */ | |
5abb8855 | 532 | return cset; |
472b1053 | 533 | } |
817929ec PM |
534 | |
535 | /* No existing cgroup group matched */ | |
536 | return NULL; | |
537 | } | |
538 | ||
69d0206c | 539 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 540 | { |
69d0206c | 541 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 542 | |
69d0206c TH |
543 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
544 | list_del(&link->cset_link); | |
36553434 LZ |
545 | kfree(link); |
546 | } | |
547 | } | |
548 | ||
69d0206c TH |
549 | /** |
550 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
551 | * @count: the number of links to allocate | |
552 | * @tmp_links: list_head the allocated links are put on | |
553 | * | |
554 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
555 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 556 | */ |
69d0206c | 557 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 558 | { |
69d0206c | 559 | struct cgrp_cset_link *link; |
817929ec | 560 | int i; |
69d0206c TH |
561 | |
562 | INIT_LIST_HEAD(tmp_links); | |
563 | ||
817929ec | 564 | for (i = 0; i < count; i++) { |
f4f4be2b | 565 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 566 | if (!link) { |
69d0206c | 567 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
568 | return -ENOMEM; |
569 | } | |
69d0206c | 570 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
571 | } |
572 | return 0; | |
573 | } | |
574 | ||
c12f65d4 LZ |
575 | /** |
576 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 577 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 578 | * @cset: the css_set to be linked |
c12f65d4 LZ |
579 | * @cgrp: the destination cgroup |
580 | */ | |
69d0206c TH |
581 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
582 | struct cgroup *cgrp) | |
c12f65d4 | 583 | { |
69d0206c | 584 | struct cgrp_cset_link *link; |
c12f65d4 | 585 | |
69d0206c TH |
586 | BUG_ON(list_empty(tmp_links)); |
587 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); | |
588 | link->cset = cset; | |
7717f7ba | 589 | link->cgrp = cgrp; |
69d0206c | 590 | list_move(&link->cset_link, &cgrp->cset_links); |
7717f7ba PM |
591 | /* |
592 | * Always add links to the tail of the list so that the list | |
593 | * is sorted by order of hierarchy creation | |
594 | */ | |
69d0206c | 595 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
596 | } |
597 | ||
b326f9d0 TH |
598 | /** |
599 | * find_css_set - return a new css_set with one cgroup updated | |
600 | * @old_cset: the baseline css_set | |
601 | * @cgrp: the cgroup to be updated | |
602 | * | |
603 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
604 | * substituted into the appropriate hierarchy. | |
817929ec | 605 | */ |
5abb8855 TH |
606 | static struct css_set *find_css_set(struct css_set *old_cset, |
607 | struct cgroup *cgrp) | |
817929ec | 608 | { |
b326f9d0 | 609 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 610 | struct css_set *cset; |
69d0206c TH |
611 | struct list_head tmp_links; |
612 | struct cgrp_cset_link *link; | |
0ac801fe | 613 | unsigned long key; |
472b1053 | 614 | |
b326f9d0 TH |
615 | lockdep_assert_held(&cgroup_mutex); |
616 | ||
817929ec PM |
617 | /* First see if we already have a cgroup group that matches |
618 | * the desired set */ | |
7e9abd89 | 619 | read_lock(&css_set_lock); |
5abb8855 TH |
620 | cset = find_existing_css_set(old_cset, cgrp, template); |
621 | if (cset) | |
622 | get_css_set(cset); | |
7e9abd89 | 623 | read_unlock(&css_set_lock); |
817929ec | 624 | |
5abb8855 TH |
625 | if (cset) |
626 | return cset; | |
817929ec | 627 | |
f4f4be2b | 628 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 629 | if (!cset) |
817929ec PM |
630 | return NULL; |
631 | ||
69d0206c | 632 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 633 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 634 | kfree(cset); |
817929ec PM |
635 | return NULL; |
636 | } | |
637 | ||
5abb8855 | 638 | atomic_set(&cset->refcount, 1); |
69d0206c | 639 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 TH |
640 | INIT_LIST_HEAD(&cset->tasks); |
641 | INIT_HLIST_NODE(&cset->hlist); | |
817929ec PM |
642 | |
643 | /* Copy the set of subsystem state objects generated in | |
644 | * find_existing_css_set() */ | |
5abb8855 | 645 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec PM |
646 | |
647 | write_lock(&css_set_lock); | |
648 | /* Add reference counts and links from the new css_set. */ | |
69d0206c | 649 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 650 | struct cgroup *c = link->cgrp; |
69d0206c | 651 | |
7717f7ba PM |
652 | if (c->root == cgrp->root) |
653 | c = cgrp; | |
69d0206c | 654 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 655 | } |
817929ec | 656 | |
69d0206c | 657 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 658 | |
817929ec | 659 | css_set_count++; |
472b1053 LZ |
660 | |
661 | /* Add this cgroup group to the hash table */ | |
5abb8855 TH |
662 | key = css_set_hash(cset->subsys); |
663 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 664 | |
817929ec PM |
665 | write_unlock(&css_set_lock); |
666 | ||
5abb8855 | 667 | return cset; |
b4f48b63 PM |
668 | } |
669 | ||
7717f7ba PM |
670 | /* |
671 | * Return the cgroup for "task" from the given hierarchy. Must be | |
672 | * called with cgroup_mutex held. | |
673 | */ | |
674 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
675 | struct cgroupfs_root *root) | |
676 | { | |
5abb8855 | 677 | struct css_set *cset; |
7717f7ba PM |
678 | struct cgroup *res = NULL; |
679 | ||
680 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
681 | read_lock(&css_set_lock); | |
682 | /* | |
683 | * No need to lock the task - since we hold cgroup_mutex the | |
684 | * task can't change groups, so the only thing that can happen | |
685 | * is that it exits and its css is set back to init_css_set. | |
686 | */ | |
a8ad805c | 687 | cset = task_css_set(task); |
5abb8855 | 688 | if (cset == &init_css_set) { |
7717f7ba PM |
689 | res = &root->top_cgroup; |
690 | } else { | |
69d0206c TH |
691 | struct cgrp_cset_link *link; |
692 | ||
693 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 694 | struct cgroup *c = link->cgrp; |
69d0206c | 695 | |
7717f7ba PM |
696 | if (c->root == root) { |
697 | res = c; | |
698 | break; | |
699 | } | |
700 | } | |
701 | } | |
702 | read_unlock(&css_set_lock); | |
703 | BUG_ON(!res); | |
704 | return res; | |
705 | } | |
706 | ||
ddbcc7e8 PM |
707 | /* |
708 | * There is one global cgroup mutex. We also require taking | |
709 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
710 | * See "The task_lock() exception", at the end of this comment. | |
711 | * | |
712 | * A task must hold cgroup_mutex to modify cgroups. | |
713 | * | |
714 | * Any task can increment and decrement the count field without lock. | |
715 | * So in general, code holding cgroup_mutex can't rely on the count | |
716 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 717 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
718 | * means that no tasks are currently attached, therefore there is no |
719 | * way a task attached to that cgroup can fork (the other way to | |
720 | * increment the count). So code holding cgroup_mutex can safely | |
721 | * assume that if the count is zero, it will stay zero. Similarly, if | |
722 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
723 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
724 | * needs that mutex. | |
725 | * | |
ddbcc7e8 PM |
726 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
727 | * (usually) take cgroup_mutex. These are the two most performance | |
728 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
729 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
730 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
731 | * to the release agent with the name of the cgroup (path relative to |
732 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
733 | * |
734 | * A cgroup can only be deleted if both its 'count' of using tasks | |
735 | * is zero, and its list of 'children' cgroups is empty. Since all | |
736 | * tasks in the system use _some_ cgroup, and since there is always at | |
737 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
738 | * always has either children cgroups and/or using tasks. So we don't | |
739 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
740 | * | |
741 | * The task_lock() exception | |
742 | * | |
743 | * The need for this exception arises from the action of | |
d0b2fdd2 | 744 | * cgroup_attach_task(), which overwrites one task's cgroup pointer with |
a043e3b2 | 745 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
746 | * several performance critical places that need to reference |
747 | * task->cgroup without the expense of grabbing a system global | |
748 | * mutex. Therefore except as noted below, when dereferencing or, as | |
d0b2fdd2 | 749 | * in cgroup_attach_task(), modifying a task's cgroup pointer we use |
ddbcc7e8 PM |
750 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
751 | * the task_struct routinely used for such matters. | |
752 | * | |
753 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 754 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
755 | */ |
756 | ||
ddbcc7e8 PM |
757 | /* |
758 | * A couple of forward declarations required, due to cyclic reference loop: | |
759 | * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir -> | |
760 | * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations | |
761 | * -> cgroup_mkdir. | |
762 | */ | |
763 | ||
18bb1db3 | 764 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode); |
ddbcc7e8 | 765 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); |
628f7cd4 | 766 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
6e1d5dcc | 767 | static const struct inode_operations cgroup_dir_inode_operations; |
828c0950 | 768 | static const struct file_operations proc_cgroupstats_operations; |
a424316c PM |
769 | |
770 | static struct backing_dev_info cgroup_backing_dev_info = { | |
d993831f | 771 | .name = "cgroup", |
e4ad08fe | 772 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
a424316c | 773 | }; |
ddbcc7e8 | 774 | |
a5e7ed32 | 775 | static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb) |
ddbcc7e8 PM |
776 | { |
777 | struct inode *inode = new_inode(sb); | |
ddbcc7e8 PM |
778 | |
779 | if (inode) { | |
85fe4025 | 780 | inode->i_ino = get_next_ino(); |
ddbcc7e8 | 781 | inode->i_mode = mode; |
76aac0e9 DH |
782 | inode->i_uid = current_fsuid(); |
783 | inode->i_gid = current_fsgid(); | |
ddbcc7e8 PM |
784 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
785 | inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info; | |
786 | } | |
787 | return inode; | |
788 | } | |
789 | ||
65dff759 LZ |
790 | static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry) |
791 | { | |
792 | struct cgroup_name *name; | |
793 | ||
794 | name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL); | |
795 | if (!name) | |
796 | return NULL; | |
797 | strcpy(name->name, dentry->d_name.name); | |
798 | return name; | |
799 | } | |
800 | ||
be445626 LZ |
801 | static void cgroup_free_fn(struct work_struct *work) |
802 | { | |
ea15f8cc | 803 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 LZ |
804 | |
805 | mutex_lock(&cgroup_mutex); | |
be445626 LZ |
806 | cgrp->root->number_of_cgroups--; |
807 | mutex_unlock(&cgroup_mutex); | |
808 | ||
415cf07a LZ |
809 | /* |
810 | * We get a ref to the parent's dentry, and put the ref when | |
811 | * this cgroup is being freed, so it's guaranteed that the | |
812 | * parent won't be destroyed before its children. | |
813 | */ | |
814 | dput(cgrp->parent->dentry); | |
815 | ||
be445626 LZ |
816 | /* |
817 | * Drop the active superblock reference that we took when we | |
cc20e01c LZ |
818 | * created the cgroup. This will free cgrp->root, if we are |
819 | * holding the last reference to @sb. | |
be445626 LZ |
820 | */ |
821 | deactivate_super(cgrp->root->sb); | |
822 | ||
b1a21367 | 823 | cgroup_pidlist_destroy_all(cgrp); |
be445626 LZ |
824 | |
825 | simple_xattrs_free(&cgrp->xattrs); | |
826 | ||
65dff759 | 827 | kfree(rcu_dereference_raw(cgrp->name)); |
be445626 LZ |
828 | kfree(cgrp); |
829 | } | |
830 | ||
831 | static void cgroup_free_rcu(struct rcu_head *head) | |
832 | { | |
833 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
834 | ||
ea15f8cc | 835 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 836 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
837 | } |
838 | ||
ddbcc7e8 PM |
839 | static void cgroup_diput(struct dentry *dentry, struct inode *inode) |
840 | { | |
841 | /* is dentry a directory ? if so, kfree() associated cgroup */ | |
842 | if (S_ISDIR(inode->i_mode)) { | |
bd89aabc | 843 | struct cgroup *cgrp = dentry->d_fsdata; |
be445626 | 844 | |
54766d4a | 845 | BUG_ON(!(cgroup_is_dead(cgrp))); |
c1a71504 LZ |
846 | |
847 | /* | |
848 | * XXX: cgrp->id is only used to look up css's. As cgroup | |
849 | * and css's lifetimes will be decoupled, it should be made | |
850 | * per-subsystem and moved to css->id so that lookups are | |
851 | * successful until the target css is released. | |
852 | */ | |
0ab02ca8 | 853 | mutex_lock(&cgroup_mutex); |
c1a71504 | 854 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
0ab02ca8 | 855 | mutex_unlock(&cgroup_mutex); |
c1a71504 LZ |
856 | cgrp->id = -1; |
857 | ||
be445626 | 858 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
05ef1d7c TH |
859 | } else { |
860 | struct cfent *cfe = __d_cfe(dentry); | |
861 | struct cgroup *cgrp = dentry->d_parent->d_fsdata; | |
862 | ||
863 | WARN_ONCE(!list_empty(&cfe->node) && | |
864 | cgrp != &cgrp->root->top_cgroup, | |
865 | "cfe still linked for %s\n", cfe->type->name); | |
712317ad | 866 | simple_xattrs_free(&cfe->xattrs); |
05ef1d7c | 867 | kfree(cfe); |
ddbcc7e8 PM |
868 | } |
869 | iput(inode); | |
870 | } | |
871 | ||
872 | static void remove_dir(struct dentry *d) | |
873 | { | |
874 | struct dentry *parent = dget(d->d_parent); | |
875 | ||
876 | d_delete(d); | |
877 | simple_rmdir(parent->d_inode, d); | |
878 | dput(parent); | |
879 | } | |
880 | ||
2739d3cc | 881 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c TH |
882 | { |
883 | struct cfent *cfe; | |
884 | ||
885 | lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex); | |
886 | lockdep_assert_held(&cgroup_mutex); | |
887 | ||
2739d3cc LZ |
888 | /* |
889 | * If we're doing cleanup due to failure of cgroup_create(), | |
890 | * the corresponding @cfe may not exist. | |
891 | */ | |
05ef1d7c TH |
892 | list_for_each_entry(cfe, &cgrp->files, node) { |
893 | struct dentry *d = cfe->dentry; | |
894 | ||
895 | if (cft && cfe->type != cft) | |
896 | continue; | |
897 | ||
898 | dget(d); | |
899 | d_delete(d); | |
ce27e317 | 900 | simple_unlink(cgrp->dentry->d_inode, d); |
05ef1d7c TH |
901 | list_del_init(&cfe->node); |
902 | dput(d); | |
903 | ||
2739d3cc | 904 | break; |
ddbcc7e8 | 905 | } |
05ef1d7c TH |
906 | } |
907 | ||
13af07df | 908 | /** |
628f7cd4 | 909 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 910 | * @cgrp: target cgroup |
13af07df AR |
911 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
912 | */ | |
628f7cd4 | 913 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 914 | { |
13af07df | 915 | struct cgroup_subsys *ss; |
b420ba7d | 916 | int i; |
05ef1d7c | 917 | |
b420ba7d | 918 | for_each_subsys(ss, i) { |
13af07df | 919 | struct cftype_set *set; |
b420ba7d TH |
920 | |
921 | if (!test_bit(i, &subsys_mask)) | |
13af07df AR |
922 | continue; |
923 | list_for_each_entry(set, &ss->cftsets, node) | |
2bb566cb | 924 | cgroup_addrm_files(cgrp, set->cfts, false); |
13af07df | 925 | } |
ddbcc7e8 PM |
926 | } |
927 | ||
928 | /* | |
929 | * NOTE : the dentry must have been dget()'ed | |
930 | */ | |
931 | static void cgroup_d_remove_dir(struct dentry *dentry) | |
932 | { | |
2fd6b7f5 | 933 | struct dentry *parent; |
ddbcc7e8 | 934 | |
2fd6b7f5 NP |
935 | parent = dentry->d_parent; |
936 | spin_lock(&parent->d_lock); | |
3ec762ad | 937 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
ddbcc7e8 | 938 | list_del_init(&dentry->d_u.d_child); |
2fd6b7f5 NP |
939 | spin_unlock(&dentry->d_lock); |
940 | spin_unlock(&parent->d_lock); | |
ddbcc7e8 PM |
941 | remove_dir(dentry); |
942 | } | |
943 | ||
944 | static int rebind_subsystems(struct cgroupfs_root *root, | |
a8a648c4 | 945 | unsigned long added_mask, unsigned removed_mask) |
ddbcc7e8 | 946 | { |
bd89aabc | 947 | struct cgroup *cgrp = &root->top_cgroup; |
30159ec7 | 948 | struct cgroup_subsys *ss; |
3126121f | 949 | int i, ret; |
ddbcc7e8 | 950 | |
aae8aab4 BB |
951 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
952 | ||
ddbcc7e8 | 953 | /* Check that any added subsystems are currently free */ |
3ed80a62 TH |
954 | for_each_subsys(ss, i) |
955 | if ((added_mask & (1 << i)) && ss->root != &cgroup_dummy_root) | |
956 | return -EBUSY; | |
ddbcc7e8 | 957 | |
3126121f TH |
958 | ret = cgroup_populate_dir(cgrp, added_mask); |
959 | if (ret) | |
3ed80a62 | 960 | return ret; |
3126121f TH |
961 | |
962 | /* | |
963 | * Nothing can fail from this point on. Remove files for the | |
964 | * removed subsystems and rebind each subsystem. | |
965 | */ | |
966 | cgroup_clear_dir(cgrp, removed_mask); | |
ddbcc7e8 | 967 | |
30159ec7 | 968 | for_each_subsys(ss, i) { |
ddbcc7e8 | 969 | unsigned long bit = 1UL << i; |
30159ec7 | 970 | |
a1a71b45 | 971 | if (bit & added_mask) { |
ddbcc7e8 | 972 | /* We're binding this subsystem to this hierarchy */ |
ca8bdcaf TH |
973 | BUG_ON(cgroup_css(cgrp, ss)); |
974 | BUG_ON(!cgroup_css(cgroup_dummy_top, ss)); | |
975 | BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top); | |
a8a648c4 | 976 | |
73e80ed8 | 977 | rcu_assign_pointer(cgrp->subsys[i], |
ca8bdcaf TH |
978 | cgroup_css(cgroup_dummy_top, ss)); |
979 | cgroup_css(cgrp, ss)->cgroup = cgrp; | |
a8a648c4 | 980 | |
b2aa30f7 | 981 | ss->root = root; |
ddbcc7e8 | 982 | if (ss->bind) |
ca8bdcaf | 983 | ss->bind(cgroup_css(cgrp, ss)); |
a8a648c4 | 984 | |
cf5d5941 | 985 | /* refcount was already taken, and we're keeping it */ |
a8a648c4 | 986 | root->subsys_mask |= bit; |
a1a71b45 | 987 | } else if (bit & removed_mask) { |
ddbcc7e8 | 988 | /* We're removing this subsystem */ |
ca8bdcaf TH |
989 | BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss)); |
990 | BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp); | |
a8a648c4 | 991 | |
ddbcc7e8 | 992 | if (ss->bind) |
ca8bdcaf | 993 | ss->bind(cgroup_css(cgroup_dummy_top, ss)); |
73e80ed8 | 994 | |
ca8bdcaf | 995 | cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top; |
73e80ed8 TH |
996 | RCU_INIT_POINTER(cgrp->subsys[i], NULL); |
997 | ||
9871bf95 | 998 | cgroup_subsys[i]->root = &cgroup_dummy_root; |
a8a648c4 | 999 | root->subsys_mask &= ~bit; |
ddbcc7e8 PM |
1000 | } |
1001 | } | |
ddbcc7e8 | 1002 | |
1672d040 TH |
1003 | /* |
1004 | * Mark @root has finished binding subsystems. @root->subsys_mask | |
1005 | * now matches the bound subsystems. | |
1006 | */ | |
1007 | root->flags |= CGRP_ROOT_SUBSYS_BOUND; | |
1008 | ||
ddbcc7e8 PM |
1009 | return 0; |
1010 | } | |
1011 | ||
34c80b1d | 1012 | static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry) |
ddbcc7e8 | 1013 | { |
34c80b1d | 1014 | struct cgroupfs_root *root = dentry->d_sb->s_fs_info; |
ddbcc7e8 | 1015 | struct cgroup_subsys *ss; |
b85d2040 | 1016 | int ssid; |
ddbcc7e8 | 1017 | |
b85d2040 TH |
1018 | for_each_subsys(ss, ssid) |
1019 | if (root->subsys_mask & (1 << ssid)) | |
1020 | seq_printf(seq, ",%s", ss->name); | |
873fe09e TH |
1021 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1022 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1023 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1024 | seq_puts(seq, ",noprefix"); |
93438629 | 1025 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1026 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1027 | |
1028 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1029 | if (strlen(root->release_agent_path)) |
1030 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1031 | spin_unlock(&release_agent_path_lock); |
1032 | ||
2260e7fc | 1033 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags)) |
97978e6d | 1034 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1035 | if (strlen(root->name)) |
1036 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1037 | return 0; |
1038 | } | |
1039 | ||
1040 | struct cgroup_sb_opts { | |
a1a71b45 | 1041 | unsigned long subsys_mask; |
ddbcc7e8 | 1042 | unsigned long flags; |
81a6a5cd | 1043 | char *release_agent; |
2260e7fc | 1044 | bool cpuset_clone_children; |
c6d57f33 | 1045 | char *name; |
2c6ab6d2 PM |
1046 | /* User explicitly requested empty subsystem */ |
1047 | bool none; | |
c6d57f33 PM |
1048 | |
1049 | struct cgroupfs_root *new_root; | |
2c6ab6d2 | 1050 | |
ddbcc7e8 PM |
1051 | }; |
1052 | ||
aae8aab4 | 1053 | /* |
9871bf95 TH |
1054 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1055 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1056 | * array. This function takes refcounts on subsystems to be used, unless it | |
1057 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1058 | */ |
cf5d5941 | 1059 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1060 | { |
32a8cf23 DL |
1061 | char *token, *o = data; |
1062 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1063 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1064 | struct cgroup_subsys *ss; |
1065 | int i; | |
f9ab5b5b | 1066 | |
aae8aab4 BB |
1067 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1068 | ||
f9ab5b5b | 1069 | #ifdef CONFIG_CPUSETS |
073219e9 | 1070 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1071 | #endif |
ddbcc7e8 | 1072 | |
c6d57f33 | 1073 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1074 | |
1075 | while ((token = strsep(&o, ",")) != NULL) { | |
1076 | if (!*token) | |
1077 | return -EINVAL; | |
32a8cf23 | 1078 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1079 | /* Explicitly have no subsystems */ |
1080 | opts->none = true; | |
32a8cf23 DL |
1081 | continue; |
1082 | } | |
1083 | if (!strcmp(token, "all")) { | |
1084 | /* Mutually exclusive option 'all' + subsystem name */ | |
1085 | if (one_ss) | |
1086 | return -EINVAL; | |
1087 | all_ss = true; | |
1088 | continue; | |
1089 | } | |
873fe09e TH |
1090 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1091 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1092 | continue; | |
1093 | } | |
32a8cf23 | 1094 | if (!strcmp(token, "noprefix")) { |
93438629 | 1095 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1096 | continue; |
1097 | } | |
1098 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1099 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1100 | continue; |
1101 | } | |
03b1cde6 | 1102 | if (!strcmp(token, "xattr")) { |
93438629 | 1103 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1104 | continue; |
1105 | } | |
32a8cf23 | 1106 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1107 | /* Specifying two release agents is forbidden */ |
1108 | if (opts->release_agent) | |
1109 | return -EINVAL; | |
c6d57f33 | 1110 | opts->release_agent = |
e400c285 | 1111 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1112 | if (!opts->release_agent) |
1113 | return -ENOMEM; | |
32a8cf23 DL |
1114 | continue; |
1115 | } | |
1116 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1117 | const char *name = token + 5; |
1118 | /* Can't specify an empty name */ | |
1119 | if (!strlen(name)) | |
1120 | return -EINVAL; | |
1121 | /* Must match [\w.-]+ */ | |
1122 | for (i = 0; i < strlen(name); i++) { | |
1123 | char c = name[i]; | |
1124 | if (isalnum(c)) | |
1125 | continue; | |
1126 | if ((c == '.') || (c == '-') || (c == '_')) | |
1127 | continue; | |
1128 | return -EINVAL; | |
1129 | } | |
1130 | /* Specifying two names is forbidden */ | |
1131 | if (opts->name) | |
1132 | return -EINVAL; | |
1133 | opts->name = kstrndup(name, | |
e400c285 | 1134 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1135 | GFP_KERNEL); |
1136 | if (!opts->name) | |
1137 | return -ENOMEM; | |
32a8cf23 DL |
1138 | |
1139 | continue; | |
1140 | } | |
1141 | ||
30159ec7 | 1142 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1143 | if (strcmp(token, ss->name)) |
1144 | continue; | |
1145 | if (ss->disabled) | |
1146 | continue; | |
1147 | ||
1148 | /* Mutually exclusive option 'all' + subsystem name */ | |
1149 | if (all_ss) | |
1150 | return -EINVAL; | |
a1a71b45 | 1151 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1152 | one_ss = true; |
1153 | ||
1154 | break; | |
1155 | } | |
1156 | if (i == CGROUP_SUBSYS_COUNT) | |
1157 | return -ENOENT; | |
1158 | } | |
1159 | ||
1160 | /* | |
1161 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1162 | * otherwise if 'none', 'name=' and a subsystem name options |
1163 | * were not specified, let's default to 'all' | |
32a8cf23 | 1164 | */ |
30159ec7 TH |
1165 | if (all_ss || (!one_ss && !opts->none && !opts->name)) |
1166 | for_each_subsys(ss, i) | |
1167 | if (!ss->disabled) | |
1168 | set_bit(i, &opts->subsys_mask); | |
ddbcc7e8 | 1169 | |
2c6ab6d2 PM |
1170 | /* Consistency checks */ |
1171 | ||
873fe09e TH |
1172 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1173 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1174 | ||
1175 | if (opts->flags & CGRP_ROOT_NOPREFIX) { | |
1176 | pr_err("cgroup: sane_behavior: noprefix is not allowed\n"); | |
1177 | return -EINVAL; | |
1178 | } | |
1179 | ||
1180 | if (opts->cpuset_clone_children) { | |
1181 | pr_err("cgroup: sane_behavior: clone_children is not allowed\n"); | |
1182 | return -EINVAL; | |
1183 | } | |
1184 | } | |
1185 | ||
f9ab5b5b LZ |
1186 | /* |
1187 | * Option noprefix was introduced just for backward compatibility | |
1188 | * with the old cpuset, so we allow noprefix only if mounting just | |
1189 | * the cpuset subsystem. | |
1190 | */ | |
93438629 | 1191 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1192 | return -EINVAL; |
1193 | ||
2c6ab6d2 PM |
1194 | |
1195 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1196 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1197 | return -EINVAL; |
1198 | ||
1199 | /* | |
1200 | * We either have to specify by name or by subsystems. (So all | |
1201 | * empty hierarchies must have a name). | |
1202 | */ | |
a1a71b45 | 1203 | if (!opts->subsys_mask && !opts->name) |
ddbcc7e8 PM |
1204 | return -EINVAL; |
1205 | ||
1206 | return 0; | |
1207 | } | |
1208 | ||
1209 | static int cgroup_remount(struct super_block *sb, int *flags, char *data) | |
1210 | { | |
1211 | int ret = 0; | |
1212 | struct cgroupfs_root *root = sb->s_fs_info; | |
bd89aabc | 1213 | struct cgroup *cgrp = &root->top_cgroup; |
ddbcc7e8 | 1214 | struct cgroup_sb_opts opts; |
a1a71b45 | 1215 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1216 | |
873fe09e TH |
1217 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1218 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1219 | return -EINVAL; | |
1220 | } | |
1221 | ||
bd89aabc | 1222 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
1223 | mutex_lock(&cgroup_mutex); |
1224 | ||
1225 | /* See what subsystems are wanted */ | |
1226 | ret = parse_cgroupfs_options(data, &opts); | |
1227 | if (ret) | |
1228 | goto out_unlock; | |
1229 | ||
a8a648c4 | 1230 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1231 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1232 | task_tgid_nr(current), current->comm); | |
1233 | ||
a1a71b45 AR |
1234 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1235 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1236 | |
cf5d5941 | 1237 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1238 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1239 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1240 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1241 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1242 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1243 | ret = -EINVAL; |
1244 | goto out_unlock; | |
1245 | } | |
1246 | ||
f172e67c TH |
1247 | /* remounting is not allowed for populated hierarchies */ |
1248 | if (root->number_of_cgroups > 1) { | |
1249 | ret = -EBUSY; | |
0670e08b | 1250 | goto out_unlock; |
cf5d5941 | 1251 | } |
ddbcc7e8 | 1252 | |
a8a648c4 | 1253 | ret = rebind_subsystems(root, added_mask, removed_mask); |
3126121f | 1254 | if (ret) |
0670e08b | 1255 | goto out_unlock; |
ddbcc7e8 | 1256 | |
69e943b7 TH |
1257 | if (opts.release_agent) { |
1258 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1259 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1260 | spin_unlock(&release_agent_path_lock); |
1261 | } | |
ddbcc7e8 | 1262 | out_unlock: |
66bdc9cf | 1263 | kfree(opts.release_agent); |
c6d57f33 | 1264 | kfree(opts.name); |
ddbcc7e8 | 1265 | mutex_unlock(&cgroup_mutex); |
bd89aabc | 1266 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
1267 | return ret; |
1268 | } | |
1269 | ||
b87221de | 1270 | static const struct super_operations cgroup_ops = { |
ddbcc7e8 PM |
1271 | .statfs = simple_statfs, |
1272 | .drop_inode = generic_delete_inode, | |
1273 | .show_options = cgroup_show_options, | |
1274 | .remount_fs = cgroup_remount, | |
1275 | }; | |
1276 | ||
cc31edce PM |
1277 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1278 | { | |
1279 | INIT_LIST_HEAD(&cgrp->sibling); | |
1280 | INIT_LIST_HEAD(&cgrp->children); | |
05ef1d7c | 1281 | INIT_LIST_HEAD(&cgrp->files); |
69d0206c | 1282 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1283 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1284 | INIT_LIST_HEAD(&cgrp->pidlists); |
1285 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1286 | cgrp->dummy_css.cgroup = cgrp; |
03b1cde6 | 1287 | simple_xattrs_init(&cgrp->xattrs); |
cc31edce | 1288 | } |
c6d57f33 | 1289 | |
ddbcc7e8 PM |
1290 | static void init_cgroup_root(struct cgroupfs_root *root) |
1291 | { | |
bd89aabc | 1292 | struct cgroup *cgrp = &root->top_cgroup; |
b0ca5a84 | 1293 | |
ddbcc7e8 PM |
1294 | INIT_LIST_HEAD(&root->root_list); |
1295 | root->number_of_cgroups = 1; | |
bd89aabc | 1296 | cgrp->root = root; |
a4ea1cc9 | 1297 | RCU_INIT_POINTER(cgrp->name, &root_cgroup_name); |
cc31edce | 1298 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1299 | idr_init(&root->cgroup_idr); |
ddbcc7e8 PM |
1300 | } |
1301 | ||
fc76df70 | 1302 | static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end) |
2c6ab6d2 | 1303 | { |
1a574231 | 1304 | int id; |
2c6ab6d2 | 1305 | |
54e7b4eb | 1306 | lockdep_assert_held(&cgroup_mutex); |
54e7b4eb | 1307 | |
fc76df70 TH |
1308 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end, |
1309 | GFP_KERNEL); | |
1a574231 TH |
1310 | if (id < 0) |
1311 | return id; | |
1312 | ||
1313 | root->hierarchy_id = id; | |
fa3ca07e TH |
1314 | return 0; |
1315 | } | |
1316 | ||
1317 | static void cgroup_exit_root_id(struct cgroupfs_root *root) | |
1318 | { | |
54e7b4eb | 1319 | lockdep_assert_held(&cgroup_mutex); |
54e7b4eb | 1320 | |
fa3ca07e | 1321 | if (root->hierarchy_id) { |
1a574231 | 1322 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
fa3ca07e TH |
1323 | root->hierarchy_id = 0; |
1324 | } | |
2c6ab6d2 PM |
1325 | } |
1326 | ||
ddbcc7e8 PM |
1327 | static int cgroup_test_super(struct super_block *sb, void *data) |
1328 | { | |
c6d57f33 | 1329 | struct cgroup_sb_opts *opts = data; |
ddbcc7e8 PM |
1330 | struct cgroupfs_root *root = sb->s_fs_info; |
1331 | ||
c6d57f33 PM |
1332 | /* If we asked for a name then it must match */ |
1333 | if (opts->name && strcmp(opts->name, root->name)) | |
1334 | return 0; | |
ddbcc7e8 | 1335 | |
2c6ab6d2 PM |
1336 | /* |
1337 | * If we asked for subsystems (or explicitly for no | |
1338 | * subsystems) then they must match | |
1339 | */ | |
a1a71b45 AR |
1340 | if ((opts->subsys_mask || opts->none) |
1341 | && (opts->subsys_mask != root->subsys_mask)) | |
ddbcc7e8 PM |
1342 | return 0; |
1343 | ||
1344 | return 1; | |
1345 | } | |
1346 | ||
c6d57f33 PM |
1347 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1348 | { | |
1349 | struct cgroupfs_root *root; | |
1350 | ||
a1a71b45 | 1351 | if (!opts->subsys_mask && !opts->none) |
c6d57f33 PM |
1352 | return NULL; |
1353 | ||
1354 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1355 | if (!root) | |
1356 | return ERR_PTR(-ENOMEM); | |
1357 | ||
1358 | init_cgroup_root(root); | |
2c6ab6d2 | 1359 | |
1672d040 TH |
1360 | /* |
1361 | * We need to set @root->subsys_mask now so that @root can be | |
1362 | * matched by cgroup_test_super() before it finishes | |
1363 | * initialization; otherwise, competing mounts with the same | |
1364 | * options may try to bind the same subsystems instead of waiting | |
1365 | * for the first one leading to unexpected mount errors. | |
1366 | * SUBSYS_BOUND will be set once actual binding is complete. | |
1367 | */ | |
a1a71b45 | 1368 | root->subsys_mask = opts->subsys_mask; |
c6d57f33 PM |
1369 | root->flags = opts->flags; |
1370 | if (opts->release_agent) | |
1371 | strcpy(root->release_agent_path, opts->release_agent); | |
1372 | if (opts->name) | |
1373 | strcpy(root->name, opts->name); | |
2260e7fc TH |
1374 | if (opts->cpuset_clone_children) |
1375 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1376 | return root; |
1377 | } | |
1378 | ||
fa3ca07e | 1379 | static void cgroup_free_root(struct cgroupfs_root *root) |
2c6ab6d2 | 1380 | { |
fa3ca07e TH |
1381 | if (root) { |
1382 | /* hierarhcy ID shoulid already have been released */ | |
1383 | WARN_ON_ONCE(root->hierarchy_id); | |
2c6ab6d2 | 1384 | |
4e96ee8e | 1385 | idr_destroy(&root->cgroup_idr); |
fa3ca07e TH |
1386 | kfree(root); |
1387 | } | |
2c6ab6d2 PM |
1388 | } |
1389 | ||
ddbcc7e8 PM |
1390 | static int cgroup_set_super(struct super_block *sb, void *data) |
1391 | { | |
1392 | int ret; | |
c6d57f33 PM |
1393 | struct cgroup_sb_opts *opts = data; |
1394 | ||
1395 | /* If we don't have a new root, we can't set up a new sb */ | |
1396 | if (!opts->new_root) | |
1397 | return -EINVAL; | |
1398 | ||
a1a71b45 | 1399 | BUG_ON(!opts->subsys_mask && !opts->none); |
ddbcc7e8 PM |
1400 | |
1401 | ret = set_anon_super(sb, NULL); | |
1402 | if (ret) | |
1403 | return ret; | |
1404 | ||
c6d57f33 PM |
1405 | sb->s_fs_info = opts->new_root; |
1406 | opts->new_root->sb = sb; | |
ddbcc7e8 PM |
1407 | |
1408 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
1409 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
1410 | sb->s_magic = CGROUP_SUPER_MAGIC; | |
1411 | sb->s_op = &cgroup_ops; | |
1412 | ||
1413 | return 0; | |
1414 | } | |
1415 | ||
1416 | static int cgroup_get_rootdir(struct super_block *sb) | |
1417 | { | |
0df6a63f AV |
1418 | static const struct dentry_operations cgroup_dops = { |
1419 | .d_iput = cgroup_diput, | |
b26d4cd3 | 1420 | .d_delete = always_delete_dentry, |
0df6a63f AV |
1421 | }; |
1422 | ||
ddbcc7e8 PM |
1423 | struct inode *inode = |
1424 | cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb); | |
ddbcc7e8 PM |
1425 | |
1426 | if (!inode) | |
1427 | return -ENOMEM; | |
1428 | ||
ddbcc7e8 PM |
1429 | inode->i_fop = &simple_dir_operations; |
1430 | inode->i_op = &cgroup_dir_inode_operations; | |
1431 | /* directories start off with i_nlink == 2 (for "." entry) */ | |
1432 | inc_nlink(inode); | |
48fde701 AV |
1433 | sb->s_root = d_make_root(inode); |
1434 | if (!sb->s_root) | |
ddbcc7e8 | 1435 | return -ENOMEM; |
0df6a63f AV |
1436 | /* for everything else we want ->d_op set */ |
1437 | sb->s_d_op = &cgroup_dops; | |
ddbcc7e8 PM |
1438 | return 0; |
1439 | } | |
1440 | ||
f7e83571 | 1441 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1442 | int flags, const char *unused_dev_name, |
f7e83571 | 1443 | void *data) |
ddbcc7e8 PM |
1444 | { |
1445 | struct cgroup_sb_opts opts; | |
c6d57f33 | 1446 | struct cgroupfs_root *root; |
ddbcc7e8 PM |
1447 | int ret = 0; |
1448 | struct super_block *sb; | |
c6d57f33 | 1449 | struct cgroupfs_root *new_root; |
3126121f | 1450 | struct list_head tmp_links; |
e25e2cbb | 1451 | struct inode *inode; |
3126121f | 1452 | const struct cred *cred; |
ddbcc7e8 PM |
1453 | |
1454 | /* First find the desired set of subsystems */ | |
aae8aab4 | 1455 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1456 | ret = parse_cgroupfs_options(data, &opts); |
aae8aab4 | 1457 | mutex_unlock(&cgroup_mutex); |
c6d57f33 PM |
1458 | if (ret) |
1459 | goto out_err; | |
ddbcc7e8 | 1460 | |
c6d57f33 PM |
1461 | /* |
1462 | * Allocate a new cgroup root. We may not need it if we're | |
1463 | * reusing an existing hierarchy. | |
1464 | */ | |
1465 | new_root = cgroup_root_from_opts(&opts); | |
1466 | if (IS_ERR(new_root)) { | |
1467 | ret = PTR_ERR(new_root); | |
1d5be6b2 | 1468 | goto out_err; |
81a6a5cd | 1469 | } |
c6d57f33 | 1470 | opts.new_root = new_root; |
ddbcc7e8 | 1471 | |
c6d57f33 | 1472 | /* Locate an existing or new sb for this hierarchy */ |
9249e17f | 1473 | sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts); |
ddbcc7e8 | 1474 | if (IS_ERR(sb)) { |
c6d57f33 | 1475 | ret = PTR_ERR(sb); |
fa3ca07e | 1476 | cgroup_free_root(opts.new_root); |
1d5be6b2 | 1477 | goto out_err; |
ddbcc7e8 PM |
1478 | } |
1479 | ||
c6d57f33 PM |
1480 | root = sb->s_fs_info; |
1481 | BUG_ON(!root); | |
1482 | if (root == opts.new_root) { | |
1483 | /* We used the new root structure, so this is a new hierarchy */ | |
c12f65d4 | 1484 | struct cgroup *root_cgrp = &root->top_cgroup; |
c6d57f33 | 1485 | struct cgroupfs_root *existing_root; |
28fd5dfc | 1486 | int i; |
5abb8855 | 1487 | struct css_set *cset; |
ddbcc7e8 PM |
1488 | |
1489 | BUG_ON(sb->s_root != NULL); | |
1490 | ||
1491 | ret = cgroup_get_rootdir(sb); | |
1492 | if (ret) | |
1493 | goto drop_new_super; | |
817929ec | 1494 | inode = sb->s_root->d_inode; |
ddbcc7e8 | 1495 | |
817929ec | 1496 | mutex_lock(&inode->i_mutex); |
ddbcc7e8 PM |
1497 | mutex_lock(&cgroup_mutex); |
1498 | ||
eb46bf89 TH |
1499 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); |
1500 | if (ret < 0) | |
4e96ee8e | 1501 | goto unlock_drop; |
eb46bf89 | 1502 | root_cgrp->id = ret; |
4e96ee8e | 1503 | |
e25e2cbb TH |
1504 | /* Check for name clashes with existing mounts */ |
1505 | ret = -EBUSY; | |
1506 | if (strlen(root->name)) | |
1507 | for_each_active_root(existing_root) | |
1508 | if (!strcmp(existing_root->name, root->name)) | |
1509 | goto unlock_drop; | |
c6d57f33 | 1510 | |
817929ec PM |
1511 | /* |
1512 | * We're accessing css_set_count without locking | |
1513 | * css_set_lock here, but that's OK - it can only be | |
1514 | * increased by someone holding cgroup_lock, and | |
1515 | * that's us. The worst that can happen is that we | |
1516 | * have some link structures left over | |
1517 | */ | |
69d0206c | 1518 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); |
e25e2cbb TH |
1519 | if (ret) |
1520 | goto unlock_drop; | |
817929ec | 1521 | |
fc76df70 TH |
1522 | /* ID 0 is reserved for dummy root, 1 for unified hierarchy */ |
1523 | ret = cgroup_init_root_id(root, 2, 0); | |
fa3ca07e TH |
1524 | if (ret) |
1525 | goto unlock_drop; | |
1526 | ||
3126121f TH |
1527 | sb->s_root->d_fsdata = root_cgrp; |
1528 | root_cgrp->dentry = sb->s_root; | |
1529 | ||
1530 | /* | |
1531 | * We're inside get_sb() and will call lookup_one_len() to | |
1532 | * create the root files, which doesn't work if SELinux is | |
1533 | * in use. The following cred dancing somehow works around | |
1534 | * it. See 2ce9738ba ("cgroupfs: use init_cred when | |
1535 | * populating new cgroupfs mount") for more details. | |
1536 | */ | |
1537 | cred = override_creds(&init_cred); | |
1538 | ||
2bb566cb | 1539 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); |
3126121f TH |
1540 | if (ret) |
1541 | goto rm_base_files; | |
1542 | ||
a8a648c4 | 1543 | ret = rebind_subsystems(root, root->subsys_mask, 0); |
3126121f TH |
1544 | if (ret) |
1545 | goto rm_base_files; | |
1546 | ||
1547 | revert_creds(cred); | |
1548 | ||
cf5d5941 BB |
1549 | /* |
1550 | * There must be no failure case after here, since rebinding | |
1551 | * takes care of subsystems' refcounts, which are explicitly | |
1552 | * dropped in the failure exit path. | |
1553 | */ | |
ddbcc7e8 | 1554 | |
9871bf95 TH |
1555 | list_add(&root->root_list, &cgroup_roots); |
1556 | cgroup_root_count++; | |
ddbcc7e8 | 1557 | |
817929ec PM |
1558 | /* Link the top cgroup in this hierarchy into all |
1559 | * the css_set objects */ | |
1560 | write_lock(&css_set_lock); | |
5abb8855 | 1561 | hash_for_each(css_set_table, i, cset, hlist) |
69d0206c | 1562 | link_css_set(&tmp_links, cset, root_cgrp); |
817929ec PM |
1563 | write_unlock(&css_set_lock); |
1564 | ||
69d0206c | 1565 | free_cgrp_cset_links(&tmp_links); |
817929ec | 1566 | |
c12f65d4 | 1567 | BUG_ON(!list_empty(&root_cgrp->children)); |
ddbcc7e8 PM |
1568 | BUG_ON(root->number_of_cgroups != 1); |
1569 | ||
ddbcc7e8 | 1570 | mutex_unlock(&cgroup_mutex); |
34f77a90 | 1571 | mutex_unlock(&inode->i_mutex); |
c6d57f33 PM |
1572 | } else { |
1573 | /* | |
1574 | * We re-used an existing hierarchy - the new root (if | |
1575 | * any) is not needed | |
1576 | */ | |
fa3ca07e | 1577 | cgroup_free_root(opts.new_root); |
873fe09e | 1578 | |
c7ba8287 | 1579 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1580 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1581 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1582 | ret = -EINVAL; | |
1583 | goto drop_new_super; | |
1584 | } else { | |
1585 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1586 | } | |
873fe09e | 1587 | } |
ddbcc7e8 PM |
1588 | } |
1589 | ||
c6d57f33 PM |
1590 | kfree(opts.release_agent); |
1591 | kfree(opts.name); | |
f7e83571 | 1592 | return dget(sb->s_root); |
ddbcc7e8 | 1593 | |
3126121f TH |
1594 | rm_base_files: |
1595 | free_cgrp_cset_links(&tmp_links); | |
2bb566cb | 1596 | cgroup_addrm_files(&root->top_cgroup, cgroup_base_files, false); |
3126121f | 1597 | revert_creds(cred); |
e25e2cbb | 1598 | unlock_drop: |
fa3ca07e | 1599 | cgroup_exit_root_id(root); |
e25e2cbb TH |
1600 | mutex_unlock(&cgroup_mutex); |
1601 | mutex_unlock(&inode->i_mutex); | |
ddbcc7e8 | 1602 | drop_new_super: |
6f5bbff9 | 1603 | deactivate_locked_super(sb); |
c6d57f33 PM |
1604 | out_err: |
1605 | kfree(opts.release_agent); | |
1606 | kfree(opts.name); | |
f7e83571 | 1607 | return ERR_PTR(ret); |
ddbcc7e8 PM |
1608 | } |
1609 | ||
dd4b0a46 SP |
1610 | static void cgroup_kill_sb(struct super_block *sb) |
1611 | { | |
ddbcc7e8 | 1612 | struct cgroupfs_root *root = sb->s_fs_info; |
bd89aabc | 1613 | struct cgroup *cgrp = &root->top_cgroup; |
69d0206c | 1614 | struct cgrp_cset_link *link, *tmp_link; |
ddbcc7e8 PM |
1615 | int ret; |
1616 | ||
1617 | BUG_ON(!root); | |
1618 | ||
1619 | BUG_ON(root->number_of_cgroups != 1); | |
bd89aabc | 1620 | BUG_ON(!list_empty(&cgrp->children)); |
ddbcc7e8 | 1621 | |
3126121f | 1622 | mutex_lock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
1623 | mutex_lock(&cgroup_mutex); |
1624 | ||
1625 | /* Rebind all subsystems back to the default hierarchy */ | |
1672d040 TH |
1626 | if (root->flags & CGRP_ROOT_SUBSYS_BOUND) { |
1627 | ret = rebind_subsystems(root, 0, root->subsys_mask); | |
1628 | /* Shouldn't be able to fail ... */ | |
1629 | BUG_ON(ret); | |
1630 | } | |
ddbcc7e8 | 1631 | |
817929ec | 1632 | /* |
69d0206c | 1633 | * Release all the links from cset_links to this hierarchy's |
817929ec PM |
1634 | * root cgroup |
1635 | */ | |
1636 | write_lock(&css_set_lock); | |
71cbb949 | 1637 | |
69d0206c TH |
1638 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { |
1639 | list_del(&link->cset_link); | |
1640 | list_del(&link->cgrp_link); | |
817929ec PM |
1641 | kfree(link); |
1642 | } | |
1643 | write_unlock(&css_set_lock); | |
1644 | ||
839ec545 PM |
1645 | if (!list_empty(&root->root_list)) { |
1646 | list_del(&root->root_list); | |
9871bf95 | 1647 | cgroup_root_count--; |
839ec545 | 1648 | } |
e5f6a860 | 1649 | |
fa3ca07e TH |
1650 | cgroup_exit_root_id(root); |
1651 | ||
ddbcc7e8 | 1652 | mutex_unlock(&cgroup_mutex); |
3126121f | 1653 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 | 1654 | |
03b1cde6 AR |
1655 | simple_xattrs_free(&cgrp->xattrs); |
1656 | ||
ddbcc7e8 | 1657 | kill_litter_super(sb); |
fa3ca07e | 1658 | cgroup_free_root(root); |
ddbcc7e8 PM |
1659 | } |
1660 | ||
1661 | static struct file_system_type cgroup_fs_type = { | |
1662 | .name = "cgroup", | |
f7e83571 | 1663 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1664 | .kill_sb = cgroup_kill_sb, |
1665 | }; | |
1666 | ||
676db4af GK |
1667 | static struct kobject *cgroup_kobj; |
1668 | ||
a043e3b2 LZ |
1669 | /** |
1670 | * cgroup_path - generate the path of a cgroup | |
1671 | * @cgrp: the cgroup in question | |
1672 | * @buf: the buffer to write the path into | |
1673 | * @buflen: the length of the buffer | |
1674 | * | |
65dff759 LZ |
1675 | * Writes path of cgroup into buf. Returns 0 on success, -errno on error. |
1676 | * | |
1677 | * We can't generate cgroup path using dentry->d_name, as accessing | |
1678 | * dentry->name must be protected by irq-unsafe dentry->d_lock or parent | |
1679 | * inode's i_mutex, while on the other hand cgroup_path() can be called | |
1680 | * with some irq-safe spinlocks held. | |
ddbcc7e8 | 1681 | */ |
bd89aabc | 1682 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
ddbcc7e8 | 1683 | { |
65dff759 | 1684 | int ret = -ENAMETOOLONG; |
ddbcc7e8 | 1685 | char *start; |
febfcef6 | 1686 | |
da1f296f TH |
1687 | if (!cgrp->parent) { |
1688 | if (strlcpy(buf, "/", buflen) >= buflen) | |
1689 | return -ENAMETOOLONG; | |
ddbcc7e8 PM |
1690 | return 0; |
1691 | } | |
1692 | ||
316eb661 | 1693 | start = buf + buflen - 1; |
316eb661 | 1694 | *start = '\0'; |
9a9686b6 | 1695 | |
65dff759 | 1696 | rcu_read_lock(); |
da1f296f | 1697 | do { |
65dff759 LZ |
1698 | const char *name = cgroup_name(cgrp); |
1699 | int len; | |
1700 | ||
1701 | len = strlen(name); | |
ddbcc7e8 | 1702 | if ((start -= len) < buf) |
65dff759 LZ |
1703 | goto out; |
1704 | memcpy(start, name, len); | |
9a9686b6 | 1705 | |
ddbcc7e8 | 1706 | if (--start < buf) |
65dff759 | 1707 | goto out; |
ddbcc7e8 | 1708 | *start = '/'; |
65dff759 LZ |
1709 | |
1710 | cgrp = cgrp->parent; | |
da1f296f | 1711 | } while (cgrp->parent); |
65dff759 | 1712 | ret = 0; |
ddbcc7e8 | 1713 | memmove(buf, start, buf + buflen - start); |
65dff759 LZ |
1714 | out: |
1715 | rcu_read_unlock(); | |
1716 | return ret; | |
ddbcc7e8 | 1717 | } |
67523c48 | 1718 | EXPORT_SYMBOL_GPL(cgroup_path); |
ddbcc7e8 | 1719 | |
857a2beb | 1720 | /** |
913ffdb5 | 1721 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1722 | * @task: target task |
857a2beb TH |
1723 | * @buf: the buffer to write the path into |
1724 | * @buflen: the length of the buffer | |
1725 | * | |
913ffdb5 TH |
1726 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1727 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1728 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1729 | * cgroup controller callbacks. | |
1730 | * | |
1731 | * Returns 0 on success, fails with -%ENAMETOOLONG if @buflen is too short. | |
857a2beb | 1732 | */ |
913ffdb5 | 1733 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb TH |
1734 | { |
1735 | struct cgroupfs_root *root; | |
913ffdb5 TH |
1736 | struct cgroup *cgrp; |
1737 | int hierarchy_id = 1, ret = 0; | |
1738 | ||
1739 | if (buflen < 2) | |
1740 | return -ENAMETOOLONG; | |
857a2beb TH |
1741 | |
1742 | mutex_lock(&cgroup_mutex); | |
1743 | ||
913ffdb5 TH |
1744 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1745 | ||
857a2beb TH |
1746 | if (root) { |
1747 | cgrp = task_cgroup_from_root(task, root); | |
1748 | ret = cgroup_path(cgrp, buf, buflen); | |
913ffdb5 TH |
1749 | } else { |
1750 | /* if no hierarchy exists, everyone is in "/" */ | |
1751 | memcpy(buf, "/", 2); | |
857a2beb TH |
1752 | } |
1753 | ||
1754 | mutex_unlock(&cgroup_mutex); | |
857a2beb TH |
1755 | return ret; |
1756 | } | |
913ffdb5 | 1757 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1758 | |
2f7ee569 TH |
1759 | /* |
1760 | * Control Group taskset | |
1761 | */ | |
134d3373 TH |
1762 | struct task_and_cgroup { |
1763 | struct task_struct *task; | |
1764 | struct cgroup *cgrp; | |
6f4b7e63 | 1765 | struct css_set *cset; |
134d3373 TH |
1766 | }; |
1767 | ||
2f7ee569 TH |
1768 | struct cgroup_taskset { |
1769 | struct task_and_cgroup single; | |
1770 | struct flex_array *tc_array; | |
1771 | int tc_array_len; | |
1772 | int idx; | |
1773 | struct cgroup *cur_cgrp; | |
1774 | }; | |
1775 | ||
1776 | /** | |
1777 | * cgroup_taskset_first - reset taskset and return the first task | |
1778 | * @tset: taskset of interest | |
1779 | * | |
1780 | * @tset iteration is initialized and the first task is returned. | |
1781 | */ | |
1782 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1783 | { | |
1784 | if (tset->tc_array) { | |
1785 | tset->idx = 0; | |
1786 | return cgroup_taskset_next(tset); | |
1787 | } else { | |
1788 | tset->cur_cgrp = tset->single.cgrp; | |
1789 | return tset->single.task; | |
1790 | } | |
1791 | } | |
1792 | EXPORT_SYMBOL_GPL(cgroup_taskset_first); | |
1793 | ||
1794 | /** | |
1795 | * cgroup_taskset_next - iterate to the next task in taskset | |
1796 | * @tset: taskset of interest | |
1797 | * | |
1798 | * Return the next task in @tset. Iteration must have been initialized | |
1799 | * with cgroup_taskset_first(). | |
1800 | */ | |
1801 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1802 | { | |
1803 | struct task_and_cgroup *tc; | |
1804 | ||
1805 | if (!tset->tc_array || tset->idx >= tset->tc_array_len) | |
1806 | return NULL; | |
1807 | ||
1808 | tc = flex_array_get(tset->tc_array, tset->idx++); | |
1809 | tset->cur_cgrp = tc->cgrp; | |
1810 | return tc->task; | |
1811 | } | |
1812 | EXPORT_SYMBOL_GPL(cgroup_taskset_next); | |
1813 | ||
1814 | /** | |
d99c8727 | 1815 | * cgroup_taskset_cur_css - return the matching css for the current task |
2f7ee569 | 1816 | * @tset: taskset of interest |
d99c8727 | 1817 | * @subsys_id: the ID of the target subsystem |
2f7ee569 | 1818 | * |
d99c8727 TH |
1819 | * Return the css for the current (last returned) task of @tset for |
1820 | * subsystem specified by @subsys_id. This function must be preceded by | |
1821 | * either cgroup_taskset_first() or cgroup_taskset_next(). | |
2f7ee569 | 1822 | */ |
d99c8727 TH |
1823 | struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset, |
1824 | int subsys_id) | |
2f7ee569 | 1825 | { |
ca8bdcaf | 1826 | return cgroup_css(tset->cur_cgrp, cgroup_subsys[subsys_id]); |
2f7ee569 | 1827 | } |
d99c8727 | 1828 | EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css); |
2f7ee569 TH |
1829 | |
1830 | /** | |
1831 | * cgroup_taskset_size - return the number of tasks in taskset | |
1832 | * @tset: taskset of interest | |
1833 | */ | |
1834 | int cgroup_taskset_size(struct cgroup_taskset *tset) | |
1835 | { | |
1836 | return tset->tc_array ? tset->tc_array_len : 1; | |
1837 | } | |
1838 | EXPORT_SYMBOL_GPL(cgroup_taskset_size); | |
1839 | ||
1840 | ||
74a1166d BB |
1841 | /* |
1842 | * cgroup_task_migrate - move a task from one cgroup to another. | |
1843 | * | |
d0b2fdd2 | 1844 | * Must be called with cgroup_mutex and threadgroup locked. |
74a1166d | 1845 | */ |
5abb8855 TH |
1846 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1847 | struct task_struct *tsk, | |
1848 | struct css_set *new_cset) | |
74a1166d | 1849 | { |
5abb8855 | 1850 | struct css_set *old_cset; |
74a1166d BB |
1851 | |
1852 | /* | |
026085ef MSB |
1853 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1854 | * setting such that we can't race against cgroup_exit() changing the | |
1855 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1856 | */ |
c84cdf75 | 1857 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1858 | old_cset = task_css_set(tsk); |
74a1166d | 1859 | |
74a1166d | 1860 | task_lock(tsk); |
5abb8855 | 1861 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d BB |
1862 | task_unlock(tsk); |
1863 | ||
1864 | /* Update the css_set linked lists if we're using them */ | |
1865 | write_lock(&css_set_lock); | |
1866 | if (!list_empty(&tsk->cg_list)) | |
5abb8855 | 1867 | list_move(&tsk->cg_list, &new_cset->tasks); |
74a1166d BB |
1868 | write_unlock(&css_set_lock); |
1869 | ||
1870 | /* | |
5abb8855 TH |
1871 | * We just gained a reference on old_cset by taking it from the |
1872 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1873 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1874 | */ |
5abb8855 TH |
1875 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
1876 | put_css_set(old_cset); | |
74a1166d BB |
1877 | } |
1878 | ||
a043e3b2 | 1879 | /** |
081aa458 | 1880 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup |
74a1166d | 1881 | * @cgrp: the cgroup to attach to |
081aa458 LZ |
1882 | * @tsk: the task or the leader of the threadgroup to be attached |
1883 | * @threadgroup: attach the whole threadgroup? | |
74a1166d | 1884 | * |
257058ae | 1885 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
081aa458 | 1886 | * task_lock of @tsk or each thread in the threadgroup individually in turn. |
74a1166d | 1887 | */ |
47cfcd09 TH |
1888 | static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk, |
1889 | bool threadgroup) | |
74a1166d BB |
1890 | { |
1891 | int retval, i, group_size; | |
74a1166d | 1892 | struct cgroupfs_root *root = cgrp->root; |
1c6727af | 1893 | struct cgroup_subsys_state *css, *failed_css = NULL; |
74a1166d | 1894 | /* threadgroup list cursor and array */ |
081aa458 | 1895 | struct task_struct *leader = tsk; |
134d3373 | 1896 | struct task_and_cgroup *tc; |
d846687d | 1897 | struct flex_array *group; |
2f7ee569 | 1898 | struct cgroup_taskset tset = { }; |
74a1166d BB |
1899 | |
1900 | /* | |
1901 | * step 0: in order to do expensive, possibly blocking operations for | |
1902 | * every thread, we cannot iterate the thread group list, since it needs | |
1903 | * rcu or tasklist locked. instead, build an array of all threads in the | |
257058ae TH |
1904 | * group - group_rwsem prevents new threads from appearing, and if |
1905 | * threads exit, this will just be an over-estimate. | |
74a1166d | 1906 | */ |
081aa458 LZ |
1907 | if (threadgroup) |
1908 | group_size = get_nr_threads(tsk); | |
1909 | else | |
1910 | group_size = 1; | |
d846687d | 1911 | /* flex_array supports very large thread-groups better than kmalloc. */ |
134d3373 | 1912 | group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL); |
74a1166d BB |
1913 | if (!group) |
1914 | return -ENOMEM; | |
d846687d | 1915 | /* pre-allocate to guarantee space while iterating in rcu read-side. */ |
3ac1707a | 1916 | retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL); |
d846687d BB |
1917 | if (retval) |
1918 | goto out_free_group_list; | |
74a1166d | 1919 | |
74a1166d | 1920 | i = 0; |
fb5d2b4c MSB |
1921 | /* |
1922 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1923 | * already PF_EXITING could be freed from underneath us unless we | |
1924 | * take an rcu_read_lock. | |
1925 | */ | |
1926 | rcu_read_lock(); | |
74a1166d | 1927 | do { |
134d3373 TH |
1928 | struct task_and_cgroup ent; |
1929 | ||
cd3d0952 TH |
1930 | /* @tsk either already exited or can't exit until the end */ |
1931 | if (tsk->flags & PF_EXITING) | |
ea84753c | 1932 | goto next; |
cd3d0952 | 1933 | |
74a1166d BB |
1934 | /* as per above, nr_threads may decrease, but not increase. */ |
1935 | BUG_ON(i >= group_size); | |
134d3373 TH |
1936 | ent.task = tsk; |
1937 | ent.cgrp = task_cgroup_from_root(tsk, root); | |
892a2b90 MSB |
1938 | /* nothing to do if this task is already in the cgroup */ |
1939 | if (ent.cgrp == cgrp) | |
ea84753c | 1940 | goto next; |
61d1d219 MSB |
1941 | /* |
1942 | * saying GFP_ATOMIC has no effect here because we did prealloc | |
1943 | * earlier, but it's good form to communicate our expectations. | |
1944 | */ | |
134d3373 | 1945 | retval = flex_array_put(group, i, &ent, GFP_ATOMIC); |
d846687d | 1946 | BUG_ON(retval != 0); |
74a1166d | 1947 | i++; |
ea84753c | 1948 | next: |
081aa458 LZ |
1949 | if (!threadgroup) |
1950 | break; | |
74a1166d | 1951 | } while_each_thread(leader, tsk); |
fb5d2b4c | 1952 | rcu_read_unlock(); |
74a1166d BB |
1953 | /* remember the number of threads in the array for later. */ |
1954 | group_size = i; | |
2f7ee569 TH |
1955 | tset.tc_array = group; |
1956 | tset.tc_array_len = group_size; | |
74a1166d | 1957 | |
134d3373 TH |
1958 | /* methods shouldn't be called if no task is actually migrating */ |
1959 | retval = 0; | |
892a2b90 | 1960 | if (!group_size) |
b07ef774 | 1961 | goto out_free_group_list; |
134d3373 | 1962 | |
74a1166d BB |
1963 | /* |
1964 | * step 1: check that we can legitimately attach to the cgroup. | |
1965 | */ | |
1c6727af TH |
1966 | for_each_css(css, i, cgrp) { |
1967 | if (css->ss->can_attach) { | |
1968 | retval = css->ss->can_attach(css, &tset); | |
74a1166d | 1969 | if (retval) { |
1c6727af | 1970 | failed_css = css; |
74a1166d BB |
1971 | goto out_cancel_attach; |
1972 | } | |
1973 | } | |
74a1166d BB |
1974 | } |
1975 | ||
1976 | /* | |
1977 | * step 2: make sure css_sets exist for all threads to be migrated. | |
1978 | * we use find_css_set, which allocates a new one if necessary. | |
1979 | */ | |
74a1166d | 1980 | for (i = 0; i < group_size; i++) { |
a8ad805c TH |
1981 | struct css_set *old_cset; |
1982 | ||
134d3373 | 1983 | tc = flex_array_get(group, i); |
a8ad805c | 1984 | old_cset = task_css_set(tc->task); |
6f4b7e63 LZ |
1985 | tc->cset = find_css_set(old_cset, cgrp); |
1986 | if (!tc->cset) { | |
61d1d219 MSB |
1987 | retval = -ENOMEM; |
1988 | goto out_put_css_set_refs; | |
74a1166d BB |
1989 | } |
1990 | } | |
1991 | ||
1992 | /* | |
494c167c TH |
1993 | * step 3: now that we're guaranteed success wrt the css_sets, |
1994 | * proceed to move all tasks to the new cgroup. There are no | |
1995 | * failure cases after here, so this is the commit point. | |
74a1166d | 1996 | */ |
74a1166d | 1997 | for (i = 0; i < group_size; i++) { |
134d3373 | 1998 | tc = flex_array_get(group, i); |
6f4b7e63 | 1999 | cgroup_task_migrate(tc->cgrp, tc->task, tc->cset); |
74a1166d BB |
2000 | } |
2001 | /* nothing is sensitive to fork() after this point. */ | |
2002 | ||
2003 | /* | |
494c167c | 2004 | * step 4: do subsystem attach callbacks. |
74a1166d | 2005 | */ |
1c6727af TH |
2006 | for_each_css(css, i, cgrp) |
2007 | if (css->ss->attach) | |
2008 | css->ss->attach(css, &tset); | |
74a1166d BB |
2009 | |
2010 | /* | |
2011 | * step 5: success! and cleanup | |
2012 | */ | |
74a1166d | 2013 | retval = 0; |
61d1d219 MSB |
2014 | out_put_css_set_refs: |
2015 | if (retval) { | |
2016 | for (i = 0; i < group_size; i++) { | |
2017 | tc = flex_array_get(group, i); | |
6f4b7e63 | 2018 | if (!tc->cset) |
61d1d219 | 2019 | break; |
6f4b7e63 | 2020 | put_css_set(tc->cset); |
61d1d219 | 2021 | } |
74a1166d BB |
2022 | } |
2023 | out_cancel_attach: | |
74a1166d | 2024 | if (retval) { |
1c6727af TH |
2025 | for_each_css(css, i, cgrp) { |
2026 | if (css == failed_css) | |
74a1166d | 2027 | break; |
1c6727af TH |
2028 | if (css->ss->cancel_attach) |
2029 | css->ss->cancel_attach(css, &tset); | |
74a1166d BB |
2030 | } |
2031 | } | |
74a1166d | 2032 | out_free_group_list: |
d846687d | 2033 | flex_array_free(group); |
74a1166d BB |
2034 | return retval; |
2035 | } | |
2036 | ||
2037 | /* | |
2038 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
2039 | * function to attach either it or all tasks in its threadgroup. Will lock |
2040 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 2041 | */ |
74a1166d | 2042 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2043 | { |
bbcb81d0 | 2044 | struct task_struct *tsk; |
c69e8d9c | 2045 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2046 | int ret; |
2047 | ||
74a1166d BB |
2048 | if (!cgroup_lock_live_group(cgrp)) |
2049 | return -ENODEV; | |
2050 | ||
b78949eb MSB |
2051 | retry_find_task: |
2052 | rcu_read_lock(); | |
bbcb81d0 | 2053 | if (pid) { |
73507f33 | 2054 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2055 | if (!tsk) { |
2056 | rcu_read_unlock(); | |
dd4b0a46 | 2057 | ret = -ESRCH; |
b78949eb | 2058 | goto out_unlock_cgroup; |
bbcb81d0 | 2059 | } |
74a1166d BB |
2060 | /* |
2061 | * even if we're attaching all tasks in the thread group, we | |
2062 | * only need to check permissions on one of them. | |
2063 | */ | |
c69e8d9c | 2064 | tcred = __task_cred(tsk); |
14a590c3 EB |
2065 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2066 | !uid_eq(cred->euid, tcred->uid) && | |
2067 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2068 | rcu_read_unlock(); |
b78949eb MSB |
2069 | ret = -EACCES; |
2070 | goto out_unlock_cgroup; | |
bbcb81d0 | 2071 | } |
b78949eb MSB |
2072 | } else |
2073 | tsk = current; | |
cd3d0952 TH |
2074 | |
2075 | if (threadgroup) | |
b78949eb | 2076 | tsk = tsk->group_leader; |
c4c27fbd MG |
2077 | |
2078 | /* | |
14a40ffc | 2079 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2080 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2081 | * with no rt_runtime allocated. Just say no. | |
2082 | */ | |
14a40ffc | 2083 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2084 | ret = -EINVAL; |
2085 | rcu_read_unlock(); | |
2086 | goto out_unlock_cgroup; | |
2087 | } | |
2088 | ||
b78949eb MSB |
2089 | get_task_struct(tsk); |
2090 | rcu_read_unlock(); | |
2091 | ||
2092 | threadgroup_lock(tsk); | |
2093 | if (threadgroup) { | |
2094 | if (!thread_group_leader(tsk)) { | |
2095 | /* | |
2096 | * a race with de_thread from another thread's exec() | |
2097 | * may strip us of our leadership, if this happens, | |
2098 | * there is no choice but to throw this task away and | |
2099 | * try again; this is | |
2100 | * "double-double-toil-and-trouble-check locking". | |
2101 | */ | |
2102 | threadgroup_unlock(tsk); | |
2103 | put_task_struct(tsk); | |
2104 | goto retry_find_task; | |
2105 | } | |
081aa458 LZ |
2106 | } |
2107 | ||
2108 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2109 | ||
cd3d0952 TH |
2110 | threadgroup_unlock(tsk); |
2111 | ||
bbcb81d0 | 2112 | put_task_struct(tsk); |
b78949eb | 2113 | out_unlock_cgroup: |
47cfcd09 | 2114 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
2115 | return ret; |
2116 | } | |
2117 | ||
7ae1bad9 TH |
2118 | /** |
2119 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2120 | * @from: attach to all cgroups of a given task | |
2121 | * @tsk: the task to be attached | |
2122 | */ | |
2123 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2124 | { | |
2125 | struct cgroupfs_root *root; | |
2126 | int retval = 0; | |
2127 | ||
47cfcd09 | 2128 | mutex_lock(&cgroup_mutex); |
7ae1bad9 | 2129 | for_each_active_root(root) { |
6f4b7e63 | 2130 | struct cgroup *from_cgrp = task_cgroup_from_root(from, root); |
7ae1bad9 | 2131 | |
6f4b7e63 | 2132 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2133 | if (retval) |
2134 | break; | |
2135 | } | |
47cfcd09 | 2136 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2137 | |
2138 | return retval; | |
2139 | } | |
2140 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2141 | ||
182446d0 TH |
2142 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2143 | struct cftype *cft, u64 pid) | |
74a1166d | 2144 | { |
182446d0 | 2145 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2146 | } |
2147 | ||
182446d0 TH |
2148 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2149 | struct cftype *cft, u64 tgid) | |
af351026 | 2150 | { |
182446d0 | 2151 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2152 | } |
2153 | ||
182446d0 TH |
2154 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
2155 | struct cftype *cft, const char *buffer) | |
e788e066 | 2156 | { |
182446d0 | 2157 | BUILD_BUG_ON(sizeof(css->cgroup->root->release_agent_path) < PATH_MAX); |
f4a2589f EK |
2158 | if (strlen(buffer) >= PATH_MAX) |
2159 | return -EINVAL; | |
182446d0 | 2160 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2161 | return -ENODEV; |
69e943b7 | 2162 | spin_lock(&release_agent_path_lock); |
182446d0 | 2163 | strcpy(css->cgroup->root->release_agent_path, buffer); |
69e943b7 | 2164 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2165 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2166 | return 0; |
2167 | } | |
2168 | ||
2da8ca82 | 2169 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2170 | { |
2da8ca82 | 2171 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2172 | |
e788e066 PM |
2173 | if (!cgroup_lock_live_group(cgrp)) |
2174 | return -ENODEV; | |
2175 | seq_puts(seq, cgrp->root->release_agent_path); | |
2176 | seq_putc(seq, '\n'); | |
47cfcd09 | 2177 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2178 | return 0; |
2179 | } | |
2180 | ||
2da8ca82 | 2181 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2182 | { |
2da8ca82 TH |
2183 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2184 | ||
2185 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2186 | return 0; |
2187 | } | |
2188 | ||
84eea842 PM |
2189 | /* A buffer size big enough for numbers or short strings */ |
2190 | #define CGROUP_LOCAL_BUFFER_SIZE 64 | |
2191 | ||
a742c59d TH |
2192 | static ssize_t cgroup_file_write(struct file *file, const char __user *userbuf, |
2193 | size_t nbytes, loff_t *ppos) | |
355e0c48 | 2194 | { |
a742c59d TH |
2195 | struct cfent *cfe = __d_cfe(file->f_dentry); |
2196 | struct cftype *cft = __d_cft(file->f_dentry); | |
2197 | struct cgroup_subsys_state *css = cfe->css; | |
2198 | size_t max_bytes = cft->max_write_len ?: CGROUP_LOCAL_BUFFER_SIZE - 1; | |
2199 | char *buf; | |
2200 | int ret; | |
355e0c48 | 2201 | |
db3b1497 | 2202 | if (nbytes >= max_bytes) |
355e0c48 | 2203 | return -E2BIG; |
355e0c48 | 2204 | |
a742c59d TH |
2205 | buf = kmalloc(nbytes + 1, GFP_KERNEL); |
2206 | if (!buf) | |
2207 | return -ENOMEM; | |
db3b1497 | 2208 | |
a742c59d TH |
2209 | if (copy_from_user(buf, userbuf, nbytes)) { |
2210 | ret = -EFAULT; | |
2211 | goto out_free; | |
2212 | } | |
ddbcc7e8 | 2213 | |
a742c59d TH |
2214 | buf[nbytes] = '\0'; |
2215 | ||
2216 | if (cft->write_string) { | |
2217 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2218 | } else if (cft->write_u64) { | |
2219 | unsigned long long v; | |
2220 | ret = kstrtoull(buf, 0, &v); | |
2221 | if (!ret) | |
2222 | ret = cft->write_u64(css, cft, v); | |
2223 | } else if (cft->write_s64) { | |
2224 | long long v; | |
2225 | ret = kstrtoll(buf, 0, &v); | |
2226 | if (!ret) | |
2227 | ret = cft->write_s64(css, cft, v); | |
2228 | } else if (cft->trigger) { | |
2229 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2230 | } else { |
a742c59d | 2231 | ret = -EINVAL; |
e73d2c61 | 2232 | } |
a742c59d TH |
2233 | out_free: |
2234 | kfree(buf); | |
2235 | return ret ?: nbytes; | |
355e0c48 PM |
2236 | } |
2237 | ||
91796569 PM |
2238 | /* |
2239 | * seqfile ops/methods for returning structured data. Currently just | |
2240 | * supports string->u64 maps, but can be extended in future. | |
2241 | */ | |
2242 | ||
6612f05b | 2243 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2244 | { |
6612f05b | 2245 | struct cftype *cft = seq_cft(seq); |
db3b1497 | 2246 | |
6612f05b TH |
2247 | if (cft->seq_start) { |
2248 | return cft->seq_start(seq, ppos); | |
2249 | } else { | |
2250 | /* | |
2251 | * The same behavior and code as single_open(). Returns | |
2252 | * !NULL if pos is at the beginning; otherwise, NULL. | |
2253 | */ | |
2254 | return NULL + !*ppos; | |
5a3eb9f6 | 2255 | } |
db3b1497 PM |
2256 | } |
2257 | ||
6612f05b | 2258 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2259 | { |
6612f05b | 2260 | struct cftype *cft = seq_cft(seq); |
ddbcc7e8 | 2261 | |
6612f05b TH |
2262 | if (cft->seq_next) { |
2263 | return cft->seq_next(seq, v, ppos); | |
2264 | } else { | |
2265 | /* | |
2266 | * The same behavior and code as single_open(), always | |
2267 | * terminate after the initial read. | |
2268 | */ | |
2269 | ++*ppos; | |
2270 | return NULL; | |
d447ea2f | 2271 | } |
ddbcc7e8 PM |
2272 | } |
2273 | ||
6612f05b | 2274 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2275 | { |
6612f05b | 2276 | struct cftype *cft = seq_cft(seq); |
ddbcc7e8 | 2277 | |
6612f05b TH |
2278 | if (cft->seq_stop) |
2279 | cft->seq_stop(seq, v); | |
ddbcc7e8 PM |
2280 | } |
2281 | ||
91796569 | 2282 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2283 | { |
7da11279 TH |
2284 | struct cftype *cft = seq_cft(m); |
2285 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2286 | |
2da8ca82 TH |
2287 | if (cft->seq_show) |
2288 | return cft->seq_show(m, arg); | |
e73d2c61 | 2289 | |
f4c753b7 | 2290 | if (cft->read_u64) |
896f5199 TH |
2291 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2292 | else if (cft->read_s64) | |
2293 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2294 | else | |
2295 | return -EINVAL; | |
2296 | return 0; | |
91796569 PM |
2297 | } |
2298 | ||
6612f05b TH |
2299 | static struct seq_operations cgroup_seq_operations = { |
2300 | .start = cgroup_seqfile_start, | |
2301 | .next = cgroup_seqfile_next, | |
2302 | .stop = cgroup_seqfile_stop, | |
2303 | .show = cgroup_seqfile_show, | |
91796569 PM |
2304 | }; |
2305 | ||
ddbcc7e8 PM |
2306 | static int cgroup_file_open(struct inode *inode, struct file *file) |
2307 | { | |
f7d58818 TH |
2308 | struct cfent *cfe = __d_cfe(file->f_dentry); |
2309 | struct cftype *cft = __d_cft(file->f_dentry); | |
105347ba TH |
2310 | struct cgroup *cgrp = __d_cgrp(cfe->dentry->d_parent); |
2311 | struct cgroup_subsys_state *css; | |
6612f05b | 2312 | struct cgroup_open_file *of; |
ddbcc7e8 | 2313 | int err; |
ddbcc7e8 PM |
2314 | |
2315 | err = generic_file_open(inode, file); | |
2316 | if (err) | |
2317 | return err; | |
75139b82 | 2318 | |
f7d58818 TH |
2319 | /* |
2320 | * If the file belongs to a subsystem, pin the css. Will be | |
2321 | * unpinned either on open failure or release. This ensures that | |
2322 | * @css stays alive for all file operations. | |
2323 | */ | |
105347ba | 2324 | rcu_read_lock(); |
ca8bdcaf TH |
2325 | css = cgroup_css(cgrp, cft->ss); |
2326 | if (cft->ss && !css_tryget(css)) | |
2327 | css = NULL; | |
105347ba | 2328 | rcu_read_unlock(); |
f4f4be2b | 2329 | |
0bfb4aa6 | 2330 | if (!css) |
f7d58818 | 2331 | return -ENODEV; |
75139b82 | 2332 | |
0bfb4aa6 TH |
2333 | /* |
2334 | * @cfe->css is used by read/write/close to determine the | |
2335 | * associated css. @file->private_data would be a better place but | |
2336 | * that's already used by seqfile. Multiple accessors may use it | |
2337 | * simultaneously which is okay as the association never changes. | |
2338 | */ | |
2339 | WARN_ON_ONCE(cfe->css && cfe->css != css); | |
2340 | cfe->css = css; | |
f4f4be2b | 2341 | |
6612f05b TH |
2342 | of = __seq_open_private(file, &cgroup_seq_operations, |
2343 | sizeof(struct cgroup_open_file)); | |
2344 | if (of) { | |
2345 | of->cfe = cfe; | |
2346 | return 0; | |
e0798ce2 | 2347 | } |
ddbcc7e8 | 2348 | |
6612f05b | 2349 | if (css->ss) |
f7d58818 | 2350 | css_put(css); |
6612f05b | 2351 | return -ENOMEM; |
ddbcc7e8 PM |
2352 | } |
2353 | ||
2354 | static int cgroup_file_release(struct inode *inode, struct file *file) | |
2355 | { | |
f7d58818 | 2356 | struct cfent *cfe = __d_cfe(file->f_dentry); |
105347ba | 2357 | struct cgroup_subsys_state *css = cfe->css; |
f7d58818 | 2358 | |
67f4c36f | 2359 | if (css->ss) |
f7d58818 | 2360 | css_put(css); |
6612f05b | 2361 | return seq_release_private(inode, file); |
ddbcc7e8 PM |
2362 | } |
2363 | ||
2364 | /* | |
2365 | * cgroup_rename - Only allow simple rename of directories in place. | |
2366 | */ | |
2367 | static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry, | |
2368 | struct inode *new_dir, struct dentry *new_dentry) | |
2369 | { | |
65dff759 LZ |
2370 | int ret; |
2371 | struct cgroup_name *name, *old_name; | |
2372 | struct cgroup *cgrp; | |
2373 | ||
2374 | /* | |
2375 | * It's convinient to use parent dir's i_mutex to protected | |
2376 | * cgrp->name. | |
2377 | */ | |
2378 | lockdep_assert_held(&old_dir->i_mutex); | |
2379 | ||
ddbcc7e8 PM |
2380 | if (!S_ISDIR(old_dentry->d_inode->i_mode)) |
2381 | return -ENOTDIR; | |
2382 | if (new_dentry->d_inode) | |
2383 | return -EEXIST; | |
2384 | if (old_dir != new_dir) | |
2385 | return -EIO; | |
65dff759 LZ |
2386 | |
2387 | cgrp = __d_cgrp(old_dentry); | |
2388 | ||
6db8e85c TH |
2389 | /* |
2390 | * This isn't a proper migration and its usefulness is very | |
2391 | * limited. Disallow if sane_behavior. | |
2392 | */ | |
2393 | if (cgroup_sane_behavior(cgrp)) | |
2394 | return -EPERM; | |
2395 | ||
65dff759 LZ |
2396 | name = cgroup_alloc_name(new_dentry); |
2397 | if (!name) | |
2398 | return -ENOMEM; | |
2399 | ||
2400 | ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry); | |
2401 | if (ret) { | |
2402 | kfree(name); | |
2403 | return ret; | |
2404 | } | |
2405 | ||
a4ea1cc9 | 2406 | old_name = rcu_dereference_protected(cgrp->name, true); |
65dff759 LZ |
2407 | rcu_assign_pointer(cgrp->name, name); |
2408 | ||
2409 | kfree_rcu(old_name, rcu_head); | |
2410 | return 0; | |
ddbcc7e8 PM |
2411 | } |
2412 | ||
03b1cde6 AR |
2413 | static struct simple_xattrs *__d_xattrs(struct dentry *dentry) |
2414 | { | |
2415 | if (S_ISDIR(dentry->d_inode->i_mode)) | |
2416 | return &__d_cgrp(dentry)->xattrs; | |
2417 | else | |
712317ad | 2418 | return &__d_cfe(dentry)->xattrs; |
03b1cde6 AR |
2419 | } |
2420 | ||
2421 | static inline int xattr_enabled(struct dentry *dentry) | |
2422 | { | |
2423 | struct cgroupfs_root *root = dentry->d_sb->s_fs_info; | |
93438629 | 2424 | return root->flags & CGRP_ROOT_XATTR; |
03b1cde6 AR |
2425 | } |
2426 | ||
2427 | static bool is_valid_xattr(const char *name) | |
2428 | { | |
2429 | if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) || | |
2430 | !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN)) | |
2431 | return true; | |
2432 | return false; | |
2433 | } | |
2434 | ||
2435 | static int cgroup_setxattr(struct dentry *dentry, const char *name, | |
2436 | const void *val, size_t size, int flags) | |
2437 | { | |
2438 | if (!xattr_enabled(dentry)) | |
2439 | return -EOPNOTSUPP; | |
2440 | if (!is_valid_xattr(name)) | |
2441 | return -EINVAL; | |
2442 | return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags); | |
2443 | } | |
2444 | ||
2445 | static int cgroup_removexattr(struct dentry *dentry, const char *name) | |
2446 | { | |
2447 | if (!xattr_enabled(dentry)) | |
2448 | return -EOPNOTSUPP; | |
2449 | if (!is_valid_xattr(name)) | |
2450 | return -EINVAL; | |
2451 | return simple_xattr_remove(__d_xattrs(dentry), name); | |
2452 | } | |
2453 | ||
2454 | static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name, | |
2455 | void *buf, size_t size) | |
2456 | { | |
2457 | if (!xattr_enabled(dentry)) | |
2458 | return -EOPNOTSUPP; | |
2459 | if (!is_valid_xattr(name)) | |
2460 | return -EINVAL; | |
2461 | return simple_xattr_get(__d_xattrs(dentry), name, buf, size); | |
2462 | } | |
2463 | ||
2464 | static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size) | |
2465 | { | |
2466 | if (!xattr_enabled(dentry)) | |
2467 | return -EOPNOTSUPP; | |
2468 | return simple_xattr_list(__d_xattrs(dentry), buf, size); | |
2469 | } | |
2470 | ||
828c0950 | 2471 | static const struct file_operations cgroup_file_operations = { |
896f5199 | 2472 | .read = seq_read, |
ddbcc7e8 PM |
2473 | .write = cgroup_file_write, |
2474 | .llseek = generic_file_llseek, | |
2475 | .open = cgroup_file_open, | |
2476 | .release = cgroup_file_release, | |
2477 | }; | |
2478 | ||
03b1cde6 AR |
2479 | static const struct inode_operations cgroup_file_inode_operations = { |
2480 | .setxattr = cgroup_setxattr, | |
2481 | .getxattr = cgroup_getxattr, | |
2482 | .listxattr = cgroup_listxattr, | |
2483 | .removexattr = cgroup_removexattr, | |
2484 | }; | |
2485 | ||
6e1d5dcc | 2486 | static const struct inode_operations cgroup_dir_inode_operations = { |
786e1448 | 2487 | .lookup = simple_lookup, |
ddbcc7e8 PM |
2488 | .mkdir = cgroup_mkdir, |
2489 | .rmdir = cgroup_rmdir, | |
2490 | .rename = cgroup_rename, | |
03b1cde6 AR |
2491 | .setxattr = cgroup_setxattr, |
2492 | .getxattr = cgroup_getxattr, | |
2493 | .listxattr = cgroup_listxattr, | |
2494 | .removexattr = cgroup_removexattr, | |
ddbcc7e8 PM |
2495 | }; |
2496 | ||
a5e7ed32 | 2497 | static int cgroup_create_file(struct dentry *dentry, umode_t mode, |
5adcee1d NP |
2498 | struct super_block *sb) |
2499 | { | |
ddbcc7e8 PM |
2500 | struct inode *inode; |
2501 | ||
2502 | if (!dentry) | |
2503 | return -ENOENT; | |
2504 | if (dentry->d_inode) | |
2505 | return -EEXIST; | |
2506 | ||
2507 | inode = cgroup_new_inode(mode, sb); | |
2508 | if (!inode) | |
2509 | return -ENOMEM; | |
2510 | ||
2511 | if (S_ISDIR(mode)) { | |
2512 | inode->i_op = &cgroup_dir_inode_operations; | |
2513 | inode->i_fop = &simple_dir_operations; | |
2514 | ||
2515 | /* start off with i_nlink == 2 (for "." entry) */ | |
2516 | inc_nlink(inode); | |
28fd6f30 | 2517 | inc_nlink(dentry->d_parent->d_inode); |
ddbcc7e8 | 2518 | |
b8a2df6a TH |
2519 | /* |
2520 | * Control reaches here with cgroup_mutex held. | |
2521 | * @inode->i_mutex should nest outside cgroup_mutex but we | |
2522 | * want to populate it immediately without releasing | |
2523 | * cgroup_mutex. As @inode isn't visible to anyone else | |
2524 | * yet, trylock will always succeed without affecting | |
2525 | * lockdep checks. | |
2526 | */ | |
2527 | WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex)); | |
ddbcc7e8 PM |
2528 | } else if (S_ISREG(mode)) { |
2529 | inode->i_size = 0; | |
2530 | inode->i_fop = &cgroup_file_operations; | |
03b1cde6 | 2531 | inode->i_op = &cgroup_file_inode_operations; |
ddbcc7e8 | 2532 | } |
ddbcc7e8 PM |
2533 | d_instantiate(dentry, inode); |
2534 | dget(dentry); /* Extra count - pin the dentry in core */ | |
2535 | return 0; | |
2536 | } | |
2537 | ||
099fca32 LZ |
2538 | /** |
2539 | * cgroup_file_mode - deduce file mode of a control file | |
2540 | * @cft: the control file in question | |
2541 | * | |
2542 | * returns cft->mode if ->mode is not 0 | |
2543 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
2544 | * returns S_IRUGO if it has only a read handler | |
2545 | * returns S_IWUSR if it has only a write hander | |
2546 | */ | |
a5e7ed32 | 2547 | static umode_t cgroup_file_mode(const struct cftype *cft) |
099fca32 | 2548 | { |
a5e7ed32 | 2549 | umode_t mode = 0; |
099fca32 LZ |
2550 | |
2551 | if (cft->mode) | |
2552 | return cft->mode; | |
2553 | ||
2da8ca82 | 2554 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
099fca32 LZ |
2555 | mode |= S_IRUGO; |
2556 | ||
6e0755b0 TH |
2557 | if (cft->write_u64 || cft->write_s64 || cft->write_string || |
2558 | cft->trigger) | |
099fca32 LZ |
2559 | mode |= S_IWUSR; |
2560 | ||
2561 | return mode; | |
2562 | } | |
2563 | ||
2bb566cb | 2564 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2565 | { |
bd89aabc | 2566 | struct dentry *dir = cgrp->dentry; |
05ef1d7c | 2567 | struct cgroup *parent = __d_cgrp(dir); |
ddbcc7e8 | 2568 | struct dentry *dentry; |
05ef1d7c | 2569 | struct cfent *cfe; |
ddbcc7e8 | 2570 | int error; |
a5e7ed32 | 2571 | umode_t mode; |
ddbcc7e8 | 2572 | char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; |
8e3f6541 | 2573 | |
9fa4db33 TH |
2574 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
2575 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) { | |
2bb566cb | 2576 | strcpy(name, cft->ss->name); |
ddbcc7e8 PM |
2577 | strcat(name, "."); |
2578 | } | |
2579 | strcat(name, cft->name); | |
05ef1d7c | 2580 | |
ddbcc7e8 | 2581 | BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); |
05ef1d7c TH |
2582 | |
2583 | cfe = kzalloc(sizeof(*cfe), GFP_KERNEL); | |
2584 | if (!cfe) | |
2585 | return -ENOMEM; | |
2586 | ||
ddbcc7e8 | 2587 | dentry = lookup_one_len(name, dir, strlen(name)); |
05ef1d7c | 2588 | if (IS_ERR(dentry)) { |
ddbcc7e8 | 2589 | error = PTR_ERR(dentry); |
05ef1d7c TH |
2590 | goto out; |
2591 | } | |
2592 | ||
d6cbf35d LZ |
2593 | cfe->type = (void *)cft; |
2594 | cfe->dentry = dentry; | |
2595 | dentry->d_fsdata = cfe; | |
2596 | simple_xattrs_init(&cfe->xattrs); | |
2597 | ||
05ef1d7c TH |
2598 | mode = cgroup_file_mode(cft); |
2599 | error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb); | |
2600 | if (!error) { | |
05ef1d7c TH |
2601 | list_add_tail(&cfe->node, &parent->files); |
2602 | cfe = NULL; | |
2603 | } | |
2604 | dput(dentry); | |
2605 | out: | |
2606 | kfree(cfe); | |
ddbcc7e8 PM |
2607 | return error; |
2608 | } | |
2609 | ||
b1f28d31 TH |
2610 | /** |
2611 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2612 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2613 | * @cfts: array of cftypes to be added |
2614 | * @is_add: whether to add or remove | |
2615 | * | |
2616 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2617 | * For removals, this function never fails. If addition fails, this |
2618 | * function doesn't remove files already added. The caller is responsible | |
2619 | * for cleaning up. | |
b1f28d31 | 2620 | */ |
2bb566cb TH |
2621 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2622 | bool is_add) | |
ddbcc7e8 | 2623 | { |
03b1cde6 | 2624 | struct cftype *cft; |
b1f28d31 TH |
2625 | int ret; |
2626 | ||
2627 | lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex); | |
2628 | lockdep_assert_held(&cgroup_mutex); | |
db0416b6 TH |
2629 | |
2630 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2631 | /* does cft->flags tell us to skip this file on @cgrp? */ |
873fe09e TH |
2632 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2633 | continue; | |
f33fddc2 G |
2634 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2635 | continue; | |
2636 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2637 | continue; | |
2638 | ||
2739d3cc | 2639 | if (is_add) { |
2bb566cb | 2640 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2641 | if (ret) { |
2739d3cc | 2642 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2643 | cft->name, ret); |
2644 | return ret; | |
2645 | } | |
2739d3cc LZ |
2646 | } else { |
2647 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2648 | } |
ddbcc7e8 | 2649 | } |
b1f28d31 | 2650 | return 0; |
ddbcc7e8 PM |
2651 | } |
2652 | ||
8e3f6541 | 2653 | static void cgroup_cfts_prepare(void) |
e8c82d20 | 2654 | __acquires(&cgroup_mutex) |
8e3f6541 TH |
2655 | { |
2656 | /* | |
2657 | * Thanks to the entanglement with vfs inode locking, we can't walk | |
2658 | * the existing cgroups under cgroup_mutex and create files. | |
492eb21b TH |
2659 | * Instead, we use css_for_each_descendant_pre() and drop RCU read |
2660 | * lock before calling cgroup_addrm_files(). | |
8e3f6541 | 2661 | */ |
8e3f6541 TH |
2662 | mutex_lock(&cgroup_mutex); |
2663 | } | |
2664 | ||
2bb566cb | 2665 | static int cgroup_cfts_commit(struct cftype *cfts, bool is_add) |
e8c82d20 | 2666 | __releases(&cgroup_mutex) |
8e3f6541 TH |
2667 | { |
2668 | LIST_HEAD(pending); | |
2bb566cb | 2669 | struct cgroup_subsys *ss = cfts[0].ss; |
492eb21b | 2670 | struct cgroup *root = &ss->root->top_cgroup; |
084457f2 | 2671 | struct super_block *sb = ss->root->sb; |
e8c82d20 LZ |
2672 | struct dentry *prev = NULL; |
2673 | struct inode *inode; | |
492eb21b | 2674 | struct cgroup_subsys_state *css; |
00356bd5 | 2675 | u64 update_before; |
9ccece80 | 2676 | int ret = 0; |
8e3f6541 TH |
2677 | |
2678 | /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */ | |
9871bf95 | 2679 | if (!cfts || ss->root == &cgroup_dummy_root || |
e8c82d20 LZ |
2680 | !atomic_inc_not_zero(&sb->s_active)) { |
2681 | mutex_unlock(&cgroup_mutex); | |
9ccece80 | 2682 | return 0; |
8e3f6541 TH |
2683 | } |
2684 | ||
8e3f6541 | 2685 | /* |
e8c82d20 LZ |
2686 | * All cgroups which are created after we drop cgroup_mutex will |
2687 | * have the updated set of files, so we only need to update the | |
00356bd5 | 2688 | * cgroups created before the current @cgroup_serial_nr_next. |
8e3f6541 | 2689 | */ |
00356bd5 | 2690 | update_before = cgroup_serial_nr_next; |
e8c82d20 | 2691 | |
e8c82d20 | 2692 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2693 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2694 | struct cgroup *cgrp = css->cgroup; |
2695 | ||
e8c82d20 LZ |
2696 | if (cgroup_is_dead(cgrp)) |
2697 | continue; | |
2698 | ||
2699 | inode = cgrp->dentry->d_inode; | |
2700 | dget(cgrp->dentry); | |
e8c82d20 LZ |
2701 | dput(prev); |
2702 | prev = cgrp->dentry; | |
8e3f6541 | 2703 | |
48573a89 | 2704 | mutex_unlock(&cgroup_mutex); |
8e3f6541 TH |
2705 | mutex_lock(&inode->i_mutex); |
2706 | mutex_lock(&cgroup_mutex); | |
00356bd5 | 2707 | if (cgrp->serial_nr < update_before && !cgroup_is_dead(cgrp)) |
2bb566cb | 2708 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
8e3f6541 | 2709 | mutex_unlock(&inode->i_mutex); |
9ccece80 TH |
2710 | if (ret) |
2711 | break; | |
8e3f6541 | 2712 | } |
48573a89 | 2713 | mutex_unlock(&cgroup_mutex); |
e8c82d20 LZ |
2714 | dput(prev); |
2715 | deactivate_super(sb); | |
9ccece80 | 2716 | return ret; |
8e3f6541 TH |
2717 | } |
2718 | ||
2719 | /** | |
2720 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2721 | * @ss: target cgroup subsystem | |
2722 | * @cfts: zero-length name terminated array of cftypes | |
2723 | * | |
2724 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2725 | * existing cgroups to which @ss is attached and all future cgroups will | |
2726 | * have them too. This function can be called anytime whether @ss is | |
2727 | * attached or not. | |
2728 | * | |
2729 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2730 | * function currently returns 0 as long as @cfts registration is successful | |
2731 | * even if some file creation attempts on existing cgroups fail. | |
2732 | */ | |
03b1cde6 | 2733 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 TH |
2734 | { |
2735 | struct cftype_set *set; | |
2bb566cb | 2736 | struct cftype *cft; |
9ccece80 | 2737 | int ret; |
8e3f6541 TH |
2738 | |
2739 | set = kzalloc(sizeof(*set), GFP_KERNEL); | |
2740 | if (!set) | |
2741 | return -ENOMEM; | |
2742 | ||
2bb566cb TH |
2743 | for (cft = cfts; cft->name[0] != '\0'; cft++) |
2744 | cft->ss = ss; | |
2745 | ||
8e3f6541 TH |
2746 | cgroup_cfts_prepare(); |
2747 | set->cfts = cfts; | |
2748 | list_add_tail(&set->node, &ss->cftsets); | |
2bb566cb | 2749 | ret = cgroup_cfts_commit(cfts, true); |
9ccece80 | 2750 | if (ret) |
2bb566cb | 2751 | cgroup_rm_cftypes(cfts); |
9ccece80 | 2752 | return ret; |
8e3f6541 TH |
2753 | } |
2754 | EXPORT_SYMBOL_GPL(cgroup_add_cftypes); | |
2755 | ||
79578621 TH |
2756 | /** |
2757 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
2758 | * @cfts: zero-length name terminated array of cftypes |
2759 | * | |
2bb566cb TH |
2760 | * Unregister @cfts. Files described by @cfts are removed from all |
2761 | * existing cgroups and all future cgroups won't have them either. This | |
2762 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
2763 | * |
2764 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 2765 | * registered. |
79578621 | 2766 | */ |
2bb566cb | 2767 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 TH |
2768 | { |
2769 | struct cftype_set *set; | |
2770 | ||
2bb566cb TH |
2771 | if (!cfts || !cfts[0].ss) |
2772 | return -ENOENT; | |
2773 | ||
79578621 TH |
2774 | cgroup_cfts_prepare(); |
2775 | ||
2bb566cb | 2776 | list_for_each_entry(set, &cfts[0].ss->cftsets, node) { |
79578621 | 2777 | if (set->cfts == cfts) { |
f57947d2 LZ |
2778 | list_del(&set->node); |
2779 | kfree(set); | |
2bb566cb | 2780 | cgroup_cfts_commit(cfts, false); |
79578621 TH |
2781 | return 0; |
2782 | } | |
2783 | } | |
2784 | ||
2bb566cb | 2785 | cgroup_cfts_commit(NULL, false); |
79578621 TH |
2786 | return -ENOENT; |
2787 | } | |
2788 | ||
a043e3b2 LZ |
2789 | /** |
2790 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2791 | * @cgrp: the cgroup in question | |
2792 | * | |
2793 | * Return the number of tasks in the cgroup. | |
2794 | */ | |
bd89aabc | 2795 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2796 | { |
2797 | int count = 0; | |
69d0206c | 2798 | struct cgrp_cset_link *link; |
817929ec PM |
2799 | |
2800 | read_lock(&css_set_lock); | |
69d0206c TH |
2801 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2802 | count += atomic_read(&link->cset->refcount); | |
817929ec | 2803 | read_unlock(&css_set_lock); |
bbcb81d0 PM |
2804 | return count; |
2805 | } | |
2806 | ||
817929ec | 2807 | /* |
0942eeee TH |
2808 | * To reduce the fork() overhead for systems that are not actually using |
2809 | * their cgroups capability, we don't maintain the lists running through | |
2810 | * each css_set to its tasks until we see the list actually used - in other | |
72ec7029 | 2811 | * words after the first call to css_task_iter_start(). |
31a7df01 | 2812 | */ |
3df91fe3 | 2813 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2814 | { |
2815 | struct task_struct *p, *g; | |
2816 | write_lock(&css_set_lock); | |
2817 | use_task_css_set_links = 1; | |
3ce3230a FW |
2818 | /* |
2819 | * We need tasklist_lock because RCU is not safe against | |
2820 | * while_each_thread(). Besides, a forking task that has passed | |
2821 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
2822 | * is not guaranteed to have its child immediately visible in the | |
2823 | * tasklist if we walk through it with RCU. | |
2824 | */ | |
2825 | read_lock(&tasklist_lock); | |
31a7df01 CW |
2826 | do_each_thread(g, p) { |
2827 | task_lock(p); | |
0e04388f LZ |
2828 | /* |
2829 | * We should check if the process is exiting, otherwise | |
2830 | * it will race with cgroup_exit() in that the list | |
2831 | * entry won't be deleted though the process has exited. | |
2832 | */ | |
2833 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
a8ad805c | 2834 | list_add(&p->cg_list, &task_css_set(p)->tasks); |
31a7df01 CW |
2835 | task_unlock(p); |
2836 | } while_each_thread(g, p); | |
3ce3230a | 2837 | read_unlock(&tasklist_lock); |
31a7df01 CW |
2838 | write_unlock(&css_set_lock); |
2839 | } | |
2840 | ||
53fa5261 | 2841 | /** |
492eb21b TH |
2842 | * css_next_child - find the next child of a given css |
2843 | * @pos_css: the current position (%NULL to initiate traversal) | |
2844 | * @parent_css: css whose children to walk | |
53fa5261 | 2845 | * |
492eb21b | 2846 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
2847 | * under either cgroup_mutex or RCU read lock. The only requirement is |
2848 | * that @parent_css and @pos_css are accessible. The next sibling is | |
2849 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 2850 | */ |
492eb21b TH |
2851 | struct cgroup_subsys_state * |
2852 | css_next_child(struct cgroup_subsys_state *pos_css, | |
2853 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 2854 | { |
492eb21b TH |
2855 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
2856 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
2857 | struct cgroup *next; |
2858 | ||
87fb54f1 | 2859 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
2860 | |
2861 | /* | |
2862 | * @pos could already have been removed. Once a cgroup is removed, | |
2863 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
2864 | * changes. As CGRP_DEAD assertion is serialized and happens |
2865 | * before the cgroup is taken off the ->sibling list, if we see it | |
2866 | * unasserted, it's guaranteed that the next sibling hasn't | |
2867 | * finished its grace period even if it's already removed, and thus | |
2868 | * safe to dereference from this RCU critical section. If | |
2869 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
2870 | * to be visible as %true here. | |
3b287a50 TH |
2871 | * |
2872 | * If @pos is dead, its next pointer can't be dereferenced; | |
2873 | * however, as each cgroup is given a monotonically increasing | |
2874 | * unique serial number and always appended to the sibling list, | |
2875 | * the next one can be found by walking the parent's children until | |
2876 | * we see a cgroup with higher serial number than @pos's. While | |
2877 | * this path can be slower, it's taken only when either the current | |
2878 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 2879 | */ |
3b287a50 TH |
2880 | if (!pos) { |
2881 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
2882 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 2883 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
2884 | } else { |
2885 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
2886 | if (next->serial_nr > pos->serial_nr) | |
2887 | break; | |
53fa5261 TH |
2888 | } |
2889 | ||
492eb21b TH |
2890 | if (&next->sibling == &cgrp->children) |
2891 | return NULL; | |
2892 | ||
ca8bdcaf | 2893 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 2894 | } |
492eb21b | 2895 | EXPORT_SYMBOL_GPL(css_next_child); |
53fa5261 | 2896 | |
574bd9f7 | 2897 | /** |
492eb21b | 2898 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 2899 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2900 | * @root: css whose descendants to walk |
574bd9f7 | 2901 | * |
492eb21b | 2902 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
2903 | * to visit for pre-order traversal of @root's descendants. @root is |
2904 | * included in the iteration and the first node to be visited. | |
75501a6d | 2905 | * |
87fb54f1 TH |
2906 | * While this function requires cgroup_mutex or RCU read locking, it |
2907 | * doesn't require the whole traversal to be contained in a single critical | |
2908 | * section. This function will return the correct next descendant as long | |
2909 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 2910 | */ |
492eb21b TH |
2911 | struct cgroup_subsys_state * |
2912 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
2913 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2914 | { |
492eb21b | 2915 | struct cgroup_subsys_state *next; |
574bd9f7 | 2916 | |
87fb54f1 | 2917 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 2918 | |
bd8815a6 | 2919 | /* if first iteration, visit @root */ |
7805d000 | 2920 | if (!pos) |
bd8815a6 | 2921 | return root; |
574bd9f7 TH |
2922 | |
2923 | /* visit the first child if exists */ | |
492eb21b | 2924 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
2925 | if (next) |
2926 | return next; | |
2927 | ||
2928 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
2929 | while (pos != root) { |
2930 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 2931 | if (next) |
574bd9f7 | 2932 | return next; |
492eb21b | 2933 | pos = css_parent(pos); |
7805d000 | 2934 | } |
574bd9f7 TH |
2935 | |
2936 | return NULL; | |
2937 | } | |
492eb21b | 2938 | EXPORT_SYMBOL_GPL(css_next_descendant_pre); |
574bd9f7 | 2939 | |
12a9d2fe | 2940 | /** |
492eb21b TH |
2941 | * css_rightmost_descendant - return the rightmost descendant of a css |
2942 | * @pos: css of interest | |
12a9d2fe | 2943 | * |
492eb21b TH |
2944 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
2945 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 2946 | * subtree of @pos. |
75501a6d | 2947 | * |
87fb54f1 TH |
2948 | * While this function requires cgroup_mutex or RCU read locking, it |
2949 | * doesn't require the whole traversal to be contained in a single critical | |
2950 | * section. This function will return the correct rightmost descendant as | |
2951 | * long as @pos is accessible. | |
12a9d2fe | 2952 | */ |
492eb21b TH |
2953 | struct cgroup_subsys_state * |
2954 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 2955 | { |
492eb21b | 2956 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 2957 | |
87fb54f1 | 2958 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
2959 | |
2960 | do { | |
2961 | last = pos; | |
2962 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
2963 | pos = NULL; | |
492eb21b | 2964 | css_for_each_child(tmp, last) |
12a9d2fe TH |
2965 | pos = tmp; |
2966 | } while (pos); | |
2967 | ||
2968 | return last; | |
2969 | } | |
492eb21b | 2970 | EXPORT_SYMBOL_GPL(css_rightmost_descendant); |
12a9d2fe | 2971 | |
492eb21b TH |
2972 | static struct cgroup_subsys_state * |
2973 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 2974 | { |
492eb21b | 2975 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
2976 | |
2977 | do { | |
2978 | last = pos; | |
492eb21b | 2979 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
2980 | } while (pos); |
2981 | ||
2982 | return last; | |
2983 | } | |
2984 | ||
2985 | /** | |
492eb21b | 2986 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 2987 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2988 | * @root: css whose descendants to walk |
574bd9f7 | 2989 | * |
492eb21b | 2990 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
2991 | * to visit for post-order traversal of @root's descendants. @root is |
2992 | * included in the iteration and the last node to be visited. | |
75501a6d | 2993 | * |
87fb54f1 TH |
2994 | * While this function requires cgroup_mutex or RCU read locking, it |
2995 | * doesn't require the whole traversal to be contained in a single critical | |
2996 | * section. This function will return the correct next descendant as long | |
2997 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
2998 | * @cgroup. | |
574bd9f7 | 2999 | */ |
492eb21b TH |
3000 | struct cgroup_subsys_state * |
3001 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3002 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3003 | { |
492eb21b | 3004 | struct cgroup_subsys_state *next; |
574bd9f7 | 3005 | |
87fb54f1 | 3006 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3007 | |
58b79a91 TH |
3008 | /* if first iteration, visit leftmost descendant which may be @root */ |
3009 | if (!pos) | |
3010 | return css_leftmost_descendant(root); | |
574bd9f7 | 3011 | |
bd8815a6 TH |
3012 | /* if we visited @root, we're done */ |
3013 | if (pos == root) | |
3014 | return NULL; | |
3015 | ||
574bd9f7 | 3016 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 3017 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 3018 | if (next) |
492eb21b | 3019 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3020 | |
3021 | /* no sibling left, visit parent */ | |
bd8815a6 | 3022 | return css_parent(pos); |
574bd9f7 | 3023 | } |
492eb21b | 3024 | EXPORT_SYMBOL_GPL(css_next_descendant_post); |
574bd9f7 | 3025 | |
0942eeee | 3026 | /** |
72ec7029 | 3027 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
3028 | * @it: the iterator to advance |
3029 | * | |
3030 | * Advance @it to the next css_set to walk. | |
d515876e | 3031 | */ |
72ec7029 | 3032 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
3033 | { |
3034 | struct list_head *l = it->cset_link; | |
3035 | struct cgrp_cset_link *link; | |
3036 | struct css_set *cset; | |
3037 | ||
3038 | /* Advance to the next non-empty css_set */ | |
3039 | do { | |
3040 | l = l->next; | |
72ec7029 | 3041 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
3042 | it->cset_link = NULL; |
3043 | return; | |
3044 | } | |
3045 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3046 | cset = link->cset; | |
3047 | } while (list_empty(&cset->tasks)); | |
3048 | it->cset_link = l; | |
3049 | it->task = cset->tasks.next; | |
3050 | } | |
3051 | ||
0942eeee | 3052 | /** |
72ec7029 TH |
3053 | * css_task_iter_start - initiate task iteration |
3054 | * @css: the css to walk tasks of | |
0942eeee TH |
3055 | * @it: the task iterator to use |
3056 | * | |
72ec7029 TH |
3057 | * Initiate iteration through the tasks of @css. The caller can call |
3058 | * css_task_iter_next() to walk through the tasks until the function | |
3059 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3060 | * called. | |
0942eeee TH |
3061 | * |
3062 | * Note that this function acquires a lock which is released when the | |
3063 | * iteration finishes. The caller can't sleep while iteration is in | |
3064 | * progress. | |
3065 | */ | |
72ec7029 TH |
3066 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3067 | struct css_task_iter *it) | |
c6ca5750 | 3068 | __acquires(css_set_lock) |
817929ec PM |
3069 | { |
3070 | /* | |
72ec7029 TH |
3071 | * The first time anyone tries to iterate across a css, we need to |
3072 | * enable the list linking each css_set to its tasks, and fix up | |
3073 | * all existing tasks. | |
817929ec | 3074 | */ |
31a7df01 CW |
3075 | if (!use_task_css_set_links) |
3076 | cgroup_enable_task_cg_lists(); | |
3077 | ||
817929ec | 3078 | read_lock(&css_set_lock); |
c59cd3d8 | 3079 | |
72ec7029 TH |
3080 | it->origin_css = css; |
3081 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 3082 | |
72ec7029 | 3083 | css_advance_task_iter(it); |
817929ec PM |
3084 | } |
3085 | ||
0942eeee | 3086 | /** |
72ec7029 | 3087 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3088 | * @it: the task iterator being iterated |
3089 | * | |
3090 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3091 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3092 | * reaches the end. | |
0942eeee | 3093 | */ |
72ec7029 | 3094 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
3095 | { |
3096 | struct task_struct *res; | |
3097 | struct list_head *l = it->task; | |
69d0206c | 3098 | struct cgrp_cset_link *link; |
817929ec PM |
3099 | |
3100 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 3101 | if (!it->cset_link) |
817929ec PM |
3102 | return NULL; |
3103 | res = list_entry(l, struct task_struct, cg_list); | |
3104 | /* Advance iterator to find next entry */ | |
3105 | l = l->next; | |
69d0206c TH |
3106 | link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link); |
3107 | if (l == &link->cset->tasks) { | |
0942eeee TH |
3108 | /* |
3109 | * We reached the end of this task list - move on to the | |
3110 | * next cgrp_cset_link. | |
3111 | */ | |
72ec7029 | 3112 | css_advance_task_iter(it); |
817929ec PM |
3113 | } else { |
3114 | it->task = l; | |
3115 | } | |
3116 | return res; | |
3117 | } | |
3118 | ||
0942eeee | 3119 | /** |
72ec7029 | 3120 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3121 | * @it: the task iterator to finish |
3122 | * | |
72ec7029 | 3123 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3124 | */ |
72ec7029 | 3125 | void css_task_iter_end(struct css_task_iter *it) |
c6ca5750 | 3126 | __releases(css_set_lock) |
817929ec PM |
3127 | { |
3128 | read_unlock(&css_set_lock); | |
3129 | } | |
3130 | ||
31a7df01 CW |
3131 | static inline int started_after_time(struct task_struct *t1, |
3132 | struct timespec *time, | |
3133 | struct task_struct *t2) | |
3134 | { | |
3135 | int start_diff = timespec_compare(&t1->start_time, time); | |
3136 | if (start_diff > 0) { | |
3137 | return 1; | |
3138 | } else if (start_diff < 0) { | |
3139 | return 0; | |
3140 | } else { | |
3141 | /* | |
3142 | * Arbitrarily, if two processes started at the same | |
3143 | * time, we'll say that the lower pointer value | |
3144 | * started first. Note that t2 may have exited by now | |
3145 | * so this may not be a valid pointer any longer, but | |
3146 | * that's fine - it still serves to distinguish | |
3147 | * between two tasks started (effectively) simultaneously. | |
3148 | */ | |
3149 | return t1 > t2; | |
3150 | } | |
3151 | } | |
3152 | ||
3153 | /* | |
3154 | * This function is a callback from heap_insert() and is used to order | |
3155 | * the heap. | |
3156 | * In this case we order the heap in descending task start time. | |
3157 | */ | |
3158 | static inline int started_after(void *p1, void *p2) | |
3159 | { | |
3160 | struct task_struct *t1 = p1; | |
3161 | struct task_struct *t2 = p2; | |
3162 | return started_after_time(t1, &t2->start_time, t2); | |
3163 | } | |
3164 | ||
3165 | /** | |
72ec7029 TH |
3166 | * css_scan_tasks - iterate though all the tasks in a css |
3167 | * @css: the css to iterate tasks of | |
e535837b TH |
3168 | * @test: optional test callback |
3169 | * @process: process callback | |
3170 | * @data: data passed to @test and @process | |
3171 | * @heap: optional pre-allocated heap used for task iteration | |
31a7df01 | 3172 | * |
72ec7029 TH |
3173 | * Iterate through all the tasks in @css, calling @test for each, and if it |
3174 | * returns %true, call @process for it also. | |
31a7df01 | 3175 | * |
e535837b | 3176 | * @test may be NULL, meaning always true (select all tasks), which |
72ec7029 | 3177 | * effectively duplicates css_task_iter_{start,next,end}() but does not |
e535837b TH |
3178 | * lock css_set_lock for the call to @process. |
3179 | * | |
3180 | * It is guaranteed that @process will act on every task that is a member | |
72ec7029 TH |
3181 | * of @css for the duration of this call. This function may or may not |
3182 | * call @process for tasks that exit or move to a different css during the | |
3183 | * call, or are forked or move into the css during the call. | |
31a7df01 | 3184 | * |
e535837b TH |
3185 | * Note that @test may be called with locks held, and may in some |
3186 | * situations be called multiple times for the same task, so it should be | |
3187 | * cheap. | |
31a7df01 | 3188 | * |
e535837b TH |
3189 | * If @heap is non-NULL, a heap has been pre-allocated and will be used for |
3190 | * heap operations (and its "gt" member will be overwritten), else a | |
3191 | * temporary heap will be used (allocation of which may cause this function | |
3192 | * to fail). | |
31a7df01 | 3193 | */ |
72ec7029 TH |
3194 | int css_scan_tasks(struct cgroup_subsys_state *css, |
3195 | bool (*test)(struct task_struct *, void *), | |
3196 | void (*process)(struct task_struct *, void *), | |
3197 | void *data, struct ptr_heap *heap) | |
31a7df01 CW |
3198 | { |
3199 | int retval, i; | |
72ec7029 | 3200 | struct css_task_iter it; |
31a7df01 CW |
3201 | struct task_struct *p, *dropped; |
3202 | /* Never dereference latest_task, since it's not refcounted */ | |
3203 | struct task_struct *latest_task = NULL; | |
3204 | struct ptr_heap tmp_heap; | |
31a7df01 CW |
3205 | struct timespec latest_time = { 0, 0 }; |
3206 | ||
e535837b | 3207 | if (heap) { |
31a7df01 | 3208 | /* The caller supplied our heap and pre-allocated its memory */ |
31a7df01 CW |
3209 | heap->gt = &started_after; |
3210 | } else { | |
3211 | /* We need to allocate our own heap memory */ | |
3212 | heap = &tmp_heap; | |
3213 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
3214 | if (retval) | |
3215 | /* cannot allocate the heap */ | |
3216 | return retval; | |
3217 | } | |
3218 | ||
3219 | again: | |
3220 | /* | |
72ec7029 | 3221 | * Scan tasks in the css, using the @test callback to determine |
e535837b TH |
3222 | * which are of interest, and invoking @process callback on the |
3223 | * ones which need an update. Since we don't want to hold any | |
3224 | * locks during the task updates, gather tasks to be processed in a | |
3225 | * heap structure. The heap is sorted by descending task start | |
3226 | * time. If the statically-sized heap fills up, we overflow tasks | |
3227 | * that started later, and in future iterations only consider tasks | |
3228 | * that started after the latest task in the previous pass. This | |
31a7df01 CW |
3229 | * guarantees forward progress and that we don't miss any tasks. |
3230 | */ | |
3231 | heap->size = 0; | |
72ec7029 TH |
3232 | css_task_iter_start(css, &it); |
3233 | while ((p = css_task_iter_next(&it))) { | |
31a7df01 CW |
3234 | /* |
3235 | * Only affect tasks that qualify per the caller's callback, | |
3236 | * if he provided one | |
3237 | */ | |
e535837b | 3238 | if (test && !test(p, data)) |
31a7df01 CW |
3239 | continue; |
3240 | /* | |
3241 | * Only process tasks that started after the last task | |
3242 | * we processed | |
3243 | */ | |
3244 | if (!started_after_time(p, &latest_time, latest_task)) | |
3245 | continue; | |
3246 | dropped = heap_insert(heap, p); | |
3247 | if (dropped == NULL) { | |
3248 | /* | |
3249 | * The new task was inserted; the heap wasn't | |
3250 | * previously full | |
3251 | */ | |
3252 | get_task_struct(p); | |
3253 | } else if (dropped != p) { | |
3254 | /* | |
3255 | * The new task was inserted, and pushed out a | |
3256 | * different task | |
3257 | */ | |
3258 | get_task_struct(p); | |
3259 | put_task_struct(dropped); | |
3260 | } | |
3261 | /* | |
3262 | * Else the new task was newer than anything already in | |
3263 | * the heap and wasn't inserted | |
3264 | */ | |
3265 | } | |
72ec7029 | 3266 | css_task_iter_end(&it); |
31a7df01 CW |
3267 | |
3268 | if (heap->size) { | |
3269 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 3270 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 3271 | if (i == 0) { |
4fe91d51 PJ |
3272 | latest_time = q->start_time; |
3273 | latest_task = q; | |
31a7df01 CW |
3274 | } |
3275 | /* Process the task per the caller's callback */ | |
e535837b | 3276 | process(q, data); |
4fe91d51 | 3277 | put_task_struct(q); |
31a7df01 CW |
3278 | } |
3279 | /* | |
3280 | * If we had to process any tasks at all, scan again | |
3281 | * in case some of them were in the middle of forking | |
3282 | * children that didn't get processed. | |
3283 | * Not the most efficient way to do it, but it avoids | |
3284 | * having to take callback_mutex in the fork path | |
3285 | */ | |
3286 | goto again; | |
3287 | } | |
3288 | if (heap == &tmp_heap) | |
3289 | heap_free(&tmp_heap); | |
3290 | return 0; | |
3291 | } | |
3292 | ||
e535837b | 3293 | static void cgroup_transfer_one_task(struct task_struct *task, void *data) |
8cc99345 | 3294 | { |
e535837b | 3295 | struct cgroup *new_cgroup = data; |
8cc99345 | 3296 | |
47cfcd09 | 3297 | mutex_lock(&cgroup_mutex); |
8cc99345 | 3298 | cgroup_attach_task(new_cgroup, task, false); |
47cfcd09 | 3299 | mutex_unlock(&cgroup_mutex); |
8cc99345 TH |
3300 | } |
3301 | ||
3302 | /** | |
3303 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
3304 | * @to: cgroup to which the tasks will be moved | |
3305 | * @from: cgroup in which the tasks currently reside | |
3306 | */ | |
3307 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
3308 | { | |
72ec7029 TH |
3309 | return css_scan_tasks(&from->dummy_css, NULL, cgroup_transfer_one_task, |
3310 | to, NULL); | |
8cc99345 TH |
3311 | } |
3312 | ||
bbcb81d0 | 3313 | /* |
102a775e | 3314 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3315 | * |
3316 | * Reading this file can return large amounts of data if a cgroup has | |
3317 | * *lots* of attached tasks. So it may need several calls to read(), | |
3318 | * but we cannot guarantee that the information we produce is correct | |
3319 | * unless we produce it entirely atomically. | |
3320 | * | |
bbcb81d0 | 3321 | */ |
bbcb81d0 | 3322 | |
24528255 LZ |
3323 | /* which pidlist file are we talking about? */ |
3324 | enum cgroup_filetype { | |
3325 | CGROUP_FILE_PROCS, | |
3326 | CGROUP_FILE_TASKS, | |
3327 | }; | |
3328 | ||
3329 | /* | |
3330 | * A pidlist is a list of pids that virtually represents the contents of one | |
3331 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3332 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3333 | * to the cgroup. | |
3334 | */ | |
3335 | struct cgroup_pidlist { | |
3336 | /* | |
3337 | * used to find which pidlist is wanted. doesn't change as long as | |
3338 | * this particular list stays in the list. | |
3339 | */ | |
3340 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3341 | /* array of xids */ | |
3342 | pid_t *list; | |
3343 | /* how many elements the above list has */ | |
3344 | int length; | |
24528255 LZ |
3345 | /* each of these stored in a list by its cgroup */ |
3346 | struct list_head links; | |
3347 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3348 | struct cgroup *owner; | |
b1a21367 TH |
3349 | /* for delayed destruction */ |
3350 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3351 | }; |
3352 | ||
d1d9fd33 BB |
3353 | /* |
3354 | * The following two functions "fix" the issue where there are more pids | |
3355 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3356 | * TODO: replace with a kernel-wide solution to this problem | |
3357 | */ | |
3358 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3359 | static void *pidlist_allocate(int count) | |
3360 | { | |
3361 | if (PIDLIST_TOO_LARGE(count)) | |
3362 | return vmalloc(count * sizeof(pid_t)); | |
3363 | else | |
3364 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3365 | } | |
b1a21367 | 3366 | |
d1d9fd33 BB |
3367 | static void pidlist_free(void *p) |
3368 | { | |
3369 | if (is_vmalloc_addr(p)) | |
3370 | vfree(p); | |
3371 | else | |
3372 | kfree(p); | |
3373 | } | |
d1d9fd33 | 3374 | |
b1a21367 TH |
3375 | /* |
3376 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3377 | * should be left afterwards. | |
3378 | */ | |
3379 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3380 | { | |
3381 | struct cgroup_pidlist *l, *tmp_l; | |
3382 | ||
3383 | mutex_lock(&cgrp->pidlist_mutex); | |
3384 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3385 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3386 | mutex_unlock(&cgrp->pidlist_mutex); | |
3387 | ||
3388 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3389 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3390 | } | |
3391 | ||
3392 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3393 | { | |
3394 | struct delayed_work *dwork = to_delayed_work(work); | |
3395 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3396 | destroy_dwork); | |
3397 | struct cgroup_pidlist *tofree = NULL; | |
3398 | ||
3399 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3400 | |
3401 | /* | |
04502365 TH |
3402 | * Destroy iff we didn't get queued again. The state won't change |
3403 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3404 | */ |
04502365 | 3405 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3406 | list_del(&l->links); |
3407 | pidlist_free(l->list); | |
3408 | put_pid_ns(l->key.ns); | |
3409 | tofree = l; | |
3410 | } | |
3411 | ||
b1a21367 TH |
3412 | mutex_unlock(&l->owner->pidlist_mutex); |
3413 | kfree(tofree); | |
3414 | } | |
3415 | ||
bbcb81d0 | 3416 | /* |
102a775e | 3417 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3418 | * Returns the number of unique elements. |
bbcb81d0 | 3419 | */ |
6ee211ad | 3420 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3421 | { |
102a775e | 3422 | int src, dest = 1; |
102a775e BB |
3423 | |
3424 | /* | |
3425 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3426 | * edge cases first; no work needs to be done for either | |
3427 | */ | |
3428 | if (length == 0 || length == 1) | |
3429 | return length; | |
3430 | /* src and dest walk down the list; dest counts unique elements */ | |
3431 | for (src = 1; src < length; src++) { | |
3432 | /* find next unique element */ | |
3433 | while (list[src] == list[src-1]) { | |
3434 | src++; | |
3435 | if (src == length) | |
3436 | goto after; | |
3437 | } | |
3438 | /* dest always points to where the next unique element goes */ | |
3439 | list[dest] = list[src]; | |
3440 | dest++; | |
3441 | } | |
3442 | after: | |
102a775e BB |
3443 | return dest; |
3444 | } | |
3445 | ||
afb2bc14 TH |
3446 | /* |
3447 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3448 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3449 | * different per namespace, each namespace needs differently sorted list, | |
3450 | * making it impossible to use, for example, single rbtree of member tasks | |
3451 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3452 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3453 | * pidlists keyed by cgroup and namespace. | |
3454 | * | |
3455 | * All this extra complexity was caused by the original implementation | |
3456 | * committing to an entirely unnecessary property. In the long term, we | |
3457 | * want to do away with it. Explicitly scramble sort order if | |
3458 | * sane_behavior so that no such expectation exists in the new interface. | |
3459 | * | |
3460 | * Scrambling is done by swapping every two consecutive bits, which is | |
3461 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3462 | */ | |
3463 | static pid_t pid_fry(pid_t pid) | |
3464 | { | |
3465 | unsigned a = pid & 0x55555555; | |
3466 | unsigned b = pid & 0xAAAAAAAA; | |
3467 | ||
3468 | return (a << 1) | (b >> 1); | |
3469 | } | |
3470 | ||
3471 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3472 | { | |
3473 | if (cgroup_sane_behavior(cgrp)) | |
3474 | return pid_fry(pid); | |
3475 | else | |
3476 | return pid; | |
3477 | } | |
3478 | ||
102a775e BB |
3479 | static int cmppid(const void *a, const void *b) |
3480 | { | |
3481 | return *(pid_t *)a - *(pid_t *)b; | |
3482 | } | |
3483 | ||
afb2bc14 TH |
3484 | static int fried_cmppid(const void *a, const void *b) |
3485 | { | |
3486 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3487 | } | |
3488 | ||
e6b81710 TH |
3489 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3490 | enum cgroup_filetype type) | |
3491 | { | |
3492 | struct cgroup_pidlist *l; | |
3493 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3494 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3495 | ||
3496 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3497 | ||
3498 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3499 | if (l->key.type == type && l->key.ns == ns) | |
3500 | return l; | |
3501 | return NULL; | |
3502 | } | |
3503 | ||
72a8cb30 BB |
3504 | /* |
3505 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3506 | * returns with the lock on that pidlist already held, and takes care | |
3507 | * of the use count, or returns NULL with no locks held if we're out of | |
3508 | * memory. | |
3509 | */ | |
e6b81710 TH |
3510 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3511 | enum cgroup_filetype type) | |
72a8cb30 BB |
3512 | { |
3513 | struct cgroup_pidlist *l; | |
b70cc5fd | 3514 | |
e6b81710 TH |
3515 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3516 | ||
3517 | l = cgroup_pidlist_find(cgrp, type); | |
3518 | if (l) | |
3519 | return l; | |
3520 | ||
72a8cb30 | 3521 | /* entry not found; create a new one */ |
f4f4be2b | 3522 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3523 | if (!l) |
72a8cb30 | 3524 | return l; |
e6b81710 | 3525 | |
b1a21367 | 3526 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3527 | l->key.type = type; |
e6b81710 TH |
3528 | /* don't need task_nsproxy() if we're looking at ourself */ |
3529 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3530 | l->owner = cgrp; |
3531 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3532 | return l; |
3533 | } | |
3534 | ||
102a775e BB |
3535 | /* |
3536 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3537 | */ | |
72a8cb30 BB |
3538 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3539 | struct cgroup_pidlist **lp) | |
102a775e BB |
3540 | { |
3541 | pid_t *array; | |
3542 | int length; | |
3543 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3544 | struct css_task_iter it; |
817929ec | 3545 | struct task_struct *tsk; |
102a775e BB |
3546 | struct cgroup_pidlist *l; |
3547 | ||
4bac00d1 TH |
3548 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3549 | ||
102a775e BB |
3550 | /* |
3551 | * If cgroup gets more users after we read count, we won't have | |
3552 | * enough space - tough. This race is indistinguishable to the | |
3553 | * caller from the case that the additional cgroup users didn't | |
3554 | * show up until sometime later on. | |
3555 | */ | |
3556 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3557 | array = pidlist_allocate(length); |
102a775e BB |
3558 | if (!array) |
3559 | return -ENOMEM; | |
3560 | /* now, populate the array */ | |
72ec7029 TH |
3561 | css_task_iter_start(&cgrp->dummy_css, &it); |
3562 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3563 | if (unlikely(n == length)) |
817929ec | 3564 | break; |
102a775e | 3565 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3566 | if (type == CGROUP_FILE_PROCS) |
3567 | pid = task_tgid_vnr(tsk); | |
3568 | else | |
3569 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3570 | if (pid > 0) /* make sure to only use valid results */ |
3571 | array[n++] = pid; | |
817929ec | 3572 | } |
72ec7029 | 3573 | css_task_iter_end(&it); |
102a775e BB |
3574 | length = n; |
3575 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3576 | if (cgroup_sane_behavior(cgrp)) |
3577 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3578 | else | |
3579 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3580 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3581 | length = pidlist_uniq(array, length); |
e6b81710 | 3582 | |
e6b81710 | 3583 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3584 | if (!l) { |
e6b81710 | 3585 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3586 | pidlist_free(array); |
72a8cb30 | 3587 | return -ENOMEM; |
102a775e | 3588 | } |
e6b81710 TH |
3589 | |
3590 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3591 | pidlist_free(l->list); |
102a775e BB |
3592 | l->list = array; |
3593 | l->length = length; | |
72a8cb30 | 3594 | *lp = l; |
102a775e | 3595 | return 0; |
bbcb81d0 PM |
3596 | } |
3597 | ||
846c7bb0 | 3598 | /** |
a043e3b2 | 3599 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3600 | * @stats: cgroupstats to fill information into |
3601 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3602 | * been requested. | |
a043e3b2 LZ |
3603 | * |
3604 | * Build and fill cgroupstats so that taskstats can export it to user | |
3605 | * space. | |
846c7bb0 BS |
3606 | */ |
3607 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3608 | { | |
3609 | int ret = -EINVAL; | |
bd89aabc | 3610 | struct cgroup *cgrp; |
72ec7029 | 3611 | struct css_task_iter it; |
846c7bb0 | 3612 | struct task_struct *tsk; |
33d283be | 3613 | |
846c7bb0 | 3614 | /* |
33d283be LZ |
3615 | * Validate dentry by checking the superblock operations, |
3616 | * and make sure it's a directory. | |
846c7bb0 | 3617 | */ |
33d283be LZ |
3618 | if (dentry->d_sb->s_op != &cgroup_ops || |
3619 | !S_ISDIR(dentry->d_inode->i_mode)) | |
846c7bb0 BS |
3620 | goto err; |
3621 | ||
3622 | ret = 0; | |
bd89aabc | 3623 | cgrp = dentry->d_fsdata; |
846c7bb0 | 3624 | |
72ec7029 TH |
3625 | css_task_iter_start(&cgrp->dummy_css, &it); |
3626 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3627 | switch (tsk->state) { |
3628 | case TASK_RUNNING: | |
3629 | stats->nr_running++; | |
3630 | break; | |
3631 | case TASK_INTERRUPTIBLE: | |
3632 | stats->nr_sleeping++; | |
3633 | break; | |
3634 | case TASK_UNINTERRUPTIBLE: | |
3635 | stats->nr_uninterruptible++; | |
3636 | break; | |
3637 | case TASK_STOPPED: | |
3638 | stats->nr_stopped++; | |
3639 | break; | |
3640 | default: | |
3641 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3642 | stats->nr_io_wait++; | |
3643 | break; | |
3644 | } | |
3645 | } | |
72ec7029 | 3646 | css_task_iter_end(&it); |
846c7bb0 | 3647 | |
846c7bb0 BS |
3648 | err: |
3649 | return ret; | |
3650 | } | |
3651 | ||
8f3ff208 | 3652 | |
bbcb81d0 | 3653 | /* |
102a775e | 3654 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3655 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3656 | * in the cgroup->l->list array. |
bbcb81d0 | 3657 | */ |
cc31edce | 3658 | |
102a775e | 3659 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3660 | { |
cc31edce PM |
3661 | /* |
3662 | * Initially we receive a position value that corresponds to | |
3663 | * one more than the last pid shown (or 0 on the first call or | |
3664 | * after a seek to the start). Use a binary-search to find the | |
3665 | * next pid to display, if any | |
3666 | */ | |
5d22444f | 3667 | struct cgroup_open_file *of = s->private; |
7da11279 | 3668 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3669 | struct cgroup_pidlist *l; |
7da11279 | 3670 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3671 | int index = 0, pid = *pos; |
4bac00d1 TH |
3672 | int *iter, ret; |
3673 | ||
3674 | mutex_lock(&cgrp->pidlist_mutex); | |
3675 | ||
3676 | /* | |
5d22444f | 3677 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3678 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3679 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3680 | * could already have been destroyed. |
3681 | */ | |
5d22444f TH |
3682 | if (of->priv) |
3683 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3684 | |
3685 | /* | |
3686 | * Either this is the first start() after open or the matching | |
3687 | * pidlist has been destroyed inbetween. Create a new one. | |
3688 | */ | |
5d22444f TH |
3689 | if (!of->priv) { |
3690 | ret = pidlist_array_load(cgrp, type, | |
3691 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3692 | if (ret) |
3693 | return ERR_PTR(ret); | |
3694 | } | |
5d22444f | 3695 | l = of->priv; |
cc31edce | 3696 | |
cc31edce | 3697 | if (pid) { |
102a775e | 3698 | int end = l->length; |
20777766 | 3699 | |
cc31edce PM |
3700 | while (index < end) { |
3701 | int mid = (index + end) / 2; | |
afb2bc14 | 3702 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3703 | index = mid; |
3704 | break; | |
afb2bc14 | 3705 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3706 | index = mid + 1; |
3707 | else | |
3708 | end = mid; | |
3709 | } | |
3710 | } | |
3711 | /* If we're off the end of the array, we're done */ | |
102a775e | 3712 | if (index >= l->length) |
cc31edce PM |
3713 | return NULL; |
3714 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3715 | iter = l->list + index; |
afb2bc14 | 3716 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3717 | return iter; |
3718 | } | |
3719 | ||
102a775e | 3720 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3721 | { |
5d22444f TH |
3722 | struct cgroup_open_file *of = s->private; |
3723 | struct cgroup_pidlist *l = of->priv; | |
62236858 | 3724 | |
5d22444f TH |
3725 | if (l) |
3726 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3727 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3728 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3729 | } |
3730 | ||
102a775e | 3731 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3732 | { |
5d22444f TH |
3733 | struct cgroup_open_file *of = s->private; |
3734 | struct cgroup_pidlist *l = of->priv; | |
102a775e BB |
3735 | pid_t *p = v; |
3736 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3737 | /* |
3738 | * Advance to the next pid in the array. If this goes off the | |
3739 | * end, we're done | |
3740 | */ | |
3741 | p++; | |
3742 | if (p >= end) { | |
3743 | return NULL; | |
3744 | } else { | |
7da11279 | 3745 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3746 | return p; |
3747 | } | |
3748 | } | |
3749 | ||
102a775e | 3750 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3751 | { |
3752 | return seq_printf(s, "%d\n", *(int *)v); | |
3753 | } | |
bbcb81d0 | 3754 | |
102a775e BB |
3755 | /* |
3756 | * seq_operations functions for iterating on pidlists through seq_file - | |
3757 | * independent of whether it's tasks or procs | |
3758 | */ | |
3759 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3760 | .start = cgroup_pidlist_start, | |
3761 | .stop = cgroup_pidlist_stop, | |
3762 | .next = cgroup_pidlist_next, | |
3763 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3764 | }; |
3765 | ||
182446d0 TH |
3766 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3767 | struct cftype *cft) | |
81a6a5cd | 3768 | { |
182446d0 | 3769 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3770 | } |
3771 | ||
182446d0 TH |
3772 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3773 | struct cftype *cft, u64 val) | |
6379c106 | 3774 | { |
182446d0 | 3775 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3776 | if (val) |
182446d0 | 3777 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3778 | else |
182446d0 | 3779 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3780 | return 0; |
3781 | } | |
3782 | ||
1c8158ee LZ |
3783 | /* |
3784 | * When dput() is called asynchronously, if umount has been done and | |
3785 | * then deactivate_super() in cgroup_free_fn() kills the superblock, | |
3786 | * there's a small window that vfs will see the root dentry with non-zero | |
3787 | * refcnt and trigger BUG(). | |
3788 | * | |
3789 | * That's why we hold a reference before dput() and drop it right after. | |
3790 | */ | |
3791 | static void cgroup_dput(struct cgroup *cgrp) | |
3792 | { | |
3793 | struct super_block *sb = cgrp->root->sb; | |
3794 | ||
3795 | atomic_inc(&sb->s_active); | |
3796 | dput(cgrp->dentry); | |
3797 | deactivate_super(sb); | |
3798 | } | |
3799 | ||
182446d0 TH |
3800 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3801 | struct cftype *cft) | |
97978e6d | 3802 | { |
182446d0 | 3803 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3804 | } |
3805 | ||
182446d0 TH |
3806 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3807 | struct cftype *cft, u64 val) | |
97978e6d DL |
3808 | { |
3809 | if (val) | |
182446d0 | 3810 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3811 | else |
182446d0 | 3812 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3813 | return 0; |
3814 | } | |
3815 | ||
d5c56ced | 3816 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3817 | { |
d5c56ced | 3818 | .name = "cgroup.procs", |
6612f05b TH |
3819 | .seq_start = cgroup_pidlist_start, |
3820 | .seq_next = cgroup_pidlist_next, | |
3821 | .seq_stop = cgroup_pidlist_stop, | |
3822 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3823 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3824 | .write_u64 = cgroup_procs_write, |
74a1166d | 3825 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3826 | }, |
97978e6d DL |
3827 | { |
3828 | .name = "cgroup.clone_children", | |
873fe09e | 3829 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3830 | .read_u64 = cgroup_clone_children_read, |
3831 | .write_u64 = cgroup_clone_children_write, | |
3832 | }, | |
873fe09e TH |
3833 | { |
3834 | .name = "cgroup.sane_behavior", | |
3835 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3836 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3837 | }, |
d5c56ced TH |
3838 | |
3839 | /* | |
3840 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3841 | * don't exist if sane_behavior. If you're depending on these, be | |
3842 | * prepared to be burned. | |
3843 | */ | |
3844 | { | |
3845 | .name = "tasks", | |
3846 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3847 | .seq_start = cgroup_pidlist_start, |
3848 | .seq_next = cgroup_pidlist_next, | |
3849 | .seq_stop = cgroup_pidlist_stop, | |
3850 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3851 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3852 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3853 | .mode = S_IRUGO | S_IWUSR, |
3854 | }, | |
3855 | { | |
3856 | .name = "notify_on_release", | |
3857 | .flags = CFTYPE_INSANE, | |
3858 | .read_u64 = cgroup_read_notify_on_release, | |
3859 | .write_u64 = cgroup_write_notify_on_release, | |
3860 | }, | |
6e6ff25b TH |
3861 | { |
3862 | .name = "release_agent", | |
cc5943a7 | 3863 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3864 | .seq_show = cgroup_release_agent_show, |
6e6ff25b TH |
3865 | .write_string = cgroup_release_agent_write, |
3866 | .max_write_len = PATH_MAX, | |
3867 | }, | |
db0416b6 | 3868 | { } /* terminate */ |
bbcb81d0 PM |
3869 | }; |
3870 | ||
13af07df | 3871 | /** |
628f7cd4 | 3872 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3873 | * @cgrp: target cgroup |
13af07df | 3874 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3875 | * |
3876 | * On failure, no file is added. | |
13af07df | 3877 | */ |
628f7cd4 | 3878 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3879 | { |
ddbcc7e8 | 3880 | struct cgroup_subsys *ss; |
b420ba7d | 3881 | int i, ret = 0; |
bbcb81d0 | 3882 | |
8e3f6541 | 3883 | /* process cftsets of each subsystem */ |
b420ba7d | 3884 | for_each_subsys(ss, i) { |
8e3f6541 | 3885 | struct cftype_set *set; |
b420ba7d TH |
3886 | |
3887 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3888 | continue; |
8e3f6541 | 3889 | |
bee55099 | 3890 | list_for_each_entry(set, &ss->cftsets, node) { |
2bb566cb | 3891 | ret = cgroup_addrm_files(cgrp, set->cfts, true); |
bee55099 TH |
3892 | if (ret < 0) |
3893 | goto err; | |
3894 | } | |
ddbcc7e8 | 3895 | } |
ddbcc7e8 | 3896 | return 0; |
bee55099 TH |
3897 | err: |
3898 | cgroup_clear_dir(cgrp, subsys_mask); | |
3899 | return ret; | |
ddbcc7e8 PM |
3900 | } |
3901 | ||
0c21ead1 TH |
3902 | /* |
3903 | * css destruction is four-stage process. | |
3904 | * | |
3905 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3906 | * Implemented in kill_css(). | |
3907 | * | |
3908 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3909 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3910 | * by invoking offline_css(). After offlining, the base ref is put. | |
3911 | * Implemented in css_killed_work_fn(). | |
3912 | * | |
3913 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3914 | * accessors are inside RCU read sections. css_release() schedules the | |
3915 | * RCU callback. | |
3916 | * | |
3917 | * 4. After the grace period, the css can be freed. Implemented in | |
3918 | * css_free_work_fn(). | |
3919 | * | |
3920 | * It is actually hairier because both step 2 and 4 require process context | |
3921 | * and thus involve punting to css->destroy_work adding two additional | |
3922 | * steps to the already complex sequence. | |
3923 | */ | |
35ef10da | 3924 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3925 | { |
3926 | struct cgroup_subsys_state *css = | |
35ef10da | 3927 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 3928 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3929 | |
0ae78e0b TH |
3930 | if (css->parent) |
3931 | css_put(css->parent); | |
3932 | ||
0c21ead1 TH |
3933 | css->ss->css_free(css); |
3934 | cgroup_dput(cgrp); | |
48ddbe19 TH |
3935 | } |
3936 | ||
0c21ead1 | 3937 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3938 | { |
3939 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3940 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3941 | |
35ef10da TH |
3942 | /* |
3943 | * css holds an extra ref to @cgrp->dentry which is put on the last | |
0c21ead1 | 3944 | * css_put(). dput() requires process context which we don't have. |
35ef10da TH |
3945 | */ |
3946 | INIT_WORK(&css->destroy_work, css_free_work_fn); | |
e5fca243 | 3947 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
3948 | } |
3949 | ||
d3daf28d TH |
3950 | static void css_release(struct percpu_ref *ref) |
3951 | { | |
3952 | struct cgroup_subsys_state *css = | |
3953 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3954 | ||
aec25020 | 3955 | rcu_assign_pointer(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 3956 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
3957 | } |
3958 | ||
623f926b TH |
3959 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
3960 | struct cgroup *cgrp) | |
ddbcc7e8 | 3961 | { |
bd89aabc | 3962 | css->cgroup = cgrp; |
72c97e54 | 3963 | css->ss = ss; |
ddbcc7e8 | 3964 | css->flags = 0; |
0ae78e0b TH |
3965 | |
3966 | if (cgrp->parent) | |
ca8bdcaf | 3967 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 3968 | else |
38b53aba | 3969 | css->flags |= CSS_ROOT; |
48ddbe19 | 3970 | |
ca8bdcaf | 3971 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
3972 | } |
3973 | ||
2a4ac633 | 3974 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 3975 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3976 | { |
623f926b | 3977 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
3978 | int ret = 0; |
3979 | ||
a31f2d3f TH |
3980 | lockdep_assert_held(&cgroup_mutex); |
3981 | ||
92fb9748 | 3982 | if (ss->css_online) |
eb95419b | 3983 | ret = ss->css_online(css); |
ae7f164a | 3984 | if (!ret) { |
eb95419b | 3985 | css->flags |= CSS_ONLINE; |
f20104de | 3986 | css->cgroup->nr_css++; |
aec25020 | 3987 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 3988 | } |
b1929db4 | 3989 | return ret; |
a31f2d3f TH |
3990 | } |
3991 | ||
2a4ac633 | 3992 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 3993 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3994 | { |
623f926b | 3995 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
3996 | |
3997 | lockdep_assert_held(&cgroup_mutex); | |
3998 | ||
3999 | if (!(css->flags & CSS_ONLINE)) | |
4000 | return; | |
4001 | ||
d7eeac19 | 4002 | if (ss->css_offline) |
eb95419b | 4003 | ss->css_offline(css); |
a31f2d3f | 4004 | |
eb95419b | 4005 | css->flags &= ~CSS_ONLINE; |
09a503ea | 4006 | css->cgroup->nr_css--; |
aec25020 | 4007 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); |
a31f2d3f TH |
4008 | } |
4009 | ||
c81c925a TH |
4010 | /** |
4011 | * create_css - create a cgroup_subsys_state | |
4012 | * @cgrp: the cgroup new css will be associated with | |
4013 | * @ss: the subsys of new css | |
4014 | * | |
4015 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
4016 | * css is online and installed in @cgrp with all interface files created. | |
4017 | * Returns 0 on success, -errno on failure. | |
4018 | */ | |
4019 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
4020 | { | |
4021 | struct cgroup *parent = cgrp->parent; | |
4022 | struct cgroup_subsys_state *css; | |
4023 | int err; | |
4024 | ||
4025 | lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex); | |
4026 | lockdep_assert_held(&cgroup_mutex); | |
4027 | ||
4028 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
4029 | if (IS_ERR(css)) | |
4030 | return PTR_ERR(css); | |
4031 | ||
4032 | err = percpu_ref_init(&css->refcnt, css_release); | |
4033 | if (err) | |
4034 | goto err_free; | |
4035 | ||
4036 | init_css(css, ss, cgrp); | |
4037 | ||
aec25020 | 4038 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a TH |
4039 | if (err) |
4040 | goto err_free; | |
4041 | ||
4042 | err = online_css(css); | |
4043 | if (err) | |
4044 | goto err_free; | |
4045 | ||
4046 | dget(cgrp->dentry); | |
4047 | css_get(css->parent); | |
4048 | ||
4049 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
4050 | parent->parent) { | |
4051 | 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", | |
4052 | current->comm, current->pid, ss->name); | |
4053 | if (!strcmp(ss->name, "memory")) | |
4054 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
4055 | ss->warned_broken_hierarchy = true; | |
4056 | } | |
4057 | ||
4058 | return 0; | |
4059 | ||
4060 | err_free: | |
4061 | percpu_ref_cancel_init(&css->refcnt); | |
4062 | ss->css_free(css); | |
4063 | return err; | |
4064 | } | |
4065 | ||
ddbcc7e8 | 4066 | /* |
a043e3b2 LZ |
4067 | * cgroup_create - create a cgroup |
4068 | * @parent: cgroup that will be parent of the new cgroup | |
4069 | * @dentry: dentry of the new cgroup | |
4070 | * @mode: mode to set on new inode | |
ddbcc7e8 | 4071 | * |
a043e3b2 | 4072 | * Must be called with the mutex on the parent inode held |
ddbcc7e8 | 4073 | */ |
ddbcc7e8 | 4074 | static long cgroup_create(struct cgroup *parent, struct dentry *dentry, |
a5e7ed32 | 4075 | umode_t mode) |
ddbcc7e8 | 4076 | { |
bd89aabc | 4077 | struct cgroup *cgrp; |
65dff759 | 4078 | struct cgroup_name *name; |
ddbcc7e8 | 4079 | struct cgroupfs_root *root = parent->root; |
b58c8998 | 4080 | int ssid, err; |
ddbcc7e8 PM |
4081 | struct cgroup_subsys *ss; |
4082 | struct super_block *sb = root->sb; | |
4083 | ||
0a950f65 | 4084 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
4085 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
4086 | if (!cgrp) | |
ddbcc7e8 PM |
4087 | return -ENOMEM; |
4088 | ||
65dff759 | 4089 | name = cgroup_alloc_name(dentry); |
b58c8998 TH |
4090 | if (!name) { |
4091 | err = -ENOMEM; | |
65dff759 | 4092 | goto err_free_cgrp; |
b58c8998 | 4093 | } |
65dff759 LZ |
4094 | rcu_assign_pointer(cgrp->name, name); |
4095 | ||
976c06bc TH |
4096 | /* |
4097 | * Only live parents can have children. Note that the liveliness | |
4098 | * check isn't strictly necessary because cgroup_mkdir() and | |
4099 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
4100 | * anyway so that locking is contained inside cgroup proper and we | |
4101 | * don't get nasty surprises if we ever grow another caller. | |
4102 | */ | |
4103 | if (!cgroup_lock_live_group(parent)) { | |
4104 | err = -ENODEV; | |
0ab02ca8 LZ |
4105 | goto err_free_name; |
4106 | } | |
4107 | ||
4108 | /* | |
4109 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4110 | * a half-baked cgroup. | |
4111 | */ | |
4112 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
4113 | if (cgrp->id < 0) { | |
4114 | err = -ENOMEM; | |
4115 | goto err_unlock; | |
976c06bc TH |
4116 | } |
4117 | ||
ddbcc7e8 PM |
4118 | /* Grab a reference on the superblock so the hierarchy doesn't |
4119 | * get deleted on unmount if there are child cgroups. This | |
4120 | * can be done outside cgroup_mutex, since the sb can't | |
4121 | * disappear while someone has an open control file on the | |
4122 | * fs */ | |
4123 | atomic_inc(&sb->s_active); | |
4124 | ||
cc31edce | 4125 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4126 | |
fe1c06ca LZ |
4127 | dentry->d_fsdata = cgrp; |
4128 | cgrp->dentry = dentry; | |
4129 | ||
bd89aabc | 4130 | cgrp->parent = parent; |
0ae78e0b | 4131 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 4132 | cgrp->root = parent->root; |
ddbcc7e8 | 4133 | |
b6abdb0e LZ |
4134 | if (notify_on_release(parent)) |
4135 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4136 | ||
2260e7fc TH |
4137 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4138 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4139 | |
4e139afc TH |
4140 | /* |
4141 | * Create directory. cgroup_create_file() returns with the new | |
4142 | * directory locked on success so that it can be populated without | |
4143 | * dropping cgroup_mutex. | |
4144 | */ | |
28fd6f30 | 4145 | err = cgroup_create_file(dentry, S_IFDIR | mode, sb); |
ddbcc7e8 | 4146 | if (err < 0) |
0ab02ca8 | 4147 | goto err_free_id; |
4e139afc | 4148 | lockdep_assert_held(&dentry->d_inode->i_mutex); |
ddbcc7e8 | 4149 | |
00356bd5 | 4150 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 4151 | |
4e139afc | 4152 | /* allocation complete, commit to creation */ |
4e139afc TH |
4153 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
4154 | root->number_of_cgroups++; | |
28fd6f30 | 4155 | |
415cf07a LZ |
4156 | /* hold a ref to the parent's dentry */ |
4157 | dget(parent->dentry); | |
4158 | ||
0d80255e TH |
4159 | /* |
4160 | * @cgrp is now fully operational. If something fails after this | |
4161 | * point, it'll be released via the normal destruction path. | |
4162 | */ | |
4e96ee8e LZ |
4163 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4164 | ||
2bb566cb | 4165 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
4166 | if (err) |
4167 | goto err_destroy; | |
4168 | ||
9d403e99 | 4169 | /* let's create and online css's */ |
b85d2040 TH |
4170 | for_each_subsys(ss, ssid) { |
4171 | if (root->subsys_mask & (1 << ssid)) { | |
4172 | err = create_css(cgrp, ss); | |
4173 | if (err) | |
4174 | goto err_destroy; | |
4175 | } | |
a8638030 | 4176 | } |
ddbcc7e8 PM |
4177 | |
4178 | mutex_unlock(&cgroup_mutex); | |
bd89aabc | 4179 | mutex_unlock(&cgrp->dentry->d_inode->i_mutex); |
ddbcc7e8 PM |
4180 | |
4181 | return 0; | |
4182 | ||
0a950f65 | 4183 | err_free_id: |
4e96ee8e | 4184 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
4185 | /* Release the reference count that we took on the superblock */ |
4186 | deactivate_super(sb); | |
4187 | err_unlock: | |
4188 | mutex_unlock(&cgroup_mutex); | |
65dff759 LZ |
4189 | err_free_name: |
4190 | kfree(rcu_dereference_raw(cgrp->name)); | |
4b8b47eb | 4191 | err_free_cgrp: |
bd89aabc | 4192 | kfree(cgrp); |
ddbcc7e8 | 4193 | return err; |
4b8b47eb TH |
4194 | |
4195 | err_destroy: | |
4196 | cgroup_destroy_locked(cgrp); | |
4197 | mutex_unlock(&cgroup_mutex); | |
4198 | mutex_unlock(&dentry->d_inode->i_mutex); | |
4199 | return err; | |
ddbcc7e8 PM |
4200 | } |
4201 | ||
18bb1db3 | 4202 | static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
ddbcc7e8 PM |
4203 | { |
4204 | struct cgroup *c_parent = dentry->d_parent->d_fsdata; | |
4205 | ||
4206 | /* the vfs holds inode->i_mutex already */ | |
4207 | return cgroup_create(c_parent, dentry, mode | S_IFDIR); | |
4208 | } | |
4209 | ||
223dbc38 TH |
4210 | /* |
4211 | * This is called when the refcnt of a css is confirmed to be killed. | |
4212 | * css_tryget() is now guaranteed to fail. | |
4213 | */ | |
4214 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4215 | { |
223dbc38 TH |
4216 | struct cgroup_subsys_state *css = |
4217 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
4218 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 4219 | |
f20104de TH |
4220 | mutex_lock(&cgroup_mutex); |
4221 | ||
09a503ea TH |
4222 | /* |
4223 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
4224 | * initate destruction. | |
4225 | */ | |
4226 | offline_css(css); | |
4227 | ||
f20104de TH |
4228 | /* |
4229 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
4230 | * be disabled before proceeding to the second phase of cgroup | |
4231 | * destruction. If we are the last one, kick it off. | |
4232 | */ | |
09a503ea | 4233 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
4234 | cgroup_destroy_css_killed(cgrp); |
4235 | ||
4236 | mutex_unlock(&cgroup_mutex); | |
09a503ea TH |
4237 | |
4238 | /* | |
4239 | * Put the css refs from kill_css(). Each css holds an extra | |
4240 | * reference to the cgroup's dentry and cgroup removal proceeds | |
4241 | * regardless of css refs. On the last put of each css, whenever | |
4242 | * that may be, the extra dentry ref is put so that dentry | |
4243 | * destruction happens only after all css's are released. | |
4244 | */ | |
4245 | css_put(css); | |
d3daf28d TH |
4246 | } |
4247 | ||
223dbc38 TH |
4248 | /* css kill confirmation processing requires process context, bounce */ |
4249 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4250 | { |
4251 | struct cgroup_subsys_state *css = | |
4252 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4253 | ||
223dbc38 | 4254 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 4255 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
4256 | } |
4257 | ||
edae0c33 TH |
4258 | /** |
4259 | * kill_css - destroy a css | |
4260 | * @css: css to destroy | |
4261 | * | |
3c14f8b4 TH |
4262 | * This function initiates destruction of @css by removing cgroup interface |
4263 | * files and putting its base reference. ->css_offline() will be invoked | |
4264 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
4265 | * reference count reaches zero, @css will be released. | |
edae0c33 TH |
4266 | */ |
4267 | static void kill_css(struct cgroup_subsys_state *css) | |
4268 | { | |
aec25020 | 4269 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 4270 | |
edae0c33 TH |
4271 | /* |
4272 | * Killing would put the base ref, but we need to keep it alive | |
4273 | * until after ->css_offline(). | |
4274 | */ | |
4275 | css_get(css); | |
4276 | ||
4277 | /* | |
4278 | * cgroup core guarantees that, by the time ->css_offline() is | |
4279 | * invoked, no new css reference will be given out via | |
4280 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
4281 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
4282 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4283 | * | |
4284 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4285 | * css is confirmed to be seen as killed on all CPUs. | |
4286 | */ | |
4287 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4288 | } |
4289 | ||
4290 | /** | |
4291 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4292 | * @cgrp: cgroup to be destroyed | |
4293 | * | |
4294 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4295 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
4296 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
4297 | * invoked. To satisfy all the requirements, destruction is implemented in | |
4298 | * the following two steps. | |
4299 | * | |
4300 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4301 | * userland visible parts and start killing the percpu refcnts of | |
4302 | * css's. Set up so that the next stage will be kicked off once all | |
4303 | * the percpu refcnts are confirmed to be killed. | |
4304 | * | |
4305 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4306 | * rest of destruction. Once all cgroup references are gone, the | |
4307 | * cgroup is RCU-freed. | |
4308 | * | |
4309 | * This function implements s1. After this step, @cgrp is gone as far as | |
4310 | * the userland is concerned and a new cgroup with the same name may be | |
4311 | * created. As cgroup doesn't care about the names internally, this | |
4312 | * doesn't cause any problem. | |
4313 | */ | |
42809dd4 TH |
4314 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4315 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4316 | { |
42809dd4 | 4317 | struct dentry *d = cgrp->dentry; |
1c6727af | 4318 | struct cgroup_subsys_state *css; |
bb78a92f | 4319 | struct cgroup *child; |
ddd69148 | 4320 | bool empty; |
1c6727af | 4321 | int ssid; |
ddbcc7e8 | 4322 | |
42809dd4 TH |
4323 | lockdep_assert_held(&d->d_inode->i_mutex); |
4324 | lockdep_assert_held(&cgroup_mutex); | |
4325 | ||
ddd69148 | 4326 | /* |
6f3d828f TH |
4327 | * css_set_lock synchronizes access to ->cset_links and prevents |
4328 | * @cgrp from being removed while __put_css_set() is in progress. | |
ddd69148 TH |
4329 | */ |
4330 | read_lock(&css_set_lock); | |
bb78a92f | 4331 | empty = list_empty(&cgrp->cset_links); |
ddd69148 TH |
4332 | read_unlock(&css_set_lock); |
4333 | if (!empty) | |
ddbcc7e8 | 4334 | return -EBUSY; |
a043e3b2 | 4335 | |
bb78a92f HD |
4336 | /* |
4337 | * Make sure there's no live children. We can't test ->children | |
4338 | * emptiness as dead children linger on it while being destroyed; | |
4339 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
4340 | */ | |
4341 | empty = true; | |
4342 | rcu_read_lock(); | |
4343 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
4344 | empty = cgroup_is_dead(child); | |
4345 | if (!empty) | |
4346 | break; | |
4347 | } | |
4348 | rcu_read_unlock(); | |
4349 | if (!empty) | |
4350 | return -EBUSY; | |
4351 | ||
88703267 | 4352 | /* |
edae0c33 TH |
4353 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
4354 | * will be invoked to perform the rest of destruction once the | |
4355 | * percpu refs of all css's are confirmed to be killed. | |
88703267 | 4356 | */ |
1c6727af TH |
4357 | for_each_css(css, ssid, cgrp) |
4358 | kill_css(css); | |
455050d2 TH |
4359 | |
4360 | /* | |
4361 | * Mark @cgrp dead. This prevents further task migration and child | |
4362 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 4363 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 4364 | * resume iteration after dropping RCU read lock. See |
492eb21b | 4365 | * css_next_child() for details. |
455050d2 | 4366 | */ |
54766d4a | 4367 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 4368 | |
455050d2 TH |
4369 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
4370 | raw_spin_lock(&release_list_lock); | |
4371 | if (!list_empty(&cgrp->release_list)) | |
4372 | list_del_init(&cgrp->release_list); | |
4373 | raw_spin_unlock(&release_list_lock); | |
4374 | ||
4375 | /* | |
f20104de TH |
4376 | * If @cgrp has css's attached, the second stage of cgroup |
4377 | * destruction is kicked off from css_killed_work_fn() after the | |
4378 | * refs of all attached css's are killed. If @cgrp doesn't have | |
4379 | * any css, we kick it off here. | |
4380 | */ | |
4381 | if (!cgrp->nr_css) | |
4382 | cgroup_destroy_css_killed(cgrp); | |
4383 | ||
455050d2 | 4384 | /* |
3c14f8b4 TH |
4385 | * Clear the base files and remove @cgrp directory. The removal |
4386 | * puts the base ref but we aren't quite done with @cgrp yet, so | |
4387 | * hold onto it. | |
455050d2 | 4388 | */ |
2bb566cb | 4389 | cgroup_addrm_files(cgrp, cgroup_base_files, false); |
455050d2 TH |
4390 | dget(d); |
4391 | cgroup_d_remove_dir(d); | |
4392 | ||
ea15f8cc TH |
4393 | return 0; |
4394 | }; | |
4395 | ||
d3daf28d | 4396 | /** |
f20104de | 4397 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
4398 | * @work: cgroup->destroy_free_work |
4399 | * | |
4400 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
4401 | * destroyed after all css's are offlined and performs the rest of |
4402 | * destruction. This is the second step of destruction described in the | |
4403 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 4404 | */ |
f20104de | 4405 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 4406 | { |
ea15f8cc TH |
4407 | struct cgroup *parent = cgrp->parent; |
4408 | struct dentry *d = cgrp->dentry; | |
ea15f8cc | 4409 | |
f20104de | 4410 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 4411 | |
999cd8a4 | 4412 | /* delete this cgroup from parent->children */ |
eb6fd504 | 4413 | list_del_rcu(&cgrp->sibling); |
ed957793 | 4414 | |
ddbcc7e8 | 4415 | dput(d); |
ddbcc7e8 | 4416 | |
bd89aabc | 4417 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 4418 | check_for_release(parent); |
ddbcc7e8 PM |
4419 | } |
4420 | ||
42809dd4 TH |
4421 | static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry) |
4422 | { | |
4423 | int ret; | |
4424 | ||
4425 | mutex_lock(&cgroup_mutex); | |
4426 | ret = cgroup_destroy_locked(dentry->d_fsdata); | |
4427 | mutex_unlock(&cgroup_mutex); | |
4428 | ||
4429 | return ret; | |
4430 | } | |
4431 | ||
3ed80a62 | 4432 | static void __init cgroup_init_cftsets(struct cgroup_subsys *ss) |
8e3f6541 TH |
4433 | { |
4434 | INIT_LIST_HEAD(&ss->cftsets); | |
4435 | ||
4436 | /* | |
4437 | * base_cftset is embedded in subsys itself, no need to worry about | |
4438 | * deregistration. | |
4439 | */ | |
4440 | if (ss->base_cftypes) { | |
2bb566cb TH |
4441 | struct cftype *cft; |
4442 | ||
4443 | for (cft = ss->base_cftypes; cft->name[0] != '\0'; cft++) | |
4444 | cft->ss = ss; | |
4445 | ||
8e3f6541 TH |
4446 | ss->base_cftset.cfts = ss->base_cftypes; |
4447 | list_add_tail(&ss->base_cftset.node, &ss->cftsets); | |
4448 | } | |
4449 | } | |
4450 | ||
06a11920 | 4451 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4452 | { |
ddbcc7e8 | 4453 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4454 | |
4455 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4456 | |
648bb56d TH |
4457 | mutex_lock(&cgroup_mutex); |
4458 | ||
8e3f6541 TH |
4459 | /* init base cftset */ |
4460 | cgroup_init_cftsets(ss); | |
4461 | ||
ddbcc7e8 | 4462 | /* Create the top cgroup state for this subsystem */ |
9871bf95 | 4463 | ss->root = &cgroup_dummy_root; |
ca8bdcaf | 4464 | css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss)); |
ddbcc7e8 PM |
4465 | /* We don't handle early failures gracefully */ |
4466 | BUG_ON(IS_ERR(css)); | |
623f926b | 4467 | init_css(css, ss, cgroup_dummy_top); |
ddbcc7e8 | 4468 | |
e8d55fde | 4469 | /* Update the init_css_set to contain a subsys |
817929ec | 4470 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
4471 | * newly registered, all tasks and hence the |
4472 | * init_css_set is in the subsystem's top cgroup. */ | |
aec25020 | 4473 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4474 | |
4475 | need_forkexit_callback |= ss->fork || ss->exit; | |
4476 | ||
e8d55fde LZ |
4477 | /* At system boot, before all subsystems have been |
4478 | * registered, no tasks have been forked, so we don't | |
4479 | * need to invoke fork callbacks here. */ | |
4480 | BUG_ON(!list_empty(&init_task.tasks)); | |
4481 | ||
ae7f164a | 4482 | BUG_ON(online_css(css)); |
a8638030 | 4483 | |
648bb56d | 4484 | mutex_unlock(&cgroup_mutex); |
e6a1105b BB |
4485 | } |
4486 | ||
ddbcc7e8 | 4487 | /** |
a043e3b2 LZ |
4488 | * cgroup_init_early - cgroup initialization at system boot |
4489 | * | |
4490 | * Initialize cgroups at system boot, and initialize any | |
4491 | * subsystems that request early init. | |
ddbcc7e8 PM |
4492 | */ |
4493 | int __init cgroup_init_early(void) | |
4494 | { | |
30159ec7 | 4495 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4496 | int i; |
30159ec7 | 4497 | |
146aa1bd | 4498 | atomic_set(&init_css_set.refcount, 1); |
69d0206c | 4499 | INIT_LIST_HEAD(&init_css_set.cgrp_links); |
817929ec | 4500 | INIT_LIST_HEAD(&init_css_set.tasks); |
472b1053 | 4501 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 4502 | css_set_count = 1; |
9871bf95 TH |
4503 | init_cgroup_root(&cgroup_dummy_root); |
4504 | cgroup_root_count = 1; | |
a4ea1cc9 | 4505 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4506 | |
69d0206c | 4507 | init_cgrp_cset_link.cset = &init_css_set; |
9871bf95 TH |
4508 | init_cgrp_cset_link.cgrp = cgroup_dummy_top; |
4509 | list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links); | |
69d0206c | 4510 | list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links); |
ddbcc7e8 | 4511 | |
3ed80a62 | 4512 | for_each_subsys(ss, i) { |
aec25020 | 4513 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4514 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4515 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4516 | ss->id, ss->name); |
073219e9 TH |
4517 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4518 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4519 | ||
aec25020 | 4520 | ss->id = i; |
073219e9 | 4521 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4522 | |
4523 | if (ss->early_init) | |
4524 | cgroup_init_subsys(ss); | |
4525 | } | |
4526 | return 0; | |
4527 | } | |
4528 | ||
4529 | /** | |
a043e3b2 LZ |
4530 | * cgroup_init - cgroup initialization |
4531 | * | |
4532 | * Register cgroup filesystem and /proc file, and initialize | |
4533 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4534 | */ |
4535 | int __init cgroup_init(void) | |
4536 | { | |
30159ec7 | 4537 | struct cgroup_subsys *ss; |
0ac801fe | 4538 | unsigned long key; |
30159ec7 | 4539 | int i, err; |
a424316c PM |
4540 | |
4541 | err = bdi_init(&cgroup_backing_dev_info); | |
4542 | if (err) | |
4543 | return err; | |
ddbcc7e8 | 4544 | |
3ed80a62 | 4545 | for_each_subsys(ss, i) { |
ddbcc7e8 PM |
4546 | if (!ss->early_init) |
4547 | cgroup_init_subsys(ss); | |
4548 | } | |
4549 | ||
fa3ca07e | 4550 | /* allocate id for the dummy hierarchy */ |
54e7b4eb | 4551 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4552 | |
82fe9b0d TH |
4553 | /* Add init_css_set to the hash table */ |
4554 | key = css_set_hash(init_css_set.subsys); | |
4555 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4556 | ||
fc76df70 | 4557 | BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1)); |
676db4af | 4558 | |
4e96ee8e LZ |
4559 | err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top, |
4560 | 0, 1, GFP_KERNEL); | |
4561 | BUG_ON(err < 0); | |
4562 | ||
54e7b4eb TH |
4563 | mutex_unlock(&cgroup_mutex); |
4564 | ||
676db4af GK |
4565 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
4566 | if (!cgroup_kobj) { | |
4567 | err = -ENOMEM; | |
4568 | goto out; | |
4569 | } | |
4570 | ||
ddbcc7e8 | 4571 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
4572 | if (err < 0) { |
4573 | kobject_put(cgroup_kobj); | |
ddbcc7e8 | 4574 | goto out; |
676db4af | 4575 | } |
ddbcc7e8 | 4576 | |
46ae220b | 4577 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
a424316c | 4578 | |
ddbcc7e8 | 4579 | out: |
a424316c PM |
4580 | if (err) |
4581 | bdi_destroy(&cgroup_backing_dev_info); | |
4582 | ||
ddbcc7e8 PM |
4583 | return err; |
4584 | } | |
b4f48b63 | 4585 | |
e5fca243 TH |
4586 | static int __init cgroup_wq_init(void) |
4587 | { | |
4588 | /* | |
4589 | * There isn't much point in executing destruction path in | |
4590 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
ab3f5faa HD |
4591 | * |
4592 | * XXX: Must be ordered to make sure parent is offlined after | |
4593 | * children. The ordering requirement is for memcg where a | |
4594 | * parent's offline may wait for a child's leading to deadlock. In | |
4595 | * the long term, this should be fixed from memcg side. | |
e5fca243 TH |
4596 | * |
4597 | * We would prefer to do this in cgroup_init() above, but that | |
4598 | * is called before init_workqueues(): so leave this until after. | |
4599 | */ | |
ab3f5faa | 4600 | cgroup_destroy_wq = alloc_ordered_workqueue("cgroup_destroy", 0); |
e5fca243 | 4601 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4602 | |
4603 | /* | |
4604 | * Used to destroy pidlists and separate to serve as flush domain. | |
4605 | * Cap @max_active to 1 too. | |
4606 | */ | |
4607 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4608 | 0, 1); | |
4609 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4610 | ||
e5fca243 TH |
4611 | return 0; |
4612 | } | |
4613 | core_initcall(cgroup_wq_init); | |
4614 | ||
a424316c PM |
4615 | /* |
4616 | * proc_cgroup_show() | |
4617 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4618 | * - Used for /proc/<pid>/cgroup. | |
4619 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4620 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4621 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4622 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4623 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4624 | * cgroup to top_cgroup. | |
4625 | */ | |
4626 | ||
4627 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4628 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4629 | { |
4630 | struct pid *pid; | |
4631 | struct task_struct *tsk; | |
4632 | char *buf; | |
4633 | int retval; | |
4634 | struct cgroupfs_root *root; | |
4635 | ||
4636 | retval = -ENOMEM; | |
4637 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4638 | if (!buf) | |
4639 | goto out; | |
4640 | ||
4641 | retval = -ESRCH; | |
4642 | pid = m->private; | |
4643 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4644 | if (!tsk) | |
4645 | goto out_free; | |
4646 | ||
4647 | retval = 0; | |
4648 | ||
4649 | mutex_lock(&cgroup_mutex); | |
4650 | ||
e5f6a860 | 4651 | for_each_active_root(root) { |
a424316c | 4652 | struct cgroup_subsys *ss; |
bd89aabc | 4653 | struct cgroup *cgrp; |
b85d2040 | 4654 | int ssid, count = 0; |
a424316c | 4655 | |
2c6ab6d2 | 4656 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 TH |
4657 | for_each_subsys(ss, ssid) |
4658 | if (root->subsys_mask & (1 << ssid)) | |
4659 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4660 | if (strlen(root->name)) |
4661 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4662 | root->name); | |
a424316c | 4663 | seq_putc(m, ':'); |
7717f7ba | 4664 | cgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 4665 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); |
a424316c PM |
4666 | if (retval < 0) |
4667 | goto out_unlock; | |
4668 | seq_puts(m, buf); | |
4669 | seq_putc(m, '\n'); | |
4670 | } | |
4671 | ||
4672 | out_unlock: | |
4673 | mutex_unlock(&cgroup_mutex); | |
4674 | put_task_struct(tsk); | |
4675 | out_free: | |
4676 | kfree(buf); | |
4677 | out: | |
4678 | return retval; | |
4679 | } | |
4680 | ||
a424316c PM |
4681 | /* Display information about each subsystem and each hierarchy */ |
4682 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4683 | { | |
30159ec7 | 4684 | struct cgroup_subsys *ss; |
a424316c | 4685 | int i; |
a424316c | 4686 | |
8bab8dde | 4687 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4688 | /* |
4689 | * ideally we don't want subsystems moving around while we do this. | |
4690 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4691 | * subsys/hierarchy state. | |
4692 | */ | |
a424316c | 4693 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4694 | |
4695 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4696 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4697 | ss->name, ss->root->hierarchy_id, | |
8bab8dde | 4698 | ss->root->number_of_cgroups, !ss->disabled); |
30159ec7 | 4699 | |
a424316c PM |
4700 | mutex_unlock(&cgroup_mutex); |
4701 | return 0; | |
4702 | } | |
4703 | ||
4704 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4705 | { | |
9dce07f1 | 4706 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4707 | } |
4708 | ||
828c0950 | 4709 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4710 | .open = cgroupstats_open, |
4711 | .read = seq_read, | |
4712 | .llseek = seq_lseek, | |
4713 | .release = single_release, | |
4714 | }; | |
4715 | ||
b4f48b63 PM |
4716 | /** |
4717 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4718 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4719 | * |
4720 | * Description: A task inherits its parent's cgroup at fork(). | |
4721 | * | |
4722 | * A pointer to the shared css_set was automatically copied in | |
4723 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
9bb71308 TH |
4724 | * it was not made under the protection of RCU or cgroup_mutex, so |
4725 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
4726 | * have already changed current->cgroups, allowing the previously | |
4727 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
4728 | * |
4729 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4730 | * task, and the passed argument 'child' points to the child task. | |
4731 | */ | |
4732 | void cgroup_fork(struct task_struct *child) | |
4733 | { | |
9bb71308 | 4734 | task_lock(current); |
a8ad805c | 4735 | get_css_set(task_css_set(current)); |
817929ec | 4736 | child->cgroups = current->cgroups; |
9bb71308 | 4737 | task_unlock(current); |
817929ec | 4738 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4739 | } |
4740 | ||
817929ec | 4741 | /** |
a043e3b2 LZ |
4742 | * cgroup_post_fork - called on a new task after adding it to the task list |
4743 | * @child: the task in question | |
4744 | * | |
5edee61e TH |
4745 | * Adds the task to the list running through its css_set if necessary and |
4746 | * call the subsystem fork() callbacks. Has to be after the task is | |
4747 | * visible on the task list in case we race with the first call to | |
0942eeee | 4748 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4749 | * list. |
a043e3b2 | 4750 | */ |
817929ec PM |
4751 | void cgroup_post_fork(struct task_struct *child) |
4752 | { | |
30159ec7 | 4753 | struct cgroup_subsys *ss; |
5edee61e TH |
4754 | int i; |
4755 | ||
3ce3230a FW |
4756 | /* |
4757 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
4758 | * under the tasklist_lock and we read it here after we added the child | |
4759 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
4760 | * yet in the tasklist when we walked through it from | |
4761 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
4762 | * should be visible now due to the paired locking and barriers implied | |
4763 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
4764 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
4765 | * lock on fork. | |
4766 | */ | |
817929ec PM |
4767 | if (use_task_css_set_links) { |
4768 | write_lock(&css_set_lock); | |
d8783832 TH |
4769 | task_lock(child); |
4770 | if (list_empty(&child->cg_list)) | |
a8ad805c | 4771 | list_add(&child->cg_list, &task_css_set(child)->tasks); |
d8783832 | 4772 | task_unlock(child); |
817929ec PM |
4773 | write_unlock(&css_set_lock); |
4774 | } | |
5edee61e TH |
4775 | |
4776 | /* | |
4777 | * Call ss->fork(). This must happen after @child is linked on | |
4778 | * css_set; otherwise, @child might change state between ->fork() | |
4779 | * and addition to css_set. | |
4780 | */ | |
4781 | if (need_forkexit_callback) { | |
3ed80a62 | 4782 | for_each_subsys(ss, i) |
5edee61e TH |
4783 | if (ss->fork) |
4784 | ss->fork(child); | |
5edee61e | 4785 | } |
817929ec | 4786 | } |
5edee61e | 4787 | |
b4f48b63 PM |
4788 | /** |
4789 | * cgroup_exit - detach cgroup from exiting task | |
4790 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4791 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4792 | * |
4793 | * Description: Detach cgroup from @tsk and release it. | |
4794 | * | |
4795 | * Note that cgroups marked notify_on_release force every task in | |
4796 | * them to take the global cgroup_mutex mutex when exiting. | |
4797 | * This could impact scaling on very large systems. Be reluctant to | |
4798 | * use notify_on_release cgroups where very high task exit scaling | |
4799 | * is required on large systems. | |
4800 | * | |
4801 | * the_top_cgroup_hack: | |
4802 | * | |
4803 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4804 | * | |
4805 | * We call cgroup_exit() while the task is still competent to | |
4806 | * handle notify_on_release(), then leave the task attached to the | |
4807 | * root cgroup in each hierarchy for the remainder of its exit. | |
4808 | * | |
4809 | * To do this properly, we would increment the reference count on | |
4810 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4811 | * code we would add a second cgroup function call, to drop that | |
4812 | * reference. This would just create an unnecessary hot spot on | |
4813 | * the top_cgroup reference count, to no avail. | |
4814 | * | |
4815 | * Normally, holding a reference to a cgroup without bumping its | |
4816 | * count is unsafe. The cgroup could go away, or someone could | |
4817 | * attach us to a different cgroup, decrementing the count on | |
4818 | * the first cgroup that we never incremented. But in this case, | |
4819 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4820 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4821 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4822 | */ |
4823 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4824 | { | |
30159ec7 | 4825 | struct cgroup_subsys *ss; |
5abb8855 | 4826 | struct css_set *cset; |
d41d5a01 | 4827 | int i; |
817929ec PM |
4828 | |
4829 | /* | |
4830 | * Unlink from the css_set task list if necessary. | |
4831 | * Optimistically check cg_list before taking | |
4832 | * css_set_lock | |
4833 | */ | |
4834 | if (!list_empty(&tsk->cg_list)) { | |
4835 | write_lock(&css_set_lock); | |
4836 | if (!list_empty(&tsk->cg_list)) | |
8d258797 | 4837 | list_del_init(&tsk->cg_list); |
817929ec PM |
4838 | write_unlock(&css_set_lock); |
4839 | } | |
4840 | ||
b4f48b63 PM |
4841 | /* Reassign the task to the init_css_set. */ |
4842 | task_lock(tsk); | |
a8ad805c TH |
4843 | cset = task_css_set(tsk); |
4844 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 PZ |
4845 | |
4846 | if (run_callbacks && need_forkexit_callback) { | |
3ed80a62 TH |
4847 | /* see cgroup_post_fork() for details */ |
4848 | for_each_subsys(ss, i) { | |
d41d5a01 | 4849 | if (ss->exit) { |
eb95419b TH |
4850 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4851 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4852 | |
eb95419b | 4853 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4854 | } |
4855 | } | |
4856 | } | |
b4f48b63 | 4857 | task_unlock(tsk); |
d41d5a01 | 4858 | |
5abb8855 | 4859 | put_css_set_taskexit(cset); |
b4f48b63 | 4860 | } |
697f4161 | 4861 | |
bd89aabc | 4862 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4863 | { |
f50daa70 | 4864 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4865 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4866 | /* |
4867 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4868 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4869 | * it now |
4870 | */ | |
81a6a5cd | 4871 | int need_schedule_work = 0; |
f50daa70 | 4872 | |
cdcc136f | 4873 | raw_spin_lock(&release_list_lock); |
54766d4a | 4874 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4875 | list_empty(&cgrp->release_list)) { |
4876 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4877 | need_schedule_work = 1; |
4878 | } | |
cdcc136f | 4879 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4880 | if (need_schedule_work) |
4881 | schedule_work(&release_agent_work); | |
4882 | } | |
4883 | } | |
4884 | ||
81a6a5cd PM |
4885 | /* |
4886 | * Notify userspace when a cgroup is released, by running the | |
4887 | * configured release agent with the name of the cgroup (path | |
4888 | * relative to the root of cgroup file system) as the argument. | |
4889 | * | |
4890 | * Most likely, this user command will try to rmdir this cgroup. | |
4891 | * | |
4892 | * This races with the possibility that some other task will be | |
4893 | * attached to this cgroup before it is removed, or that some other | |
4894 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4895 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4896 | * unused, and this cgroup will be reprieved from its death sentence, | |
4897 | * to continue to serve a useful existence. Next time it's released, | |
4898 | * we will get notified again, if it still has 'notify_on_release' set. | |
4899 | * | |
4900 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4901 | * means only wait until the task is successfully execve()'d. The | |
4902 | * separate release agent task is forked by call_usermodehelper(), | |
4903 | * then control in this thread returns here, without waiting for the | |
4904 | * release agent task. We don't bother to wait because the caller of | |
4905 | * this routine has no use for the exit status of the release agent | |
4906 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4907 | */ |
81a6a5cd PM |
4908 | static void cgroup_release_agent(struct work_struct *work) |
4909 | { | |
4910 | BUG_ON(work != &release_agent_work); | |
4911 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4912 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4913 | while (!list_empty(&release_list)) { |
4914 | char *argv[3], *envp[3]; | |
4915 | int i; | |
e788e066 | 4916 | char *pathbuf = NULL, *agentbuf = NULL; |
bd89aabc | 4917 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4918 | struct cgroup, |
4919 | release_list); | |
bd89aabc | 4920 | list_del_init(&cgrp->release_list); |
cdcc136f | 4921 | raw_spin_unlock(&release_list_lock); |
81a6a5cd | 4922 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
e788e066 PM |
4923 | if (!pathbuf) |
4924 | goto continue_free; | |
4925 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
4926 | goto continue_free; | |
4927 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4928 | if (!agentbuf) | |
4929 | goto continue_free; | |
81a6a5cd PM |
4930 | |
4931 | i = 0; | |
e788e066 PM |
4932 | argv[i++] = agentbuf; |
4933 | argv[i++] = pathbuf; | |
81a6a5cd PM |
4934 | argv[i] = NULL; |
4935 | ||
4936 | i = 0; | |
4937 | /* minimal command environment */ | |
4938 | envp[i++] = "HOME=/"; | |
4939 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4940 | envp[i] = NULL; | |
4941 | ||
4942 | /* Drop the lock while we invoke the usermode helper, | |
4943 | * since the exec could involve hitting disk and hence | |
4944 | * be a slow process */ | |
4945 | mutex_unlock(&cgroup_mutex); | |
4946 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4947 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4948 | continue_free: |
4949 | kfree(pathbuf); | |
4950 | kfree(agentbuf); | |
cdcc136f | 4951 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4952 | } |
cdcc136f | 4953 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4954 | mutex_unlock(&cgroup_mutex); |
4955 | } | |
8bab8dde PM |
4956 | |
4957 | static int __init cgroup_disable(char *str) | |
4958 | { | |
30159ec7 | 4959 | struct cgroup_subsys *ss; |
8bab8dde | 4960 | char *token; |
30159ec7 | 4961 | int i; |
8bab8dde PM |
4962 | |
4963 | while ((token = strsep(&str, ",")) != NULL) { | |
4964 | if (!*token) | |
4965 | continue; | |
be45c900 | 4966 | |
3ed80a62 | 4967 | for_each_subsys(ss, i) { |
8bab8dde PM |
4968 | if (!strcmp(token, ss->name)) { |
4969 | ss->disabled = 1; | |
4970 | printk(KERN_INFO "Disabling %s control group" | |
4971 | " subsystem\n", ss->name); | |
4972 | break; | |
4973 | } | |
4974 | } | |
4975 | } | |
4976 | return 1; | |
4977 | } | |
4978 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 4979 | |
b77d7b60 | 4980 | /** |
5a17f543 | 4981 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
4982 | * @dentry: directory dentry of interest |
4983 | * @ss: subsystem of interest | |
b77d7b60 | 4984 | * |
5a17f543 TH |
4985 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
4986 | * to get the corresponding css and return it. If such css doesn't exist | |
4987 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 4988 | */ |
5a17f543 TH |
4989 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
4990 | struct cgroup_subsys *ss) | |
e5d1367f SE |
4991 | { |
4992 | struct cgroup *cgrp; | |
5a17f543 | 4993 | struct cgroup_subsys_state *css; |
b77d7b60 | 4994 | |
35cf0836 TH |
4995 | /* is @dentry a cgroup dir? */ |
4996 | if (!dentry->d_inode || | |
4997 | dentry->d_inode->i_op != &cgroup_dir_inode_operations) | |
e5d1367f SE |
4998 | return ERR_PTR(-EBADF); |
4999 | ||
5a17f543 TH |
5000 | rcu_read_lock(); |
5001 | ||
35cf0836 | 5002 | cgrp = __d_cgrp(dentry); |
5a17f543 TH |
5003 | css = cgroup_css(cgrp, ss); |
5004 | ||
5005 | if (!css || !css_tryget(css)) | |
5006 | css = ERR_PTR(-ENOENT); | |
5007 | ||
5008 | rcu_read_unlock(); | |
5009 | return css; | |
e5d1367f | 5010 | } |
e5d1367f | 5011 | |
1cb650b9 LZ |
5012 | /** |
5013 | * css_from_id - lookup css by id | |
5014 | * @id: the cgroup id | |
5015 | * @ss: cgroup subsys to be looked into | |
5016 | * | |
5017 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5018 | * Should be called under rcu_read_lock(). | |
5019 | */ | |
5020 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5021 | { | |
5022 | struct cgroup *cgrp; | |
5023 | ||
87fb54f1 | 5024 | cgroup_assert_mutex_or_rcu_locked(); |
1cb650b9 LZ |
5025 | |
5026 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
5027 | if (cgrp) | |
d1625964 | 5028 | return cgroup_css(cgrp, ss); |
1cb650b9 | 5029 | return NULL; |
e5d1367f SE |
5030 | } |
5031 | ||
fe693435 | 5032 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5033 | static struct cgroup_subsys_state * |
5034 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5035 | { |
5036 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5037 | ||
5038 | if (!css) | |
5039 | return ERR_PTR(-ENOMEM); | |
5040 | ||
5041 | return css; | |
5042 | } | |
5043 | ||
eb95419b | 5044 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5045 | { |
eb95419b | 5046 | kfree(css); |
fe693435 PM |
5047 | } |
5048 | ||
182446d0 TH |
5049 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5050 | struct cftype *cft) | |
fe693435 | 5051 | { |
182446d0 | 5052 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5053 | } |
5054 | ||
182446d0 TH |
5055 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5056 | struct cftype *cft) | |
fe693435 PM |
5057 | { |
5058 | return (u64)(unsigned long)current->cgroups; | |
5059 | } | |
5060 | ||
182446d0 | 5061 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5062 | struct cftype *cft) |
fe693435 PM |
5063 | { |
5064 | u64 count; | |
5065 | ||
5066 | rcu_read_lock(); | |
a8ad805c | 5067 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5068 | rcu_read_unlock(); |
5069 | return count; | |
5070 | } | |
5071 | ||
2da8ca82 | 5072 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5073 | { |
69d0206c | 5074 | struct cgrp_cset_link *link; |
5abb8855 | 5075 | struct css_set *cset; |
7717f7ba PM |
5076 | |
5077 | read_lock(&css_set_lock); | |
5078 | rcu_read_lock(); | |
5abb8855 | 5079 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5080 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba PM |
5081 | struct cgroup *c = link->cgrp; |
5082 | const char *name; | |
5083 | ||
5084 | if (c->dentry) | |
5085 | name = c->dentry->d_name.name; | |
5086 | else | |
5087 | name = "?"; | |
2c6ab6d2 PM |
5088 | seq_printf(seq, "Root %d group %s\n", |
5089 | c->root->hierarchy_id, name); | |
7717f7ba PM |
5090 | } |
5091 | rcu_read_unlock(); | |
5092 | read_unlock(&css_set_lock); | |
5093 | return 0; | |
5094 | } | |
5095 | ||
5096 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 5097 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5098 | { |
2da8ca82 | 5099 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 5100 | struct cgrp_cset_link *link; |
7717f7ba PM |
5101 | |
5102 | read_lock(&css_set_lock); | |
182446d0 | 5103 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5104 | struct css_set *cset = link->cset; |
7717f7ba PM |
5105 | struct task_struct *task; |
5106 | int count = 0; | |
5abb8855 TH |
5107 | seq_printf(seq, "css_set %p\n", cset); |
5108 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
7717f7ba PM |
5109 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { |
5110 | seq_puts(seq, " ...\n"); | |
5111 | break; | |
5112 | } else { | |
5113 | seq_printf(seq, " task %d\n", | |
5114 | task_pid_vnr(task)); | |
5115 | } | |
5116 | } | |
5117 | } | |
5118 | read_unlock(&css_set_lock); | |
5119 | return 0; | |
5120 | } | |
5121 | ||
182446d0 | 5122 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5123 | { |
182446d0 | 5124 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
5125 | } |
5126 | ||
5127 | static struct cftype debug_files[] = { | |
fe693435 PM |
5128 | { |
5129 | .name = "taskcount", | |
5130 | .read_u64 = debug_taskcount_read, | |
5131 | }, | |
5132 | ||
5133 | { | |
5134 | .name = "current_css_set", | |
5135 | .read_u64 = current_css_set_read, | |
5136 | }, | |
5137 | ||
5138 | { | |
5139 | .name = "current_css_set_refcount", | |
5140 | .read_u64 = current_css_set_refcount_read, | |
5141 | }, | |
5142 | ||
7717f7ba PM |
5143 | { |
5144 | .name = "current_css_set_cg_links", | |
2da8ca82 | 5145 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
5146 | }, |
5147 | ||
5148 | { | |
5149 | .name = "cgroup_css_links", | |
2da8ca82 | 5150 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
5151 | }, |
5152 | ||
fe693435 PM |
5153 | { |
5154 | .name = "releasable", | |
5155 | .read_u64 = releasable_read, | |
5156 | }, | |
fe693435 | 5157 | |
4baf6e33 TH |
5158 | { } /* terminate */ |
5159 | }; | |
fe693435 | 5160 | |
073219e9 | 5161 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
5162 | .css_alloc = debug_css_alloc, |
5163 | .css_free = debug_css_free, | |
4baf6e33 | 5164 | .base_cftypes = debug_files, |
fe693435 PM |
5165 | }; |
5166 | #endif /* CONFIG_CGROUP_DEBUG */ |