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> | |
ddbcc7e8 | 43 | #include <linux/slab.h> |
ddbcc7e8 PM |
44 | #include <linux/spinlock.h> |
45 | #include <linux/string.h> | |
bbcb81d0 | 46 | #include <linux/sort.h> |
81a6a5cd | 47 | #include <linux/kmod.h> |
846c7bb0 BS |
48 | #include <linux/delayacct.h> |
49 | #include <linux/cgroupstats.h> | |
0ac801fe | 50 | #include <linux/hashtable.h> |
096b7fe0 | 51 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 52 | #include <linux/idr.h> |
d1d9fd33 | 53 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
081aa458 | 54 | #include <linux/flex_array.h> /* used in cgroup_attach_task */ |
c4c27fbd | 55 | #include <linux/kthread.h> |
846c7bb0 | 56 | |
60063497 | 57 | #include <linux/atomic.h> |
ddbcc7e8 | 58 | |
b1a21367 TH |
59 | /* |
60 | * pidlists linger the following amount before being destroyed. The goal | |
61 | * is avoiding frequent destruction in the middle of consecutive read calls | |
62 | * Expiring in the middle is a performance problem not a correctness one. | |
63 | * 1 sec should be enough. | |
64 | */ | |
65 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
66 | ||
8d7e6fb0 TH |
67 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
68 | MAX_CFTYPE_NAME + 2) | |
69 | ||
ace2bee8 TH |
70 | /* |
71 | * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file | |
72 | * creation/removal and hierarchy changing operations including cgroup | |
73 | * creation, removal, css association and controller rebinding. This outer | |
74 | * lock is needed mainly to resolve the circular dependency between kernfs | |
75 | * active ref and cgroup_mutex. cgroup_tree_mutex nests above both. | |
76 | */ | |
77 | static DEFINE_MUTEX(cgroup_tree_mutex); | |
78 | ||
e25e2cbb TH |
79 | /* |
80 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
81 | * hierarchy must be performed while holding it. | |
e25e2cbb | 82 | */ |
2219449a TH |
83 | #ifdef CONFIG_PROVE_RCU |
84 | DEFINE_MUTEX(cgroup_mutex); | |
8af01f56 | 85 | EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for lockdep */ |
2219449a | 86 | #else |
81a6a5cd | 87 | static DEFINE_MUTEX(cgroup_mutex); |
2219449a TH |
88 | #endif |
89 | ||
69e943b7 TH |
90 | /* |
91 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
92 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
93 | */ | |
94 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
95 | ||
ace2bee8 | 96 | #define cgroup_assert_mutexes_or_rcu_locked() \ |
87fb54f1 | 97 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
ace2bee8 | 98 | lockdep_is_held(&cgroup_tree_mutex) || \ |
87fb54f1 | 99 | lockdep_is_held(&cgroup_mutex), \ |
ace2bee8 | 100 | "cgroup_[tree_]mutex or RCU read lock required"); |
87fb54f1 | 101 | |
e5fca243 TH |
102 | /* |
103 | * cgroup destruction makes heavy use of work items and there can be a lot | |
104 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
105 | * destruction work items don't end up filling up max_active of system_wq | |
106 | * which may lead to deadlock. | |
107 | */ | |
108 | static struct workqueue_struct *cgroup_destroy_wq; | |
109 | ||
b1a21367 TH |
110 | /* |
111 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
112 | * separate workqueue as flush domain. | |
113 | */ | |
114 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
115 | ||
3ed80a62 | 116 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 117 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 118 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
119 | #include <linux/cgroup_subsys.h> |
120 | }; | |
073219e9 TH |
121 | #undef SUBSYS |
122 | ||
123 | /* array of cgroup subsystem names */ | |
124 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
125 | static const char *cgroup_subsys_name[] = { | |
126 | #include <linux/cgroup_subsys.h> | |
127 | }; | |
128 | #undef SUBSYS | |
ddbcc7e8 | 129 | |
ddbcc7e8 | 130 | /* |
9871bf95 TH |
131 | * The dummy hierarchy, reserved for the subsystems that are otherwise |
132 | * unattached - it never has more than a single cgroup, and all tasks are | |
133 | * part of that cgroup. | |
ddbcc7e8 | 134 | */ |
9871bf95 TH |
135 | static struct cgroupfs_root cgroup_dummy_root; |
136 | ||
137 | /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */ | |
138 | static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup; | |
ddbcc7e8 PM |
139 | |
140 | /* The list of hierarchy roots */ | |
141 | ||
9871bf95 TH |
142 | static LIST_HEAD(cgroup_roots); |
143 | static int cgroup_root_count; | |
ddbcc7e8 | 144 | |
3417ae1f | 145 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 146 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 147 | |
794611a1 LZ |
148 | /* |
149 | * Assign a monotonically increasing serial number to cgroups. It | |
150 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
151 | * numbers. Also, as cgroups are always appended to the parent's | |
152 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
153 | * the ascending serial number order on the list. Protected by |
154 | * cgroup_mutex. | |
794611a1 | 155 | */ |
00356bd5 | 156 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 157 | |
ddbcc7e8 | 158 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
159 | * check for fork/exit handlers to call. This avoids us having to do |
160 | * extra work in the fork/exit path if none of the subsystems need to | |
161 | * be called. | |
ddbcc7e8 | 162 | */ |
8947f9d5 | 163 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 164 | |
628f7cd4 TH |
165 | static struct cftype cgroup_base_files[]; |
166 | ||
59f5296b | 167 | static void cgroup_put(struct cgroup *cgrp); |
f2e85d57 TH |
168 | static int rebind_subsystems(struct cgroupfs_root *root, |
169 | unsigned long added_mask, unsigned removed_mask); | |
f20104de | 170 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 171 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
2bb566cb TH |
172 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
173 | bool is_add); | |
b1a21367 | 174 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 175 | |
95109b62 TH |
176 | /** |
177 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
178 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 179 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 180 | * |
ca8bdcaf TH |
181 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
182 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
183 | * the caller is responsible for pinning the returned css if it wants to | |
184 | * keep accessing it outside the said locks. This function may return | |
185 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
186 | */ |
187 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 188 | struct cgroup_subsys *ss) |
95109b62 | 189 | { |
ca8bdcaf | 190 | if (ss) |
aec25020 | 191 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 TH |
192 | lockdep_is_held(&cgroup_tree_mutex) || |
193 | lockdep_is_held(&cgroup_mutex)); | |
ca8bdcaf TH |
194 | else |
195 | return &cgrp->dummy_css; | |
95109b62 | 196 | } |
42809dd4 | 197 | |
ddbcc7e8 | 198 | /* convenient tests for these bits */ |
54766d4a | 199 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 200 | { |
54766d4a | 201 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
202 | } |
203 | ||
59f5296b TH |
204 | struct cgroup_subsys_state *seq_css(struct seq_file *seq) |
205 | { | |
2bd59d48 TH |
206 | struct kernfs_open_file *of = seq->private; |
207 | struct cgroup *cgrp = of->kn->parent->priv; | |
208 | struct cftype *cft = seq_cft(seq); | |
209 | ||
210 | /* | |
211 | * This is open and unprotected implementation of cgroup_css(). | |
212 | * seq_css() is only called from a kernfs file operation which has | |
213 | * an active reference on the file. Because all the subsystem | |
214 | * files are drained before a css is disassociated with a cgroup, | |
215 | * the matching css from the cgroup's subsys table is guaranteed to | |
216 | * be and stay valid until the enclosing operation is complete. | |
217 | */ | |
218 | if (cft->ss) | |
219 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
220 | else | |
221 | return &cgrp->dummy_css; | |
59f5296b TH |
222 | } |
223 | EXPORT_SYMBOL_GPL(seq_css); | |
224 | ||
78574cf9 LZ |
225 | /** |
226 | * cgroup_is_descendant - test ancestry | |
227 | * @cgrp: the cgroup to be tested | |
228 | * @ancestor: possible ancestor of @cgrp | |
229 | * | |
230 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
231 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
232 | * and @ancestor are accessible. | |
233 | */ | |
234 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
235 | { | |
236 | while (cgrp) { | |
237 | if (cgrp == ancestor) | |
238 | return true; | |
239 | cgrp = cgrp->parent; | |
240 | } | |
241 | return false; | |
242 | } | |
243 | EXPORT_SYMBOL_GPL(cgroup_is_descendant); | |
ddbcc7e8 | 244 | |
e9685a03 | 245 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
246 | { |
247 | const int bits = | |
bd89aabc PM |
248 | (1 << CGRP_RELEASABLE) | |
249 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
250 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
251 | } |
252 | ||
e9685a03 | 253 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 254 | { |
bd89aabc | 255 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
256 | } |
257 | ||
1c6727af TH |
258 | /** |
259 | * for_each_css - iterate all css's of a cgroup | |
260 | * @css: the iteration cursor | |
261 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
262 | * @cgrp: the target cgroup to iterate css's of | |
263 | * | |
264 | * Should be called under cgroup_mutex. | |
265 | */ | |
266 | #define for_each_css(css, ssid, cgrp) \ | |
267 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
268 | if (!((css) = rcu_dereference_check( \ | |
269 | (cgrp)->subsys[(ssid)], \ | |
ace2bee8 | 270 | lockdep_is_held(&cgroup_tree_mutex) || \ |
1c6727af TH |
271 | lockdep_is_held(&cgroup_mutex)))) { } \ |
272 | else | |
273 | ||
30159ec7 | 274 | /** |
3ed80a62 | 275 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 276 | * @ss: the iteration cursor |
780cd8b3 | 277 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 278 | */ |
780cd8b3 | 279 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
280 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
281 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 282 | |
5549c497 TH |
283 | /* iterate across the active hierarchies */ |
284 | #define for_each_active_root(root) \ | |
285 | list_for_each_entry((root), &cgroup_roots, root_list) | |
ddbcc7e8 | 286 | |
7ae1bad9 TH |
287 | /** |
288 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
289 | * @cgrp: the cgroup to be checked for liveness | |
290 | * | |
47cfcd09 TH |
291 | * On success, returns true; the mutex should be later unlocked. On |
292 | * failure returns false with no lock held. | |
7ae1bad9 | 293 | */ |
b9777cf8 | 294 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
295 | { |
296 | mutex_lock(&cgroup_mutex); | |
54766d4a | 297 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
298 | mutex_unlock(&cgroup_mutex); |
299 | return false; | |
300 | } | |
301 | return true; | |
302 | } | |
7ae1bad9 | 303 | |
81a6a5cd PM |
304 | /* the list of cgroups eligible for automatic release. Protected by |
305 | * release_list_lock */ | |
306 | static LIST_HEAD(release_list); | |
cdcc136f | 307 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
308 | static void cgroup_release_agent(struct work_struct *work); |
309 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 310 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 311 | |
69d0206c TH |
312 | /* |
313 | * A cgroup can be associated with multiple css_sets as different tasks may | |
314 | * belong to different cgroups on different hierarchies. In the other | |
315 | * direction, a css_set is naturally associated with multiple cgroups. | |
316 | * This M:N relationship is represented by the following link structure | |
317 | * which exists for each association and allows traversing the associations | |
318 | * from both sides. | |
319 | */ | |
320 | struct cgrp_cset_link { | |
321 | /* the cgroup and css_set this link associates */ | |
322 | struct cgroup *cgrp; | |
323 | struct css_set *cset; | |
324 | ||
325 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
326 | struct list_head cset_link; | |
327 | ||
328 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
329 | struct list_head cgrp_link; | |
817929ec PM |
330 | }; |
331 | ||
332 | /* The default css_set - used by init and its children prior to any | |
333 | * hierarchies being mounted. It contains a pointer to the root state | |
334 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
335 | * reference-counted, to improve performance when child cgroups | |
336 | * haven't been created. | |
337 | */ | |
338 | ||
339 | static struct css_set init_css_set; | |
69d0206c | 340 | static struct cgrp_cset_link init_cgrp_cset_link; |
817929ec | 341 | |
0942eeee TH |
342 | /* |
343 | * css_set_lock protects the list of css_set objects, and the chain of | |
344 | * tasks off each css_set. Nests outside task->alloc_lock due to | |
72ec7029 | 345 | * css_task_iter_start(). |
0942eeee | 346 | */ |
817929ec PM |
347 | static DEFINE_RWLOCK(css_set_lock); |
348 | static int css_set_count; | |
349 | ||
7717f7ba PM |
350 | /* |
351 | * hash table for cgroup groups. This improves the performance to find | |
352 | * an existing css_set. This hash doesn't (currently) take into | |
353 | * account cgroups in empty hierarchies. | |
354 | */ | |
472b1053 | 355 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 356 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 357 | |
0ac801fe | 358 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 359 | { |
0ac801fe | 360 | unsigned long key = 0UL; |
30159ec7 TH |
361 | struct cgroup_subsys *ss; |
362 | int i; | |
472b1053 | 363 | |
30159ec7 | 364 | for_each_subsys(ss, i) |
0ac801fe LZ |
365 | key += (unsigned long)css[i]; |
366 | key = (key >> 16) ^ key; | |
472b1053 | 367 | |
0ac801fe | 368 | return key; |
472b1053 LZ |
369 | } |
370 | ||
0942eeee TH |
371 | /* |
372 | * We don't maintain the lists running through each css_set to its task | |
72ec7029 TH |
373 | * until after the first call to css_task_iter_start(). This reduces the |
374 | * fork()/exit() overhead for people who have cgroups compiled into their | |
375 | * kernel but not actually in use. | |
0942eeee | 376 | */ |
8947f9d5 | 377 | static int use_task_css_set_links __read_mostly; |
817929ec | 378 | |
5abb8855 | 379 | static void __put_css_set(struct css_set *cset, int taskexit) |
b4f48b63 | 380 | { |
69d0206c | 381 | struct cgrp_cset_link *link, *tmp_link; |
5abb8855 | 382 | |
146aa1bd LJ |
383 | /* |
384 | * Ensure that the refcount doesn't hit zero while any readers | |
385 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
386 | * rwlock | |
387 | */ | |
5abb8855 | 388 | if (atomic_add_unless(&cset->refcount, -1, 1)) |
146aa1bd LJ |
389 | return; |
390 | write_lock(&css_set_lock); | |
5abb8855 | 391 | if (!atomic_dec_and_test(&cset->refcount)) { |
146aa1bd LJ |
392 | write_unlock(&css_set_lock); |
393 | return; | |
394 | } | |
81a6a5cd | 395 | |
2c6ab6d2 | 396 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
5abb8855 | 397 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
398 | css_set_count--; |
399 | ||
69d0206c | 400 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 401 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 402 | |
69d0206c TH |
403 | list_del(&link->cset_link); |
404 | list_del(&link->cgrp_link); | |
71b5707e | 405 | |
ddd69148 | 406 | /* @cgrp can't go away while we're holding css_set_lock */ |
6f3d828f | 407 | if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { |
81a6a5cd | 408 | if (taskexit) |
bd89aabc PM |
409 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
410 | check_for_release(cgrp); | |
81a6a5cd | 411 | } |
2c6ab6d2 PM |
412 | |
413 | kfree(link); | |
81a6a5cd | 414 | } |
2c6ab6d2 PM |
415 | |
416 | write_unlock(&css_set_lock); | |
5abb8855 | 417 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
418 | } |
419 | ||
817929ec PM |
420 | /* |
421 | * refcounted get/put for css_set objects | |
422 | */ | |
5abb8855 | 423 | static inline void get_css_set(struct css_set *cset) |
817929ec | 424 | { |
5abb8855 | 425 | atomic_inc(&cset->refcount); |
817929ec PM |
426 | } |
427 | ||
5abb8855 | 428 | static inline void put_css_set(struct css_set *cset) |
817929ec | 429 | { |
5abb8855 | 430 | __put_css_set(cset, 0); |
817929ec PM |
431 | } |
432 | ||
5abb8855 | 433 | static inline void put_css_set_taskexit(struct css_set *cset) |
81a6a5cd | 434 | { |
5abb8855 | 435 | __put_css_set(cset, 1); |
81a6a5cd PM |
436 | } |
437 | ||
b326f9d0 | 438 | /** |
7717f7ba | 439 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
440 | * @cset: candidate css_set being tested |
441 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
442 | * @new_cgrp: cgroup that's being entered by the task |
443 | * @template: desired set of css pointers in css_set (pre-calculated) | |
444 | * | |
6f4b7e63 | 445 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
446 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
447 | */ | |
5abb8855 TH |
448 | static bool compare_css_sets(struct css_set *cset, |
449 | struct css_set *old_cset, | |
7717f7ba PM |
450 | struct cgroup *new_cgrp, |
451 | struct cgroup_subsys_state *template[]) | |
452 | { | |
453 | struct list_head *l1, *l2; | |
454 | ||
5abb8855 | 455 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) { |
7717f7ba PM |
456 | /* Not all subsystems matched */ |
457 | return false; | |
458 | } | |
459 | ||
460 | /* | |
461 | * Compare cgroup pointers in order to distinguish between | |
462 | * different cgroups in heirarchies with no subsystems. We | |
463 | * could get by with just this check alone (and skip the | |
464 | * memcmp above) but on most setups the memcmp check will | |
465 | * avoid the need for this more expensive check on almost all | |
466 | * candidates. | |
467 | */ | |
468 | ||
69d0206c TH |
469 | l1 = &cset->cgrp_links; |
470 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 471 | while (1) { |
69d0206c | 472 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 473 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
474 | |
475 | l1 = l1->next; | |
476 | l2 = l2->next; | |
477 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
478 | if (l1 == &cset->cgrp_links) { |
479 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
480 | break; |
481 | } else { | |
69d0206c | 482 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
483 | } |
484 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
485 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
486 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
487 | cgrp1 = link1->cgrp; | |
488 | cgrp2 = link2->cgrp; | |
7717f7ba | 489 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 490 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
491 | |
492 | /* | |
493 | * If this hierarchy is the hierarchy of the cgroup | |
494 | * that's changing, then we need to check that this | |
495 | * css_set points to the new cgroup; if it's any other | |
496 | * hierarchy, then this css_set should point to the | |
497 | * same cgroup as the old css_set. | |
498 | */ | |
5abb8855 TH |
499 | if (cgrp1->root == new_cgrp->root) { |
500 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
501 | return false; |
502 | } else { | |
5abb8855 | 503 | if (cgrp1 != cgrp2) |
7717f7ba PM |
504 | return false; |
505 | } | |
506 | } | |
507 | return true; | |
508 | } | |
509 | ||
b326f9d0 TH |
510 | /** |
511 | * find_existing_css_set - init css array and find the matching css_set | |
512 | * @old_cset: the css_set that we're using before the cgroup transition | |
513 | * @cgrp: the cgroup that we're moving into | |
514 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 515 | */ |
5abb8855 TH |
516 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
517 | struct cgroup *cgrp, | |
518 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 519 | { |
bd89aabc | 520 | struct cgroupfs_root *root = cgrp->root; |
30159ec7 | 521 | struct cgroup_subsys *ss; |
5abb8855 | 522 | struct css_set *cset; |
0ac801fe | 523 | unsigned long key; |
b326f9d0 | 524 | int i; |
817929ec | 525 | |
aae8aab4 BB |
526 | /* |
527 | * Build the set of subsystem state objects that we want to see in the | |
528 | * new css_set. while subsystems can change globally, the entries here | |
529 | * won't change, so no need for locking. | |
530 | */ | |
30159ec7 | 531 | for_each_subsys(ss, i) { |
a1a71b45 | 532 | if (root->subsys_mask & (1UL << i)) { |
817929ec PM |
533 | /* Subsystem is in this hierarchy. So we want |
534 | * the subsystem state from the new | |
535 | * cgroup */ | |
ca8bdcaf | 536 | template[i] = cgroup_css(cgrp, ss); |
817929ec PM |
537 | } else { |
538 | /* Subsystem is not in this hierarchy, so we | |
539 | * don't want to change the subsystem state */ | |
5abb8855 | 540 | template[i] = old_cset->subsys[i]; |
817929ec PM |
541 | } |
542 | } | |
543 | ||
0ac801fe | 544 | key = css_set_hash(template); |
5abb8855 TH |
545 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
546 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
547 | continue; |
548 | ||
549 | /* This css_set matches what we need */ | |
5abb8855 | 550 | return cset; |
472b1053 | 551 | } |
817929ec PM |
552 | |
553 | /* No existing cgroup group matched */ | |
554 | return NULL; | |
555 | } | |
556 | ||
69d0206c | 557 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 558 | { |
69d0206c | 559 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 560 | |
69d0206c TH |
561 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
562 | list_del(&link->cset_link); | |
36553434 LZ |
563 | kfree(link); |
564 | } | |
565 | } | |
566 | ||
69d0206c TH |
567 | /** |
568 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
569 | * @count: the number of links to allocate | |
570 | * @tmp_links: list_head the allocated links are put on | |
571 | * | |
572 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
573 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 574 | */ |
69d0206c | 575 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 576 | { |
69d0206c | 577 | struct cgrp_cset_link *link; |
817929ec | 578 | int i; |
69d0206c TH |
579 | |
580 | INIT_LIST_HEAD(tmp_links); | |
581 | ||
817929ec | 582 | for (i = 0; i < count; i++) { |
f4f4be2b | 583 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 584 | if (!link) { |
69d0206c | 585 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
586 | return -ENOMEM; |
587 | } | |
69d0206c | 588 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
589 | } |
590 | return 0; | |
591 | } | |
592 | ||
c12f65d4 LZ |
593 | /** |
594 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 595 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 596 | * @cset: the css_set to be linked |
c12f65d4 LZ |
597 | * @cgrp: the destination cgroup |
598 | */ | |
69d0206c TH |
599 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
600 | struct cgroup *cgrp) | |
c12f65d4 | 601 | { |
69d0206c | 602 | struct cgrp_cset_link *link; |
c12f65d4 | 603 | |
69d0206c TH |
604 | BUG_ON(list_empty(tmp_links)); |
605 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); | |
606 | link->cset = cset; | |
7717f7ba | 607 | link->cgrp = cgrp; |
69d0206c | 608 | list_move(&link->cset_link, &cgrp->cset_links); |
7717f7ba PM |
609 | /* |
610 | * Always add links to the tail of the list so that the list | |
611 | * is sorted by order of hierarchy creation | |
612 | */ | |
69d0206c | 613 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
614 | } |
615 | ||
b326f9d0 TH |
616 | /** |
617 | * find_css_set - return a new css_set with one cgroup updated | |
618 | * @old_cset: the baseline css_set | |
619 | * @cgrp: the cgroup to be updated | |
620 | * | |
621 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
622 | * substituted into the appropriate hierarchy. | |
817929ec | 623 | */ |
5abb8855 TH |
624 | static struct css_set *find_css_set(struct css_set *old_cset, |
625 | struct cgroup *cgrp) | |
817929ec | 626 | { |
b326f9d0 | 627 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 628 | struct css_set *cset; |
69d0206c TH |
629 | struct list_head tmp_links; |
630 | struct cgrp_cset_link *link; | |
0ac801fe | 631 | unsigned long key; |
472b1053 | 632 | |
b326f9d0 TH |
633 | lockdep_assert_held(&cgroup_mutex); |
634 | ||
817929ec PM |
635 | /* First see if we already have a cgroup group that matches |
636 | * the desired set */ | |
7e9abd89 | 637 | read_lock(&css_set_lock); |
5abb8855 TH |
638 | cset = find_existing_css_set(old_cset, cgrp, template); |
639 | if (cset) | |
640 | get_css_set(cset); | |
7e9abd89 | 641 | read_unlock(&css_set_lock); |
817929ec | 642 | |
5abb8855 TH |
643 | if (cset) |
644 | return cset; | |
817929ec | 645 | |
f4f4be2b | 646 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 647 | if (!cset) |
817929ec PM |
648 | return NULL; |
649 | ||
69d0206c | 650 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 651 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 652 | kfree(cset); |
817929ec PM |
653 | return NULL; |
654 | } | |
655 | ||
5abb8855 | 656 | atomic_set(&cset->refcount, 1); |
69d0206c | 657 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 TH |
658 | INIT_LIST_HEAD(&cset->tasks); |
659 | INIT_HLIST_NODE(&cset->hlist); | |
817929ec PM |
660 | |
661 | /* Copy the set of subsystem state objects generated in | |
662 | * find_existing_css_set() */ | |
5abb8855 | 663 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec PM |
664 | |
665 | write_lock(&css_set_lock); | |
666 | /* Add reference counts and links from the new css_set. */ | |
69d0206c | 667 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 668 | struct cgroup *c = link->cgrp; |
69d0206c | 669 | |
7717f7ba PM |
670 | if (c->root == cgrp->root) |
671 | c = cgrp; | |
69d0206c | 672 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 673 | } |
817929ec | 674 | |
69d0206c | 675 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 676 | |
817929ec | 677 | css_set_count++; |
472b1053 LZ |
678 | |
679 | /* Add this cgroup group to the hash table */ | |
5abb8855 TH |
680 | key = css_set_hash(cset->subsys); |
681 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 682 | |
817929ec PM |
683 | write_unlock(&css_set_lock); |
684 | ||
5abb8855 | 685 | return cset; |
b4f48b63 PM |
686 | } |
687 | ||
2bd59d48 TH |
688 | static struct cgroupfs_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
689 | { | |
690 | struct cgroup *top_cgrp = kf_root->kn->priv; | |
691 | ||
692 | return top_cgrp->root; | |
693 | } | |
694 | ||
f2e85d57 TH |
695 | static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end) |
696 | { | |
697 | int id; | |
698 | ||
699 | lockdep_assert_held(&cgroup_mutex); | |
700 | ||
701 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end, | |
702 | GFP_KERNEL); | |
703 | if (id < 0) | |
704 | return id; | |
705 | ||
706 | root->hierarchy_id = id; | |
707 | return 0; | |
708 | } | |
709 | ||
710 | static void cgroup_exit_root_id(struct cgroupfs_root *root) | |
711 | { | |
712 | lockdep_assert_held(&cgroup_mutex); | |
713 | ||
714 | if (root->hierarchy_id) { | |
715 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
716 | root->hierarchy_id = 0; | |
717 | } | |
718 | } | |
719 | ||
720 | static void cgroup_free_root(struct cgroupfs_root *root) | |
721 | { | |
722 | if (root) { | |
723 | /* hierarhcy ID shoulid already have been released */ | |
724 | WARN_ON_ONCE(root->hierarchy_id); | |
725 | ||
726 | idr_destroy(&root->cgroup_idr); | |
727 | kfree(root); | |
728 | } | |
729 | } | |
730 | ||
59f5296b TH |
731 | static void cgroup_get_root(struct cgroupfs_root *root) |
732 | { | |
2bd59d48 TH |
733 | /* |
734 | * The caller must ensure that @root is alive, which can be | |
735 | * achieved by holding a ref on one of the member cgroups or | |
736 | * following a registered reference to @root while holding | |
737 | * cgroup_tree_mutex. | |
738 | */ | |
739 | WARN_ON_ONCE(atomic_read(&root->refcnt) <= 0); | |
740 | atomic_inc(&root->refcnt); | |
59f5296b TH |
741 | } |
742 | ||
743 | static void cgroup_put_root(struct cgroupfs_root *root) | |
744 | { | |
f2e85d57 TH |
745 | struct cgroup *cgrp = &root->top_cgroup; |
746 | struct cgrp_cset_link *link, *tmp_link; | |
747 | int ret; | |
748 | ||
2bd59d48 TH |
749 | /* |
750 | * @root's refcnt reaching zero and its deregistration should be | |
751 | * atomic w.r.t. cgroup_tree_mutex. This ensures that | |
752 | * cgroup_get_root() is safe to invoke if @root is registered. | |
753 | */ | |
754 | mutex_lock(&cgroup_tree_mutex); | |
755 | if (!atomic_dec_and_test(&root->refcnt)) { | |
756 | mutex_unlock(&cgroup_tree_mutex); | |
757 | return; | |
758 | } | |
759 | mutex_lock(&cgroup_mutex); | |
f2e85d57 TH |
760 | |
761 | BUG_ON(root->number_of_cgroups != 1); | |
762 | BUG_ON(!list_empty(&cgrp->children)); | |
763 | ||
f2e85d57 TH |
764 | /* Rebind all subsystems back to the default hierarchy */ |
765 | if (root->flags & CGRP_ROOT_SUBSYS_BOUND) { | |
766 | ret = rebind_subsystems(root, 0, root->subsys_mask); | |
767 | /* Shouldn't be able to fail ... */ | |
768 | BUG_ON(ret); | |
769 | } | |
770 | ||
771 | /* | |
772 | * Release all the links from cset_links to this hierarchy's | |
773 | * root cgroup | |
774 | */ | |
775 | write_lock(&css_set_lock); | |
776 | ||
777 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
778 | list_del(&link->cset_link); | |
779 | list_del(&link->cgrp_link); | |
780 | kfree(link); | |
781 | } | |
782 | write_unlock(&css_set_lock); | |
783 | ||
784 | if (!list_empty(&root->root_list)) { | |
785 | list_del(&root->root_list); | |
786 | cgroup_root_count--; | |
787 | } | |
788 | ||
789 | cgroup_exit_root_id(root); | |
790 | ||
791 | mutex_unlock(&cgroup_mutex); | |
792 | mutex_unlock(&cgroup_tree_mutex); | |
f2e85d57 | 793 | |
2bd59d48 | 794 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
795 | cgroup_free_root(root); |
796 | } | |
797 | ||
7717f7ba PM |
798 | /* |
799 | * Return the cgroup for "task" from the given hierarchy. Must be | |
800 | * called with cgroup_mutex held. | |
801 | */ | |
802 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
803 | struct cgroupfs_root *root) | |
804 | { | |
5abb8855 | 805 | struct css_set *cset; |
7717f7ba PM |
806 | struct cgroup *res = NULL; |
807 | ||
808 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
809 | read_lock(&css_set_lock); | |
810 | /* | |
811 | * No need to lock the task - since we hold cgroup_mutex the | |
812 | * task can't change groups, so the only thing that can happen | |
813 | * is that it exits and its css is set back to init_css_set. | |
814 | */ | |
a8ad805c | 815 | cset = task_css_set(task); |
5abb8855 | 816 | if (cset == &init_css_set) { |
7717f7ba PM |
817 | res = &root->top_cgroup; |
818 | } else { | |
69d0206c TH |
819 | struct cgrp_cset_link *link; |
820 | ||
821 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 822 | struct cgroup *c = link->cgrp; |
69d0206c | 823 | |
7717f7ba PM |
824 | if (c->root == root) { |
825 | res = c; | |
826 | break; | |
827 | } | |
828 | } | |
829 | } | |
830 | read_unlock(&css_set_lock); | |
831 | BUG_ON(!res); | |
832 | return res; | |
833 | } | |
834 | ||
ddbcc7e8 PM |
835 | /* |
836 | * There is one global cgroup mutex. We also require taking | |
837 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
838 | * See "The task_lock() exception", at the end of this comment. | |
839 | * | |
840 | * A task must hold cgroup_mutex to modify cgroups. | |
841 | * | |
842 | * Any task can increment and decrement the count field without lock. | |
843 | * So in general, code holding cgroup_mutex can't rely on the count | |
844 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 845 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
846 | * means that no tasks are currently attached, therefore there is no |
847 | * way a task attached to that cgroup can fork (the other way to | |
848 | * increment the count). So code holding cgroup_mutex can safely | |
849 | * assume that if the count is zero, it will stay zero. Similarly, if | |
850 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
851 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
852 | * needs that mutex. | |
853 | * | |
ddbcc7e8 PM |
854 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
855 | * (usually) take cgroup_mutex. These are the two most performance | |
856 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
857 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
858 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
859 | * to the release agent with the name of the cgroup (path relative to |
860 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
861 | * |
862 | * A cgroup can only be deleted if both its 'count' of using tasks | |
863 | * is zero, and its list of 'children' cgroups is empty. Since all | |
864 | * tasks in the system use _some_ cgroup, and since there is always at | |
865 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
866 | * always has either children cgroups and/or using tasks. So we don't | |
867 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
868 | * | |
869 | * The task_lock() exception | |
870 | * | |
871 | * The need for this exception arises from the action of | |
d0b2fdd2 | 872 | * cgroup_attach_task(), which overwrites one task's cgroup pointer with |
a043e3b2 | 873 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
874 | * several performance critical places that need to reference |
875 | * task->cgroup without the expense of grabbing a system global | |
876 | * mutex. Therefore except as noted below, when dereferencing or, as | |
d0b2fdd2 | 877 | * in cgroup_attach_task(), modifying a task's cgroup pointer we use |
ddbcc7e8 PM |
878 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
879 | * the task_struct routinely used for such matters. | |
880 | * | |
881 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 882 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
883 | */ |
884 | ||
628f7cd4 | 885 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
2bd59d48 | 886 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 887 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 888 | |
8d7e6fb0 TH |
889 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
890 | char *buf) | |
891 | { | |
892 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && | |
893 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
894 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
895 | cft->ss->name, cft->name); | |
896 | else | |
897 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
898 | return buf; | |
899 | } | |
900 | ||
f2e85d57 TH |
901 | /** |
902 | * cgroup_file_mode - deduce file mode of a control file | |
903 | * @cft: the control file in question | |
904 | * | |
905 | * returns cft->mode if ->mode is not 0 | |
906 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
907 | * returns S_IRUGO if it has only a read handler | |
908 | * returns S_IWUSR if it has only a write hander | |
909 | */ | |
910 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
911 | { | |
912 | umode_t mode = 0; | |
913 | ||
914 | if (cft->mode) | |
915 | return cft->mode; | |
916 | ||
917 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
918 | mode |= S_IRUGO; | |
919 | ||
920 | if (cft->write_u64 || cft->write_s64 || cft->write_string || | |
921 | cft->trigger) | |
922 | mode |= S_IWUSR; | |
923 | ||
924 | return mode; | |
925 | } | |
926 | ||
be445626 LZ |
927 | static void cgroup_free_fn(struct work_struct *work) |
928 | { | |
ea15f8cc | 929 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 LZ |
930 | |
931 | mutex_lock(&cgroup_mutex); | |
be445626 LZ |
932 | cgrp->root->number_of_cgroups--; |
933 | mutex_unlock(&cgroup_mutex); | |
934 | ||
415cf07a | 935 | /* |
59f5296b TH |
936 | * We get a ref to the parent, and put the ref when this cgroup is |
937 | * being freed, so it's guaranteed that the parent won't be | |
938 | * destroyed before its children. | |
415cf07a | 939 | */ |
59f5296b | 940 | cgroup_put(cgrp->parent); |
415cf07a | 941 | |
59f5296b TH |
942 | /* put the root reference that we took when we created the cgroup */ |
943 | cgroup_put_root(cgrp->root); | |
be445626 | 944 | |
b1a21367 | 945 | cgroup_pidlist_destroy_all(cgrp); |
be445626 | 946 | |
6f30558f | 947 | kernfs_put(cgrp->kn); |
be445626 LZ |
948 | kfree(cgrp); |
949 | } | |
950 | ||
951 | static void cgroup_free_rcu(struct rcu_head *head) | |
952 | { | |
953 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
954 | ||
ea15f8cc | 955 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 956 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
957 | } |
958 | ||
59f5296b TH |
959 | static void cgroup_get(struct cgroup *cgrp) |
960 | { | |
2bd59d48 TH |
961 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
962 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
963 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
964 | } |
965 | ||
59f5296b TH |
966 | static void cgroup_put(struct cgroup *cgrp) |
967 | { | |
2bd59d48 TH |
968 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
969 | return; | |
970 | if (WARN_ON_ONCE(!cgroup_is_dead(cgrp))) | |
971 | return; | |
59f5296b | 972 | |
2bd59d48 TH |
973 | /* |
974 | * XXX: cgrp->id is only used to look up css's. As cgroup and | |
975 | * css's lifetimes will be decoupled, it should be made | |
976 | * per-subsystem and moved to css->id so that lookups are | |
977 | * successful until the target css is released. | |
978 | */ | |
979 | mutex_lock(&cgroup_mutex); | |
980 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
981 | mutex_unlock(&cgroup_mutex); | |
982 | cgrp->id = -1; | |
ddbcc7e8 | 983 | |
2bd59d48 | 984 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 PM |
985 | } |
986 | ||
2739d3cc | 987 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 988 | { |
2bd59d48 | 989 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 990 | |
ace2bee8 | 991 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 992 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
993 | } |
994 | ||
13af07df | 995 | /** |
628f7cd4 | 996 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 997 | * @cgrp: target cgroup |
13af07df AR |
998 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
999 | */ | |
628f7cd4 | 1000 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 1001 | { |
13af07df | 1002 | struct cgroup_subsys *ss; |
b420ba7d | 1003 | int i; |
05ef1d7c | 1004 | |
b420ba7d | 1005 | for_each_subsys(ss, i) { |
0adb0704 | 1006 | struct cftype *cfts; |
b420ba7d TH |
1007 | |
1008 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 1009 | continue; |
0adb0704 TH |
1010 | list_for_each_entry(cfts, &ss->cfts, node) |
1011 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 1012 | } |
ddbcc7e8 PM |
1013 | } |
1014 | ||
ddbcc7e8 | 1015 | static int rebind_subsystems(struct cgroupfs_root *root, |
a8a648c4 | 1016 | unsigned long added_mask, unsigned removed_mask) |
ddbcc7e8 | 1017 | { |
bd89aabc | 1018 | struct cgroup *cgrp = &root->top_cgroup; |
30159ec7 | 1019 | struct cgroup_subsys *ss; |
3126121f | 1020 | int i, ret; |
ddbcc7e8 | 1021 | |
ace2bee8 TH |
1022 | lockdep_assert_held(&cgroup_tree_mutex); |
1023 | lockdep_assert_held(&cgroup_mutex); | |
aae8aab4 | 1024 | |
ddbcc7e8 | 1025 | /* Check that any added subsystems are currently free */ |
3ed80a62 TH |
1026 | for_each_subsys(ss, i) |
1027 | if ((added_mask & (1 << i)) && ss->root != &cgroup_dummy_root) | |
1028 | return -EBUSY; | |
ddbcc7e8 | 1029 | |
3126121f TH |
1030 | ret = cgroup_populate_dir(cgrp, added_mask); |
1031 | if (ret) | |
3ed80a62 | 1032 | return ret; |
3126121f TH |
1033 | |
1034 | /* | |
1035 | * Nothing can fail from this point on. Remove files for the | |
1036 | * removed subsystems and rebind each subsystem. | |
1037 | */ | |
4ac06017 | 1038 | mutex_unlock(&cgroup_mutex); |
3126121f | 1039 | cgroup_clear_dir(cgrp, removed_mask); |
4ac06017 | 1040 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1041 | |
30159ec7 | 1042 | for_each_subsys(ss, i) { |
ddbcc7e8 | 1043 | unsigned long bit = 1UL << i; |
30159ec7 | 1044 | |
a1a71b45 | 1045 | if (bit & added_mask) { |
ddbcc7e8 | 1046 | /* We're binding this subsystem to this hierarchy */ |
ca8bdcaf TH |
1047 | BUG_ON(cgroup_css(cgrp, ss)); |
1048 | BUG_ON(!cgroup_css(cgroup_dummy_top, ss)); | |
1049 | BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top); | |
a8a648c4 | 1050 | |
73e80ed8 | 1051 | rcu_assign_pointer(cgrp->subsys[i], |
ca8bdcaf TH |
1052 | cgroup_css(cgroup_dummy_top, ss)); |
1053 | cgroup_css(cgrp, ss)->cgroup = cgrp; | |
a8a648c4 | 1054 | |
b2aa30f7 | 1055 | ss->root = root; |
ddbcc7e8 | 1056 | if (ss->bind) |
ca8bdcaf | 1057 | ss->bind(cgroup_css(cgrp, ss)); |
a8a648c4 | 1058 | |
cf5d5941 | 1059 | /* refcount was already taken, and we're keeping it */ |
a8a648c4 | 1060 | root->subsys_mask |= bit; |
a1a71b45 | 1061 | } else if (bit & removed_mask) { |
ddbcc7e8 | 1062 | /* We're removing this subsystem */ |
ca8bdcaf TH |
1063 | BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss)); |
1064 | BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp); | |
a8a648c4 | 1065 | |
ddbcc7e8 | 1066 | if (ss->bind) |
ca8bdcaf | 1067 | ss->bind(cgroup_css(cgroup_dummy_top, ss)); |
73e80ed8 | 1068 | |
ca8bdcaf | 1069 | cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top; |
73e80ed8 TH |
1070 | RCU_INIT_POINTER(cgrp->subsys[i], NULL); |
1071 | ||
9871bf95 | 1072 | cgroup_subsys[i]->root = &cgroup_dummy_root; |
a8a648c4 | 1073 | root->subsys_mask &= ~bit; |
ddbcc7e8 PM |
1074 | } |
1075 | } | |
ddbcc7e8 | 1076 | |
1672d040 TH |
1077 | /* |
1078 | * Mark @root has finished binding subsystems. @root->subsys_mask | |
1079 | * now matches the bound subsystems. | |
1080 | */ | |
1081 | root->flags |= CGRP_ROOT_SUBSYS_BOUND; | |
2bd59d48 | 1082 | kernfs_activate(cgrp->kn); |
1672d040 | 1083 | |
ddbcc7e8 PM |
1084 | return 0; |
1085 | } | |
1086 | ||
2bd59d48 TH |
1087 | static int cgroup_show_options(struct seq_file *seq, |
1088 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1089 | { |
2bd59d48 | 1090 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1091 | struct cgroup_subsys *ss; |
b85d2040 | 1092 | int ssid; |
ddbcc7e8 | 1093 | |
b85d2040 TH |
1094 | for_each_subsys(ss, ssid) |
1095 | if (root->subsys_mask & (1 << ssid)) | |
1096 | seq_printf(seq, ",%s", ss->name); | |
873fe09e TH |
1097 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1098 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1099 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1100 | seq_puts(seq, ",noprefix"); |
93438629 | 1101 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1102 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1103 | |
1104 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1105 | if (strlen(root->release_agent_path)) |
1106 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1107 | spin_unlock(&release_agent_path_lock); |
1108 | ||
2260e7fc | 1109 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags)) |
97978e6d | 1110 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1111 | if (strlen(root->name)) |
1112 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1113 | return 0; |
1114 | } | |
1115 | ||
1116 | struct cgroup_sb_opts { | |
a1a71b45 | 1117 | unsigned long subsys_mask; |
ddbcc7e8 | 1118 | unsigned long flags; |
81a6a5cd | 1119 | char *release_agent; |
2260e7fc | 1120 | bool cpuset_clone_children; |
c6d57f33 | 1121 | char *name; |
2c6ab6d2 PM |
1122 | /* User explicitly requested empty subsystem */ |
1123 | bool none; | |
ddbcc7e8 PM |
1124 | }; |
1125 | ||
aae8aab4 | 1126 | /* |
9871bf95 TH |
1127 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1128 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1129 | * array. This function takes refcounts on subsystems to be used, unless it | |
1130 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1131 | */ |
cf5d5941 | 1132 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1133 | { |
32a8cf23 DL |
1134 | char *token, *o = data; |
1135 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1136 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1137 | struct cgroup_subsys *ss; |
1138 | int i; | |
f9ab5b5b | 1139 | |
aae8aab4 BB |
1140 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1141 | ||
f9ab5b5b | 1142 | #ifdef CONFIG_CPUSETS |
073219e9 | 1143 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1144 | #endif |
ddbcc7e8 | 1145 | |
c6d57f33 | 1146 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1147 | |
1148 | while ((token = strsep(&o, ",")) != NULL) { | |
1149 | if (!*token) | |
1150 | return -EINVAL; | |
32a8cf23 | 1151 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1152 | /* Explicitly have no subsystems */ |
1153 | opts->none = true; | |
32a8cf23 DL |
1154 | continue; |
1155 | } | |
1156 | if (!strcmp(token, "all")) { | |
1157 | /* Mutually exclusive option 'all' + subsystem name */ | |
1158 | if (one_ss) | |
1159 | return -EINVAL; | |
1160 | all_ss = true; | |
1161 | continue; | |
1162 | } | |
873fe09e TH |
1163 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1164 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1165 | continue; | |
1166 | } | |
32a8cf23 | 1167 | if (!strcmp(token, "noprefix")) { |
93438629 | 1168 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1169 | continue; |
1170 | } | |
1171 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1172 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1173 | continue; |
1174 | } | |
03b1cde6 | 1175 | if (!strcmp(token, "xattr")) { |
93438629 | 1176 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1177 | continue; |
1178 | } | |
32a8cf23 | 1179 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1180 | /* Specifying two release agents is forbidden */ |
1181 | if (opts->release_agent) | |
1182 | return -EINVAL; | |
c6d57f33 | 1183 | opts->release_agent = |
e400c285 | 1184 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1185 | if (!opts->release_agent) |
1186 | return -ENOMEM; | |
32a8cf23 DL |
1187 | continue; |
1188 | } | |
1189 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1190 | const char *name = token + 5; |
1191 | /* Can't specify an empty name */ | |
1192 | if (!strlen(name)) | |
1193 | return -EINVAL; | |
1194 | /* Must match [\w.-]+ */ | |
1195 | for (i = 0; i < strlen(name); i++) { | |
1196 | char c = name[i]; | |
1197 | if (isalnum(c)) | |
1198 | continue; | |
1199 | if ((c == '.') || (c == '-') || (c == '_')) | |
1200 | continue; | |
1201 | return -EINVAL; | |
1202 | } | |
1203 | /* Specifying two names is forbidden */ | |
1204 | if (opts->name) | |
1205 | return -EINVAL; | |
1206 | opts->name = kstrndup(name, | |
e400c285 | 1207 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1208 | GFP_KERNEL); |
1209 | if (!opts->name) | |
1210 | return -ENOMEM; | |
32a8cf23 DL |
1211 | |
1212 | continue; | |
1213 | } | |
1214 | ||
30159ec7 | 1215 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1216 | if (strcmp(token, ss->name)) |
1217 | continue; | |
1218 | if (ss->disabled) | |
1219 | continue; | |
1220 | ||
1221 | /* Mutually exclusive option 'all' + subsystem name */ | |
1222 | if (all_ss) | |
1223 | return -EINVAL; | |
a1a71b45 | 1224 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1225 | one_ss = true; |
1226 | ||
1227 | break; | |
1228 | } | |
1229 | if (i == CGROUP_SUBSYS_COUNT) | |
1230 | return -ENOENT; | |
1231 | } | |
1232 | ||
1233 | /* | |
1234 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1235 | * otherwise if 'none', 'name=' and a subsystem name options |
1236 | * were not specified, let's default to 'all' | |
32a8cf23 | 1237 | */ |
30159ec7 TH |
1238 | if (all_ss || (!one_ss && !opts->none && !opts->name)) |
1239 | for_each_subsys(ss, i) | |
1240 | if (!ss->disabled) | |
1241 | set_bit(i, &opts->subsys_mask); | |
ddbcc7e8 | 1242 | |
2c6ab6d2 PM |
1243 | /* Consistency checks */ |
1244 | ||
873fe09e TH |
1245 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1246 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1247 | ||
1248 | if (opts->flags & CGRP_ROOT_NOPREFIX) { | |
1249 | pr_err("cgroup: sane_behavior: noprefix is not allowed\n"); | |
1250 | return -EINVAL; | |
1251 | } | |
1252 | ||
1253 | if (opts->cpuset_clone_children) { | |
1254 | pr_err("cgroup: sane_behavior: clone_children is not allowed\n"); | |
1255 | return -EINVAL; | |
1256 | } | |
86bf4b68 TH |
1257 | |
1258 | if (opts->flags & CGRP_ROOT_XATTR) | |
1259 | pr_warning("cgroup: sane_behavior: xattr is always available, flag unnecessary\n"); | |
873fe09e TH |
1260 | } |
1261 | ||
f9ab5b5b LZ |
1262 | /* |
1263 | * Option noprefix was introduced just for backward compatibility | |
1264 | * with the old cpuset, so we allow noprefix only if mounting just | |
1265 | * the cpuset subsystem. | |
1266 | */ | |
93438629 | 1267 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1268 | return -EINVAL; |
1269 | ||
2c6ab6d2 PM |
1270 | |
1271 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1272 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1273 | return -EINVAL; |
1274 | ||
1275 | /* | |
1276 | * We either have to specify by name or by subsystems. (So all | |
1277 | * empty hierarchies must have a name). | |
1278 | */ | |
a1a71b45 | 1279 | if (!opts->subsys_mask && !opts->name) |
ddbcc7e8 PM |
1280 | return -EINVAL; |
1281 | ||
1282 | return 0; | |
1283 | } | |
1284 | ||
2bd59d48 | 1285 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1286 | { |
1287 | int ret = 0; | |
2bd59d48 | 1288 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1289 | struct cgroup_sb_opts opts; |
a1a71b45 | 1290 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1291 | |
873fe09e TH |
1292 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1293 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1294 | return -EINVAL; | |
1295 | } | |
1296 | ||
ace2bee8 | 1297 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1298 | mutex_lock(&cgroup_mutex); |
1299 | ||
1300 | /* See what subsystems are wanted */ | |
1301 | ret = parse_cgroupfs_options(data, &opts); | |
1302 | if (ret) | |
1303 | goto out_unlock; | |
1304 | ||
a8a648c4 | 1305 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1306 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1307 | task_tgid_nr(current), current->comm); | |
1308 | ||
a1a71b45 AR |
1309 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1310 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1311 | |
cf5d5941 | 1312 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1313 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1314 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1315 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1316 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1317 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1318 | ret = -EINVAL; |
1319 | goto out_unlock; | |
1320 | } | |
1321 | ||
f172e67c TH |
1322 | /* remounting is not allowed for populated hierarchies */ |
1323 | if (root->number_of_cgroups > 1) { | |
1324 | ret = -EBUSY; | |
0670e08b | 1325 | goto out_unlock; |
cf5d5941 | 1326 | } |
ddbcc7e8 | 1327 | |
a8a648c4 | 1328 | ret = rebind_subsystems(root, added_mask, removed_mask); |
3126121f | 1329 | if (ret) |
0670e08b | 1330 | goto out_unlock; |
ddbcc7e8 | 1331 | |
69e943b7 TH |
1332 | if (opts.release_agent) { |
1333 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1334 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1335 | spin_unlock(&release_agent_path_lock); |
1336 | } | |
ddbcc7e8 | 1337 | out_unlock: |
66bdc9cf | 1338 | kfree(opts.release_agent); |
c6d57f33 | 1339 | kfree(opts.name); |
ddbcc7e8 | 1340 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1341 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1342 | return ret; |
1343 | } | |
1344 | ||
cc31edce PM |
1345 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1346 | { | |
2bd59d48 | 1347 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1348 | INIT_LIST_HEAD(&cgrp->sibling); |
1349 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1350 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1351 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1352 | INIT_LIST_HEAD(&cgrp->pidlists); |
1353 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1354 | cgrp->dummy_css.cgroup = cgrp; |
cc31edce | 1355 | } |
c6d57f33 | 1356 | |
ddbcc7e8 PM |
1357 | static void init_cgroup_root(struct cgroupfs_root *root) |
1358 | { | |
bd89aabc | 1359 | struct cgroup *cgrp = &root->top_cgroup; |
b0ca5a84 | 1360 | |
2bd59d48 | 1361 | atomic_set(&root->refcnt, 1); |
ddbcc7e8 PM |
1362 | INIT_LIST_HEAD(&root->root_list); |
1363 | root->number_of_cgroups = 1; | |
bd89aabc | 1364 | cgrp->root = root; |
cc31edce | 1365 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1366 | idr_init(&root->cgroup_idr); |
ddbcc7e8 PM |
1367 | } |
1368 | ||
c6d57f33 PM |
1369 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1370 | { | |
1371 | struct cgroupfs_root *root; | |
1372 | ||
a1a71b45 | 1373 | if (!opts->subsys_mask && !opts->none) |
2bd59d48 | 1374 | return ERR_PTR(-EINVAL); |
c6d57f33 PM |
1375 | |
1376 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1377 | if (!root) | |
1378 | return ERR_PTR(-ENOMEM); | |
1379 | ||
1380 | init_cgroup_root(root); | |
2c6ab6d2 | 1381 | |
1672d040 TH |
1382 | /* |
1383 | * We need to set @root->subsys_mask now so that @root can be | |
1384 | * matched by cgroup_test_super() before it finishes | |
1385 | * initialization; otherwise, competing mounts with the same | |
1386 | * options may try to bind the same subsystems instead of waiting | |
1387 | * for the first one leading to unexpected mount errors. | |
1388 | * SUBSYS_BOUND will be set once actual binding is complete. | |
1389 | */ | |
a1a71b45 | 1390 | root->subsys_mask = opts->subsys_mask; |
c6d57f33 PM |
1391 | root->flags = opts->flags; |
1392 | if (opts->release_agent) | |
1393 | strcpy(root->release_agent_path, opts->release_agent); | |
1394 | if (opts->name) | |
1395 | strcpy(root->name, opts->name); | |
2260e7fc TH |
1396 | if (opts->cpuset_clone_children) |
1397 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1398 | return root; |
1399 | } | |
1400 | ||
d427dfeb TH |
1401 | static int cgroup_setup_root(struct cgroupfs_root *root) |
1402 | { | |
1403 | LIST_HEAD(tmp_links); | |
d427dfeb | 1404 | struct cgroup *root_cgrp = &root->top_cgroup; |
d427dfeb | 1405 | struct css_set *cset; |
d427dfeb TH |
1406 | int i, ret; |
1407 | ||
1408 | lockdep_assert_held(&cgroup_tree_mutex); | |
1409 | lockdep_assert_held(&cgroup_mutex); | |
d427dfeb TH |
1410 | |
1411 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); | |
1412 | if (ret < 0) | |
2bd59d48 | 1413 | goto out; |
d427dfeb TH |
1414 | root_cgrp->id = ret; |
1415 | ||
d427dfeb TH |
1416 | /* |
1417 | * We're accessing css_set_count without locking css_set_lock here, | |
1418 | * but that's OK - it can only be increased by someone holding | |
1419 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1420 | * have some link structures left over | |
1421 | */ | |
1422 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1423 | if (ret) | |
2bd59d48 | 1424 | goto out; |
d427dfeb TH |
1425 | |
1426 | /* ID 0 is reserved for dummy root, 1 for unified hierarchy */ | |
1427 | ret = cgroup_init_root_id(root, 2, 0); | |
1428 | if (ret) | |
2bd59d48 | 1429 | goto out; |
d427dfeb | 1430 | |
2bd59d48 TH |
1431 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1432 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1433 | root_cgrp); | |
1434 | if (IS_ERR(root->kf_root)) { | |
1435 | ret = PTR_ERR(root->kf_root); | |
1436 | goto exit_root_id; | |
1437 | } | |
1438 | root_cgrp->kn = root->kf_root->kn; | |
d427dfeb TH |
1439 | |
1440 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); | |
1441 | if (ret) | |
2bd59d48 | 1442 | goto destroy_root; |
d427dfeb TH |
1443 | |
1444 | ret = rebind_subsystems(root, root->subsys_mask, 0); | |
1445 | if (ret) | |
2bd59d48 | 1446 | goto destroy_root; |
d427dfeb TH |
1447 | |
1448 | /* | |
1449 | * There must be no failure case after here, since rebinding takes | |
1450 | * care of subsystems' refcounts, which are explicitly dropped in | |
1451 | * the failure exit path. | |
1452 | */ | |
1453 | list_add(&root->root_list, &cgroup_roots); | |
1454 | cgroup_root_count++; | |
1455 | ||
1456 | /* | |
1457 | * Link the top cgroup in this hierarchy into all the css_set | |
1458 | * objects. | |
1459 | */ | |
1460 | write_lock(&css_set_lock); | |
1461 | hash_for_each(css_set_table, i, cset, hlist) | |
1462 | link_css_set(&tmp_links, cset, root_cgrp); | |
1463 | write_unlock(&css_set_lock); | |
1464 | ||
1465 | BUG_ON(!list_empty(&root_cgrp->children)); | |
1466 | BUG_ON(root->number_of_cgroups != 1); | |
1467 | ||
2bd59d48 | 1468 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1469 | ret = 0; |
2bd59d48 | 1470 | goto out; |
d427dfeb | 1471 | |
2bd59d48 TH |
1472 | destroy_root: |
1473 | kernfs_destroy_root(root->kf_root); | |
1474 | root->kf_root = NULL; | |
1475 | exit_root_id: | |
d427dfeb | 1476 | cgroup_exit_root_id(root); |
2bd59d48 | 1477 | out: |
d427dfeb TH |
1478 | free_cgrp_cset_links(&tmp_links); |
1479 | return ret; | |
1480 | } | |
1481 | ||
f7e83571 | 1482 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1483 | int flags, const char *unused_dev_name, |
f7e83571 | 1484 | void *data) |
ddbcc7e8 | 1485 | { |
2bd59d48 | 1486 | struct cgroupfs_root *root; |
ddbcc7e8 | 1487 | struct cgroup_sb_opts opts; |
2bd59d48 | 1488 | struct dentry *dentry; |
8e30e2b8 | 1489 | int ret; |
ddbcc7e8 | 1490 | |
8e30e2b8 | 1491 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1492 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1493 | |
1494 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1495 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1496 | if (ret) |
8e30e2b8 | 1497 | goto out_unlock; |
ddbcc7e8 | 1498 | |
2bd59d48 TH |
1499 | /* look for a matching existing root */ |
1500 | for_each_active_root(root) { | |
1501 | bool name_match = false; | |
ddbcc7e8 | 1502 | |
2bd59d48 TH |
1503 | /* |
1504 | * If we asked for a name then it must match. Also, if | |
1505 | * name matches but sybsys_mask doesn't, we should fail. | |
1506 | * Remember whether name matched. | |
1507 | */ | |
1508 | if (opts.name) { | |
1509 | if (strcmp(opts.name, root->name)) | |
1510 | continue; | |
1511 | name_match = true; | |
1512 | } | |
ddbcc7e8 | 1513 | |
c6d57f33 | 1514 | /* |
2bd59d48 TH |
1515 | * If we asked for subsystems (or explicitly for no |
1516 | * subsystems) then they must match. | |
c6d57f33 | 1517 | */ |
2bd59d48 TH |
1518 | if ((opts.subsys_mask || opts.none) && |
1519 | (opts.subsys_mask != root->subsys_mask)) { | |
1520 | if (!name_match) | |
1521 | continue; | |
1522 | ret = -EBUSY; | |
1523 | goto out_unlock; | |
1524 | } | |
873fe09e | 1525 | |
c7ba8287 | 1526 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1527 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1528 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1529 | ret = -EINVAL; | |
8e30e2b8 | 1530 | goto out_unlock; |
2a0ff3fb JL |
1531 | } else { |
1532 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1533 | } | |
873fe09e | 1534 | } |
2bd59d48 TH |
1535 | |
1536 | cgroup_get_root(root); | |
1537 | goto out_unlock; | |
ddbcc7e8 PM |
1538 | } |
1539 | ||
2bd59d48 TH |
1540 | /* no such thing, create a new one */ |
1541 | root = cgroup_root_from_opts(&opts); | |
1542 | if (IS_ERR(root)) { | |
1543 | ret = PTR_ERR(root); | |
1544 | goto out_unlock; | |
1545 | } | |
1546 | ||
1547 | ret = cgroup_setup_root(root); | |
1548 | if (ret) | |
1549 | cgroup_free_root(root); | |
1550 | ||
8e30e2b8 | 1551 | out_unlock: |
e25e2cbb | 1552 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1553 | mutex_unlock(&cgroup_tree_mutex); |
8e30e2b8 | 1554 | |
c6d57f33 PM |
1555 | kfree(opts.release_agent); |
1556 | kfree(opts.name); | |
8e30e2b8 | 1557 | |
2bd59d48 | 1558 | if (ret) |
8e30e2b8 | 1559 | return ERR_PTR(ret); |
2bd59d48 TH |
1560 | |
1561 | dentry = kernfs_mount(fs_type, flags, root->kf_root); | |
1562 | if (IS_ERR(dentry)) | |
1563 | cgroup_put_root(root); | |
1564 | return dentry; | |
1565 | } | |
1566 | ||
1567 | static void cgroup_kill_sb(struct super_block *sb) | |
1568 | { | |
1569 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
1570 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); | |
1571 | ||
1572 | cgroup_put_root(root); | |
1573 | kernfs_kill_sb(sb); | |
ddbcc7e8 PM |
1574 | } |
1575 | ||
ddbcc7e8 PM |
1576 | static struct file_system_type cgroup_fs_type = { |
1577 | .name = "cgroup", | |
f7e83571 | 1578 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1579 | .kill_sb = cgroup_kill_sb, |
1580 | }; | |
1581 | ||
676db4af GK |
1582 | static struct kobject *cgroup_kobj; |
1583 | ||
857a2beb | 1584 | /** |
913ffdb5 | 1585 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1586 | * @task: target task |
857a2beb TH |
1587 | * @buf: the buffer to write the path into |
1588 | * @buflen: the length of the buffer | |
1589 | * | |
913ffdb5 TH |
1590 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1591 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1592 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1593 | * cgroup controller callbacks. | |
1594 | * | |
e61734c5 | 1595 | * Return value is the same as kernfs_path(). |
857a2beb | 1596 | */ |
e61734c5 | 1597 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb TH |
1598 | { |
1599 | struct cgroupfs_root *root; | |
913ffdb5 | 1600 | struct cgroup *cgrp; |
e61734c5 TH |
1601 | int hierarchy_id = 1; |
1602 | char *path = NULL; | |
857a2beb TH |
1603 | |
1604 | mutex_lock(&cgroup_mutex); | |
1605 | ||
913ffdb5 TH |
1606 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1607 | ||
857a2beb TH |
1608 | if (root) { |
1609 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1610 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1611 | } else { |
1612 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1613 | if (strlcpy(buf, "/", buflen) < buflen) |
1614 | path = buf; | |
857a2beb TH |
1615 | } |
1616 | ||
1617 | mutex_unlock(&cgroup_mutex); | |
e61734c5 | 1618 | return path; |
857a2beb | 1619 | } |
913ffdb5 | 1620 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1621 | |
2f7ee569 TH |
1622 | /* |
1623 | * Control Group taskset | |
1624 | */ | |
134d3373 TH |
1625 | struct task_and_cgroup { |
1626 | struct task_struct *task; | |
1627 | struct cgroup *cgrp; | |
6f4b7e63 | 1628 | struct css_set *cset; |
134d3373 TH |
1629 | }; |
1630 | ||
2f7ee569 TH |
1631 | struct cgroup_taskset { |
1632 | struct task_and_cgroup single; | |
1633 | struct flex_array *tc_array; | |
1634 | int tc_array_len; | |
1635 | int idx; | |
1636 | struct cgroup *cur_cgrp; | |
1637 | }; | |
1638 | ||
1639 | /** | |
1640 | * cgroup_taskset_first - reset taskset and return the first task | |
1641 | * @tset: taskset of interest | |
1642 | * | |
1643 | * @tset iteration is initialized and the first task is returned. | |
1644 | */ | |
1645 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1646 | { | |
1647 | if (tset->tc_array) { | |
1648 | tset->idx = 0; | |
1649 | return cgroup_taskset_next(tset); | |
1650 | } else { | |
1651 | tset->cur_cgrp = tset->single.cgrp; | |
1652 | return tset->single.task; | |
1653 | } | |
1654 | } | |
1655 | EXPORT_SYMBOL_GPL(cgroup_taskset_first); | |
1656 | ||
1657 | /** | |
1658 | * cgroup_taskset_next - iterate to the next task in taskset | |
1659 | * @tset: taskset of interest | |
1660 | * | |
1661 | * Return the next task in @tset. Iteration must have been initialized | |
1662 | * with cgroup_taskset_first(). | |
1663 | */ | |
1664 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1665 | { | |
1666 | struct task_and_cgroup *tc; | |
1667 | ||
1668 | if (!tset->tc_array || tset->idx >= tset->tc_array_len) | |
1669 | return NULL; | |
1670 | ||
1671 | tc = flex_array_get(tset->tc_array, tset->idx++); | |
1672 | tset->cur_cgrp = tc->cgrp; | |
1673 | return tc->task; | |
1674 | } | |
1675 | EXPORT_SYMBOL_GPL(cgroup_taskset_next); | |
1676 | ||
1677 | /** | |
d99c8727 | 1678 | * cgroup_taskset_cur_css - return the matching css for the current task |
2f7ee569 | 1679 | * @tset: taskset of interest |
d99c8727 | 1680 | * @subsys_id: the ID of the target subsystem |
2f7ee569 | 1681 | * |
d99c8727 TH |
1682 | * Return the css for the current (last returned) task of @tset for |
1683 | * subsystem specified by @subsys_id. This function must be preceded by | |
1684 | * either cgroup_taskset_first() or cgroup_taskset_next(). | |
2f7ee569 | 1685 | */ |
d99c8727 TH |
1686 | struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset, |
1687 | int subsys_id) | |
2f7ee569 | 1688 | { |
ca8bdcaf | 1689 | return cgroup_css(tset->cur_cgrp, cgroup_subsys[subsys_id]); |
2f7ee569 | 1690 | } |
d99c8727 | 1691 | EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css); |
2f7ee569 TH |
1692 | |
1693 | /** | |
1694 | * cgroup_taskset_size - return the number of tasks in taskset | |
1695 | * @tset: taskset of interest | |
1696 | */ | |
1697 | int cgroup_taskset_size(struct cgroup_taskset *tset) | |
1698 | { | |
1699 | return tset->tc_array ? tset->tc_array_len : 1; | |
1700 | } | |
1701 | EXPORT_SYMBOL_GPL(cgroup_taskset_size); | |
1702 | ||
1703 | ||
74a1166d BB |
1704 | /* |
1705 | * cgroup_task_migrate - move a task from one cgroup to another. | |
1706 | * | |
d0b2fdd2 | 1707 | * Must be called with cgroup_mutex and threadgroup locked. |
74a1166d | 1708 | */ |
5abb8855 TH |
1709 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1710 | struct task_struct *tsk, | |
1711 | struct css_set *new_cset) | |
74a1166d | 1712 | { |
5abb8855 | 1713 | struct css_set *old_cset; |
74a1166d BB |
1714 | |
1715 | /* | |
026085ef MSB |
1716 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1717 | * setting such that we can't race against cgroup_exit() changing the | |
1718 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1719 | */ |
c84cdf75 | 1720 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1721 | old_cset = task_css_set(tsk); |
74a1166d | 1722 | |
74a1166d | 1723 | task_lock(tsk); |
5abb8855 | 1724 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d BB |
1725 | task_unlock(tsk); |
1726 | ||
1727 | /* Update the css_set linked lists if we're using them */ | |
1728 | write_lock(&css_set_lock); | |
1729 | if (!list_empty(&tsk->cg_list)) | |
5abb8855 | 1730 | list_move(&tsk->cg_list, &new_cset->tasks); |
74a1166d BB |
1731 | write_unlock(&css_set_lock); |
1732 | ||
1733 | /* | |
5abb8855 TH |
1734 | * We just gained a reference on old_cset by taking it from the |
1735 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1736 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1737 | */ |
5abb8855 TH |
1738 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
1739 | put_css_set(old_cset); | |
74a1166d BB |
1740 | } |
1741 | ||
a043e3b2 | 1742 | /** |
081aa458 | 1743 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup |
74a1166d | 1744 | * @cgrp: the cgroup to attach to |
081aa458 LZ |
1745 | * @tsk: the task or the leader of the threadgroup to be attached |
1746 | * @threadgroup: attach the whole threadgroup? | |
74a1166d | 1747 | * |
257058ae | 1748 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
081aa458 | 1749 | * task_lock of @tsk or each thread in the threadgroup individually in turn. |
74a1166d | 1750 | */ |
47cfcd09 TH |
1751 | static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk, |
1752 | bool threadgroup) | |
74a1166d BB |
1753 | { |
1754 | int retval, i, group_size; | |
74a1166d | 1755 | struct cgroupfs_root *root = cgrp->root; |
1c6727af | 1756 | struct cgroup_subsys_state *css, *failed_css = NULL; |
74a1166d | 1757 | /* threadgroup list cursor and array */ |
081aa458 | 1758 | struct task_struct *leader = tsk; |
134d3373 | 1759 | struct task_and_cgroup *tc; |
d846687d | 1760 | struct flex_array *group; |
2f7ee569 | 1761 | struct cgroup_taskset tset = { }; |
74a1166d BB |
1762 | |
1763 | /* | |
1764 | * step 0: in order to do expensive, possibly blocking operations for | |
1765 | * every thread, we cannot iterate the thread group list, since it needs | |
1766 | * rcu or tasklist locked. instead, build an array of all threads in the | |
257058ae TH |
1767 | * group - group_rwsem prevents new threads from appearing, and if |
1768 | * threads exit, this will just be an over-estimate. | |
74a1166d | 1769 | */ |
081aa458 LZ |
1770 | if (threadgroup) |
1771 | group_size = get_nr_threads(tsk); | |
1772 | else | |
1773 | group_size = 1; | |
d846687d | 1774 | /* flex_array supports very large thread-groups better than kmalloc. */ |
134d3373 | 1775 | group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL); |
74a1166d BB |
1776 | if (!group) |
1777 | return -ENOMEM; | |
d846687d | 1778 | /* pre-allocate to guarantee space while iterating in rcu read-side. */ |
3ac1707a | 1779 | retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL); |
d846687d BB |
1780 | if (retval) |
1781 | goto out_free_group_list; | |
74a1166d | 1782 | |
74a1166d | 1783 | i = 0; |
fb5d2b4c MSB |
1784 | /* |
1785 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1786 | * already PF_EXITING could be freed from underneath us unless we | |
1787 | * take an rcu_read_lock. | |
1788 | */ | |
1789 | rcu_read_lock(); | |
74a1166d | 1790 | do { |
134d3373 TH |
1791 | struct task_and_cgroup ent; |
1792 | ||
cd3d0952 TH |
1793 | /* @tsk either already exited or can't exit until the end */ |
1794 | if (tsk->flags & PF_EXITING) | |
ea84753c | 1795 | goto next; |
cd3d0952 | 1796 | |
74a1166d BB |
1797 | /* as per above, nr_threads may decrease, but not increase. */ |
1798 | BUG_ON(i >= group_size); | |
134d3373 TH |
1799 | ent.task = tsk; |
1800 | ent.cgrp = task_cgroup_from_root(tsk, root); | |
892a2b90 MSB |
1801 | /* nothing to do if this task is already in the cgroup */ |
1802 | if (ent.cgrp == cgrp) | |
ea84753c | 1803 | goto next; |
61d1d219 MSB |
1804 | /* |
1805 | * saying GFP_ATOMIC has no effect here because we did prealloc | |
1806 | * earlier, but it's good form to communicate our expectations. | |
1807 | */ | |
134d3373 | 1808 | retval = flex_array_put(group, i, &ent, GFP_ATOMIC); |
d846687d | 1809 | BUG_ON(retval != 0); |
74a1166d | 1810 | i++; |
ea84753c | 1811 | next: |
081aa458 LZ |
1812 | if (!threadgroup) |
1813 | break; | |
74a1166d | 1814 | } while_each_thread(leader, tsk); |
fb5d2b4c | 1815 | rcu_read_unlock(); |
74a1166d BB |
1816 | /* remember the number of threads in the array for later. */ |
1817 | group_size = i; | |
2f7ee569 TH |
1818 | tset.tc_array = group; |
1819 | tset.tc_array_len = group_size; | |
74a1166d | 1820 | |
134d3373 TH |
1821 | /* methods shouldn't be called if no task is actually migrating */ |
1822 | retval = 0; | |
892a2b90 | 1823 | if (!group_size) |
b07ef774 | 1824 | goto out_free_group_list; |
134d3373 | 1825 | |
74a1166d BB |
1826 | /* |
1827 | * step 1: check that we can legitimately attach to the cgroup. | |
1828 | */ | |
1c6727af TH |
1829 | for_each_css(css, i, cgrp) { |
1830 | if (css->ss->can_attach) { | |
1831 | retval = css->ss->can_attach(css, &tset); | |
74a1166d | 1832 | if (retval) { |
1c6727af | 1833 | failed_css = css; |
74a1166d BB |
1834 | goto out_cancel_attach; |
1835 | } | |
1836 | } | |
74a1166d BB |
1837 | } |
1838 | ||
1839 | /* | |
1840 | * step 2: make sure css_sets exist for all threads to be migrated. | |
1841 | * we use find_css_set, which allocates a new one if necessary. | |
1842 | */ | |
74a1166d | 1843 | for (i = 0; i < group_size; i++) { |
a8ad805c TH |
1844 | struct css_set *old_cset; |
1845 | ||
134d3373 | 1846 | tc = flex_array_get(group, i); |
a8ad805c | 1847 | old_cset = task_css_set(tc->task); |
6f4b7e63 LZ |
1848 | tc->cset = find_css_set(old_cset, cgrp); |
1849 | if (!tc->cset) { | |
61d1d219 MSB |
1850 | retval = -ENOMEM; |
1851 | goto out_put_css_set_refs; | |
74a1166d BB |
1852 | } |
1853 | } | |
1854 | ||
1855 | /* | |
494c167c TH |
1856 | * step 3: now that we're guaranteed success wrt the css_sets, |
1857 | * proceed to move all tasks to the new cgroup. There are no | |
1858 | * failure cases after here, so this is the commit point. | |
74a1166d | 1859 | */ |
74a1166d | 1860 | for (i = 0; i < group_size; i++) { |
134d3373 | 1861 | tc = flex_array_get(group, i); |
6f4b7e63 | 1862 | cgroup_task_migrate(tc->cgrp, tc->task, tc->cset); |
74a1166d BB |
1863 | } |
1864 | /* nothing is sensitive to fork() after this point. */ | |
1865 | ||
1866 | /* | |
494c167c | 1867 | * step 4: do subsystem attach callbacks. |
74a1166d | 1868 | */ |
1c6727af TH |
1869 | for_each_css(css, i, cgrp) |
1870 | if (css->ss->attach) | |
1871 | css->ss->attach(css, &tset); | |
74a1166d BB |
1872 | |
1873 | /* | |
1874 | * step 5: success! and cleanup | |
1875 | */ | |
74a1166d | 1876 | retval = 0; |
61d1d219 MSB |
1877 | out_put_css_set_refs: |
1878 | if (retval) { | |
1879 | for (i = 0; i < group_size; i++) { | |
1880 | tc = flex_array_get(group, i); | |
6f4b7e63 | 1881 | if (!tc->cset) |
61d1d219 | 1882 | break; |
6f4b7e63 | 1883 | put_css_set(tc->cset); |
61d1d219 | 1884 | } |
74a1166d BB |
1885 | } |
1886 | out_cancel_attach: | |
74a1166d | 1887 | if (retval) { |
1c6727af TH |
1888 | for_each_css(css, i, cgrp) { |
1889 | if (css == failed_css) | |
74a1166d | 1890 | break; |
1c6727af TH |
1891 | if (css->ss->cancel_attach) |
1892 | css->ss->cancel_attach(css, &tset); | |
74a1166d BB |
1893 | } |
1894 | } | |
74a1166d | 1895 | out_free_group_list: |
d846687d | 1896 | flex_array_free(group); |
74a1166d BB |
1897 | return retval; |
1898 | } | |
1899 | ||
1900 | /* | |
1901 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
1902 | * function to attach either it or all tasks in its threadgroup. Will lock |
1903 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 1904 | */ |
74a1166d | 1905 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 1906 | { |
bbcb81d0 | 1907 | struct task_struct *tsk; |
c69e8d9c | 1908 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
1909 | int ret; |
1910 | ||
74a1166d BB |
1911 | if (!cgroup_lock_live_group(cgrp)) |
1912 | return -ENODEV; | |
1913 | ||
b78949eb MSB |
1914 | retry_find_task: |
1915 | rcu_read_lock(); | |
bbcb81d0 | 1916 | if (pid) { |
73507f33 | 1917 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
1918 | if (!tsk) { |
1919 | rcu_read_unlock(); | |
dd4b0a46 | 1920 | ret = -ESRCH; |
b78949eb | 1921 | goto out_unlock_cgroup; |
bbcb81d0 | 1922 | } |
74a1166d BB |
1923 | /* |
1924 | * even if we're attaching all tasks in the thread group, we | |
1925 | * only need to check permissions on one of them. | |
1926 | */ | |
c69e8d9c | 1927 | tcred = __task_cred(tsk); |
14a590c3 EB |
1928 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
1929 | !uid_eq(cred->euid, tcred->uid) && | |
1930 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 1931 | rcu_read_unlock(); |
b78949eb MSB |
1932 | ret = -EACCES; |
1933 | goto out_unlock_cgroup; | |
bbcb81d0 | 1934 | } |
b78949eb MSB |
1935 | } else |
1936 | tsk = current; | |
cd3d0952 TH |
1937 | |
1938 | if (threadgroup) | |
b78949eb | 1939 | tsk = tsk->group_leader; |
c4c27fbd MG |
1940 | |
1941 | /* | |
14a40ffc | 1942 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
1943 | * trapped in a cpuset, or RT worker may be born in a cgroup |
1944 | * with no rt_runtime allocated. Just say no. | |
1945 | */ | |
14a40ffc | 1946 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
1947 | ret = -EINVAL; |
1948 | rcu_read_unlock(); | |
1949 | goto out_unlock_cgroup; | |
1950 | } | |
1951 | ||
b78949eb MSB |
1952 | get_task_struct(tsk); |
1953 | rcu_read_unlock(); | |
1954 | ||
1955 | threadgroup_lock(tsk); | |
1956 | if (threadgroup) { | |
1957 | if (!thread_group_leader(tsk)) { | |
1958 | /* | |
1959 | * a race with de_thread from another thread's exec() | |
1960 | * may strip us of our leadership, if this happens, | |
1961 | * there is no choice but to throw this task away and | |
1962 | * try again; this is | |
1963 | * "double-double-toil-and-trouble-check locking". | |
1964 | */ | |
1965 | threadgroup_unlock(tsk); | |
1966 | put_task_struct(tsk); | |
1967 | goto retry_find_task; | |
1968 | } | |
081aa458 LZ |
1969 | } |
1970 | ||
1971 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
1972 | ||
cd3d0952 TH |
1973 | threadgroup_unlock(tsk); |
1974 | ||
bbcb81d0 | 1975 | put_task_struct(tsk); |
b78949eb | 1976 | out_unlock_cgroup: |
47cfcd09 | 1977 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
1978 | return ret; |
1979 | } | |
1980 | ||
7ae1bad9 TH |
1981 | /** |
1982 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
1983 | * @from: attach to all cgroups of a given task | |
1984 | * @tsk: the task to be attached | |
1985 | */ | |
1986 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
1987 | { | |
1988 | struct cgroupfs_root *root; | |
1989 | int retval = 0; | |
1990 | ||
47cfcd09 | 1991 | mutex_lock(&cgroup_mutex); |
7ae1bad9 | 1992 | for_each_active_root(root) { |
6f4b7e63 | 1993 | struct cgroup *from_cgrp = task_cgroup_from_root(from, root); |
7ae1bad9 | 1994 | |
6f4b7e63 | 1995 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
1996 | if (retval) |
1997 | break; | |
1998 | } | |
47cfcd09 | 1999 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2000 | |
2001 | return retval; | |
2002 | } | |
2003 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2004 | ||
182446d0 TH |
2005 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2006 | struct cftype *cft, u64 pid) | |
74a1166d | 2007 | { |
182446d0 | 2008 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2009 | } |
2010 | ||
182446d0 TH |
2011 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2012 | struct cftype *cft, u64 tgid) | |
af351026 | 2013 | { |
182446d0 | 2014 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2015 | } |
2016 | ||
182446d0 TH |
2017 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
2018 | struct cftype *cft, const char *buffer) | |
e788e066 | 2019 | { |
5f469907 TH |
2020 | struct cgroupfs_root *root = css->cgroup->root; |
2021 | ||
2022 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
182446d0 | 2023 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2024 | return -ENODEV; |
69e943b7 | 2025 | spin_lock(&release_agent_path_lock); |
5f469907 TH |
2026 | strlcpy(root->release_agent_path, buffer, |
2027 | sizeof(root->release_agent_path)); | |
69e943b7 | 2028 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2029 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2030 | return 0; |
2031 | } | |
2032 | ||
2da8ca82 | 2033 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2034 | { |
2da8ca82 | 2035 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2036 | |
e788e066 PM |
2037 | if (!cgroup_lock_live_group(cgrp)) |
2038 | return -ENODEV; | |
2039 | seq_puts(seq, cgrp->root->release_agent_path); | |
2040 | seq_putc(seq, '\n'); | |
47cfcd09 | 2041 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2042 | return 0; |
2043 | } | |
2044 | ||
2da8ca82 | 2045 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2046 | { |
2da8ca82 TH |
2047 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2048 | ||
2049 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2050 | return 0; |
2051 | } | |
2052 | ||
2bd59d48 TH |
2053 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2054 | size_t nbytes, loff_t off) | |
355e0c48 | 2055 | { |
2bd59d48 TH |
2056 | struct cgroup *cgrp = of->kn->parent->priv; |
2057 | struct cftype *cft = of->kn->priv; | |
2058 | struct cgroup_subsys_state *css; | |
a742c59d | 2059 | int ret; |
355e0c48 | 2060 | |
2bd59d48 TH |
2061 | /* |
2062 | * kernfs guarantees that a file isn't deleted with operations in | |
2063 | * flight, which means that the matching css is and stays alive and | |
2064 | * doesn't need to be pinned. The RCU locking is not necessary | |
2065 | * either. It's just for the convenience of using cgroup_css(). | |
2066 | */ | |
2067 | rcu_read_lock(); | |
2068 | css = cgroup_css(cgrp, cft->ss); | |
2069 | rcu_read_unlock(); | |
a742c59d TH |
2070 | |
2071 | if (cft->write_string) { | |
2072 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2073 | } else if (cft->write_u64) { | |
2074 | unsigned long long v; | |
2075 | ret = kstrtoull(buf, 0, &v); | |
2076 | if (!ret) | |
2077 | ret = cft->write_u64(css, cft, v); | |
2078 | } else if (cft->write_s64) { | |
2079 | long long v; | |
2080 | ret = kstrtoll(buf, 0, &v); | |
2081 | if (!ret) | |
2082 | ret = cft->write_s64(css, cft, v); | |
2083 | } else if (cft->trigger) { | |
2084 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2085 | } else { |
a742c59d | 2086 | ret = -EINVAL; |
e73d2c61 | 2087 | } |
2bd59d48 | 2088 | |
a742c59d | 2089 | return ret ?: nbytes; |
355e0c48 PM |
2090 | } |
2091 | ||
6612f05b | 2092 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2093 | { |
2bd59d48 | 2094 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2095 | } |
2096 | ||
6612f05b | 2097 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2098 | { |
2bd59d48 | 2099 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2100 | } |
2101 | ||
6612f05b | 2102 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2103 | { |
2bd59d48 | 2104 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2105 | } |
2106 | ||
91796569 | 2107 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2108 | { |
7da11279 TH |
2109 | struct cftype *cft = seq_cft(m); |
2110 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2111 | |
2da8ca82 TH |
2112 | if (cft->seq_show) |
2113 | return cft->seq_show(m, arg); | |
e73d2c61 | 2114 | |
f4c753b7 | 2115 | if (cft->read_u64) |
896f5199 TH |
2116 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2117 | else if (cft->read_s64) | |
2118 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2119 | else | |
2120 | return -EINVAL; | |
2121 | return 0; | |
91796569 PM |
2122 | } |
2123 | ||
2bd59d48 TH |
2124 | static struct kernfs_ops cgroup_kf_single_ops = { |
2125 | .atomic_write_len = PAGE_SIZE, | |
2126 | .write = cgroup_file_write, | |
2127 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2128 | }; |
2129 | ||
2bd59d48 TH |
2130 | static struct kernfs_ops cgroup_kf_ops = { |
2131 | .atomic_write_len = PAGE_SIZE, | |
2132 | .write = cgroup_file_write, | |
2133 | .seq_start = cgroup_seqfile_start, | |
2134 | .seq_next = cgroup_seqfile_next, | |
2135 | .seq_stop = cgroup_seqfile_stop, | |
2136 | .seq_show = cgroup_seqfile_show, | |
2137 | }; | |
ddbcc7e8 PM |
2138 | |
2139 | /* | |
2140 | * cgroup_rename - Only allow simple rename of directories in place. | |
2141 | */ | |
2bd59d48 TH |
2142 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2143 | const char *new_name_str) | |
ddbcc7e8 | 2144 | { |
2bd59d48 | 2145 | struct cgroup *cgrp = kn->priv; |
2bd59d48 | 2146 | int ret; |
65dff759 | 2147 | |
2bd59d48 | 2148 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2149 | return -ENOTDIR; |
2bd59d48 | 2150 | if (kn->parent != new_parent) |
ddbcc7e8 | 2151 | return -EIO; |
65dff759 | 2152 | |
6db8e85c TH |
2153 | /* |
2154 | * This isn't a proper migration and its usefulness is very | |
2155 | * limited. Disallow if sane_behavior. | |
2156 | */ | |
2157 | if (cgroup_sane_behavior(cgrp)) | |
2158 | return -EPERM; | |
2159 | ||
2bd59d48 TH |
2160 | mutex_lock(&cgroup_tree_mutex); |
2161 | mutex_lock(&cgroup_mutex); | |
2162 | ||
2163 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
65dff759 | 2164 | |
2bd59d48 TH |
2165 | mutex_unlock(&cgroup_mutex); |
2166 | mutex_unlock(&cgroup_tree_mutex); | |
2bd59d48 | 2167 | return ret; |
ddbcc7e8 PM |
2168 | } |
2169 | ||
2bb566cb | 2170 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2171 | { |
8d7e6fb0 | 2172 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2173 | struct kernfs_node *kn; |
2174 | struct lock_class_key *key = NULL; | |
05ef1d7c | 2175 | |
2bd59d48 TH |
2176 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2177 | key = &cft->lockdep_key; | |
2178 | #endif | |
2179 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2180 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2181 | NULL, false, key); | |
2182 | if (IS_ERR(kn)) | |
2183 | return PTR_ERR(kn); | |
2184 | return 0; | |
ddbcc7e8 PM |
2185 | } |
2186 | ||
b1f28d31 TH |
2187 | /** |
2188 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2189 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2190 | * @cfts: array of cftypes to be added |
2191 | * @is_add: whether to add or remove | |
2192 | * | |
2193 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2194 | * For removals, this function never fails. If addition fails, this |
2195 | * function doesn't remove files already added. The caller is responsible | |
2196 | * for cleaning up. | |
b1f28d31 | 2197 | */ |
2bb566cb TH |
2198 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2199 | bool is_add) | |
ddbcc7e8 | 2200 | { |
03b1cde6 | 2201 | struct cftype *cft; |
b1f28d31 TH |
2202 | int ret; |
2203 | ||
ace2bee8 | 2204 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2205 | |
2206 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2207 | /* does cft->flags tell us to skip this file on @cgrp? */ |
873fe09e TH |
2208 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2209 | continue; | |
f33fddc2 G |
2210 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2211 | continue; | |
2212 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2213 | continue; | |
2214 | ||
2739d3cc | 2215 | if (is_add) { |
2bb566cb | 2216 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2217 | if (ret) { |
2739d3cc | 2218 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2219 | cft->name, ret); |
2220 | return ret; | |
2221 | } | |
2739d3cc LZ |
2222 | } else { |
2223 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2224 | } |
ddbcc7e8 | 2225 | } |
b1f28d31 | 2226 | return 0; |
ddbcc7e8 PM |
2227 | } |
2228 | ||
21a2d343 | 2229 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2230 | { |
2231 | LIST_HEAD(pending); | |
2bb566cb | 2232 | struct cgroup_subsys *ss = cfts[0].ss; |
492eb21b | 2233 | struct cgroup *root = &ss->root->top_cgroup; |
492eb21b | 2234 | struct cgroup_subsys_state *css; |
9ccece80 | 2235 | int ret = 0; |
8e3f6541 | 2236 | |
21a2d343 | 2237 | lockdep_assert_held(&cgroup_tree_mutex); |
4ac06017 | 2238 | |
21a2d343 TH |
2239 | /* don't bother if @ss isn't attached */ |
2240 | if (ss->root == &cgroup_dummy_root) | |
9ccece80 | 2241 | return 0; |
e8c82d20 | 2242 | |
e8c82d20 | 2243 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2244 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2245 | struct cgroup *cgrp = css->cgroup; |
2246 | ||
e8c82d20 LZ |
2247 | if (cgroup_is_dead(cgrp)) |
2248 | continue; | |
2249 | ||
21a2d343 | 2250 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2251 | if (ret) |
2252 | break; | |
8e3f6541 | 2253 | } |
21a2d343 TH |
2254 | |
2255 | if (is_add && !ret) | |
2256 | kernfs_activate(root->kn); | |
9ccece80 | 2257 | return ret; |
8e3f6541 TH |
2258 | } |
2259 | ||
2da440a2 TH |
2260 | static void cgroup_exit_cftypes(struct cftype *cfts) |
2261 | { | |
2262 | struct cftype *cft; | |
2263 | ||
2bd59d48 TH |
2264 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2265 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2266 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2267 | kfree(cft->kf_ops); | |
2268 | cft->kf_ops = NULL; | |
2da440a2 | 2269 | cft->ss = NULL; |
2bd59d48 | 2270 | } |
2da440a2 TH |
2271 | } |
2272 | ||
2bd59d48 | 2273 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2274 | { |
2275 | struct cftype *cft; | |
2276 | ||
2bd59d48 TH |
2277 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2278 | struct kernfs_ops *kf_ops; | |
2279 | ||
0adb0704 TH |
2280 | WARN_ON(cft->ss || cft->kf_ops); |
2281 | ||
2bd59d48 TH |
2282 | if (cft->seq_start) |
2283 | kf_ops = &cgroup_kf_ops; | |
2284 | else | |
2285 | kf_ops = &cgroup_kf_single_ops; | |
2286 | ||
2287 | /* | |
2288 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2289 | * make a copy of kf_ops to set its atomic_write_len. | |
2290 | */ | |
2291 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2292 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2293 | if (!kf_ops) { | |
2294 | cgroup_exit_cftypes(cfts); | |
2295 | return -ENOMEM; | |
2296 | } | |
2297 | kf_ops->atomic_write_len = cft->max_write_len; | |
2298 | } | |
2299 | ||
2300 | cft->kf_ops = kf_ops; | |
2da440a2 | 2301 | cft->ss = ss; |
2bd59d48 TH |
2302 | } |
2303 | ||
2304 | return 0; | |
2da440a2 TH |
2305 | } |
2306 | ||
21a2d343 TH |
2307 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
2308 | { | |
2309 | lockdep_assert_held(&cgroup_tree_mutex); | |
2310 | ||
2311 | if (!cfts || !cfts[0].ss) | |
2312 | return -ENOENT; | |
2313 | ||
2314 | list_del(&cfts->node); | |
2315 | cgroup_apply_cftypes(cfts, false); | |
2316 | cgroup_exit_cftypes(cfts); | |
2317 | return 0; | |
2318 | } | |
2319 | ||
80b13586 TH |
2320 | /** |
2321 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
2322 | * @cfts: zero-length name terminated array of cftypes | |
2323 | * | |
2324 | * Unregister @cfts. Files described by @cfts are removed from all | |
2325 | * existing cgroups and all future cgroups won't have them either. This | |
2326 | * function can be called anytime whether @cfts' subsys is attached or not. | |
2327 | * | |
2328 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2329 | * registered. | |
2330 | */ | |
2331 | int cgroup_rm_cftypes(struct cftype *cfts) | |
2332 | { | |
21a2d343 | 2333 | int ret; |
80b13586 | 2334 | |
21a2d343 TH |
2335 | mutex_lock(&cgroup_tree_mutex); |
2336 | ret = cgroup_rm_cftypes_locked(cfts); | |
2337 | mutex_unlock(&cgroup_tree_mutex); | |
2338 | return ret; | |
80b13586 TH |
2339 | } |
2340 | ||
8e3f6541 TH |
2341 | /** |
2342 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2343 | * @ss: target cgroup subsystem | |
2344 | * @cfts: zero-length name terminated array of cftypes | |
2345 | * | |
2346 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2347 | * existing cgroups to which @ss is attached and all future cgroups will | |
2348 | * have them too. This function can be called anytime whether @ss is | |
2349 | * attached or not. | |
2350 | * | |
2351 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2352 | * function currently returns 0 as long as @cfts registration is successful | |
2353 | * even if some file creation attempts on existing cgroups fail. | |
2354 | */ | |
03b1cde6 | 2355 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 2356 | { |
9ccece80 | 2357 | int ret; |
8e3f6541 | 2358 | |
2bd59d48 TH |
2359 | ret = cgroup_init_cftypes(ss, cfts); |
2360 | if (ret) | |
2361 | return ret; | |
2bb566cb | 2362 | |
21a2d343 TH |
2363 | mutex_lock(&cgroup_tree_mutex); |
2364 | ||
0adb0704 | 2365 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 2366 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 2367 | if (ret) |
21a2d343 TH |
2368 | cgroup_rm_cftypes_locked(cfts); |
2369 | ||
2370 | mutex_unlock(&cgroup_tree_mutex); | |
9ccece80 | 2371 | return ret; |
8e3f6541 TH |
2372 | } |
2373 | EXPORT_SYMBOL_GPL(cgroup_add_cftypes); | |
2374 | ||
a043e3b2 LZ |
2375 | /** |
2376 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2377 | * @cgrp: the cgroup in question | |
2378 | * | |
2379 | * Return the number of tasks in the cgroup. | |
2380 | */ | |
bd89aabc | 2381 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2382 | { |
2383 | int count = 0; | |
69d0206c | 2384 | struct cgrp_cset_link *link; |
817929ec PM |
2385 | |
2386 | read_lock(&css_set_lock); | |
69d0206c TH |
2387 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2388 | count += atomic_read(&link->cset->refcount); | |
817929ec | 2389 | read_unlock(&css_set_lock); |
bbcb81d0 PM |
2390 | return count; |
2391 | } | |
2392 | ||
817929ec | 2393 | /* |
0942eeee TH |
2394 | * To reduce the fork() overhead for systems that are not actually using |
2395 | * their cgroups capability, we don't maintain the lists running through | |
2396 | * each css_set to its tasks until we see the list actually used - in other | |
72ec7029 | 2397 | * words after the first call to css_task_iter_start(). |
31a7df01 | 2398 | */ |
3df91fe3 | 2399 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2400 | { |
2401 | struct task_struct *p, *g; | |
2402 | write_lock(&css_set_lock); | |
2403 | use_task_css_set_links = 1; | |
3ce3230a FW |
2404 | /* |
2405 | * We need tasklist_lock because RCU is not safe against | |
2406 | * while_each_thread(). Besides, a forking task that has passed | |
2407 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
2408 | * is not guaranteed to have its child immediately visible in the | |
2409 | * tasklist if we walk through it with RCU. | |
2410 | */ | |
2411 | read_lock(&tasklist_lock); | |
31a7df01 CW |
2412 | do_each_thread(g, p) { |
2413 | task_lock(p); | |
0e04388f LZ |
2414 | /* |
2415 | * We should check if the process is exiting, otherwise | |
2416 | * it will race with cgroup_exit() in that the list | |
2417 | * entry won't be deleted though the process has exited. | |
2418 | */ | |
2419 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
a8ad805c | 2420 | list_add(&p->cg_list, &task_css_set(p)->tasks); |
31a7df01 CW |
2421 | task_unlock(p); |
2422 | } while_each_thread(g, p); | |
3ce3230a | 2423 | read_unlock(&tasklist_lock); |
31a7df01 CW |
2424 | write_unlock(&css_set_lock); |
2425 | } | |
2426 | ||
53fa5261 | 2427 | /** |
492eb21b TH |
2428 | * css_next_child - find the next child of a given css |
2429 | * @pos_css: the current position (%NULL to initiate traversal) | |
2430 | * @parent_css: css whose children to walk | |
53fa5261 | 2431 | * |
492eb21b | 2432 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
2433 | * under either cgroup_mutex or RCU read lock. The only requirement is |
2434 | * that @parent_css and @pos_css are accessible. The next sibling is | |
2435 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 2436 | */ |
492eb21b TH |
2437 | struct cgroup_subsys_state * |
2438 | css_next_child(struct cgroup_subsys_state *pos_css, | |
2439 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 2440 | { |
492eb21b TH |
2441 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
2442 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
2443 | struct cgroup *next; |
2444 | ||
ace2bee8 | 2445 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
2446 | |
2447 | /* | |
2448 | * @pos could already have been removed. Once a cgroup is removed, | |
2449 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
2450 | * changes. As CGRP_DEAD assertion is serialized and happens |
2451 | * before the cgroup is taken off the ->sibling list, if we see it | |
2452 | * unasserted, it's guaranteed that the next sibling hasn't | |
2453 | * finished its grace period even if it's already removed, and thus | |
2454 | * safe to dereference from this RCU critical section. If | |
2455 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
2456 | * to be visible as %true here. | |
3b287a50 TH |
2457 | * |
2458 | * If @pos is dead, its next pointer can't be dereferenced; | |
2459 | * however, as each cgroup is given a monotonically increasing | |
2460 | * unique serial number and always appended to the sibling list, | |
2461 | * the next one can be found by walking the parent's children until | |
2462 | * we see a cgroup with higher serial number than @pos's. While | |
2463 | * this path can be slower, it's taken only when either the current | |
2464 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 2465 | */ |
3b287a50 TH |
2466 | if (!pos) { |
2467 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
2468 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 2469 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
2470 | } else { |
2471 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
2472 | if (next->serial_nr > pos->serial_nr) | |
2473 | break; | |
53fa5261 TH |
2474 | } |
2475 | ||
492eb21b TH |
2476 | if (&next->sibling == &cgrp->children) |
2477 | return NULL; | |
2478 | ||
ca8bdcaf | 2479 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 2480 | } |
492eb21b | 2481 | EXPORT_SYMBOL_GPL(css_next_child); |
53fa5261 | 2482 | |
574bd9f7 | 2483 | /** |
492eb21b | 2484 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 2485 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2486 | * @root: css whose descendants to walk |
574bd9f7 | 2487 | * |
492eb21b | 2488 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
2489 | * to visit for pre-order traversal of @root's descendants. @root is |
2490 | * included in the iteration and the first node to be visited. | |
75501a6d | 2491 | * |
87fb54f1 TH |
2492 | * While this function requires cgroup_mutex or RCU read locking, it |
2493 | * doesn't require the whole traversal to be contained in a single critical | |
2494 | * section. This function will return the correct next descendant as long | |
2495 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 2496 | */ |
492eb21b TH |
2497 | struct cgroup_subsys_state * |
2498 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
2499 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2500 | { |
492eb21b | 2501 | struct cgroup_subsys_state *next; |
574bd9f7 | 2502 | |
ace2bee8 | 2503 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2504 | |
bd8815a6 | 2505 | /* if first iteration, visit @root */ |
7805d000 | 2506 | if (!pos) |
bd8815a6 | 2507 | return root; |
574bd9f7 TH |
2508 | |
2509 | /* visit the first child if exists */ | |
492eb21b | 2510 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
2511 | if (next) |
2512 | return next; | |
2513 | ||
2514 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
2515 | while (pos != root) { |
2516 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 2517 | if (next) |
574bd9f7 | 2518 | return next; |
492eb21b | 2519 | pos = css_parent(pos); |
7805d000 | 2520 | } |
574bd9f7 TH |
2521 | |
2522 | return NULL; | |
2523 | } | |
492eb21b | 2524 | EXPORT_SYMBOL_GPL(css_next_descendant_pre); |
574bd9f7 | 2525 | |
12a9d2fe | 2526 | /** |
492eb21b TH |
2527 | * css_rightmost_descendant - return the rightmost descendant of a css |
2528 | * @pos: css of interest | |
12a9d2fe | 2529 | * |
492eb21b TH |
2530 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
2531 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 2532 | * subtree of @pos. |
75501a6d | 2533 | * |
87fb54f1 TH |
2534 | * While this function requires cgroup_mutex or RCU read locking, it |
2535 | * doesn't require the whole traversal to be contained in a single critical | |
2536 | * section. This function will return the correct rightmost descendant as | |
2537 | * long as @pos is accessible. | |
12a9d2fe | 2538 | */ |
492eb21b TH |
2539 | struct cgroup_subsys_state * |
2540 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 2541 | { |
492eb21b | 2542 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 2543 | |
ace2bee8 | 2544 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
2545 | |
2546 | do { | |
2547 | last = pos; | |
2548 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
2549 | pos = NULL; | |
492eb21b | 2550 | css_for_each_child(tmp, last) |
12a9d2fe TH |
2551 | pos = tmp; |
2552 | } while (pos); | |
2553 | ||
2554 | return last; | |
2555 | } | |
492eb21b | 2556 | EXPORT_SYMBOL_GPL(css_rightmost_descendant); |
12a9d2fe | 2557 | |
492eb21b TH |
2558 | static struct cgroup_subsys_state * |
2559 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 2560 | { |
492eb21b | 2561 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
2562 | |
2563 | do { | |
2564 | last = pos; | |
492eb21b | 2565 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
2566 | } while (pos); |
2567 | ||
2568 | return last; | |
2569 | } | |
2570 | ||
2571 | /** | |
492eb21b | 2572 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 2573 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2574 | * @root: css whose descendants to walk |
574bd9f7 | 2575 | * |
492eb21b | 2576 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
2577 | * to visit for post-order traversal of @root's descendants. @root is |
2578 | * included in the iteration and the last node to be visited. | |
75501a6d | 2579 | * |
87fb54f1 TH |
2580 | * While this function requires cgroup_mutex or RCU read locking, it |
2581 | * doesn't require the whole traversal to be contained in a single critical | |
2582 | * section. This function will return the correct next descendant as long | |
2583 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
2584 | * @cgroup. | |
574bd9f7 | 2585 | */ |
492eb21b TH |
2586 | struct cgroup_subsys_state * |
2587 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
2588 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2589 | { |
492eb21b | 2590 | struct cgroup_subsys_state *next; |
574bd9f7 | 2591 | |
ace2bee8 | 2592 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2593 | |
58b79a91 TH |
2594 | /* if first iteration, visit leftmost descendant which may be @root */ |
2595 | if (!pos) | |
2596 | return css_leftmost_descendant(root); | |
574bd9f7 | 2597 | |
bd8815a6 TH |
2598 | /* if we visited @root, we're done */ |
2599 | if (pos == root) | |
2600 | return NULL; | |
2601 | ||
574bd9f7 | 2602 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 2603 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 2604 | if (next) |
492eb21b | 2605 | return css_leftmost_descendant(next); |
574bd9f7 TH |
2606 | |
2607 | /* no sibling left, visit parent */ | |
bd8815a6 | 2608 | return css_parent(pos); |
574bd9f7 | 2609 | } |
492eb21b | 2610 | EXPORT_SYMBOL_GPL(css_next_descendant_post); |
574bd9f7 | 2611 | |
0942eeee | 2612 | /** |
72ec7029 | 2613 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
2614 | * @it: the iterator to advance |
2615 | * | |
2616 | * Advance @it to the next css_set to walk. | |
d515876e | 2617 | */ |
72ec7029 | 2618 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
2619 | { |
2620 | struct list_head *l = it->cset_link; | |
2621 | struct cgrp_cset_link *link; | |
2622 | struct css_set *cset; | |
2623 | ||
2624 | /* Advance to the next non-empty css_set */ | |
2625 | do { | |
2626 | l = l->next; | |
72ec7029 | 2627 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
2628 | it->cset_link = NULL; |
2629 | return; | |
2630 | } | |
2631 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
2632 | cset = link->cset; | |
2633 | } while (list_empty(&cset->tasks)); | |
2634 | it->cset_link = l; | |
2635 | it->task = cset->tasks.next; | |
2636 | } | |
2637 | ||
0942eeee | 2638 | /** |
72ec7029 TH |
2639 | * css_task_iter_start - initiate task iteration |
2640 | * @css: the css to walk tasks of | |
0942eeee TH |
2641 | * @it: the task iterator to use |
2642 | * | |
72ec7029 TH |
2643 | * Initiate iteration through the tasks of @css. The caller can call |
2644 | * css_task_iter_next() to walk through the tasks until the function | |
2645 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
2646 | * called. | |
0942eeee TH |
2647 | * |
2648 | * Note that this function acquires a lock which is released when the | |
2649 | * iteration finishes. The caller can't sleep while iteration is in | |
2650 | * progress. | |
2651 | */ | |
72ec7029 TH |
2652 | void css_task_iter_start(struct cgroup_subsys_state *css, |
2653 | struct css_task_iter *it) | |
c6ca5750 | 2654 | __acquires(css_set_lock) |
817929ec PM |
2655 | { |
2656 | /* | |
72ec7029 TH |
2657 | * The first time anyone tries to iterate across a css, we need to |
2658 | * enable the list linking each css_set to its tasks, and fix up | |
2659 | * all existing tasks. | |
817929ec | 2660 | */ |
31a7df01 CW |
2661 | if (!use_task_css_set_links) |
2662 | cgroup_enable_task_cg_lists(); | |
2663 | ||
817929ec | 2664 | read_lock(&css_set_lock); |
c59cd3d8 | 2665 | |
72ec7029 TH |
2666 | it->origin_css = css; |
2667 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 2668 | |
72ec7029 | 2669 | css_advance_task_iter(it); |
817929ec PM |
2670 | } |
2671 | ||
0942eeee | 2672 | /** |
72ec7029 | 2673 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
2674 | * @it: the task iterator being iterated |
2675 | * | |
2676 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
2677 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
2678 | * reaches the end. | |
0942eeee | 2679 | */ |
72ec7029 | 2680 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
2681 | { |
2682 | struct task_struct *res; | |
2683 | struct list_head *l = it->task; | |
69d0206c | 2684 | struct cgrp_cset_link *link; |
817929ec PM |
2685 | |
2686 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 2687 | if (!it->cset_link) |
817929ec PM |
2688 | return NULL; |
2689 | res = list_entry(l, struct task_struct, cg_list); | |
2690 | /* Advance iterator to find next entry */ | |
2691 | l = l->next; | |
69d0206c TH |
2692 | link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link); |
2693 | if (l == &link->cset->tasks) { | |
0942eeee TH |
2694 | /* |
2695 | * We reached the end of this task list - move on to the | |
2696 | * next cgrp_cset_link. | |
2697 | */ | |
72ec7029 | 2698 | css_advance_task_iter(it); |
817929ec PM |
2699 | } else { |
2700 | it->task = l; | |
2701 | } | |
2702 | return res; | |
2703 | } | |
2704 | ||
0942eeee | 2705 | /** |
72ec7029 | 2706 | * css_task_iter_end - finish task iteration |
0942eeee TH |
2707 | * @it: the task iterator to finish |
2708 | * | |
72ec7029 | 2709 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 2710 | */ |
72ec7029 | 2711 | void css_task_iter_end(struct css_task_iter *it) |
c6ca5750 | 2712 | __releases(css_set_lock) |
817929ec PM |
2713 | { |
2714 | read_unlock(&css_set_lock); | |
2715 | } | |
2716 | ||
31a7df01 CW |
2717 | static inline int started_after_time(struct task_struct *t1, |
2718 | struct timespec *time, | |
2719 | struct task_struct *t2) | |
2720 | { | |
2721 | int start_diff = timespec_compare(&t1->start_time, time); | |
2722 | if (start_diff > 0) { | |
2723 | return 1; | |
2724 | } else if (start_diff < 0) { | |
2725 | return 0; | |
2726 | } else { | |
2727 | /* | |
2728 | * Arbitrarily, if two processes started at the same | |
2729 | * time, we'll say that the lower pointer value | |
2730 | * started first. Note that t2 may have exited by now | |
2731 | * so this may not be a valid pointer any longer, but | |
2732 | * that's fine - it still serves to distinguish | |
2733 | * between two tasks started (effectively) simultaneously. | |
2734 | */ | |
2735 | return t1 > t2; | |
2736 | } | |
2737 | } | |
2738 | ||
2739 | /* | |
2740 | * This function is a callback from heap_insert() and is used to order | |
2741 | * the heap. | |
2742 | * In this case we order the heap in descending task start time. | |
2743 | */ | |
2744 | static inline int started_after(void *p1, void *p2) | |
2745 | { | |
2746 | struct task_struct *t1 = p1; | |
2747 | struct task_struct *t2 = p2; | |
2748 | return started_after_time(t1, &t2->start_time, t2); | |
2749 | } | |
2750 | ||
2751 | /** | |
72ec7029 TH |
2752 | * css_scan_tasks - iterate though all the tasks in a css |
2753 | * @css: the css to iterate tasks of | |
e535837b TH |
2754 | * @test: optional test callback |
2755 | * @process: process callback | |
2756 | * @data: data passed to @test and @process | |
2757 | * @heap: optional pre-allocated heap used for task iteration | |
31a7df01 | 2758 | * |
72ec7029 TH |
2759 | * Iterate through all the tasks in @css, calling @test for each, and if it |
2760 | * returns %true, call @process for it also. | |
31a7df01 | 2761 | * |
e535837b | 2762 | * @test may be NULL, meaning always true (select all tasks), which |
72ec7029 | 2763 | * effectively duplicates css_task_iter_{start,next,end}() but does not |
e535837b TH |
2764 | * lock css_set_lock for the call to @process. |
2765 | * | |
2766 | * It is guaranteed that @process will act on every task that is a member | |
72ec7029 TH |
2767 | * of @css for the duration of this call. This function may or may not |
2768 | * call @process for tasks that exit or move to a different css during the | |
2769 | * call, or are forked or move into the css during the call. | |
31a7df01 | 2770 | * |
e535837b TH |
2771 | * Note that @test may be called with locks held, and may in some |
2772 | * situations be called multiple times for the same task, so it should be | |
2773 | * cheap. | |
31a7df01 | 2774 | * |
e535837b TH |
2775 | * If @heap is non-NULL, a heap has been pre-allocated and will be used for |
2776 | * heap operations (and its "gt" member will be overwritten), else a | |
2777 | * temporary heap will be used (allocation of which may cause this function | |
2778 | * to fail). | |
31a7df01 | 2779 | */ |
72ec7029 TH |
2780 | int css_scan_tasks(struct cgroup_subsys_state *css, |
2781 | bool (*test)(struct task_struct *, void *), | |
2782 | void (*process)(struct task_struct *, void *), | |
2783 | void *data, struct ptr_heap *heap) | |
31a7df01 CW |
2784 | { |
2785 | int retval, i; | |
72ec7029 | 2786 | struct css_task_iter it; |
31a7df01 CW |
2787 | struct task_struct *p, *dropped; |
2788 | /* Never dereference latest_task, since it's not refcounted */ | |
2789 | struct task_struct *latest_task = NULL; | |
2790 | struct ptr_heap tmp_heap; | |
31a7df01 CW |
2791 | struct timespec latest_time = { 0, 0 }; |
2792 | ||
e535837b | 2793 | if (heap) { |
31a7df01 | 2794 | /* The caller supplied our heap and pre-allocated its memory */ |
31a7df01 CW |
2795 | heap->gt = &started_after; |
2796 | } else { | |
2797 | /* We need to allocate our own heap memory */ | |
2798 | heap = &tmp_heap; | |
2799 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
2800 | if (retval) | |
2801 | /* cannot allocate the heap */ | |
2802 | return retval; | |
2803 | } | |
2804 | ||
2805 | again: | |
2806 | /* | |
72ec7029 | 2807 | * Scan tasks in the css, using the @test callback to determine |
e535837b TH |
2808 | * which are of interest, and invoking @process callback on the |
2809 | * ones which need an update. Since we don't want to hold any | |
2810 | * locks during the task updates, gather tasks to be processed in a | |
2811 | * heap structure. The heap is sorted by descending task start | |
2812 | * time. If the statically-sized heap fills up, we overflow tasks | |
2813 | * that started later, and in future iterations only consider tasks | |
2814 | * that started after the latest task in the previous pass. This | |
31a7df01 CW |
2815 | * guarantees forward progress and that we don't miss any tasks. |
2816 | */ | |
2817 | heap->size = 0; | |
72ec7029 TH |
2818 | css_task_iter_start(css, &it); |
2819 | while ((p = css_task_iter_next(&it))) { | |
31a7df01 CW |
2820 | /* |
2821 | * Only affect tasks that qualify per the caller's callback, | |
2822 | * if he provided one | |
2823 | */ | |
e535837b | 2824 | if (test && !test(p, data)) |
31a7df01 CW |
2825 | continue; |
2826 | /* | |
2827 | * Only process tasks that started after the last task | |
2828 | * we processed | |
2829 | */ | |
2830 | if (!started_after_time(p, &latest_time, latest_task)) | |
2831 | continue; | |
2832 | dropped = heap_insert(heap, p); | |
2833 | if (dropped == NULL) { | |
2834 | /* | |
2835 | * The new task was inserted; the heap wasn't | |
2836 | * previously full | |
2837 | */ | |
2838 | get_task_struct(p); | |
2839 | } else if (dropped != p) { | |
2840 | /* | |
2841 | * The new task was inserted, and pushed out a | |
2842 | * different task | |
2843 | */ | |
2844 | get_task_struct(p); | |
2845 | put_task_struct(dropped); | |
2846 | } | |
2847 | /* | |
2848 | * Else the new task was newer than anything already in | |
2849 | * the heap and wasn't inserted | |
2850 | */ | |
2851 | } | |
72ec7029 | 2852 | css_task_iter_end(&it); |
31a7df01 CW |
2853 | |
2854 | if (heap->size) { | |
2855 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 2856 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 2857 | if (i == 0) { |
4fe91d51 PJ |
2858 | latest_time = q->start_time; |
2859 | latest_task = q; | |
31a7df01 CW |
2860 | } |
2861 | /* Process the task per the caller's callback */ | |
e535837b | 2862 | process(q, data); |
4fe91d51 | 2863 | put_task_struct(q); |
31a7df01 CW |
2864 | } |
2865 | /* | |
2866 | * If we had to process any tasks at all, scan again | |
2867 | * in case some of them were in the middle of forking | |
2868 | * children that didn't get processed. | |
2869 | * Not the most efficient way to do it, but it avoids | |
2870 | * having to take callback_mutex in the fork path | |
2871 | */ | |
2872 | goto again; | |
2873 | } | |
2874 | if (heap == &tmp_heap) | |
2875 | heap_free(&tmp_heap); | |
2876 | return 0; | |
2877 | } | |
2878 | ||
e535837b | 2879 | static void cgroup_transfer_one_task(struct task_struct *task, void *data) |
8cc99345 | 2880 | { |
e535837b | 2881 | struct cgroup *new_cgroup = data; |
8cc99345 | 2882 | |
47cfcd09 | 2883 | mutex_lock(&cgroup_mutex); |
8cc99345 | 2884 | cgroup_attach_task(new_cgroup, task, false); |
47cfcd09 | 2885 | mutex_unlock(&cgroup_mutex); |
8cc99345 TH |
2886 | } |
2887 | ||
2888 | /** | |
2889 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
2890 | * @to: cgroup to which the tasks will be moved | |
2891 | * @from: cgroup in which the tasks currently reside | |
2892 | */ | |
2893 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
2894 | { | |
72ec7029 TH |
2895 | return css_scan_tasks(&from->dummy_css, NULL, cgroup_transfer_one_task, |
2896 | to, NULL); | |
8cc99345 TH |
2897 | } |
2898 | ||
bbcb81d0 | 2899 | /* |
102a775e | 2900 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
2901 | * |
2902 | * Reading this file can return large amounts of data if a cgroup has | |
2903 | * *lots* of attached tasks. So it may need several calls to read(), | |
2904 | * but we cannot guarantee that the information we produce is correct | |
2905 | * unless we produce it entirely atomically. | |
2906 | * | |
bbcb81d0 | 2907 | */ |
bbcb81d0 | 2908 | |
24528255 LZ |
2909 | /* which pidlist file are we talking about? */ |
2910 | enum cgroup_filetype { | |
2911 | CGROUP_FILE_PROCS, | |
2912 | CGROUP_FILE_TASKS, | |
2913 | }; | |
2914 | ||
2915 | /* | |
2916 | * A pidlist is a list of pids that virtually represents the contents of one | |
2917 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
2918 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
2919 | * to the cgroup. | |
2920 | */ | |
2921 | struct cgroup_pidlist { | |
2922 | /* | |
2923 | * used to find which pidlist is wanted. doesn't change as long as | |
2924 | * this particular list stays in the list. | |
2925 | */ | |
2926 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
2927 | /* array of xids */ | |
2928 | pid_t *list; | |
2929 | /* how many elements the above list has */ | |
2930 | int length; | |
24528255 LZ |
2931 | /* each of these stored in a list by its cgroup */ |
2932 | struct list_head links; | |
2933 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
2934 | struct cgroup *owner; | |
b1a21367 TH |
2935 | /* for delayed destruction */ |
2936 | struct delayed_work destroy_dwork; | |
24528255 LZ |
2937 | }; |
2938 | ||
d1d9fd33 BB |
2939 | /* |
2940 | * The following two functions "fix" the issue where there are more pids | |
2941 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
2942 | * TODO: replace with a kernel-wide solution to this problem | |
2943 | */ | |
2944 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
2945 | static void *pidlist_allocate(int count) | |
2946 | { | |
2947 | if (PIDLIST_TOO_LARGE(count)) | |
2948 | return vmalloc(count * sizeof(pid_t)); | |
2949 | else | |
2950 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
2951 | } | |
b1a21367 | 2952 | |
d1d9fd33 BB |
2953 | static void pidlist_free(void *p) |
2954 | { | |
2955 | if (is_vmalloc_addr(p)) | |
2956 | vfree(p); | |
2957 | else | |
2958 | kfree(p); | |
2959 | } | |
d1d9fd33 | 2960 | |
b1a21367 TH |
2961 | /* |
2962 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
2963 | * should be left afterwards. | |
2964 | */ | |
2965 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
2966 | { | |
2967 | struct cgroup_pidlist *l, *tmp_l; | |
2968 | ||
2969 | mutex_lock(&cgrp->pidlist_mutex); | |
2970 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
2971 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
2972 | mutex_unlock(&cgrp->pidlist_mutex); | |
2973 | ||
2974 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
2975 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
2976 | } | |
2977 | ||
2978 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
2979 | { | |
2980 | struct delayed_work *dwork = to_delayed_work(work); | |
2981 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
2982 | destroy_dwork); | |
2983 | struct cgroup_pidlist *tofree = NULL; | |
2984 | ||
2985 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
2986 | |
2987 | /* | |
04502365 TH |
2988 | * Destroy iff we didn't get queued again. The state won't change |
2989 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 2990 | */ |
04502365 | 2991 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
2992 | list_del(&l->links); |
2993 | pidlist_free(l->list); | |
2994 | put_pid_ns(l->key.ns); | |
2995 | tofree = l; | |
2996 | } | |
2997 | ||
b1a21367 TH |
2998 | mutex_unlock(&l->owner->pidlist_mutex); |
2999 | kfree(tofree); | |
3000 | } | |
3001 | ||
bbcb81d0 | 3002 | /* |
102a775e | 3003 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3004 | * Returns the number of unique elements. |
bbcb81d0 | 3005 | */ |
6ee211ad | 3006 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3007 | { |
102a775e | 3008 | int src, dest = 1; |
102a775e BB |
3009 | |
3010 | /* | |
3011 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3012 | * edge cases first; no work needs to be done for either | |
3013 | */ | |
3014 | if (length == 0 || length == 1) | |
3015 | return length; | |
3016 | /* src and dest walk down the list; dest counts unique elements */ | |
3017 | for (src = 1; src < length; src++) { | |
3018 | /* find next unique element */ | |
3019 | while (list[src] == list[src-1]) { | |
3020 | src++; | |
3021 | if (src == length) | |
3022 | goto after; | |
3023 | } | |
3024 | /* dest always points to where the next unique element goes */ | |
3025 | list[dest] = list[src]; | |
3026 | dest++; | |
3027 | } | |
3028 | after: | |
102a775e BB |
3029 | return dest; |
3030 | } | |
3031 | ||
afb2bc14 TH |
3032 | /* |
3033 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3034 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3035 | * different per namespace, each namespace needs differently sorted list, | |
3036 | * making it impossible to use, for example, single rbtree of member tasks | |
3037 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3038 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3039 | * pidlists keyed by cgroup and namespace. | |
3040 | * | |
3041 | * All this extra complexity was caused by the original implementation | |
3042 | * committing to an entirely unnecessary property. In the long term, we | |
3043 | * want to do away with it. Explicitly scramble sort order if | |
3044 | * sane_behavior so that no such expectation exists in the new interface. | |
3045 | * | |
3046 | * Scrambling is done by swapping every two consecutive bits, which is | |
3047 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3048 | */ | |
3049 | static pid_t pid_fry(pid_t pid) | |
3050 | { | |
3051 | unsigned a = pid & 0x55555555; | |
3052 | unsigned b = pid & 0xAAAAAAAA; | |
3053 | ||
3054 | return (a << 1) | (b >> 1); | |
3055 | } | |
3056 | ||
3057 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3058 | { | |
3059 | if (cgroup_sane_behavior(cgrp)) | |
3060 | return pid_fry(pid); | |
3061 | else | |
3062 | return pid; | |
3063 | } | |
3064 | ||
102a775e BB |
3065 | static int cmppid(const void *a, const void *b) |
3066 | { | |
3067 | return *(pid_t *)a - *(pid_t *)b; | |
3068 | } | |
3069 | ||
afb2bc14 TH |
3070 | static int fried_cmppid(const void *a, const void *b) |
3071 | { | |
3072 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3073 | } | |
3074 | ||
e6b81710 TH |
3075 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3076 | enum cgroup_filetype type) | |
3077 | { | |
3078 | struct cgroup_pidlist *l; | |
3079 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3080 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3081 | ||
3082 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3083 | ||
3084 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3085 | if (l->key.type == type && l->key.ns == ns) | |
3086 | return l; | |
3087 | return NULL; | |
3088 | } | |
3089 | ||
72a8cb30 BB |
3090 | /* |
3091 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3092 | * returns with the lock on that pidlist already held, and takes care | |
3093 | * of the use count, or returns NULL with no locks held if we're out of | |
3094 | * memory. | |
3095 | */ | |
e6b81710 TH |
3096 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3097 | enum cgroup_filetype type) | |
72a8cb30 BB |
3098 | { |
3099 | struct cgroup_pidlist *l; | |
b70cc5fd | 3100 | |
e6b81710 TH |
3101 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3102 | ||
3103 | l = cgroup_pidlist_find(cgrp, type); | |
3104 | if (l) | |
3105 | return l; | |
3106 | ||
72a8cb30 | 3107 | /* entry not found; create a new one */ |
f4f4be2b | 3108 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3109 | if (!l) |
72a8cb30 | 3110 | return l; |
e6b81710 | 3111 | |
b1a21367 | 3112 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3113 | l->key.type = type; |
e6b81710 TH |
3114 | /* don't need task_nsproxy() if we're looking at ourself */ |
3115 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3116 | l->owner = cgrp; |
3117 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3118 | return l; |
3119 | } | |
3120 | ||
102a775e BB |
3121 | /* |
3122 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3123 | */ | |
72a8cb30 BB |
3124 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3125 | struct cgroup_pidlist **lp) | |
102a775e BB |
3126 | { |
3127 | pid_t *array; | |
3128 | int length; | |
3129 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3130 | struct css_task_iter it; |
817929ec | 3131 | struct task_struct *tsk; |
102a775e BB |
3132 | struct cgroup_pidlist *l; |
3133 | ||
4bac00d1 TH |
3134 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3135 | ||
102a775e BB |
3136 | /* |
3137 | * If cgroup gets more users after we read count, we won't have | |
3138 | * enough space - tough. This race is indistinguishable to the | |
3139 | * caller from the case that the additional cgroup users didn't | |
3140 | * show up until sometime later on. | |
3141 | */ | |
3142 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3143 | array = pidlist_allocate(length); |
102a775e BB |
3144 | if (!array) |
3145 | return -ENOMEM; | |
3146 | /* now, populate the array */ | |
72ec7029 TH |
3147 | css_task_iter_start(&cgrp->dummy_css, &it); |
3148 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3149 | if (unlikely(n == length)) |
817929ec | 3150 | break; |
102a775e | 3151 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3152 | if (type == CGROUP_FILE_PROCS) |
3153 | pid = task_tgid_vnr(tsk); | |
3154 | else | |
3155 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3156 | if (pid > 0) /* make sure to only use valid results */ |
3157 | array[n++] = pid; | |
817929ec | 3158 | } |
72ec7029 | 3159 | css_task_iter_end(&it); |
102a775e BB |
3160 | length = n; |
3161 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3162 | if (cgroup_sane_behavior(cgrp)) |
3163 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3164 | else | |
3165 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3166 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3167 | length = pidlist_uniq(array, length); |
e6b81710 | 3168 | |
e6b81710 | 3169 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3170 | if (!l) { |
e6b81710 | 3171 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3172 | pidlist_free(array); |
72a8cb30 | 3173 | return -ENOMEM; |
102a775e | 3174 | } |
e6b81710 TH |
3175 | |
3176 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3177 | pidlist_free(l->list); |
102a775e BB |
3178 | l->list = array; |
3179 | l->length = length; | |
72a8cb30 | 3180 | *lp = l; |
102a775e | 3181 | return 0; |
bbcb81d0 PM |
3182 | } |
3183 | ||
846c7bb0 | 3184 | /** |
a043e3b2 | 3185 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3186 | * @stats: cgroupstats to fill information into |
3187 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3188 | * been requested. | |
a043e3b2 LZ |
3189 | * |
3190 | * Build and fill cgroupstats so that taskstats can export it to user | |
3191 | * space. | |
846c7bb0 BS |
3192 | */ |
3193 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3194 | { | |
2bd59d48 | 3195 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3196 | struct cgroup *cgrp; |
72ec7029 | 3197 | struct css_task_iter it; |
846c7bb0 | 3198 | struct task_struct *tsk; |
33d283be | 3199 | |
2bd59d48 TH |
3200 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3201 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3202 | kernfs_type(kn) != KERNFS_DIR) | |
3203 | return -EINVAL; | |
3204 | ||
846c7bb0 | 3205 | /* |
2bd59d48 TH |
3206 | * We aren't being called from kernfs and there's no guarantee on |
3207 | * @kn->priv's validity. For this and css_tryget_from_dir(), | |
3208 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
846c7bb0 | 3209 | */ |
2bd59d48 TH |
3210 | rcu_read_lock(); |
3211 | cgrp = rcu_dereference(kn->priv); | |
3212 | if (!cgrp) { | |
3213 | rcu_read_unlock(); | |
3214 | return -ENOENT; | |
3215 | } | |
846c7bb0 | 3216 | |
72ec7029 TH |
3217 | css_task_iter_start(&cgrp->dummy_css, &it); |
3218 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3219 | switch (tsk->state) { |
3220 | case TASK_RUNNING: | |
3221 | stats->nr_running++; | |
3222 | break; | |
3223 | case TASK_INTERRUPTIBLE: | |
3224 | stats->nr_sleeping++; | |
3225 | break; | |
3226 | case TASK_UNINTERRUPTIBLE: | |
3227 | stats->nr_uninterruptible++; | |
3228 | break; | |
3229 | case TASK_STOPPED: | |
3230 | stats->nr_stopped++; | |
3231 | break; | |
3232 | default: | |
3233 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3234 | stats->nr_io_wait++; | |
3235 | break; | |
3236 | } | |
3237 | } | |
72ec7029 | 3238 | css_task_iter_end(&it); |
846c7bb0 | 3239 | |
2bd59d48 TH |
3240 | rcu_read_unlock(); |
3241 | return 0; | |
846c7bb0 BS |
3242 | } |
3243 | ||
8f3ff208 | 3244 | |
bbcb81d0 | 3245 | /* |
102a775e | 3246 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3247 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3248 | * in the cgroup->l->list array. |
bbcb81d0 | 3249 | */ |
cc31edce | 3250 | |
102a775e | 3251 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3252 | { |
cc31edce PM |
3253 | /* |
3254 | * Initially we receive a position value that corresponds to | |
3255 | * one more than the last pid shown (or 0 on the first call or | |
3256 | * after a seek to the start). Use a binary-search to find the | |
3257 | * next pid to display, if any | |
3258 | */ | |
2bd59d48 | 3259 | struct kernfs_open_file *of = s->private; |
7da11279 | 3260 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3261 | struct cgroup_pidlist *l; |
7da11279 | 3262 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3263 | int index = 0, pid = *pos; |
4bac00d1 TH |
3264 | int *iter, ret; |
3265 | ||
3266 | mutex_lock(&cgrp->pidlist_mutex); | |
3267 | ||
3268 | /* | |
5d22444f | 3269 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3270 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3271 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3272 | * could already have been destroyed. |
3273 | */ | |
5d22444f TH |
3274 | if (of->priv) |
3275 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3276 | |
3277 | /* | |
3278 | * Either this is the first start() after open or the matching | |
3279 | * pidlist has been destroyed inbetween. Create a new one. | |
3280 | */ | |
5d22444f TH |
3281 | if (!of->priv) { |
3282 | ret = pidlist_array_load(cgrp, type, | |
3283 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3284 | if (ret) |
3285 | return ERR_PTR(ret); | |
3286 | } | |
5d22444f | 3287 | l = of->priv; |
cc31edce | 3288 | |
cc31edce | 3289 | if (pid) { |
102a775e | 3290 | int end = l->length; |
20777766 | 3291 | |
cc31edce PM |
3292 | while (index < end) { |
3293 | int mid = (index + end) / 2; | |
afb2bc14 | 3294 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3295 | index = mid; |
3296 | break; | |
afb2bc14 | 3297 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3298 | index = mid + 1; |
3299 | else | |
3300 | end = mid; | |
3301 | } | |
3302 | } | |
3303 | /* If we're off the end of the array, we're done */ | |
102a775e | 3304 | if (index >= l->length) |
cc31edce PM |
3305 | return NULL; |
3306 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3307 | iter = l->list + index; |
afb2bc14 | 3308 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3309 | return iter; |
3310 | } | |
3311 | ||
102a775e | 3312 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3313 | { |
2bd59d48 | 3314 | struct kernfs_open_file *of = s->private; |
5d22444f | 3315 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3316 | |
5d22444f TH |
3317 | if (l) |
3318 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3319 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3320 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3321 | } |
3322 | ||
102a775e | 3323 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3324 | { |
2bd59d48 | 3325 | struct kernfs_open_file *of = s->private; |
5d22444f | 3326 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3327 | pid_t *p = v; |
3328 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3329 | /* |
3330 | * Advance to the next pid in the array. If this goes off the | |
3331 | * end, we're done | |
3332 | */ | |
3333 | p++; | |
3334 | if (p >= end) { | |
3335 | return NULL; | |
3336 | } else { | |
7da11279 | 3337 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3338 | return p; |
3339 | } | |
3340 | } | |
3341 | ||
102a775e | 3342 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3343 | { |
3344 | return seq_printf(s, "%d\n", *(int *)v); | |
3345 | } | |
bbcb81d0 | 3346 | |
102a775e BB |
3347 | /* |
3348 | * seq_operations functions for iterating on pidlists through seq_file - | |
3349 | * independent of whether it's tasks or procs | |
3350 | */ | |
3351 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3352 | .start = cgroup_pidlist_start, | |
3353 | .stop = cgroup_pidlist_stop, | |
3354 | .next = cgroup_pidlist_next, | |
3355 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3356 | }; |
3357 | ||
182446d0 TH |
3358 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3359 | struct cftype *cft) | |
81a6a5cd | 3360 | { |
182446d0 | 3361 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3362 | } |
3363 | ||
182446d0 TH |
3364 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3365 | struct cftype *cft, u64 val) | |
6379c106 | 3366 | { |
182446d0 | 3367 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3368 | if (val) |
182446d0 | 3369 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3370 | else |
182446d0 | 3371 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3372 | return 0; |
3373 | } | |
3374 | ||
182446d0 TH |
3375 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3376 | struct cftype *cft) | |
97978e6d | 3377 | { |
182446d0 | 3378 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3379 | } |
3380 | ||
182446d0 TH |
3381 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3382 | struct cftype *cft, u64 val) | |
97978e6d DL |
3383 | { |
3384 | if (val) | |
182446d0 | 3385 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3386 | else |
182446d0 | 3387 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3388 | return 0; |
3389 | } | |
3390 | ||
d5c56ced | 3391 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3392 | { |
d5c56ced | 3393 | .name = "cgroup.procs", |
6612f05b TH |
3394 | .seq_start = cgroup_pidlist_start, |
3395 | .seq_next = cgroup_pidlist_next, | |
3396 | .seq_stop = cgroup_pidlist_stop, | |
3397 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3398 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3399 | .write_u64 = cgroup_procs_write, |
74a1166d | 3400 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3401 | }, |
97978e6d DL |
3402 | { |
3403 | .name = "cgroup.clone_children", | |
873fe09e | 3404 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3405 | .read_u64 = cgroup_clone_children_read, |
3406 | .write_u64 = cgroup_clone_children_write, | |
3407 | }, | |
873fe09e TH |
3408 | { |
3409 | .name = "cgroup.sane_behavior", | |
3410 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3411 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3412 | }, |
d5c56ced TH |
3413 | |
3414 | /* | |
3415 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3416 | * don't exist if sane_behavior. If you're depending on these, be | |
3417 | * prepared to be burned. | |
3418 | */ | |
3419 | { | |
3420 | .name = "tasks", | |
3421 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3422 | .seq_start = cgroup_pidlist_start, |
3423 | .seq_next = cgroup_pidlist_next, | |
3424 | .seq_stop = cgroup_pidlist_stop, | |
3425 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3426 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3427 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3428 | .mode = S_IRUGO | S_IWUSR, |
3429 | }, | |
3430 | { | |
3431 | .name = "notify_on_release", | |
3432 | .flags = CFTYPE_INSANE, | |
3433 | .read_u64 = cgroup_read_notify_on_release, | |
3434 | .write_u64 = cgroup_write_notify_on_release, | |
3435 | }, | |
6e6ff25b TH |
3436 | { |
3437 | .name = "release_agent", | |
cc5943a7 | 3438 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3439 | .seq_show = cgroup_release_agent_show, |
6e6ff25b | 3440 | .write_string = cgroup_release_agent_write, |
5f469907 | 3441 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 3442 | }, |
db0416b6 | 3443 | { } /* terminate */ |
bbcb81d0 PM |
3444 | }; |
3445 | ||
13af07df | 3446 | /** |
628f7cd4 | 3447 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3448 | * @cgrp: target cgroup |
13af07df | 3449 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3450 | * |
3451 | * On failure, no file is added. | |
13af07df | 3452 | */ |
628f7cd4 | 3453 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3454 | { |
ddbcc7e8 | 3455 | struct cgroup_subsys *ss; |
b420ba7d | 3456 | int i, ret = 0; |
bbcb81d0 | 3457 | |
8e3f6541 | 3458 | /* process cftsets of each subsystem */ |
b420ba7d | 3459 | for_each_subsys(ss, i) { |
0adb0704 | 3460 | struct cftype *cfts; |
b420ba7d TH |
3461 | |
3462 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3463 | continue; |
8e3f6541 | 3464 | |
0adb0704 TH |
3465 | list_for_each_entry(cfts, &ss->cfts, node) { |
3466 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
3467 | if (ret < 0) |
3468 | goto err; | |
3469 | } | |
ddbcc7e8 | 3470 | } |
ddbcc7e8 | 3471 | return 0; |
bee55099 TH |
3472 | err: |
3473 | cgroup_clear_dir(cgrp, subsys_mask); | |
3474 | return ret; | |
ddbcc7e8 PM |
3475 | } |
3476 | ||
0c21ead1 TH |
3477 | /* |
3478 | * css destruction is four-stage process. | |
3479 | * | |
3480 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3481 | * Implemented in kill_css(). | |
3482 | * | |
3483 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3484 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3485 | * by invoking offline_css(). After offlining, the base ref is put. | |
3486 | * Implemented in css_killed_work_fn(). | |
3487 | * | |
3488 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3489 | * accessors are inside RCU read sections. css_release() schedules the | |
3490 | * RCU callback. | |
3491 | * | |
3492 | * 4. After the grace period, the css can be freed. Implemented in | |
3493 | * css_free_work_fn(). | |
3494 | * | |
3495 | * It is actually hairier because both step 2 and 4 require process context | |
3496 | * and thus involve punting to css->destroy_work adding two additional | |
3497 | * steps to the already complex sequence. | |
3498 | */ | |
35ef10da | 3499 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3500 | { |
3501 | struct cgroup_subsys_state *css = | |
35ef10da | 3502 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 3503 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3504 | |
0ae78e0b TH |
3505 | if (css->parent) |
3506 | css_put(css->parent); | |
3507 | ||
0c21ead1 | 3508 | css->ss->css_free(css); |
2bd59d48 | 3509 | cgroup_put(cgrp); |
48ddbe19 TH |
3510 | } |
3511 | ||
0c21ead1 | 3512 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3513 | { |
3514 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3515 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3516 | |
35ef10da | 3517 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 3518 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
3519 | } |
3520 | ||
d3daf28d TH |
3521 | static void css_release(struct percpu_ref *ref) |
3522 | { | |
3523 | struct cgroup_subsys_state *css = | |
3524 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3525 | ||
aec25020 | 3526 | rcu_assign_pointer(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 3527 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
3528 | } |
3529 | ||
623f926b TH |
3530 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
3531 | struct cgroup *cgrp) | |
ddbcc7e8 | 3532 | { |
bd89aabc | 3533 | css->cgroup = cgrp; |
72c97e54 | 3534 | css->ss = ss; |
ddbcc7e8 | 3535 | css->flags = 0; |
0ae78e0b TH |
3536 | |
3537 | if (cgrp->parent) | |
ca8bdcaf | 3538 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 3539 | else |
38b53aba | 3540 | css->flags |= CSS_ROOT; |
48ddbe19 | 3541 | |
ca8bdcaf | 3542 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
3543 | } |
3544 | ||
2a4ac633 | 3545 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 3546 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3547 | { |
623f926b | 3548 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
3549 | int ret = 0; |
3550 | ||
ace2bee8 | 3551 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3552 | lockdep_assert_held(&cgroup_mutex); |
3553 | ||
92fb9748 | 3554 | if (ss->css_online) |
eb95419b | 3555 | ret = ss->css_online(css); |
ae7f164a | 3556 | if (!ret) { |
eb95419b | 3557 | css->flags |= CSS_ONLINE; |
f20104de | 3558 | css->cgroup->nr_css++; |
aec25020 | 3559 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 3560 | } |
b1929db4 | 3561 | return ret; |
a31f2d3f TH |
3562 | } |
3563 | ||
2a4ac633 | 3564 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 3565 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3566 | { |
623f926b | 3567 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 3568 | |
ace2bee8 | 3569 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3570 | lockdep_assert_held(&cgroup_mutex); |
3571 | ||
3572 | if (!(css->flags & CSS_ONLINE)) | |
3573 | return; | |
3574 | ||
d7eeac19 | 3575 | if (ss->css_offline) |
eb95419b | 3576 | ss->css_offline(css); |
a31f2d3f | 3577 | |
eb95419b | 3578 | css->flags &= ~CSS_ONLINE; |
09a503ea | 3579 | css->cgroup->nr_css--; |
aec25020 | 3580 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); |
a31f2d3f TH |
3581 | } |
3582 | ||
c81c925a TH |
3583 | /** |
3584 | * create_css - create a cgroup_subsys_state | |
3585 | * @cgrp: the cgroup new css will be associated with | |
3586 | * @ss: the subsys of new css | |
3587 | * | |
3588 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
3589 | * css is online and installed in @cgrp with all interface files created. | |
3590 | * Returns 0 on success, -errno on failure. | |
3591 | */ | |
3592 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
3593 | { | |
3594 | struct cgroup *parent = cgrp->parent; | |
3595 | struct cgroup_subsys_state *css; | |
3596 | int err; | |
3597 | ||
c81c925a TH |
3598 | lockdep_assert_held(&cgroup_mutex); |
3599 | ||
3600 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
3601 | if (IS_ERR(css)) | |
3602 | return PTR_ERR(css); | |
3603 | ||
3604 | err = percpu_ref_init(&css->refcnt, css_release); | |
3605 | if (err) | |
3606 | goto err_free; | |
3607 | ||
3608 | init_css(css, ss, cgrp); | |
3609 | ||
aec25020 | 3610 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a TH |
3611 | if (err) |
3612 | goto err_free; | |
3613 | ||
3614 | err = online_css(css); | |
3615 | if (err) | |
3616 | goto err_free; | |
3617 | ||
59f5296b | 3618 | cgroup_get(cgrp); |
c81c925a TH |
3619 | css_get(css->parent); |
3620 | ||
3621 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
3622 | parent->parent) { | |
3623 | 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", | |
3624 | current->comm, current->pid, ss->name); | |
3625 | if (!strcmp(ss->name, "memory")) | |
3626 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
3627 | ss->warned_broken_hierarchy = true; | |
3628 | } | |
3629 | ||
3630 | return 0; | |
3631 | ||
3632 | err_free: | |
3633 | percpu_ref_cancel_init(&css->refcnt); | |
3634 | ss->css_free(css); | |
3635 | return err; | |
3636 | } | |
3637 | ||
2bd59d48 | 3638 | /** |
a043e3b2 LZ |
3639 | * cgroup_create - create a cgroup |
3640 | * @parent: cgroup that will be parent of the new cgroup | |
e61734c5 | 3641 | * @name: name of the new cgroup |
2bd59d48 | 3642 | * @mode: mode to set on new cgroup |
ddbcc7e8 | 3643 | */ |
e61734c5 | 3644 | static long cgroup_create(struct cgroup *parent, const char *name, |
2bd59d48 | 3645 | umode_t mode) |
ddbcc7e8 | 3646 | { |
bd89aabc | 3647 | struct cgroup *cgrp; |
ddbcc7e8 | 3648 | struct cgroupfs_root *root = parent->root; |
b58c8998 | 3649 | int ssid, err; |
ddbcc7e8 | 3650 | struct cgroup_subsys *ss; |
2bd59d48 | 3651 | struct kernfs_node *kn; |
ddbcc7e8 | 3652 | |
0a950f65 | 3653 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
3654 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
3655 | if (!cgrp) | |
ddbcc7e8 PM |
3656 | return -ENOMEM; |
3657 | ||
ace2bee8 TH |
3658 | mutex_lock(&cgroup_tree_mutex); |
3659 | ||
976c06bc TH |
3660 | /* |
3661 | * Only live parents can have children. Note that the liveliness | |
3662 | * check isn't strictly necessary because cgroup_mkdir() and | |
3663 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
3664 | * anyway so that locking is contained inside cgroup proper and we | |
3665 | * don't get nasty surprises if we ever grow another caller. | |
3666 | */ | |
3667 | if (!cgroup_lock_live_group(parent)) { | |
3668 | err = -ENODEV; | |
ace2bee8 | 3669 | goto err_unlock_tree; |
0ab02ca8 LZ |
3670 | } |
3671 | ||
3672 | /* | |
3673 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
3674 | * a half-baked cgroup. | |
3675 | */ | |
3676 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
3677 | if (cgrp->id < 0) { | |
3678 | err = -ENOMEM; | |
3679 | goto err_unlock; | |
976c06bc TH |
3680 | } |
3681 | ||
cc31edce | 3682 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3683 | |
bd89aabc | 3684 | cgrp->parent = parent; |
0ae78e0b | 3685 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 3686 | cgrp->root = parent->root; |
ddbcc7e8 | 3687 | |
b6abdb0e LZ |
3688 | if (notify_on_release(parent)) |
3689 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3690 | ||
2260e7fc TH |
3691 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
3692 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 3693 | |
2bd59d48 | 3694 | /* create the directory */ |
e61734c5 | 3695 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 TH |
3696 | if (IS_ERR(kn)) { |
3697 | err = PTR_ERR(kn); | |
0ab02ca8 | 3698 | goto err_free_id; |
2bd59d48 TH |
3699 | } |
3700 | cgrp->kn = kn; | |
ddbcc7e8 | 3701 | |
6f30558f TH |
3702 | /* |
3703 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
3704 | * that @cgrp->kn is always accessible. | |
3705 | */ | |
3706 | kernfs_get(kn); | |
3707 | ||
00356bd5 | 3708 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 3709 | |
4e139afc | 3710 | /* allocation complete, commit to creation */ |
4e139afc TH |
3711 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3712 | root->number_of_cgroups++; | |
28fd6f30 | 3713 | |
2bd59d48 TH |
3714 | /* |
3715 | * Grab a reference on the root and parent so that they don't get | |
3716 | * deleted while there are child cgroups. | |
3717 | */ | |
3718 | cgroup_get_root(root); | |
59f5296b | 3719 | cgroup_get(parent); |
415cf07a | 3720 | |
0d80255e TH |
3721 | /* |
3722 | * @cgrp is now fully operational. If something fails after this | |
3723 | * point, it'll be released via the normal destruction path. | |
3724 | */ | |
4e96ee8e LZ |
3725 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
3726 | ||
2bb566cb | 3727 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
3728 | if (err) |
3729 | goto err_destroy; | |
3730 | ||
9d403e99 | 3731 | /* let's create and online css's */ |
b85d2040 TH |
3732 | for_each_subsys(ss, ssid) { |
3733 | if (root->subsys_mask & (1 << ssid)) { | |
3734 | err = create_css(cgrp, ss); | |
3735 | if (err) | |
3736 | goto err_destroy; | |
3737 | } | |
a8638030 | 3738 | } |
ddbcc7e8 | 3739 | |
2bd59d48 TH |
3740 | kernfs_activate(kn); |
3741 | ||
ddbcc7e8 | 3742 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3743 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
3744 | |
3745 | return 0; | |
3746 | ||
0a950f65 | 3747 | err_free_id: |
4e96ee8e | 3748 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
3749 | err_unlock: |
3750 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
3751 | err_unlock_tree: |
3752 | mutex_unlock(&cgroup_tree_mutex); | |
bd89aabc | 3753 | kfree(cgrp); |
ddbcc7e8 | 3754 | return err; |
4b8b47eb TH |
3755 | |
3756 | err_destroy: | |
3757 | cgroup_destroy_locked(cgrp); | |
3758 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3759 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 3760 | return err; |
ddbcc7e8 PM |
3761 | } |
3762 | ||
2bd59d48 TH |
3763 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
3764 | umode_t mode) | |
ddbcc7e8 | 3765 | { |
2bd59d48 | 3766 | struct cgroup *parent = parent_kn->priv; |
ddbcc7e8 | 3767 | |
2bd59d48 | 3768 | return cgroup_create(parent, name, mode); |
ddbcc7e8 PM |
3769 | } |
3770 | ||
223dbc38 TH |
3771 | /* |
3772 | * This is called when the refcnt of a css is confirmed to be killed. | |
3773 | * css_tryget() is now guaranteed to fail. | |
3774 | */ | |
3775 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 3776 | { |
223dbc38 TH |
3777 | struct cgroup_subsys_state *css = |
3778 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
3779 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 3780 | |
ace2bee8 | 3781 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
3782 | mutex_lock(&cgroup_mutex); |
3783 | ||
09a503ea TH |
3784 | /* |
3785 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
3786 | * initate destruction. | |
3787 | */ | |
3788 | offline_css(css); | |
3789 | ||
f20104de TH |
3790 | /* |
3791 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
3792 | * be disabled before proceeding to the second phase of cgroup | |
3793 | * destruction. If we are the last one, kick it off. | |
3794 | */ | |
09a503ea | 3795 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
3796 | cgroup_destroy_css_killed(cgrp); |
3797 | ||
3798 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3799 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
3800 | |
3801 | /* | |
3802 | * Put the css refs from kill_css(). Each css holds an extra | |
3803 | * reference to the cgroup's dentry and cgroup removal proceeds | |
3804 | * regardless of css refs. On the last put of each css, whenever | |
3805 | * that may be, the extra dentry ref is put so that dentry | |
3806 | * destruction happens only after all css's are released. | |
3807 | */ | |
3808 | css_put(css); | |
d3daf28d TH |
3809 | } |
3810 | ||
223dbc38 TH |
3811 | /* css kill confirmation processing requires process context, bounce */ |
3812 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
3813 | { |
3814 | struct cgroup_subsys_state *css = | |
3815 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3816 | ||
223dbc38 | 3817 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 3818 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
3819 | } |
3820 | ||
edae0c33 TH |
3821 | /** |
3822 | * kill_css - destroy a css | |
3823 | * @css: css to destroy | |
3824 | * | |
3c14f8b4 TH |
3825 | * This function initiates destruction of @css by removing cgroup interface |
3826 | * files and putting its base reference. ->css_offline() will be invoked | |
3827 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
3828 | * reference count reaches zero, @css will be released. | |
edae0c33 TH |
3829 | */ |
3830 | static void kill_css(struct cgroup_subsys_state *css) | |
3831 | { | |
2bd59d48 TH |
3832 | /* |
3833 | * This must happen before css is disassociated with its cgroup. | |
3834 | * See seq_css() for details. | |
3835 | */ | |
aec25020 | 3836 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 3837 | |
edae0c33 TH |
3838 | /* |
3839 | * Killing would put the base ref, but we need to keep it alive | |
3840 | * until after ->css_offline(). | |
3841 | */ | |
3842 | css_get(css); | |
3843 | ||
3844 | /* | |
3845 | * cgroup core guarantees that, by the time ->css_offline() is | |
3846 | * invoked, no new css reference will be given out via | |
3847 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
3848 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
3849 | * guarantee that the ref is seen as killed on all CPUs on return. | |
3850 | * | |
3851 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
3852 | * css is confirmed to be seen as killed on all CPUs. | |
3853 | */ | |
3854 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
3855 | } |
3856 | ||
3857 | /** | |
3858 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
3859 | * @cgrp: cgroup to be destroyed | |
3860 | * | |
3861 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
3862 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
3863 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
3864 | * invoked. To satisfy all the requirements, destruction is implemented in | |
3865 | * the following two steps. | |
3866 | * | |
3867 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
3868 | * userland visible parts and start killing the percpu refcnts of | |
3869 | * css's. Set up so that the next stage will be kicked off once all | |
3870 | * the percpu refcnts are confirmed to be killed. | |
3871 | * | |
3872 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
3873 | * rest of destruction. Once all cgroup references are gone, the | |
3874 | * cgroup is RCU-freed. | |
3875 | * | |
3876 | * This function implements s1. After this step, @cgrp is gone as far as | |
3877 | * the userland is concerned and a new cgroup with the same name may be | |
3878 | * created. As cgroup doesn't care about the names internally, this | |
3879 | * doesn't cause any problem. | |
3880 | */ | |
42809dd4 TH |
3881 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
3882 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 3883 | { |
bb78a92f | 3884 | struct cgroup *child; |
2bd59d48 | 3885 | struct cgroup_subsys_state *css; |
ddd69148 | 3886 | bool empty; |
1c6727af | 3887 | int ssid; |
ddbcc7e8 | 3888 | |
ace2bee8 | 3889 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
3890 | lockdep_assert_held(&cgroup_mutex); |
3891 | ||
ddd69148 | 3892 | /* |
6f3d828f TH |
3893 | * css_set_lock synchronizes access to ->cset_links and prevents |
3894 | * @cgrp from being removed while __put_css_set() is in progress. | |
ddd69148 TH |
3895 | */ |
3896 | read_lock(&css_set_lock); | |
bb78a92f | 3897 | empty = list_empty(&cgrp->cset_links); |
ddd69148 TH |
3898 | read_unlock(&css_set_lock); |
3899 | if (!empty) | |
ddbcc7e8 | 3900 | return -EBUSY; |
a043e3b2 | 3901 | |
bb78a92f HD |
3902 | /* |
3903 | * Make sure there's no live children. We can't test ->children | |
3904 | * emptiness as dead children linger on it while being destroyed; | |
3905 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
3906 | */ | |
3907 | empty = true; | |
3908 | rcu_read_lock(); | |
3909 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
3910 | empty = cgroup_is_dead(child); | |
3911 | if (!empty) | |
3912 | break; | |
3913 | } | |
3914 | rcu_read_unlock(); | |
3915 | if (!empty) | |
3916 | return -EBUSY; | |
3917 | ||
88703267 | 3918 | /* |
edae0c33 TH |
3919 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
3920 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
3921 | * percpu refs of all css's are confirmed to be killed. This |
3922 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 3923 | */ |
4ac06017 | 3924 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
3925 | for_each_css(css, ssid, cgrp) |
3926 | kill_css(css); | |
4ac06017 | 3927 | mutex_lock(&cgroup_mutex); |
455050d2 TH |
3928 | |
3929 | /* | |
3930 | * Mark @cgrp dead. This prevents further task migration and child | |
3931 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 3932 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 3933 | * resume iteration after dropping RCU read lock. See |
492eb21b | 3934 | * css_next_child() for details. |
455050d2 | 3935 | */ |
54766d4a | 3936 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 3937 | |
455050d2 TH |
3938 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
3939 | raw_spin_lock(&release_list_lock); | |
3940 | if (!list_empty(&cgrp->release_list)) | |
3941 | list_del_init(&cgrp->release_list); | |
3942 | raw_spin_unlock(&release_list_lock); | |
3943 | ||
3944 | /* | |
f20104de TH |
3945 | * If @cgrp has css's attached, the second stage of cgroup |
3946 | * destruction is kicked off from css_killed_work_fn() after the | |
3947 | * refs of all attached css's are killed. If @cgrp doesn't have | |
3948 | * any css, we kick it off here. | |
3949 | */ | |
3950 | if (!cgrp->nr_css) | |
3951 | cgroup_destroy_css_killed(cgrp); | |
3952 | ||
2bd59d48 TH |
3953 | /* remove @cgrp directory along with the base files */ |
3954 | mutex_unlock(&cgroup_mutex); | |
3955 | ||
455050d2 | 3956 | /* |
2bd59d48 TH |
3957 | * There are two control paths which try to determine cgroup from |
3958 | * dentry without going through kernfs - cgroupstats_build() and | |
3959 | * css_tryget_from_dir(). Those are supported by RCU protecting | |
3960 | * clearing of cgrp->kn->priv backpointer, which should happen | |
3961 | * after all files under it have been removed. | |
455050d2 | 3962 | */ |
6f30558f | 3963 | kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ |
2bd59d48 | 3964 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); |
2bd59d48 | 3965 | |
4ac06017 | 3966 | mutex_lock(&cgroup_mutex); |
455050d2 | 3967 | |
ea15f8cc TH |
3968 | return 0; |
3969 | }; | |
3970 | ||
d3daf28d | 3971 | /** |
f20104de | 3972 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
3973 | * @work: cgroup->destroy_free_work |
3974 | * | |
3975 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
3976 | * destroyed after all css's are offlined and performs the rest of |
3977 | * destruction. This is the second step of destruction described in the | |
3978 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 3979 | */ |
f20104de | 3980 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 3981 | { |
ea15f8cc | 3982 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 3983 | |
ace2bee8 | 3984 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 3985 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 3986 | |
999cd8a4 | 3987 | /* delete this cgroup from parent->children */ |
eb6fd504 | 3988 | list_del_rcu(&cgrp->sibling); |
ed957793 | 3989 | |
59f5296b | 3990 | cgroup_put(cgrp); |
ddbcc7e8 | 3991 | |
bd89aabc | 3992 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 3993 | check_for_release(parent); |
ddbcc7e8 PM |
3994 | } |
3995 | ||
2bd59d48 | 3996 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 3997 | { |
2bd59d48 TH |
3998 | struct cgroup *cgrp = kn->priv; |
3999 | int ret = 0; | |
4000 | ||
4001 | /* | |
4002 | * This is self-destruction but @kn can't be removed while this | |
4003 | * callback is in progress. Let's break active protection. Once | |
4004 | * the protection is broken, @cgrp can be destroyed at any point. | |
4005 | * Pin it so that it stays accessible. | |
4006 | */ | |
4007 | cgroup_get(cgrp); | |
4008 | kernfs_break_active_protection(kn); | |
42809dd4 | 4009 | |
ace2bee8 | 4010 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 4011 | mutex_lock(&cgroup_mutex); |
2bd59d48 TH |
4012 | |
4013 | /* | |
4014 | * @cgrp might already have been destroyed while we're trying to | |
4015 | * grab the mutexes. | |
4016 | */ | |
4017 | if (!cgroup_is_dead(cgrp)) | |
4018 | ret = cgroup_destroy_locked(cgrp); | |
4019 | ||
42809dd4 | 4020 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4021 | mutex_unlock(&cgroup_tree_mutex); |
42809dd4 | 4022 | |
2bd59d48 TH |
4023 | kernfs_unbreak_active_protection(kn); |
4024 | cgroup_put(cgrp); | |
42809dd4 TH |
4025 | return ret; |
4026 | } | |
4027 | ||
2bd59d48 TH |
4028 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4029 | .remount_fs = cgroup_remount, | |
4030 | .show_options = cgroup_show_options, | |
4031 | .mkdir = cgroup_mkdir, | |
4032 | .rmdir = cgroup_rmdir, | |
4033 | .rename = cgroup_rename, | |
4034 | }; | |
4035 | ||
06a11920 | 4036 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4037 | { |
ddbcc7e8 | 4038 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4039 | |
4040 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4041 | |
ace2bee8 | 4042 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
4043 | mutex_lock(&cgroup_mutex); |
4044 | ||
0adb0704 | 4045 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4046 | |
ddbcc7e8 | 4047 | /* Create the top cgroup state for this subsystem */ |
9871bf95 | 4048 | ss->root = &cgroup_dummy_root; |
ca8bdcaf | 4049 | css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss)); |
ddbcc7e8 PM |
4050 | /* We don't handle early failures gracefully */ |
4051 | BUG_ON(IS_ERR(css)); | |
623f926b | 4052 | init_css(css, ss, cgroup_dummy_top); |
ddbcc7e8 | 4053 | |
e8d55fde | 4054 | /* Update the init_css_set to contain a subsys |
817929ec | 4055 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
4056 | * newly registered, all tasks and hence the |
4057 | * init_css_set is in the subsystem's top cgroup. */ | |
aec25020 | 4058 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4059 | |
4060 | need_forkexit_callback |= ss->fork || ss->exit; | |
4061 | ||
e8d55fde LZ |
4062 | /* At system boot, before all subsystems have been |
4063 | * registered, no tasks have been forked, so we don't | |
4064 | * need to invoke fork callbacks here. */ | |
4065 | BUG_ON(!list_empty(&init_task.tasks)); | |
4066 | ||
ae7f164a | 4067 | BUG_ON(online_css(css)); |
a8638030 | 4068 | |
648bb56d | 4069 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4070 | mutex_unlock(&cgroup_tree_mutex); |
e6a1105b BB |
4071 | } |
4072 | ||
ddbcc7e8 | 4073 | /** |
a043e3b2 LZ |
4074 | * cgroup_init_early - cgroup initialization at system boot |
4075 | * | |
4076 | * Initialize cgroups at system boot, and initialize any | |
4077 | * subsystems that request early init. | |
ddbcc7e8 PM |
4078 | */ |
4079 | int __init cgroup_init_early(void) | |
4080 | { | |
30159ec7 | 4081 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4082 | int i; |
30159ec7 | 4083 | |
146aa1bd | 4084 | atomic_set(&init_css_set.refcount, 1); |
69d0206c | 4085 | INIT_LIST_HEAD(&init_css_set.cgrp_links); |
817929ec | 4086 | INIT_LIST_HEAD(&init_css_set.tasks); |
472b1053 | 4087 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 4088 | css_set_count = 1; |
9871bf95 TH |
4089 | init_cgroup_root(&cgroup_dummy_root); |
4090 | cgroup_root_count = 1; | |
a4ea1cc9 | 4091 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4092 | |
69d0206c | 4093 | init_cgrp_cset_link.cset = &init_css_set; |
9871bf95 TH |
4094 | init_cgrp_cset_link.cgrp = cgroup_dummy_top; |
4095 | list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links); | |
69d0206c | 4096 | list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links); |
ddbcc7e8 | 4097 | |
3ed80a62 | 4098 | for_each_subsys(ss, i) { |
aec25020 | 4099 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4100 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4101 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4102 | ss->id, ss->name); |
073219e9 TH |
4103 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4104 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4105 | ||
aec25020 | 4106 | ss->id = i; |
073219e9 | 4107 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4108 | |
4109 | if (ss->early_init) | |
4110 | cgroup_init_subsys(ss); | |
4111 | } | |
4112 | return 0; | |
4113 | } | |
4114 | ||
4115 | /** | |
a043e3b2 LZ |
4116 | * cgroup_init - cgroup initialization |
4117 | * | |
4118 | * Register cgroup filesystem and /proc file, and initialize | |
4119 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4120 | */ |
4121 | int __init cgroup_init(void) | |
4122 | { | |
30159ec7 | 4123 | struct cgroup_subsys *ss; |
0ac801fe | 4124 | unsigned long key; |
30159ec7 | 4125 | int i, err; |
a424316c | 4126 | |
2bd59d48 | 4127 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
2da440a2 | 4128 | |
3ed80a62 | 4129 | for_each_subsys(ss, i) { |
ddbcc7e8 PM |
4130 | if (!ss->early_init) |
4131 | cgroup_init_subsys(ss); | |
de00ffa5 TH |
4132 | |
4133 | /* | |
4134 | * cftype registration needs kmalloc and can't be done | |
4135 | * during early_init. Register base cftypes separately. | |
4136 | */ | |
4137 | if (ss->base_cftypes) | |
4138 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
ddbcc7e8 PM |
4139 | } |
4140 | ||
fa3ca07e | 4141 | /* allocate id for the dummy hierarchy */ |
54e7b4eb | 4142 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4143 | |
82fe9b0d TH |
4144 | /* Add init_css_set to the hash table */ |
4145 | key = css_set_hash(init_css_set.subsys); | |
4146 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4147 | ||
fc76df70 | 4148 | BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1)); |
676db4af | 4149 | |
4e96ee8e LZ |
4150 | err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top, |
4151 | 0, 1, GFP_KERNEL); | |
4152 | BUG_ON(err < 0); | |
4153 | ||
54e7b4eb TH |
4154 | mutex_unlock(&cgroup_mutex); |
4155 | ||
676db4af | 4156 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4157 | if (!cgroup_kobj) |
4158 | return -ENOMEM; | |
676db4af | 4159 | |
ddbcc7e8 | 4160 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
4161 | if (err < 0) { |
4162 | kobject_put(cgroup_kobj); | |
2bd59d48 | 4163 | return err; |
676db4af | 4164 | } |
ddbcc7e8 | 4165 | |
46ae220b | 4166 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 4167 | return 0; |
ddbcc7e8 | 4168 | } |
b4f48b63 | 4169 | |
e5fca243 TH |
4170 | static int __init cgroup_wq_init(void) |
4171 | { | |
4172 | /* | |
4173 | * There isn't much point in executing destruction path in | |
4174 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
ab3f5faa HD |
4175 | * |
4176 | * XXX: Must be ordered to make sure parent is offlined after | |
4177 | * children. The ordering requirement is for memcg where a | |
4178 | * parent's offline may wait for a child's leading to deadlock. In | |
4179 | * the long term, this should be fixed from memcg side. | |
e5fca243 TH |
4180 | * |
4181 | * We would prefer to do this in cgroup_init() above, but that | |
4182 | * is called before init_workqueues(): so leave this until after. | |
4183 | */ | |
ab3f5faa | 4184 | cgroup_destroy_wq = alloc_ordered_workqueue("cgroup_destroy", 0); |
e5fca243 | 4185 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4186 | |
4187 | /* | |
4188 | * Used to destroy pidlists and separate to serve as flush domain. | |
4189 | * Cap @max_active to 1 too. | |
4190 | */ | |
4191 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4192 | 0, 1); | |
4193 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4194 | ||
e5fca243 TH |
4195 | return 0; |
4196 | } | |
4197 | core_initcall(cgroup_wq_init); | |
4198 | ||
a424316c PM |
4199 | /* |
4200 | * proc_cgroup_show() | |
4201 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4202 | * - Used for /proc/<pid>/cgroup. | |
4203 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4204 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4205 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4206 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4207 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4208 | * cgroup to top_cgroup. | |
4209 | */ | |
4210 | ||
4211 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4212 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4213 | { |
4214 | struct pid *pid; | |
4215 | struct task_struct *tsk; | |
e61734c5 | 4216 | char *buf, *path; |
a424316c PM |
4217 | int retval; |
4218 | struct cgroupfs_root *root; | |
4219 | ||
4220 | retval = -ENOMEM; | |
e61734c5 | 4221 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4222 | if (!buf) |
4223 | goto out; | |
4224 | ||
4225 | retval = -ESRCH; | |
4226 | pid = m->private; | |
4227 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4228 | if (!tsk) | |
4229 | goto out_free; | |
4230 | ||
4231 | retval = 0; | |
4232 | ||
4233 | mutex_lock(&cgroup_mutex); | |
4234 | ||
e5f6a860 | 4235 | for_each_active_root(root) { |
a424316c | 4236 | struct cgroup_subsys *ss; |
bd89aabc | 4237 | struct cgroup *cgrp; |
b85d2040 | 4238 | int ssid, count = 0; |
a424316c | 4239 | |
2c6ab6d2 | 4240 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 TH |
4241 | for_each_subsys(ss, ssid) |
4242 | if (root->subsys_mask & (1 << ssid)) | |
4243 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4244 | if (strlen(root->name)) |
4245 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4246 | root->name); | |
a424316c | 4247 | seq_putc(m, ':'); |
7717f7ba | 4248 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
4249 | path = cgroup_path(cgrp, buf, PATH_MAX); |
4250 | if (!path) { | |
4251 | retval = -ENAMETOOLONG; | |
a424316c | 4252 | goto out_unlock; |
e61734c5 TH |
4253 | } |
4254 | seq_puts(m, path); | |
a424316c PM |
4255 | seq_putc(m, '\n'); |
4256 | } | |
4257 | ||
4258 | out_unlock: | |
4259 | mutex_unlock(&cgroup_mutex); | |
4260 | put_task_struct(tsk); | |
4261 | out_free: | |
4262 | kfree(buf); | |
4263 | out: | |
4264 | return retval; | |
4265 | } | |
4266 | ||
a424316c PM |
4267 | /* Display information about each subsystem and each hierarchy */ |
4268 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4269 | { | |
30159ec7 | 4270 | struct cgroup_subsys *ss; |
a424316c | 4271 | int i; |
a424316c | 4272 | |
8bab8dde | 4273 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4274 | /* |
4275 | * ideally we don't want subsystems moving around while we do this. | |
4276 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4277 | * subsys/hierarchy state. | |
4278 | */ | |
a424316c | 4279 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4280 | |
4281 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4282 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4283 | ss->name, ss->root->hierarchy_id, | |
8bab8dde | 4284 | ss->root->number_of_cgroups, !ss->disabled); |
30159ec7 | 4285 | |
a424316c PM |
4286 | mutex_unlock(&cgroup_mutex); |
4287 | return 0; | |
4288 | } | |
4289 | ||
4290 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4291 | { | |
9dce07f1 | 4292 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4293 | } |
4294 | ||
828c0950 | 4295 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4296 | .open = cgroupstats_open, |
4297 | .read = seq_read, | |
4298 | .llseek = seq_lseek, | |
4299 | .release = single_release, | |
4300 | }; | |
4301 | ||
b4f48b63 PM |
4302 | /** |
4303 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4304 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4305 | * |
4306 | * Description: A task inherits its parent's cgroup at fork(). | |
4307 | * | |
4308 | * A pointer to the shared css_set was automatically copied in | |
4309 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
9bb71308 TH |
4310 | * it was not made under the protection of RCU or cgroup_mutex, so |
4311 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
4312 | * have already changed current->cgroups, allowing the previously | |
4313 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
4314 | * |
4315 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4316 | * task, and the passed argument 'child' points to the child task. | |
4317 | */ | |
4318 | void cgroup_fork(struct task_struct *child) | |
4319 | { | |
9bb71308 | 4320 | task_lock(current); |
a8ad805c | 4321 | get_css_set(task_css_set(current)); |
817929ec | 4322 | child->cgroups = current->cgroups; |
9bb71308 | 4323 | task_unlock(current); |
817929ec | 4324 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4325 | } |
4326 | ||
817929ec | 4327 | /** |
a043e3b2 LZ |
4328 | * cgroup_post_fork - called on a new task after adding it to the task list |
4329 | * @child: the task in question | |
4330 | * | |
5edee61e TH |
4331 | * Adds the task to the list running through its css_set if necessary and |
4332 | * call the subsystem fork() callbacks. Has to be after the task is | |
4333 | * visible on the task list in case we race with the first call to | |
0942eeee | 4334 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4335 | * list. |
a043e3b2 | 4336 | */ |
817929ec PM |
4337 | void cgroup_post_fork(struct task_struct *child) |
4338 | { | |
30159ec7 | 4339 | struct cgroup_subsys *ss; |
5edee61e TH |
4340 | int i; |
4341 | ||
3ce3230a FW |
4342 | /* |
4343 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
4344 | * under the tasklist_lock and we read it here after we added the child | |
4345 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
4346 | * yet in the tasklist when we walked through it from | |
4347 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
4348 | * should be visible now due to the paired locking and barriers implied | |
4349 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
4350 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
4351 | * lock on fork. | |
4352 | */ | |
817929ec PM |
4353 | if (use_task_css_set_links) { |
4354 | write_lock(&css_set_lock); | |
d8783832 TH |
4355 | task_lock(child); |
4356 | if (list_empty(&child->cg_list)) | |
a8ad805c | 4357 | list_add(&child->cg_list, &task_css_set(child)->tasks); |
d8783832 | 4358 | task_unlock(child); |
817929ec PM |
4359 | write_unlock(&css_set_lock); |
4360 | } | |
5edee61e TH |
4361 | |
4362 | /* | |
4363 | * Call ss->fork(). This must happen after @child is linked on | |
4364 | * css_set; otherwise, @child might change state between ->fork() | |
4365 | * and addition to css_set. | |
4366 | */ | |
4367 | if (need_forkexit_callback) { | |
3ed80a62 | 4368 | for_each_subsys(ss, i) |
5edee61e TH |
4369 | if (ss->fork) |
4370 | ss->fork(child); | |
5edee61e | 4371 | } |
817929ec | 4372 | } |
5edee61e | 4373 | |
b4f48b63 PM |
4374 | /** |
4375 | * cgroup_exit - detach cgroup from exiting task | |
4376 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4377 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4378 | * |
4379 | * Description: Detach cgroup from @tsk and release it. | |
4380 | * | |
4381 | * Note that cgroups marked notify_on_release force every task in | |
4382 | * them to take the global cgroup_mutex mutex when exiting. | |
4383 | * This could impact scaling on very large systems. Be reluctant to | |
4384 | * use notify_on_release cgroups where very high task exit scaling | |
4385 | * is required on large systems. | |
4386 | * | |
4387 | * the_top_cgroup_hack: | |
4388 | * | |
4389 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4390 | * | |
4391 | * We call cgroup_exit() while the task is still competent to | |
4392 | * handle notify_on_release(), then leave the task attached to the | |
4393 | * root cgroup in each hierarchy for the remainder of its exit. | |
4394 | * | |
4395 | * To do this properly, we would increment the reference count on | |
4396 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4397 | * code we would add a second cgroup function call, to drop that | |
4398 | * reference. This would just create an unnecessary hot spot on | |
4399 | * the top_cgroup reference count, to no avail. | |
4400 | * | |
4401 | * Normally, holding a reference to a cgroup without bumping its | |
4402 | * count is unsafe. The cgroup could go away, or someone could | |
4403 | * attach us to a different cgroup, decrementing the count on | |
4404 | * the first cgroup that we never incremented. But in this case, | |
4405 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4406 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4407 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4408 | */ |
4409 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4410 | { | |
30159ec7 | 4411 | struct cgroup_subsys *ss; |
5abb8855 | 4412 | struct css_set *cset; |
d41d5a01 | 4413 | int i; |
817929ec PM |
4414 | |
4415 | /* | |
4416 | * Unlink from the css_set task list if necessary. | |
4417 | * Optimistically check cg_list before taking | |
4418 | * css_set_lock | |
4419 | */ | |
4420 | if (!list_empty(&tsk->cg_list)) { | |
4421 | write_lock(&css_set_lock); | |
4422 | if (!list_empty(&tsk->cg_list)) | |
8d258797 | 4423 | list_del_init(&tsk->cg_list); |
817929ec PM |
4424 | write_unlock(&css_set_lock); |
4425 | } | |
4426 | ||
b4f48b63 PM |
4427 | /* Reassign the task to the init_css_set. */ |
4428 | task_lock(tsk); | |
a8ad805c TH |
4429 | cset = task_css_set(tsk); |
4430 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 PZ |
4431 | |
4432 | if (run_callbacks && need_forkexit_callback) { | |
3ed80a62 TH |
4433 | /* see cgroup_post_fork() for details */ |
4434 | for_each_subsys(ss, i) { | |
d41d5a01 | 4435 | if (ss->exit) { |
eb95419b TH |
4436 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4437 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4438 | |
eb95419b | 4439 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4440 | } |
4441 | } | |
4442 | } | |
b4f48b63 | 4443 | task_unlock(tsk); |
d41d5a01 | 4444 | |
5abb8855 | 4445 | put_css_set_taskexit(cset); |
b4f48b63 | 4446 | } |
697f4161 | 4447 | |
bd89aabc | 4448 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4449 | { |
f50daa70 | 4450 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4451 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4452 | /* |
4453 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4454 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4455 | * it now |
4456 | */ | |
81a6a5cd | 4457 | int need_schedule_work = 0; |
f50daa70 | 4458 | |
cdcc136f | 4459 | raw_spin_lock(&release_list_lock); |
54766d4a | 4460 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4461 | list_empty(&cgrp->release_list)) { |
4462 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4463 | need_schedule_work = 1; |
4464 | } | |
cdcc136f | 4465 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4466 | if (need_schedule_work) |
4467 | schedule_work(&release_agent_work); | |
4468 | } | |
4469 | } | |
4470 | ||
81a6a5cd PM |
4471 | /* |
4472 | * Notify userspace when a cgroup is released, by running the | |
4473 | * configured release agent with the name of the cgroup (path | |
4474 | * relative to the root of cgroup file system) as the argument. | |
4475 | * | |
4476 | * Most likely, this user command will try to rmdir this cgroup. | |
4477 | * | |
4478 | * This races with the possibility that some other task will be | |
4479 | * attached to this cgroup before it is removed, or that some other | |
4480 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4481 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4482 | * unused, and this cgroup will be reprieved from its death sentence, | |
4483 | * to continue to serve a useful existence. Next time it's released, | |
4484 | * we will get notified again, if it still has 'notify_on_release' set. | |
4485 | * | |
4486 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4487 | * means only wait until the task is successfully execve()'d. The | |
4488 | * separate release agent task is forked by call_usermodehelper(), | |
4489 | * then control in this thread returns here, without waiting for the | |
4490 | * release agent task. We don't bother to wait because the caller of | |
4491 | * this routine has no use for the exit status of the release agent | |
4492 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4493 | */ |
81a6a5cd PM |
4494 | static void cgroup_release_agent(struct work_struct *work) |
4495 | { | |
4496 | BUG_ON(work != &release_agent_work); | |
4497 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4498 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4499 | while (!list_empty(&release_list)) { |
4500 | char *argv[3], *envp[3]; | |
4501 | int i; | |
e61734c5 | 4502 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 4503 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4504 | struct cgroup, |
4505 | release_list); | |
bd89aabc | 4506 | list_del_init(&cgrp->release_list); |
cdcc136f | 4507 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 4508 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
4509 | if (!pathbuf) |
4510 | goto continue_free; | |
e61734c5 TH |
4511 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
4512 | if (!path) | |
e788e066 PM |
4513 | goto continue_free; |
4514 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4515 | if (!agentbuf) | |
4516 | goto continue_free; | |
81a6a5cd PM |
4517 | |
4518 | i = 0; | |
e788e066 | 4519 | argv[i++] = agentbuf; |
e61734c5 | 4520 | argv[i++] = path; |
81a6a5cd PM |
4521 | argv[i] = NULL; |
4522 | ||
4523 | i = 0; | |
4524 | /* minimal command environment */ | |
4525 | envp[i++] = "HOME=/"; | |
4526 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4527 | envp[i] = NULL; | |
4528 | ||
4529 | /* Drop the lock while we invoke the usermode helper, | |
4530 | * since the exec could involve hitting disk and hence | |
4531 | * be a slow process */ | |
4532 | mutex_unlock(&cgroup_mutex); | |
4533 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4534 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4535 | continue_free: |
4536 | kfree(pathbuf); | |
4537 | kfree(agentbuf); | |
cdcc136f | 4538 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4539 | } |
cdcc136f | 4540 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4541 | mutex_unlock(&cgroup_mutex); |
4542 | } | |
8bab8dde PM |
4543 | |
4544 | static int __init cgroup_disable(char *str) | |
4545 | { | |
30159ec7 | 4546 | struct cgroup_subsys *ss; |
8bab8dde | 4547 | char *token; |
30159ec7 | 4548 | int i; |
8bab8dde PM |
4549 | |
4550 | while ((token = strsep(&str, ",")) != NULL) { | |
4551 | if (!*token) | |
4552 | continue; | |
be45c900 | 4553 | |
3ed80a62 | 4554 | for_each_subsys(ss, i) { |
8bab8dde PM |
4555 | if (!strcmp(token, ss->name)) { |
4556 | ss->disabled = 1; | |
4557 | printk(KERN_INFO "Disabling %s control group" | |
4558 | " subsystem\n", ss->name); | |
4559 | break; | |
4560 | } | |
4561 | } | |
4562 | } | |
4563 | return 1; | |
4564 | } | |
4565 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 4566 | |
b77d7b60 | 4567 | /** |
5a17f543 | 4568 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
4569 | * @dentry: directory dentry of interest |
4570 | * @ss: subsystem of interest | |
b77d7b60 | 4571 | * |
5a17f543 TH |
4572 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
4573 | * to get the corresponding css and return it. If such css doesn't exist | |
4574 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 4575 | */ |
5a17f543 TH |
4576 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
4577 | struct cgroup_subsys *ss) | |
e5d1367f | 4578 | { |
2bd59d48 TH |
4579 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
4580 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 4581 | struct cgroup *cgrp; |
b77d7b60 | 4582 | |
35cf0836 | 4583 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
4584 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
4585 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
4586 | return ERR_PTR(-EBADF); |
4587 | ||
5a17f543 TH |
4588 | rcu_read_lock(); |
4589 | ||
2bd59d48 TH |
4590 | /* |
4591 | * This path doesn't originate from kernfs and @kn could already | |
4592 | * have been or be removed at any point. @kn->priv is RCU | |
4593 | * protected for this access. See destroy_locked() for details. | |
4594 | */ | |
4595 | cgrp = rcu_dereference(kn->priv); | |
4596 | if (cgrp) | |
4597 | css = cgroup_css(cgrp, ss); | |
5a17f543 TH |
4598 | |
4599 | if (!css || !css_tryget(css)) | |
4600 | css = ERR_PTR(-ENOENT); | |
4601 | ||
4602 | rcu_read_unlock(); | |
4603 | return css; | |
e5d1367f | 4604 | } |
e5d1367f | 4605 | |
1cb650b9 LZ |
4606 | /** |
4607 | * css_from_id - lookup css by id | |
4608 | * @id: the cgroup id | |
4609 | * @ss: cgroup subsys to be looked into | |
4610 | * | |
4611 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
4612 | * Should be called under rcu_read_lock(). | |
4613 | */ | |
4614 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
4615 | { | |
4616 | struct cgroup *cgrp; | |
4617 | ||
ace2bee8 | 4618 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
4619 | |
4620 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
4621 | if (cgrp) | |
d1625964 | 4622 | return cgroup_css(cgrp, ss); |
1cb650b9 | 4623 | return NULL; |
e5d1367f SE |
4624 | } |
4625 | ||
fe693435 | 4626 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
4627 | static struct cgroup_subsys_state * |
4628 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
4629 | { |
4630 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
4631 | ||
4632 | if (!css) | |
4633 | return ERR_PTR(-ENOMEM); | |
4634 | ||
4635 | return css; | |
4636 | } | |
4637 | ||
eb95419b | 4638 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 4639 | { |
eb95419b | 4640 | kfree(css); |
fe693435 PM |
4641 | } |
4642 | ||
182446d0 TH |
4643 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
4644 | struct cftype *cft) | |
fe693435 | 4645 | { |
182446d0 | 4646 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
4647 | } |
4648 | ||
182446d0 TH |
4649 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
4650 | struct cftype *cft) | |
fe693435 PM |
4651 | { |
4652 | return (u64)(unsigned long)current->cgroups; | |
4653 | } | |
4654 | ||
182446d0 | 4655 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 4656 | struct cftype *cft) |
fe693435 PM |
4657 | { |
4658 | u64 count; | |
4659 | ||
4660 | rcu_read_lock(); | |
a8ad805c | 4661 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
4662 | rcu_read_unlock(); |
4663 | return count; | |
4664 | } | |
4665 | ||
2da8ca82 | 4666 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4667 | { |
69d0206c | 4668 | struct cgrp_cset_link *link; |
5abb8855 | 4669 | struct css_set *cset; |
e61734c5 TH |
4670 | char *name_buf; |
4671 | ||
4672 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
4673 | if (!name_buf) | |
4674 | return -ENOMEM; | |
7717f7ba PM |
4675 | |
4676 | read_lock(&css_set_lock); | |
4677 | rcu_read_lock(); | |
5abb8855 | 4678 | cset = rcu_dereference(current->cgroups); |
69d0206c | 4679 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 4680 | struct cgroup *c = link->cgrp; |
59f5296b TH |
4681 | const char *name = "?"; |
4682 | ||
e61734c5 TH |
4683 | if (c != cgroup_dummy_top) { |
4684 | cgroup_name(c, name_buf, NAME_MAX + 1); | |
4685 | name = name_buf; | |
4686 | } | |
7717f7ba | 4687 | |
2c6ab6d2 PM |
4688 | seq_printf(seq, "Root %d group %s\n", |
4689 | c->root->hierarchy_id, name); | |
7717f7ba PM |
4690 | } |
4691 | rcu_read_unlock(); | |
4692 | read_unlock(&css_set_lock); | |
e61734c5 | 4693 | kfree(name_buf); |
7717f7ba PM |
4694 | return 0; |
4695 | } | |
4696 | ||
4697 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 4698 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4699 | { |
2da8ca82 | 4700 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 4701 | struct cgrp_cset_link *link; |
7717f7ba PM |
4702 | |
4703 | read_lock(&css_set_lock); | |
182446d0 | 4704 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 4705 | struct css_set *cset = link->cset; |
7717f7ba PM |
4706 | struct task_struct *task; |
4707 | int count = 0; | |
5abb8855 TH |
4708 | seq_printf(seq, "css_set %p\n", cset); |
4709 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
7717f7ba PM |
4710 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { |
4711 | seq_puts(seq, " ...\n"); | |
4712 | break; | |
4713 | } else { | |
4714 | seq_printf(seq, " task %d\n", | |
4715 | task_pid_vnr(task)); | |
4716 | } | |
4717 | } | |
4718 | } | |
4719 | read_unlock(&css_set_lock); | |
4720 | return 0; | |
4721 | } | |
4722 | ||
182446d0 | 4723 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 4724 | { |
182446d0 | 4725 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
4726 | } |
4727 | ||
4728 | static struct cftype debug_files[] = { | |
fe693435 PM |
4729 | { |
4730 | .name = "taskcount", | |
4731 | .read_u64 = debug_taskcount_read, | |
4732 | }, | |
4733 | ||
4734 | { | |
4735 | .name = "current_css_set", | |
4736 | .read_u64 = current_css_set_read, | |
4737 | }, | |
4738 | ||
4739 | { | |
4740 | .name = "current_css_set_refcount", | |
4741 | .read_u64 = current_css_set_refcount_read, | |
4742 | }, | |
4743 | ||
7717f7ba PM |
4744 | { |
4745 | .name = "current_css_set_cg_links", | |
2da8ca82 | 4746 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
4747 | }, |
4748 | ||
4749 | { | |
4750 | .name = "cgroup_css_links", | |
2da8ca82 | 4751 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
4752 | }, |
4753 | ||
fe693435 PM |
4754 | { |
4755 | .name = "releasable", | |
4756 | .read_u64 = releasable_read, | |
4757 | }, | |
fe693435 | 4758 | |
4baf6e33 TH |
4759 | { } /* terminate */ |
4760 | }; | |
fe693435 | 4761 | |
073219e9 | 4762 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
4763 | .css_alloc = debug_css_alloc, |
4764 | .css_free = debug_css_free, | |
4baf6e33 | 4765 | .base_cftypes = debug_files, |
fe693435 PM |
4766 | }; |
4767 | #endif /* CONFIG_CGROUP_DEBUG */ |