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 | 760 | |
3c9c825b | 761 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
f2e85d57 TH |
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 | 930 | |
3c9c825b | 931 | atomic_dec(&cgrp->root->nr_cgrps); |
be445626 | 932 | |
415cf07a | 933 | /* |
59f5296b TH |
934 | * We get a ref to the parent, and put the ref when this cgroup is |
935 | * being freed, so it's guaranteed that the parent won't be | |
936 | * destroyed before its children. | |
415cf07a | 937 | */ |
59f5296b | 938 | cgroup_put(cgrp->parent); |
415cf07a | 939 | |
59f5296b TH |
940 | /* put the root reference that we took when we created the cgroup */ |
941 | cgroup_put_root(cgrp->root); | |
be445626 | 942 | |
b1a21367 | 943 | cgroup_pidlist_destroy_all(cgrp); |
be445626 | 944 | |
6f30558f | 945 | kernfs_put(cgrp->kn); |
be445626 LZ |
946 | kfree(cgrp); |
947 | } | |
948 | ||
949 | static void cgroup_free_rcu(struct rcu_head *head) | |
950 | { | |
951 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
952 | ||
ea15f8cc | 953 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 954 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
955 | } |
956 | ||
59f5296b TH |
957 | static void cgroup_get(struct cgroup *cgrp) |
958 | { | |
2bd59d48 TH |
959 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
960 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
961 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
962 | } |
963 | ||
59f5296b TH |
964 | static void cgroup_put(struct cgroup *cgrp) |
965 | { | |
2bd59d48 TH |
966 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
967 | return; | |
968 | if (WARN_ON_ONCE(!cgroup_is_dead(cgrp))) | |
969 | return; | |
59f5296b | 970 | |
2bd59d48 TH |
971 | /* |
972 | * XXX: cgrp->id is only used to look up css's. As cgroup and | |
973 | * css's lifetimes will be decoupled, it should be made | |
974 | * per-subsystem and moved to css->id so that lookups are | |
975 | * successful until the target css is released. | |
976 | */ | |
977 | mutex_lock(&cgroup_mutex); | |
978 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
979 | mutex_unlock(&cgroup_mutex); | |
980 | cgrp->id = -1; | |
ddbcc7e8 | 981 | |
2bd59d48 | 982 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 PM |
983 | } |
984 | ||
2739d3cc | 985 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 986 | { |
2bd59d48 | 987 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 988 | |
ace2bee8 | 989 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 990 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
991 | } |
992 | ||
13af07df | 993 | /** |
628f7cd4 | 994 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 995 | * @cgrp: target cgroup |
13af07df AR |
996 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
997 | */ | |
628f7cd4 | 998 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 999 | { |
13af07df | 1000 | struct cgroup_subsys *ss; |
b420ba7d | 1001 | int i; |
05ef1d7c | 1002 | |
b420ba7d | 1003 | for_each_subsys(ss, i) { |
0adb0704 | 1004 | struct cftype *cfts; |
b420ba7d TH |
1005 | |
1006 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 1007 | continue; |
0adb0704 TH |
1008 | list_for_each_entry(cfts, &ss->cfts, node) |
1009 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 1010 | } |
ddbcc7e8 PM |
1011 | } |
1012 | ||
ddbcc7e8 | 1013 | static int rebind_subsystems(struct cgroupfs_root *root, |
a8a648c4 | 1014 | unsigned long added_mask, unsigned removed_mask) |
ddbcc7e8 | 1015 | { |
bd89aabc | 1016 | struct cgroup *cgrp = &root->top_cgroup; |
30159ec7 | 1017 | struct cgroup_subsys *ss; |
3126121f | 1018 | int i, ret; |
ddbcc7e8 | 1019 | |
ace2bee8 TH |
1020 | lockdep_assert_held(&cgroup_tree_mutex); |
1021 | lockdep_assert_held(&cgroup_mutex); | |
aae8aab4 | 1022 | |
ddbcc7e8 | 1023 | /* Check that any added subsystems are currently free */ |
3ed80a62 TH |
1024 | for_each_subsys(ss, i) |
1025 | if ((added_mask & (1 << i)) && ss->root != &cgroup_dummy_root) | |
1026 | return -EBUSY; | |
ddbcc7e8 | 1027 | |
3126121f TH |
1028 | ret = cgroup_populate_dir(cgrp, added_mask); |
1029 | if (ret) | |
3ed80a62 | 1030 | return ret; |
3126121f TH |
1031 | |
1032 | /* | |
1033 | * Nothing can fail from this point on. Remove files for the | |
1034 | * removed subsystems and rebind each subsystem. | |
1035 | */ | |
4ac06017 | 1036 | mutex_unlock(&cgroup_mutex); |
3126121f | 1037 | cgroup_clear_dir(cgrp, removed_mask); |
4ac06017 | 1038 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1039 | |
30159ec7 | 1040 | for_each_subsys(ss, i) { |
ddbcc7e8 | 1041 | unsigned long bit = 1UL << i; |
30159ec7 | 1042 | |
a1a71b45 | 1043 | if (bit & added_mask) { |
ddbcc7e8 | 1044 | /* We're binding this subsystem to this hierarchy */ |
ca8bdcaf TH |
1045 | BUG_ON(cgroup_css(cgrp, ss)); |
1046 | BUG_ON(!cgroup_css(cgroup_dummy_top, ss)); | |
1047 | BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top); | |
a8a648c4 | 1048 | |
73e80ed8 | 1049 | rcu_assign_pointer(cgrp->subsys[i], |
ca8bdcaf TH |
1050 | cgroup_css(cgroup_dummy_top, ss)); |
1051 | cgroup_css(cgrp, ss)->cgroup = cgrp; | |
a8a648c4 | 1052 | |
b2aa30f7 | 1053 | ss->root = root; |
ddbcc7e8 | 1054 | if (ss->bind) |
ca8bdcaf | 1055 | ss->bind(cgroup_css(cgrp, ss)); |
a8a648c4 | 1056 | |
cf5d5941 | 1057 | /* refcount was already taken, and we're keeping it */ |
a8a648c4 | 1058 | root->subsys_mask |= bit; |
a1a71b45 | 1059 | } else if (bit & removed_mask) { |
ddbcc7e8 | 1060 | /* We're removing this subsystem */ |
ca8bdcaf TH |
1061 | BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss)); |
1062 | BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp); | |
a8a648c4 | 1063 | |
ddbcc7e8 | 1064 | if (ss->bind) |
ca8bdcaf | 1065 | ss->bind(cgroup_css(cgroup_dummy_top, ss)); |
73e80ed8 | 1066 | |
ca8bdcaf | 1067 | cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top; |
73e80ed8 TH |
1068 | RCU_INIT_POINTER(cgrp->subsys[i], NULL); |
1069 | ||
9871bf95 | 1070 | cgroup_subsys[i]->root = &cgroup_dummy_root; |
a8a648c4 | 1071 | root->subsys_mask &= ~bit; |
ddbcc7e8 PM |
1072 | } |
1073 | } | |
ddbcc7e8 | 1074 | |
1672d040 TH |
1075 | /* |
1076 | * Mark @root has finished binding subsystems. @root->subsys_mask | |
1077 | * now matches the bound subsystems. | |
1078 | */ | |
1079 | root->flags |= CGRP_ROOT_SUBSYS_BOUND; | |
2bd59d48 | 1080 | kernfs_activate(cgrp->kn); |
1672d040 | 1081 | |
ddbcc7e8 PM |
1082 | return 0; |
1083 | } | |
1084 | ||
2bd59d48 TH |
1085 | static int cgroup_show_options(struct seq_file *seq, |
1086 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1087 | { |
2bd59d48 | 1088 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1089 | struct cgroup_subsys *ss; |
b85d2040 | 1090 | int ssid; |
ddbcc7e8 | 1091 | |
b85d2040 TH |
1092 | for_each_subsys(ss, ssid) |
1093 | if (root->subsys_mask & (1 << ssid)) | |
1094 | seq_printf(seq, ",%s", ss->name); | |
873fe09e TH |
1095 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1096 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1097 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1098 | seq_puts(seq, ",noprefix"); |
93438629 | 1099 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1100 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1101 | |
1102 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1103 | if (strlen(root->release_agent_path)) |
1104 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1105 | spin_unlock(&release_agent_path_lock); |
1106 | ||
2260e7fc | 1107 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags)) |
97978e6d | 1108 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1109 | if (strlen(root->name)) |
1110 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1111 | return 0; |
1112 | } | |
1113 | ||
1114 | struct cgroup_sb_opts { | |
a1a71b45 | 1115 | unsigned long subsys_mask; |
ddbcc7e8 | 1116 | unsigned long flags; |
81a6a5cd | 1117 | char *release_agent; |
2260e7fc | 1118 | bool cpuset_clone_children; |
c6d57f33 | 1119 | char *name; |
2c6ab6d2 PM |
1120 | /* User explicitly requested empty subsystem */ |
1121 | bool none; | |
ddbcc7e8 PM |
1122 | }; |
1123 | ||
aae8aab4 | 1124 | /* |
9871bf95 TH |
1125 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1126 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1127 | * array. This function takes refcounts on subsystems to be used, unless it | |
1128 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1129 | */ |
cf5d5941 | 1130 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1131 | { |
32a8cf23 DL |
1132 | char *token, *o = data; |
1133 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1134 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1135 | struct cgroup_subsys *ss; |
1136 | int i; | |
f9ab5b5b | 1137 | |
aae8aab4 BB |
1138 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1139 | ||
f9ab5b5b | 1140 | #ifdef CONFIG_CPUSETS |
073219e9 | 1141 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1142 | #endif |
ddbcc7e8 | 1143 | |
c6d57f33 | 1144 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1145 | |
1146 | while ((token = strsep(&o, ",")) != NULL) { | |
1147 | if (!*token) | |
1148 | return -EINVAL; | |
32a8cf23 | 1149 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1150 | /* Explicitly have no subsystems */ |
1151 | opts->none = true; | |
32a8cf23 DL |
1152 | continue; |
1153 | } | |
1154 | if (!strcmp(token, "all")) { | |
1155 | /* Mutually exclusive option 'all' + subsystem name */ | |
1156 | if (one_ss) | |
1157 | return -EINVAL; | |
1158 | all_ss = true; | |
1159 | continue; | |
1160 | } | |
873fe09e TH |
1161 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1162 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1163 | continue; | |
1164 | } | |
32a8cf23 | 1165 | if (!strcmp(token, "noprefix")) { |
93438629 | 1166 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1167 | continue; |
1168 | } | |
1169 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1170 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1171 | continue; |
1172 | } | |
03b1cde6 | 1173 | if (!strcmp(token, "xattr")) { |
93438629 | 1174 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1175 | continue; |
1176 | } | |
32a8cf23 | 1177 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1178 | /* Specifying two release agents is forbidden */ |
1179 | if (opts->release_agent) | |
1180 | return -EINVAL; | |
c6d57f33 | 1181 | opts->release_agent = |
e400c285 | 1182 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1183 | if (!opts->release_agent) |
1184 | return -ENOMEM; | |
32a8cf23 DL |
1185 | continue; |
1186 | } | |
1187 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1188 | const char *name = token + 5; |
1189 | /* Can't specify an empty name */ | |
1190 | if (!strlen(name)) | |
1191 | return -EINVAL; | |
1192 | /* Must match [\w.-]+ */ | |
1193 | for (i = 0; i < strlen(name); i++) { | |
1194 | char c = name[i]; | |
1195 | if (isalnum(c)) | |
1196 | continue; | |
1197 | if ((c == '.') || (c == '-') || (c == '_')) | |
1198 | continue; | |
1199 | return -EINVAL; | |
1200 | } | |
1201 | /* Specifying two names is forbidden */ | |
1202 | if (opts->name) | |
1203 | return -EINVAL; | |
1204 | opts->name = kstrndup(name, | |
e400c285 | 1205 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1206 | GFP_KERNEL); |
1207 | if (!opts->name) | |
1208 | return -ENOMEM; | |
32a8cf23 DL |
1209 | |
1210 | continue; | |
1211 | } | |
1212 | ||
30159ec7 | 1213 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1214 | if (strcmp(token, ss->name)) |
1215 | continue; | |
1216 | if (ss->disabled) | |
1217 | continue; | |
1218 | ||
1219 | /* Mutually exclusive option 'all' + subsystem name */ | |
1220 | if (all_ss) | |
1221 | return -EINVAL; | |
a1a71b45 | 1222 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1223 | one_ss = true; |
1224 | ||
1225 | break; | |
1226 | } | |
1227 | if (i == CGROUP_SUBSYS_COUNT) | |
1228 | return -ENOENT; | |
1229 | } | |
1230 | ||
1231 | /* | |
1232 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1233 | * otherwise if 'none', 'name=' and a subsystem name options |
1234 | * were not specified, let's default to 'all' | |
32a8cf23 | 1235 | */ |
30159ec7 TH |
1236 | if (all_ss || (!one_ss && !opts->none && !opts->name)) |
1237 | for_each_subsys(ss, i) | |
1238 | if (!ss->disabled) | |
1239 | set_bit(i, &opts->subsys_mask); | |
ddbcc7e8 | 1240 | |
2c6ab6d2 PM |
1241 | /* Consistency checks */ |
1242 | ||
873fe09e TH |
1243 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1244 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1245 | ||
1246 | if (opts->flags & CGRP_ROOT_NOPREFIX) { | |
1247 | pr_err("cgroup: sane_behavior: noprefix is not allowed\n"); | |
1248 | return -EINVAL; | |
1249 | } | |
1250 | ||
1251 | if (opts->cpuset_clone_children) { | |
1252 | pr_err("cgroup: sane_behavior: clone_children is not allowed\n"); | |
1253 | return -EINVAL; | |
1254 | } | |
86bf4b68 TH |
1255 | |
1256 | if (opts->flags & CGRP_ROOT_XATTR) | |
1257 | pr_warning("cgroup: sane_behavior: xattr is always available, flag unnecessary\n"); | |
873fe09e TH |
1258 | } |
1259 | ||
f9ab5b5b LZ |
1260 | /* |
1261 | * Option noprefix was introduced just for backward compatibility | |
1262 | * with the old cpuset, so we allow noprefix only if mounting just | |
1263 | * the cpuset subsystem. | |
1264 | */ | |
93438629 | 1265 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1266 | return -EINVAL; |
1267 | ||
2c6ab6d2 PM |
1268 | |
1269 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1270 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1271 | return -EINVAL; |
1272 | ||
1273 | /* | |
1274 | * We either have to specify by name or by subsystems. (So all | |
1275 | * empty hierarchies must have a name). | |
1276 | */ | |
a1a71b45 | 1277 | if (!opts->subsys_mask && !opts->name) |
ddbcc7e8 PM |
1278 | return -EINVAL; |
1279 | ||
1280 | return 0; | |
1281 | } | |
1282 | ||
2bd59d48 | 1283 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1284 | { |
1285 | int ret = 0; | |
2bd59d48 | 1286 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1287 | struct cgroup_sb_opts opts; |
a1a71b45 | 1288 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1289 | |
873fe09e TH |
1290 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1291 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1292 | return -EINVAL; | |
1293 | } | |
1294 | ||
ace2bee8 | 1295 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1296 | mutex_lock(&cgroup_mutex); |
1297 | ||
1298 | /* See what subsystems are wanted */ | |
1299 | ret = parse_cgroupfs_options(data, &opts); | |
1300 | if (ret) | |
1301 | goto out_unlock; | |
1302 | ||
a8a648c4 | 1303 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1304 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1305 | task_tgid_nr(current), current->comm); | |
1306 | ||
a1a71b45 AR |
1307 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1308 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1309 | |
cf5d5941 | 1310 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1311 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1312 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1313 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1314 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1315 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1316 | ret = -EINVAL; |
1317 | goto out_unlock; | |
1318 | } | |
1319 | ||
f172e67c | 1320 | /* remounting is not allowed for populated hierarchies */ |
3c9c825b | 1321 | if (!list_empty(&root->top_cgroup.children)) { |
f172e67c | 1322 | ret = -EBUSY; |
0670e08b | 1323 | goto out_unlock; |
cf5d5941 | 1324 | } |
ddbcc7e8 | 1325 | |
a8a648c4 | 1326 | ret = rebind_subsystems(root, added_mask, removed_mask); |
3126121f | 1327 | if (ret) |
0670e08b | 1328 | goto out_unlock; |
ddbcc7e8 | 1329 | |
69e943b7 TH |
1330 | if (opts.release_agent) { |
1331 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1332 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1333 | spin_unlock(&release_agent_path_lock); |
1334 | } | |
ddbcc7e8 | 1335 | out_unlock: |
66bdc9cf | 1336 | kfree(opts.release_agent); |
c6d57f33 | 1337 | kfree(opts.name); |
ddbcc7e8 | 1338 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1339 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1340 | return ret; |
1341 | } | |
1342 | ||
cc31edce PM |
1343 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1344 | { | |
2bd59d48 | 1345 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1346 | INIT_LIST_HEAD(&cgrp->sibling); |
1347 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1348 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1349 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1350 | INIT_LIST_HEAD(&cgrp->pidlists); |
1351 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1352 | cgrp->dummy_css.cgroup = cgrp; |
cc31edce | 1353 | } |
c6d57f33 | 1354 | |
ddbcc7e8 PM |
1355 | static void init_cgroup_root(struct cgroupfs_root *root) |
1356 | { | |
bd89aabc | 1357 | struct cgroup *cgrp = &root->top_cgroup; |
b0ca5a84 | 1358 | |
2bd59d48 | 1359 | atomic_set(&root->refcnt, 1); |
ddbcc7e8 | 1360 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1361 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1362 | cgrp->root = root; |
cc31edce | 1363 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1364 | idr_init(&root->cgroup_idr); |
ddbcc7e8 PM |
1365 | } |
1366 | ||
c6d57f33 PM |
1367 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1368 | { | |
1369 | struct cgroupfs_root *root; | |
1370 | ||
a1a71b45 | 1371 | if (!opts->subsys_mask && !opts->none) |
2bd59d48 | 1372 | return ERR_PTR(-EINVAL); |
c6d57f33 PM |
1373 | |
1374 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1375 | if (!root) | |
1376 | return ERR_PTR(-ENOMEM); | |
1377 | ||
1378 | init_cgroup_root(root); | |
2c6ab6d2 | 1379 | |
1672d040 TH |
1380 | /* |
1381 | * We need to set @root->subsys_mask now so that @root can be | |
1382 | * matched by cgroup_test_super() before it finishes | |
1383 | * initialization; otherwise, competing mounts with the same | |
1384 | * options may try to bind the same subsystems instead of waiting | |
1385 | * for the first one leading to unexpected mount errors. | |
1386 | * SUBSYS_BOUND will be set once actual binding is complete. | |
1387 | */ | |
a1a71b45 | 1388 | root->subsys_mask = opts->subsys_mask; |
c6d57f33 PM |
1389 | root->flags = opts->flags; |
1390 | if (opts->release_agent) | |
1391 | strcpy(root->release_agent_path, opts->release_agent); | |
1392 | if (opts->name) | |
1393 | strcpy(root->name, opts->name); | |
2260e7fc TH |
1394 | if (opts->cpuset_clone_children) |
1395 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1396 | return root; |
1397 | } | |
1398 | ||
d427dfeb TH |
1399 | static int cgroup_setup_root(struct cgroupfs_root *root) |
1400 | { | |
1401 | LIST_HEAD(tmp_links); | |
d427dfeb | 1402 | struct cgroup *root_cgrp = &root->top_cgroup; |
d427dfeb | 1403 | struct css_set *cset; |
d427dfeb TH |
1404 | int i, ret; |
1405 | ||
1406 | lockdep_assert_held(&cgroup_tree_mutex); | |
1407 | lockdep_assert_held(&cgroup_mutex); | |
d427dfeb TH |
1408 | |
1409 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); | |
1410 | if (ret < 0) | |
2bd59d48 | 1411 | goto out; |
d427dfeb TH |
1412 | root_cgrp->id = ret; |
1413 | ||
d427dfeb TH |
1414 | /* |
1415 | * We're accessing css_set_count without locking css_set_lock here, | |
1416 | * but that's OK - it can only be increased by someone holding | |
1417 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1418 | * have some link structures left over | |
1419 | */ | |
1420 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1421 | if (ret) | |
2bd59d48 | 1422 | goto out; |
d427dfeb TH |
1423 | |
1424 | /* ID 0 is reserved for dummy root, 1 for unified hierarchy */ | |
1425 | ret = cgroup_init_root_id(root, 2, 0); | |
1426 | if (ret) | |
2bd59d48 | 1427 | goto out; |
d427dfeb | 1428 | |
2bd59d48 TH |
1429 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1430 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1431 | root_cgrp); | |
1432 | if (IS_ERR(root->kf_root)) { | |
1433 | ret = PTR_ERR(root->kf_root); | |
1434 | goto exit_root_id; | |
1435 | } | |
1436 | root_cgrp->kn = root->kf_root->kn; | |
d427dfeb TH |
1437 | |
1438 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); | |
1439 | if (ret) | |
2bd59d48 | 1440 | goto destroy_root; |
d427dfeb TH |
1441 | |
1442 | ret = rebind_subsystems(root, root->subsys_mask, 0); | |
1443 | if (ret) | |
2bd59d48 | 1444 | goto destroy_root; |
d427dfeb TH |
1445 | |
1446 | /* | |
1447 | * There must be no failure case after here, since rebinding takes | |
1448 | * care of subsystems' refcounts, which are explicitly dropped in | |
1449 | * the failure exit path. | |
1450 | */ | |
1451 | list_add(&root->root_list, &cgroup_roots); | |
1452 | cgroup_root_count++; | |
1453 | ||
1454 | /* | |
1455 | * Link the top cgroup in this hierarchy into all the css_set | |
1456 | * objects. | |
1457 | */ | |
1458 | write_lock(&css_set_lock); | |
1459 | hash_for_each(css_set_table, i, cset, hlist) | |
1460 | link_css_set(&tmp_links, cset, root_cgrp); | |
1461 | write_unlock(&css_set_lock); | |
1462 | ||
1463 | BUG_ON(!list_empty(&root_cgrp->children)); | |
3c9c825b | 1464 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
d427dfeb | 1465 | |
2bd59d48 | 1466 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1467 | ret = 0; |
2bd59d48 | 1468 | goto out; |
d427dfeb | 1469 | |
2bd59d48 TH |
1470 | destroy_root: |
1471 | kernfs_destroy_root(root->kf_root); | |
1472 | root->kf_root = NULL; | |
1473 | exit_root_id: | |
d427dfeb | 1474 | cgroup_exit_root_id(root); |
2bd59d48 | 1475 | out: |
d427dfeb TH |
1476 | free_cgrp_cset_links(&tmp_links); |
1477 | return ret; | |
1478 | } | |
1479 | ||
f7e83571 | 1480 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1481 | int flags, const char *unused_dev_name, |
f7e83571 | 1482 | void *data) |
ddbcc7e8 | 1483 | { |
2bd59d48 | 1484 | struct cgroupfs_root *root; |
ddbcc7e8 | 1485 | struct cgroup_sb_opts opts; |
2bd59d48 | 1486 | struct dentry *dentry; |
8e30e2b8 | 1487 | int ret; |
ddbcc7e8 | 1488 | |
8e30e2b8 | 1489 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1490 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1491 | |
1492 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1493 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1494 | if (ret) |
8e30e2b8 | 1495 | goto out_unlock; |
ddbcc7e8 | 1496 | |
2bd59d48 TH |
1497 | /* look for a matching existing root */ |
1498 | for_each_active_root(root) { | |
1499 | bool name_match = false; | |
ddbcc7e8 | 1500 | |
2bd59d48 TH |
1501 | /* |
1502 | * If we asked for a name then it must match. Also, if | |
1503 | * name matches but sybsys_mask doesn't, we should fail. | |
1504 | * Remember whether name matched. | |
1505 | */ | |
1506 | if (opts.name) { | |
1507 | if (strcmp(opts.name, root->name)) | |
1508 | continue; | |
1509 | name_match = true; | |
1510 | } | |
ddbcc7e8 | 1511 | |
c6d57f33 | 1512 | /* |
2bd59d48 TH |
1513 | * If we asked for subsystems (or explicitly for no |
1514 | * subsystems) then they must match. | |
c6d57f33 | 1515 | */ |
2bd59d48 TH |
1516 | if ((opts.subsys_mask || opts.none) && |
1517 | (opts.subsys_mask != root->subsys_mask)) { | |
1518 | if (!name_match) | |
1519 | continue; | |
1520 | ret = -EBUSY; | |
1521 | goto out_unlock; | |
1522 | } | |
873fe09e | 1523 | |
c7ba8287 | 1524 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1525 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1526 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1527 | ret = -EINVAL; | |
8e30e2b8 | 1528 | goto out_unlock; |
2a0ff3fb JL |
1529 | } else { |
1530 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1531 | } | |
873fe09e | 1532 | } |
2bd59d48 TH |
1533 | |
1534 | cgroup_get_root(root); | |
1535 | goto out_unlock; | |
ddbcc7e8 PM |
1536 | } |
1537 | ||
2bd59d48 TH |
1538 | /* no such thing, create a new one */ |
1539 | root = cgroup_root_from_opts(&opts); | |
1540 | if (IS_ERR(root)) { | |
1541 | ret = PTR_ERR(root); | |
1542 | goto out_unlock; | |
1543 | } | |
1544 | ||
1545 | ret = cgroup_setup_root(root); | |
1546 | if (ret) | |
1547 | cgroup_free_root(root); | |
1548 | ||
8e30e2b8 | 1549 | out_unlock: |
e25e2cbb | 1550 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1551 | mutex_unlock(&cgroup_tree_mutex); |
8e30e2b8 | 1552 | |
c6d57f33 PM |
1553 | kfree(opts.release_agent); |
1554 | kfree(opts.name); | |
8e30e2b8 | 1555 | |
2bd59d48 | 1556 | if (ret) |
8e30e2b8 | 1557 | return ERR_PTR(ret); |
2bd59d48 TH |
1558 | |
1559 | dentry = kernfs_mount(fs_type, flags, root->kf_root); | |
1560 | if (IS_ERR(dentry)) | |
1561 | cgroup_put_root(root); | |
1562 | return dentry; | |
1563 | } | |
1564 | ||
1565 | static void cgroup_kill_sb(struct super_block *sb) | |
1566 | { | |
1567 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
1568 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); | |
1569 | ||
1570 | cgroup_put_root(root); | |
1571 | kernfs_kill_sb(sb); | |
ddbcc7e8 PM |
1572 | } |
1573 | ||
ddbcc7e8 PM |
1574 | static struct file_system_type cgroup_fs_type = { |
1575 | .name = "cgroup", | |
f7e83571 | 1576 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1577 | .kill_sb = cgroup_kill_sb, |
1578 | }; | |
1579 | ||
676db4af GK |
1580 | static struct kobject *cgroup_kobj; |
1581 | ||
857a2beb | 1582 | /** |
913ffdb5 | 1583 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1584 | * @task: target task |
857a2beb TH |
1585 | * @buf: the buffer to write the path into |
1586 | * @buflen: the length of the buffer | |
1587 | * | |
913ffdb5 TH |
1588 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1589 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1590 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1591 | * cgroup controller callbacks. | |
1592 | * | |
e61734c5 | 1593 | * Return value is the same as kernfs_path(). |
857a2beb | 1594 | */ |
e61734c5 | 1595 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb TH |
1596 | { |
1597 | struct cgroupfs_root *root; | |
913ffdb5 | 1598 | struct cgroup *cgrp; |
e61734c5 TH |
1599 | int hierarchy_id = 1; |
1600 | char *path = NULL; | |
857a2beb TH |
1601 | |
1602 | mutex_lock(&cgroup_mutex); | |
1603 | ||
913ffdb5 TH |
1604 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1605 | ||
857a2beb TH |
1606 | if (root) { |
1607 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1608 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1609 | } else { |
1610 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1611 | if (strlcpy(buf, "/", buflen) < buflen) |
1612 | path = buf; | |
857a2beb TH |
1613 | } |
1614 | ||
1615 | mutex_unlock(&cgroup_mutex); | |
e61734c5 | 1616 | return path; |
857a2beb | 1617 | } |
913ffdb5 | 1618 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1619 | |
2f7ee569 TH |
1620 | /* |
1621 | * Control Group taskset | |
1622 | */ | |
134d3373 TH |
1623 | struct task_and_cgroup { |
1624 | struct task_struct *task; | |
1625 | struct cgroup *cgrp; | |
6f4b7e63 | 1626 | struct css_set *cset; |
134d3373 TH |
1627 | }; |
1628 | ||
2f7ee569 TH |
1629 | struct cgroup_taskset { |
1630 | struct task_and_cgroup single; | |
1631 | struct flex_array *tc_array; | |
1632 | int tc_array_len; | |
1633 | int idx; | |
1634 | struct cgroup *cur_cgrp; | |
1635 | }; | |
1636 | ||
1637 | /** | |
1638 | * cgroup_taskset_first - reset taskset and return the first task | |
1639 | * @tset: taskset of interest | |
1640 | * | |
1641 | * @tset iteration is initialized and the first task is returned. | |
1642 | */ | |
1643 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1644 | { | |
1645 | if (tset->tc_array) { | |
1646 | tset->idx = 0; | |
1647 | return cgroup_taskset_next(tset); | |
1648 | } else { | |
1649 | tset->cur_cgrp = tset->single.cgrp; | |
1650 | return tset->single.task; | |
1651 | } | |
1652 | } | |
1653 | EXPORT_SYMBOL_GPL(cgroup_taskset_first); | |
1654 | ||
1655 | /** | |
1656 | * cgroup_taskset_next - iterate to the next task in taskset | |
1657 | * @tset: taskset of interest | |
1658 | * | |
1659 | * Return the next task in @tset. Iteration must have been initialized | |
1660 | * with cgroup_taskset_first(). | |
1661 | */ | |
1662 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1663 | { | |
1664 | struct task_and_cgroup *tc; | |
1665 | ||
1666 | if (!tset->tc_array || tset->idx >= tset->tc_array_len) | |
1667 | return NULL; | |
1668 | ||
1669 | tc = flex_array_get(tset->tc_array, tset->idx++); | |
1670 | tset->cur_cgrp = tc->cgrp; | |
1671 | return tc->task; | |
1672 | } | |
1673 | EXPORT_SYMBOL_GPL(cgroup_taskset_next); | |
1674 | ||
1675 | /** | |
d99c8727 | 1676 | * cgroup_taskset_cur_css - return the matching css for the current task |
2f7ee569 | 1677 | * @tset: taskset of interest |
d99c8727 | 1678 | * @subsys_id: the ID of the target subsystem |
2f7ee569 | 1679 | * |
d99c8727 TH |
1680 | * Return the css for the current (last returned) task of @tset for |
1681 | * subsystem specified by @subsys_id. This function must be preceded by | |
1682 | * either cgroup_taskset_first() or cgroup_taskset_next(). | |
2f7ee569 | 1683 | */ |
d99c8727 TH |
1684 | struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset, |
1685 | int subsys_id) | |
2f7ee569 | 1686 | { |
ca8bdcaf | 1687 | return cgroup_css(tset->cur_cgrp, cgroup_subsys[subsys_id]); |
2f7ee569 | 1688 | } |
d99c8727 | 1689 | EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css); |
2f7ee569 TH |
1690 | |
1691 | /** | |
1692 | * cgroup_taskset_size - return the number of tasks in taskset | |
1693 | * @tset: taskset of interest | |
1694 | */ | |
1695 | int cgroup_taskset_size(struct cgroup_taskset *tset) | |
1696 | { | |
1697 | return tset->tc_array ? tset->tc_array_len : 1; | |
1698 | } | |
1699 | EXPORT_SYMBOL_GPL(cgroup_taskset_size); | |
1700 | ||
1701 | ||
74a1166d BB |
1702 | /* |
1703 | * cgroup_task_migrate - move a task from one cgroup to another. | |
1704 | * | |
d0b2fdd2 | 1705 | * Must be called with cgroup_mutex and threadgroup locked. |
74a1166d | 1706 | */ |
5abb8855 TH |
1707 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1708 | struct task_struct *tsk, | |
1709 | struct css_set *new_cset) | |
74a1166d | 1710 | { |
5abb8855 | 1711 | struct css_set *old_cset; |
74a1166d BB |
1712 | |
1713 | /* | |
026085ef MSB |
1714 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1715 | * setting such that we can't race against cgroup_exit() changing the | |
1716 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1717 | */ |
c84cdf75 | 1718 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1719 | old_cset = task_css_set(tsk); |
74a1166d | 1720 | |
74a1166d | 1721 | task_lock(tsk); |
5abb8855 | 1722 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d BB |
1723 | task_unlock(tsk); |
1724 | ||
1725 | /* Update the css_set linked lists if we're using them */ | |
1726 | write_lock(&css_set_lock); | |
1727 | if (!list_empty(&tsk->cg_list)) | |
5abb8855 | 1728 | list_move(&tsk->cg_list, &new_cset->tasks); |
74a1166d BB |
1729 | write_unlock(&css_set_lock); |
1730 | ||
1731 | /* | |
5abb8855 TH |
1732 | * We just gained a reference on old_cset by taking it from the |
1733 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1734 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1735 | */ |
5abb8855 TH |
1736 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
1737 | put_css_set(old_cset); | |
74a1166d BB |
1738 | } |
1739 | ||
a043e3b2 | 1740 | /** |
081aa458 | 1741 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup |
74a1166d | 1742 | * @cgrp: the cgroup to attach to |
081aa458 LZ |
1743 | * @tsk: the task or the leader of the threadgroup to be attached |
1744 | * @threadgroup: attach the whole threadgroup? | |
74a1166d | 1745 | * |
257058ae | 1746 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
081aa458 | 1747 | * task_lock of @tsk or each thread in the threadgroup individually in turn. |
74a1166d | 1748 | */ |
47cfcd09 TH |
1749 | static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk, |
1750 | bool threadgroup) | |
74a1166d BB |
1751 | { |
1752 | int retval, i, group_size; | |
74a1166d | 1753 | struct cgroupfs_root *root = cgrp->root; |
1c6727af | 1754 | struct cgroup_subsys_state *css, *failed_css = NULL; |
74a1166d | 1755 | /* threadgroup list cursor and array */ |
081aa458 | 1756 | struct task_struct *leader = tsk; |
134d3373 | 1757 | struct task_and_cgroup *tc; |
d846687d | 1758 | struct flex_array *group; |
2f7ee569 | 1759 | struct cgroup_taskset tset = { }; |
74a1166d BB |
1760 | |
1761 | /* | |
1762 | * step 0: in order to do expensive, possibly blocking operations for | |
1763 | * every thread, we cannot iterate the thread group list, since it needs | |
1764 | * rcu or tasklist locked. instead, build an array of all threads in the | |
257058ae TH |
1765 | * group - group_rwsem prevents new threads from appearing, and if |
1766 | * threads exit, this will just be an over-estimate. | |
74a1166d | 1767 | */ |
081aa458 LZ |
1768 | if (threadgroup) |
1769 | group_size = get_nr_threads(tsk); | |
1770 | else | |
1771 | group_size = 1; | |
d846687d | 1772 | /* flex_array supports very large thread-groups better than kmalloc. */ |
134d3373 | 1773 | group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL); |
74a1166d BB |
1774 | if (!group) |
1775 | return -ENOMEM; | |
d846687d | 1776 | /* pre-allocate to guarantee space while iterating in rcu read-side. */ |
3ac1707a | 1777 | retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL); |
d846687d BB |
1778 | if (retval) |
1779 | goto out_free_group_list; | |
74a1166d | 1780 | |
74a1166d | 1781 | i = 0; |
fb5d2b4c MSB |
1782 | /* |
1783 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1784 | * already PF_EXITING could be freed from underneath us unless we | |
1785 | * take an rcu_read_lock. | |
1786 | */ | |
1787 | rcu_read_lock(); | |
74a1166d | 1788 | do { |
134d3373 TH |
1789 | struct task_and_cgroup ent; |
1790 | ||
cd3d0952 TH |
1791 | /* @tsk either already exited or can't exit until the end */ |
1792 | if (tsk->flags & PF_EXITING) | |
ea84753c | 1793 | goto next; |
cd3d0952 | 1794 | |
74a1166d BB |
1795 | /* as per above, nr_threads may decrease, but not increase. */ |
1796 | BUG_ON(i >= group_size); | |
134d3373 TH |
1797 | ent.task = tsk; |
1798 | ent.cgrp = task_cgroup_from_root(tsk, root); | |
892a2b90 MSB |
1799 | /* nothing to do if this task is already in the cgroup */ |
1800 | if (ent.cgrp == cgrp) | |
ea84753c | 1801 | goto next; |
61d1d219 MSB |
1802 | /* |
1803 | * saying GFP_ATOMIC has no effect here because we did prealloc | |
1804 | * earlier, but it's good form to communicate our expectations. | |
1805 | */ | |
134d3373 | 1806 | retval = flex_array_put(group, i, &ent, GFP_ATOMIC); |
d846687d | 1807 | BUG_ON(retval != 0); |
74a1166d | 1808 | i++; |
ea84753c | 1809 | next: |
081aa458 LZ |
1810 | if (!threadgroup) |
1811 | break; | |
74a1166d | 1812 | } while_each_thread(leader, tsk); |
fb5d2b4c | 1813 | rcu_read_unlock(); |
74a1166d BB |
1814 | /* remember the number of threads in the array for later. */ |
1815 | group_size = i; | |
2f7ee569 TH |
1816 | tset.tc_array = group; |
1817 | tset.tc_array_len = group_size; | |
74a1166d | 1818 | |
134d3373 TH |
1819 | /* methods shouldn't be called if no task is actually migrating */ |
1820 | retval = 0; | |
892a2b90 | 1821 | if (!group_size) |
b07ef774 | 1822 | goto out_free_group_list; |
134d3373 | 1823 | |
74a1166d BB |
1824 | /* |
1825 | * step 1: check that we can legitimately attach to the cgroup. | |
1826 | */ | |
1c6727af TH |
1827 | for_each_css(css, i, cgrp) { |
1828 | if (css->ss->can_attach) { | |
1829 | retval = css->ss->can_attach(css, &tset); | |
74a1166d | 1830 | if (retval) { |
1c6727af | 1831 | failed_css = css; |
74a1166d BB |
1832 | goto out_cancel_attach; |
1833 | } | |
1834 | } | |
74a1166d BB |
1835 | } |
1836 | ||
1837 | /* | |
1838 | * step 2: make sure css_sets exist for all threads to be migrated. | |
1839 | * we use find_css_set, which allocates a new one if necessary. | |
1840 | */ | |
74a1166d | 1841 | for (i = 0; i < group_size; i++) { |
a8ad805c TH |
1842 | struct css_set *old_cset; |
1843 | ||
134d3373 | 1844 | tc = flex_array_get(group, i); |
a8ad805c | 1845 | old_cset = task_css_set(tc->task); |
6f4b7e63 LZ |
1846 | tc->cset = find_css_set(old_cset, cgrp); |
1847 | if (!tc->cset) { | |
61d1d219 MSB |
1848 | retval = -ENOMEM; |
1849 | goto out_put_css_set_refs; | |
74a1166d BB |
1850 | } |
1851 | } | |
1852 | ||
1853 | /* | |
494c167c TH |
1854 | * step 3: now that we're guaranteed success wrt the css_sets, |
1855 | * proceed to move all tasks to the new cgroup. There are no | |
1856 | * failure cases after here, so this is the commit point. | |
74a1166d | 1857 | */ |
74a1166d | 1858 | for (i = 0; i < group_size; i++) { |
134d3373 | 1859 | tc = flex_array_get(group, i); |
6f4b7e63 | 1860 | cgroup_task_migrate(tc->cgrp, tc->task, tc->cset); |
74a1166d BB |
1861 | } |
1862 | /* nothing is sensitive to fork() after this point. */ | |
1863 | ||
1864 | /* | |
494c167c | 1865 | * step 4: do subsystem attach callbacks. |
74a1166d | 1866 | */ |
1c6727af TH |
1867 | for_each_css(css, i, cgrp) |
1868 | if (css->ss->attach) | |
1869 | css->ss->attach(css, &tset); | |
74a1166d BB |
1870 | |
1871 | /* | |
1872 | * step 5: success! and cleanup | |
1873 | */ | |
74a1166d | 1874 | retval = 0; |
61d1d219 MSB |
1875 | out_put_css_set_refs: |
1876 | if (retval) { | |
1877 | for (i = 0; i < group_size; i++) { | |
1878 | tc = flex_array_get(group, i); | |
6f4b7e63 | 1879 | if (!tc->cset) |
61d1d219 | 1880 | break; |
6f4b7e63 | 1881 | put_css_set(tc->cset); |
61d1d219 | 1882 | } |
74a1166d BB |
1883 | } |
1884 | out_cancel_attach: | |
74a1166d | 1885 | if (retval) { |
1c6727af TH |
1886 | for_each_css(css, i, cgrp) { |
1887 | if (css == failed_css) | |
74a1166d | 1888 | break; |
1c6727af TH |
1889 | if (css->ss->cancel_attach) |
1890 | css->ss->cancel_attach(css, &tset); | |
74a1166d BB |
1891 | } |
1892 | } | |
74a1166d | 1893 | out_free_group_list: |
d846687d | 1894 | flex_array_free(group); |
74a1166d BB |
1895 | return retval; |
1896 | } | |
1897 | ||
1898 | /* | |
1899 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
1900 | * function to attach either it or all tasks in its threadgroup. Will lock |
1901 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 1902 | */ |
74a1166d | 1903 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 1904 | { |
bbcb81d0 | 1905 | struct task_struct *tsk; |
c69e8d9c | 1906 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
1907 | int ret; |
1908 | ||
74a1166d BB |
1909 | if (!cgroup_lock_live_group(cgrp)) |
1910 | return -ENODEV; | |
1911 | ||
b78949eb MSB |
1912 | retry_find_task: |
1913 | rcu_read_lock(); | |
bbcb81d0 | 1914 | if (pid) { |
73507f33 | 1915 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
1916 | if (!tsk) { |
1917 | rcu_read_unlock(); | |
dd4b0a46 | 1918 | ret = -ESRCH; |
b78949eb | 1919 | goto out_unlock_cgroup; |
bbcb81d0 | 1920 | } |
74a1166d BB |
1921 | /* |
1922 | * even if we're attaching all tasks in the thread group, we | |
1923 | * only need to check permissions on one of them. | |
1924 | */ | |
c69e8d9c | 1925 | tcred = __task_cred(tsk); |
14a590c3 EB |
1926 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
1927 | !uid_eq(cred->euid, tcred->uid) && | |
1928 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 1929 | rcu_read_unlock(); |
b78949eb MSB |
1930 | ret = -EACCES; |
1931 | goto out_unlock_cgroup; | |
bbcb81d0 | 1932 | } |
b78949eb MSB |
1933 | } else |
1934 | tsk = current; | |
cd3d0952 TH |
1935 | |
1936 | if (threadgroup) | |
b78949eb | 1937 | tsk = tsk->group_leader; |
c4c27fbd MG |
1938 | |
1939 | /* | |
14a40ffc | 1940 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
1941 | * trapped in a cpuset, or RT worker may be born in a cgroup |
1942 | * with no rt_runtime allocated. Just say no. | |
1943 | */ | |
14a40ffc | 1944 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
1945 | ret = -EINVAL; |
1946 | rcu_read_unlock(); | |
1947 | goto out_unlock_cgroup; | |
1948 | } | |
1949 | ||
b78949eb MSB |
1950 | get_task_struct(tsk); |
1951 | rcu_read_unlock(); | |
1952 | ||
1953 | threadgroup_lock(tsk); | |
1954 | if (threadgroup) { | |
1955 | if (!thread_group_leader(tsk)) { | |
1956 | /* | |
1957 | * a race with de_thread from another thread's exec() | |
1958 | * may strip us of our leadership, if this happens, | |
1959 | * there is no choice but to throw this task away and | |
1960 | * try again; this is | |
1961 | * "double-double-toil-and-trouble-check locking". | |
1962 | */ | |
1963 | threadgroup_unlock(tsk); | |
1964 | put_task_struct(tsk); | |
1965 | goto retry_find_task; | |
1966 | } | |
081aa458 LZ |
1967 | } |
1968 | ||
1969 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
1970 | ||
cd3d0952 TH |
1971 | threadgroup_unlock(tsk); |
1972 | ||
bbcb81d0 | 1973 | put_task_struct(tsk); |
b78949eb | 1974 | out_unlock_cgroup: |
47cfcd09 | 1975 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
1976 | return ret; |
1977 | } | |
1978 | ||
7ae1bad9 TH |
1979 | /** |
1980 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
1981 | * @from: attach to all cgroups of a given task | |
1982 | * @tsk: the task to be attached | |
1983 | */ | |
1984 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
1985 | { | |
1986 | struct cgroupfs_root *root; | |
1987 | int retval = 0; | |
1988 | ||
47cfcd09 | 1989 | mutex_lock(&cgroup_mutex); |
7ae1bad9 | 1990 | for_each_active_root(root) { |
6f4b7e63 | 1991 | struct cgroup *from_cgrp = task_cgroup_from_root(from, root); |
7ae1bad9 | 1992 | |
6f4b7e63 | 1993 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
1994 | if (retval) |
1995 | break; | |
1996 | } | |
47cfcd09 | 1997 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
1998 | |
1999 | return retval; | |
2000 | } | |
2001 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2002 | ||
182446d0 TH |
2003 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2004 | struct cftype *cft, u64 pid) | |
74a1166d | 2005 | { |
182446d0 | 2006 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2007 | } |
2008 | ||
182446d0 TH |
2009 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2010 | struct cftype *cft, u64 tgid) | |
af351026 | 2011 | { |
182446d0 | 2012 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2013 | } |
2014 | ||
182446d0 TH |
2015 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
2016 | struct cftype *cft, const char *buffer) | |
e788e066 | 2017 | { |
5f469907 TH |
2018 | struct cgroupfs_root *root = css->cgroup->root; |
2019 | ||
2020 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
182446d0 | 2021 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2022 | return -ENODEV; |
69e943b7 | 2023 | spin_lock(&release_agent_path_lock); |
5f469907 TH |
2024 | strlcpy(root->release_agent_path, buffer, |
2025 | sizeof(root->release_agent_path)); | |
69e943b7 | 2026 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2027 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2028 | return 0; |
2029 | } | |
2030 | ||
2da8ca82 | 2031 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2032 | { |
2da8ca82 | 2033 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2034 | |
e788e066 PM |
2035 | if (!cgroup_lock_live_group(cgrp)) |
2036 | return -ENODEV; | |
2037 | seq_puts(seq, cgrp->root->release_agent_path); | |
2038 | seq_putc(seq, '\n'); | |
47cfcd09 | 2039 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2040 | return 0; |
2041 | } | |
2042 | ||
2da8ca82 | 2043 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2044 | { |
2da8ca82 TH |
2045 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2046 | ||
2047 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2048 | return 0; |
2049 | } | |
2050 | ||
2bd59d48 TH |
2051 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2052 | size_t nbytes, loff_t off) | |
355e0c48 | 2053 | { |
2bd59d48 TH |
2054 | struct cgroup *cgrp = of->kn->parent->priv; |
2055 | struct cftype *cft = of->kn->priv; | |
2056 | struct cgroup_subsys_state *css; | |
a742c59d | 2057 | int ret; |
355e0c48 | 2058 | |
2bd59d48 TH |
2059 | /* |
2060 | * kernfs guarantees that a file isn't deleted with operations in | |
2061 | * flight, which means that the matching css is and stays alive and | |
2062 | * doesn't need to be pinned. The RCU locking is not necessary | |
2063 | * either. It's just for the convenience of using cgroup_css(). | |
2064 | */ | |
2065 | rcu_read_lock(); | |
2066 | css = cgroup_css(cgrp, cft->ss); | |
2067 | rcu_read_unlock(); | |
a742c59d TH |
2068 | |
2069 | if (cft->write_string) { | |
2070 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2071 | } else if (cft->write_u64) { | |
2072 | unsigned long long v; | |
2073 | ret = kstrtoull(buf, 0, &v); | |
2074 | if (!ret) | |
2075 | ret = cft->write_u64(css, cft, v); | |
2076 | } else if (cft->write_s64) { | |
2077 | long long v; | |
2078 | ret = kstrtoll(buf, 0, &v); | |
2079 | if (!ret) | |
2080 | ret = cft->write_s64(css, cft, v); | |
2081 | } else if (cft->trigger) { | |
2082 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2083 | } else { |
a742c59d | 2084 | ret = -EINVAL; |
e73d2c61 | 2085 | } |
2bd59d48 | 2086 | |
a742c59d | 2087 | return ret ?: nbytes; |
355e0c48 PM |
2088 | } |
2089 | ||
6612f05b | 2090 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2091 | { |
2bd59d48 | 2092 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2093 | } |
2094 | ||
6612f05b | 2095 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2096 | { |
2bd59d48 | 2097 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2098 | } |
2099 | ||
6612f05b | 2100 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2101 | { |
2bd59d48 | 2102 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2103 | } |
2104 | ||
91796569 | 2105 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2106 | { |
7da11279 TH |
2107 | struct cftype *cft = seq_cft(m); |
2108 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2109 | |
2da8ca82 TH |
2110 | if (cft->seq_show) |
2111 | return cft->seq_show(m, arg); | |
e73d2c61 | 2112 | |
f4c753b7 | 2113 | if (cft->read_u64) |
896f5199 TH |
2114 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2115 | else if (cft->read_s64) | |
2116 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2117 | else | |
2118 | return -EINVAL; | |
2119 | return 0; | |
91796569 PM |
2120 | } |
2121 | ||
2bd59d48 TH |
2122 | static struct kernfs_ops cgroup_kf_single_ops = { |
2123 | .atomic_write_len = PAGE_SIZE, | |
2124 | .write = cgroup_file_write, | |
2125 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2126 | }; |
2127 | ||
2bd59d48 TH |
2128 | static struct kernfs_ops cgroup_kf_ops = { |
2129 | .atomic_write_len = PAGE_SIZE, | |
2130 | .write = cgroup_file_write, | |
2131 | .seq_start = cgroup_seqfile_start, | |
2132 | .seq_next = cgroup_seqfile_next, | |
2133 | .seq_stop = cgroup_seqfile_stop, | |
2134 | .seq_show = cgroup_seqfile_show, | |
2135 | }; | |
ddbcc7e8 PM |
2136 | |
2137 | /* | |
2138 | * cgroup_rename - Only allow simple rename of directories in place. | |
2139 | */ | |
2bd59d48 TH |
2140 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2141 | const char *new_name_str) | |
ddbcc7e8 | 2142 | { |
2bd59d48 | 2143 | struct cgroup *cgrp = kn->priv; |
2bd59d48 | 2144 | int ret; |
65dff759 | 2145 | |
2bd59d48 | 2146 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2147 | return -ENOTDIR; |
2bd59d48 | 2148 | if (kn->parent != new_parent) |
ddbcc7e8 | 2149 | return -EIO; |
65dff759 | 2150 | |
6db8e85c TH |
2151 | /* |
2152 | * This isn't a proper migration and its usefulness is very | |
2153 | * limited. Disallow if sane_behavior. | |
2154 | */ | |
2155 | if (cgroup_sane_behavior(cgrp)) | |
2156 | return -EPERM; | |
2157 | ||
2bd59d48 TH |
2158 | mutex_lock(&cgroup_tree_mutex); |
2159 | mutex_lock(&cgroup_mutex); | |
2160 | ||
2161 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
65dff759 | 2162 | |
2bd59d48 TH |
2163 | mutex_unlock(&cgroup_mutex); |
2164 | mutex_unlock(&cgroup_tree_mutex); | |
2bd59d48 | 2165 | return ret; |
ddbcc7e8 PM |
2166 | } |
2167 | ||
2bb566cb | 2168 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2169 | { |
8d7e6fb0 | 2170 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2171 | struct kernfs_node *kn; |
2172 | struct lock_class_key *key = NULL; | |
05ef1d7c | 2173 | |
2bd59d48 TH |
2174 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2175 | key = &cft->lockdep_key; | |
2176 | #endif | |
2177 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2178 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2179 | NULL, false, key); | |
2180 | if (IS_ERR(kn)) | |
2181 | return PTR_ERR(kn); | |
2182 | return 0; | |
ddbcc7e8 PM |
2183 | } |
2184 | ||
b1f28d31 TH |
2185 | /** |
2186 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2187 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2188 | * @cfts: array of cftypes to be added |
2189 | * @is_add: whether to add or remove | |
2190 | * | |
2191 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2192 | * For removals, this function never fails. If addition fails, this |
2193 | * function doesn't remove files already added. The caller is responsible | |
2194 | * for cleaning up. | |
b1f28d31 | 2195 | */ |
2bb566cb TH |
2196 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2197 | bool is_add) | |
ddbcc7e8 | 2198 | { |
03b1cde6 | 2199 | struct cftype *cft; |
b1f28d31 TH |
2200 | int ret; |
2201 | ||
ace2bee8 | 2202 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2203 | |
2204 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2205 | /* does cft->flags tell us to skip this file on @cgrp? */ |
873fe09e TH |
2206 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2207 | continue; | |
f33fddc2 G |
2208 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2209 | continue; | |
2210 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2211 | continue; | |
2212 | ||
2739d3cc | 2213 | if (is_add) { |
2bb566cb | 2214 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2215 | if (ret) { |
2739d3cc | 2216 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2217 | cft->name, ret); |
2218 | return ret; | |
2219 | } | |
2739d3cc LZ |
2220 | } else { |
2221 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2222 | } |
ddbcc7e8 | 2223 | } |
b1f28d31 | 2224 | return 0; |
ddbcc7e8 PM |
2225 | } |
2226 | ||
21a2d343 | 2227 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2228 | { |
2229 | LIST_HEAD(pending); | |
2bb566cb | 2230 | struct cgroup_subsys *ss = cfts[0].ss; |
492eb21b | 2231 | struct cgroup *root = &ss->root->top_cgroup; |
492eb21b | 2232 | struct cgroup_subsys_state *css; |
9ccece80 | 2233 | int ret = 0; |
8e3f6541 | 2234 | |
21a2d343 | 2235 | lockdep_assert_held(&cgroup_tree_mutex); |
4ac06017 | 2236 | |
21a2d343 TH |
2237 | /* don't bother if @ss isn't attached */ |
2238 | if (ss->root == &cgroup_dummy_root) | |
9ccece80 | 2239 | return 0; |
e8c82d20 | 2240 | |
e8c82d20 | 2241 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2242 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2243 | struct cgroup *cgrp = css->cgroup; |
2244 | ||
e8c82d20 LZ |
2245 | if (cgroup_is_dead(cgrp)) |
2246 | continue; | |
2247 | ||
21a2d343 | 2248 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2249 | if (ret) |
2250 | break; | |
8e3f6541 | 2251 | } |
21a2d343 TH |
2252 | |
2253 | if (is_add && !ret) | |
2254 | kernfs_activate(root->kn); | |
9ccece80 | 2255 | return ret; |
8e3f6541 TH |
2256 | } |
2257 | ||
2da440a2 TH |
2258 | static void cgroup_exit_cftypes(struct cftype *cfts) |
2259 | { | |
2260 | struct cftype *cft; | |
2261 | ||
2bd59d48 TH |
2262 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2263 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2264 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2265 | kfree(cft->kf_ops); | |
2266 | cft->kf_ops = NULL; | |
2da440a2 | 2267 | cft->ss = NULL; |
2bd59d48 | 2268 | } |
2da440a2 TH |
2269 | } |
2270 | ||
2bd59d48 | 2271 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2272 | { |
2273 | struct cftype *cft; | |
2274 | ||
2bd59d48 TH |
2275 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2276 | struct kernfs_ops *kf_ops; | |
2277 | ||
0adb0704 TH |
2278 | WARN_ON(cft->ss || cft->kf_ops); |
2279 | ||
2bd59d48 TH |
2280 | if (cft->seq_start) |
2281 | kf_ops = &cgroup_kf_ops; | |
2282 | else | |
2283 | kf_ops = &cgroup_kf_single_ops; | |
2284 | ||
2285 | /* | |
2286 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2287 | * make a copy of kf_ops to set its atomic_write_len. | |
2288 | */ | |
2289 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2290 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2291 | if (!kf_ops) { | |
2292 | cgroup_exit_cftypes(cfts); | |
2293 | return -ENOMEM; | |
2294 | } | |
2295 | kf_ops->atomic_write_len = cft->max_write_len; | |
2296 | } | |
2297 | ||
2298 | cft->kf_ops = kf_ops; | |
2da440a2 | 2299 | cft->ss = ss; |
2bd59d48 TH |
2300 | } |
2301 | ||
2302 | return 0; | |
2da440a2 TH |
2303 | } |
2304 | ||
21a2d343 TH |
2305 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
2306 | { | |
2307 | lockdep_assert_held(&cgroup_tree_mutex); | |
2308 | ||
2309 | if (!cfts || !cfts[0].ss) | |
2310 | return -ENOENT; | |
2311 | ||
2312 | list_del(&cfts->node); | |
2313 | cgroup_apply_cftypes(cfts, false); | |
2314 | cgroup_exit_cftypes(cfts); | |
2315 | return 0; | |
2316 | } | |
2317 | ||
80b13586 TH |
2318 | /** |
2319 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
2320 | * @cfts: zero-length name terminated array of cftypes | |
2321 | * | |
2322 | * Unregister @cfts. Files described by @cfts are removed from all | |
2323 | * existing cgroups and all future cgroups won't have them either. This | |
2324 | * function can be called anytime whether @cfts' subsys is attached or not. | |
2325 | * | |
2326 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2327 | * registered. | |
2328 | */ | |
2329 | int cgroup_rm_cftypes(struct cftype *cfts) | |
2330 | { | |
21a2d343 | 2331 | int ret; |
80b13586 | 2332 | |
21a2d343 TH |
2333 | mutex_lock(&cgroup_tree_mutex); |
2334 | ret = cgroup_rm_cftypes_locked(cfts); | |
2335 | mutex_unlock(&cgroup_tree_mutex); | |
2336 | return ret; | |
80b13586 TH |
2337 | } |
2338 | ||
8e3f6541 TH |
2339 | /** |
2340 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2341 | * @ss: target cgroup subsystem | |
2342 | * @cfts: zero-length name terminated array of cftypes | |
2343 | * | |
2344 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2345 | * existing cgroups to which @ss is attached and all future cgroups will | |
2346 | * have them too. This function can be called anytime whether @ss is | |
2347 | * attached or not. | |
2348 | * | |
2349 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2350 | * function currently returns 0 as long as @cfts registration is successful | |
2351 | * even if some file creation attempts on existing cgroups fail. | |
2352 | */ | |
03b1cde6 | 2353 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 2354 | { |
9ccece80 | 2355 | int ret; |
8e3f6541 | 2356 | |
2bd59d48 TH |
2357 | ret = cgroup_init_cftypes(ss, cfts); |
2358 | if (ret) | |
2359 | return ret; | |
2bb566cb | 2360 | |
21a2d343 TH |
2361 | mutex_lock(&cgroup_tree_mutex); |
2362 | ||
0adb0704 | 2363 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 2364 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 2365 | if (ret) |
21a2d343 TH |
2366 | cgroup_rm_cftypes_locked(cfts); |
2367 | ||
2368 | mutex_unlock(&cgroup_tree_mutex); | |
9ccece80 | 2369 | return ret; |
8e3f6541 TH |
2370 | } |
2371 | EXPORT_SYMBOL_GPL(cgroup_add_cftypes); | |
2372 | ||
a043e3b2 LZ |
2373 | /** |
2374 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2375 | * @cgrp: the cgroup in question | |
2376 | * | |
2377 | * Return the number of tasks in the cgroup. | |
2378 | */ | |
bd89aabc | 2379 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2380 | { |
2381 | int count = 0; | |
69d0206c | 2382 | struct cgrp_cset_link *link; |
817929ec PM |
2383 | |
2384 | read_lock(&css_set_lock); | |
69d0206c TH |
2385 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2386 | count += atomic_read(&link->cset->refcount); | |
817929ec | 2387 | read_unlock(&css_set_lock); |
bbcb81d0 PM |
2388 | return count; |
2389 | } | |
2390 | ||
817929ec | 2391 | /* |
0942eeee TH |
2392 | * To reduce the fork() overhead for systems that are not actually using |
2393 | * their cgroups capability, we don't maintain the lists running through | |
2394 | * each css_set to its tasks until we see the list actually used - in other | |
72ec7029 | 2395 | * words after the first call to css_task_iter_start(). |
31a7df01 | 2396 | */ |
3df91fe3 | 2397 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2398 | { |
2399 | struct task_struct *p, *g; | |
2400 | write_lock(&css_set_lock); | |
2401 | use_task_css_set_links = 1; | |
3ce3230a FW |
2402 | /* |
2403 | * We need tasklist_lock because RCU is not safe against | |
2404 | * while_each_thread(). Besides, a forking task that has passed | |
2405 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
2406 | * is not guaranteed to have its child immediately visible in the | |
2407 | * tasklist if we walk through it with RCU. | |
2408 | */ | |
2409 | read_lock(&tasklist_lock); | |
31a7df01 CW |
2410 | do_each_thread(g, p) { |
2411 | task_lock(p); | |
0e04388f LZ |
2412 | /* |
2413 | * We should check if the process is exiting, otherwise | |
2414 | * it will race with cgroup_exit() in that the list | |
2415 | * entry won't be deleted though the process has exited. | |
2416 | */ | |
2417 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
a8ad805c | 2418 | list_add(&p->cg_list, &task_css_set(p)->tasks); |
31a7df01 CW |
2419 | task_unlock(p); |
2420 | } while_each_thread(g, p); | |
3ce3230a | 2421 | read_unlock(&tasklist_lock); |
31a7df01 CW |
2422 | write_unlock(&css_set_lock); |
2423 | } | |
2424 | ||
53fa5261 | 2425 | /** |
492eb21b TH |
2426 | * css_next_child - find the next child of a given css |
2427 | * @pos_css: the current position (%NULL to initiate traversal) | |
2428 | * @parent_css: css whose children to walk | |
53fa5261 | 2429 | * |
492eb21b | 2430 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
2431 | * under either cgroup_mutex or RCU read lock. The only requirement is |
2432 | * that @parent_css and @pos_css are accessible. The next sibling is | |
2433 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 2434 | */ |
492eb21b TH |
2435 | struct cgroup_subsys_state * |
2436 | css_next_child(struct cgroup_subsys_state *pos_css, | |
2437 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 2438 | { |
492eb21b TH |
2439 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
2440 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
2441 | struct cgroup *next; |
2442 | ||
ace2bee8 | 2443 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
2444 | |
2445 | /* | |
2446 | * @pos could already have been removed. Once a cgroup is removed, | |
2447 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
2448 | * changes. As CGRP_DEAD assertion is serialized and happens |
2449 | * before the cgroup is taken off the ->sibling list, if we see it | |
2450 | * unasserted, it's guaranteed that the next sibling hasn't | |
2451 | * finished its grace period even if it's already removed, and thus | |
2452 | * safe to dereference from this RCU critical section. If | |
2453 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
2454 | * to be visible as %true here. | |
3b287a50 TH |
2455 | * |
2456 | * If @pos is dead, its next pointer can't be dereferenced; | |
2457 | * however, as each cgroup is given a monotonically increasing | |
2458 | * unique serial number and always appended to the sibling list, | |
2459 | * the next one can be found by walking the parent's children until | |
2460 | * we see a cgroup with higher serial number than @pos's. While | |
2461 | * this path can be slower, it's taken only when either the current | |
2462 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 2463 | */ |
3b287a50 TH |
2464 | if (!pos) { |
2465 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
2466 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 2467 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
2468 | } else { |
2469 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
2470 | if (next->serial_nr > pos->serial_nr) | |
2471 | break; | |
53fa5261 TH |
2472 | } |
2473 | ||
492eb21b TH |
2474 | if (&next->sibling == &cgrp->children) |
2475 | return NULL; | |
2476 | ||
ca8bdcaf | 2477 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 2478 | } |
492eb21b | 2479 | EXPORT_SYMBOL_GPL(css_next_child); |
53fa5261 | 2480 | |
574bd9f7 | 2481 | /** |
492eb21b | 2482 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 2483 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2484 | * @root: css whose descendants to walk |
574bd9f7 | 2485 | * |
492eb21b | 2486 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
2487 | * to visit for pre-order traversal of @root's descendants. @root is |
2488 | * included in the iteration and the first node to be visited. | |
75501a6d | 2489 | * |
87fb54f1 TH |
2490 | * While this function requires cgroup_mutex or RCU read locking, it |
2491 | * doesn't require the whole traversal to be contained in a single critical | |
2492 | * section. This function will return the correct next descendant as long | |
2493 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 2494 | */ |
492eb21b TH |
2495 | struct cgroup_subsys_state * |
2496 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
2497 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2498 | { |
492eb21b | 2499 | struct cgroup_subsys_state *next; |
574bd9f7 | 2500 | |
ace2bee8 | 2501 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2502 | |
bd8815a6 | 2503 | /* if first iteration, visit @root */ |
7805d000 | 2504 | if (!pos) |
bd8815a6 | 2505 | return root; |
574bd9f7 TH |
2506 | |
2507 | /* visit the first child if exists */ | |
492eb21b | 2508 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
2509 | if (next) |
2510 | return next; | |
2511 | ||
2512 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
2513 | while (pos != root) { |
2514 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 2515 | if (next) |
574bd9f7 | 2516 | return next; |
492eb21b | 2517 | pos = css_parent(pos); |
7805d000 | 2518 | } |
574bd9f7 TH |
2519 | |
2520 | return NULL; | |
2521 | } | |
492eb21b | 2522 | EXPORT_SYMBOL_GPL(css_next_descendant_pre); |
574bd9f7 | 2523 | |
12a9d2fe | 2524 | /** |
492eb21b TH |
2525 | * css_rightmost_descendant - return the rightmost descendant of a css |
2526 | * @pos: css of interest | |
12a9d2fe | 2527 | * |
492eb21b TH |
2528 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
2529 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 2530 | * subtree of @pos. |
75501a6d | 2531 | * |
87fb54f1 TH |
2532 | * While this function requires cgroup_mutex or RCU read locking, it |
2533 | * doesn't require the whole traversal to be contained in a single critical | |
2534 | * section. This function will return the correct rightmost descendant as | |
2535 | * long as @pos is accessible. | |
12a9d2fe | 2536 | */ |
492eb21b TH |
2537 | struct cgroup_subsys_state * |
2538 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 2539 | { |
492eb21b | 2540 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 2541 | |
ace2bee8 | 2542 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
2543 | |
2544 | do { | |
2545 | last = pos; | |
2546 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
2547 | pos = NULL; | |
492eb21b | 2548 | css_for_each_child(tmp, last) |
12a9d2fe TH |
2549 | pos = tmp; |
2550 | } while (pos); | |
2551 | ||
2552 | return last; | |
2553 | } | |
492eb21b | 2554 | EXPORT_SYMBOL_GPL(css_rightmost_descendant); |
12a9d2fe | 2555 | |
492eb21b TH |
2556 | static struct cgroup_subsys_state * |
2557 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 2558 | { |
492eb21b | 2559 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
2560 | |
2561 | do { | |
2562 | last = pos; | |
492eb21b | 2563 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
2564 | } while (pos); |
2565 | ||
2566 | return last; | |
2567 | } | |
2568 | ||
2569 | /** | |
492eb21b | 2570 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 2571 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2572 | * @root: css whose descendants to walk |
574bd9f7 | 2573 | * |
492eb21b | 2574 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
2575 | * to visit for post-order traversal of @root's descendants. @root is |
2576 | * included in the iteration and the last node to be visited. | |
75501a6d | 2577 | * |
87fb54f1 TH |
2578 | * While this function requires cgroup_mutex or RCU read locking, it |
2579 | * doesn't require the whole traversal to be contained in a single critical | |
2580 | * section. This function will return the correct next descendant as long | |
2581 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
2582 | * @cgroup. | |
574bd9f7 | 2583 | */ |
492eb21b TH |
2584 | struct cgroup_subsys_state * |
2585 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
2586 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2587 | { |
492eb21b | 2588 | struct cgroup_subsys_state *next; |
574bd9f7 | 2589 | |
ace2bee8 | 2590 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2591 | |
58b79a91 TH |
2592 | /* if first iteration, visit leftmost descendant which may be @root */ |
2593 | if (!pos) | |
2594 | return css_leftmost_descendant(root); | |
574bd9f7 | 2595 | |
bd8815a6 TH |
2596 | /* if we visited @root, we're done */ |
2597 | if (pos == root) | |
2598 | return NULL; | |
2599 | ||
574bd9f7 | 2600 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 2601 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 2602 | if (next) |
492eb21b | 2603 | return css_leftmost_descendant(next); |
574bd9f7 TH |
2604 | |
2605 | /* no sibling left, visit parent */ | |
bd8815a6 | 2606 | return css_parent(pos); |
574bd9f7 | 2607 | } |
492eb21b | 2608 | EXPORT_SYMBOL_GPL(css_next_descendant_post); |
574bd9f7 | 2609 | |
0942eeee | 2610 | /** |
72ec7029 | 2611 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
2612 | * @it: the iterator to advance |
2613 | * | |
2614 | * Advance @it to the next css_set to walk. | |
d515876e | 2615 | */ |
72ec7029 | 2616 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
2617 | { |
2618 | struct list_head *l = it->cset_link; | |
2619 | struct cgrp_cset_link *link; | |
2620 | struct css_set *cset; | |
2621 | ||
2622 | /* Advance to the next non-empty css_set */ | |
2623 | do { | |
2624 | l = l->next; | |
72ec7029 | 2625 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
2626 | it->cset_link = NULL; |
2627 | return; | |
2628 | } | |
2629 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
2630 | cset = link->cset; | |
2631 | } while (list_empty(&cset->tasks)); | |
2632 | it->cset_link = l; | |
2633 | it->task = cset->tasks.next; | |
2634 | } | |
2635 | ||
0942eeee | 2636 | /** |
72ec7029 TH |
2637 | * css_task_iter_start - initiate task iteration |
2638 | * @css: the css to walk tasks of | |
0942eeee TH |
2639 | * @it: the task iterator to use |
2640 | * | |
72ec7029 TH |
2641 | * Initiate iteration through the tasks of @css. The caller can call |
2642 | * css_task_iter_next() to walk through the tasks until the function | |
2643 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
2644 | * called. | |
0942eeee TH |
2645 | * |
2646 | * Note that this function acquires a lock which is released when the | |
2647 | * iteration finishes. The caller can't sleep while iteration is in | |
2648 | * progress. | |
2649 | */ | |
72ec7029 TH |
2650 | void css_task_iter_start(struct cgroup_subsys_state *css, |
2651 | struct css_task_iter *it) | |
c6ca5750 | 2652 | __acquires(css_set_lock) |
817929ec PM |
2653 | { |
2654 | /* | |
72ec7029 TH |
2655 | * The first time anyone tries to iterate across a css, we need to |
2656 | * enable the list linking each css_set to its tasks, and fix up | |
2657 | * all existing tasks. | |
817929ec | 2658 | */ |
31a7df01 CW |
2659 | if (!use_task_css_set_links) |
2660 | cgroup_enable_task_cg_lists(); | |
2661 | ||
817929ec | 2662 | read_lock(&css_set_lock); |
c59cd3d8 | 2663 | |
72ec7029 TH |
2664 | it->origin_css = css; |
2665 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 2666 | |
72ec7029 | 2667 | css_advance_task_iter(it); |
817929ec PM |
2668 | } |
2669 | ||
0942eeee | 2670 | /** |
72ec7029 | 2671 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
2672 | * @it: the task iterator being iterated |
2673 | * | |
2674 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
2675 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
2676 | * reaches the end. | |
0942eeee | 2677 | */ |
72ec7029 | 2678 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
2679 | { |
2680 | struct task_struct *res; | |
2681 | struct list_head *l = it->task; | |
69d0206c | 2682 | struct cgrp_cset_link *link; |
817929ec PM |
2683 | |
2684 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 2685 | if (!it->cset_link) |
817929ec PM |
2686 | return NULL; |
2687 | res = list_entry(l, struct task_struct, cg_list); | |
2688 | /* Advance iterator to find next entry */ | |
2689 | l = l->next; | |
69d0206c TH |
2690 | link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link); |
2691 | if (l == &link->cset->tasks) { | |
0942eeee TH |
2692 | /* |
2693 | * We reached the end of this task list - move on to the | |
2694 | * next cgrp_cset_link. | |
2695 | */ | |
72ec7029 | 2696 | css_advance_task_iter(it); |
817929ec PM |
2697 | } else { |
2698 | it->task = l; | |
2699 | } | |
2700 | return res; | |
2701 | } | |
2702 | ||
0942eeee | 2703 | /** |
72ec7029 | 2704 | * css_task_iter_end - finish task iteration |
0942eeee TH |
2705 | * @it: the task iterator to finish |
2706 | * | |
72ec7029 | 2707 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 2708 | */ |
72ec7029 | 2709 | void css_task_iter_end(struct css_task_iter *it) |
c6ca5750 | 2710 | __releases(css_set_lock) |
817929ec PM |
2711 | { |
2712 | read_unlock(&css_set_lock); | |
2713 | } | |
2714 | ||
31a7df01 CW |
2715 | static inline int started_after_time(struct task_struct *t1, |
2716 | struct timespec *time, | |
2717 | struct task_struct *t2) | |
2718 | { | |
2719 | int start_diff = timespec_compare(&t1->start_time, time); | |
2720 | if (start_diff > 0) { | |
2721 | return 1; | |
2722 | } else if (start_diff < 0) { | |
2723 | return 0; | |
2724 | } else { | |
2725 | /* | |
2726 | * Arbitrarily, if two processes started at the same | |
2727 | * time, we'll say that the lower pointer value | |
2728 | * started first. Note that t2 may have exited by now | |
2729 | * so this may not be a valid pointer any longer, but | |
2730 | * that's fine - it still serves to distinguish | |
2731 | * between two tasks started (effectively) simultaneously. | |
2732 | */ | |
2733 | return t1 > t2; | |
2734 | } | |
2735 | } | |
2736 | ||
2737 | /* | |
2738 | * This function is a callback from heap_insert() and is used to order | |
2739 | * the heap. | |
2740 | * In this case we order the heap in descending task start time. | |
2741 | */ | |
2742 | static inline int started_after(void *p1, void *p2) | |
2743 | { | |
2744 | struct task_struct *t1 = p1; | |
2745 | struct task_struct *t2 = p2; | |
2746 | return started_after_time(t1, &t2->start_time, t2); | |
2747 | } | |
2748 | ||
2749 | /** | |
72ec7029 TH |
2750 | * css_scan_tasks - iterate though all the tasks in a css |
2751 | * @css: the css to iterate tasks of | |
e535837b TH |
2752 | * @test: optional test callback |
2753 | * @process: process callback | |
2754 | * @data: data passed to @test and @process | |
2755 | * @heap: optional pre-allocated heap used for task iteration | |
31a7df01 | 2756 | * |
72ec7029 TH |
2757 | * Iterate through all the tasks in @css, calling @test for each, and if it |
2758 | * returns %true, call @process for it also. | |
31a7df01 | 2759 | * |
e535837b | 2760 | * @test may be NULL, meaning always true (select all tasks), which |
72ec7029 | 2761 | * effectively duplicates css_task_iter_{start,next,end}() but does not |
e535837b TH |
2762 | * lock css_set_lock for the call to @process. |
2763 | * | |
2764 | * It is guaranteed that @process will act on every task that is a member | |
72ec7029 TH |
2765 | * of @css for the duration of this call. This function may or may not |
2766 | * call @process for tasks that exit or move to a different css during the | |
2767 | * call, or are forked or move into the css during the call. | |
31a7df01 | 2768 | * |
e535837b TH |
2769 | * Note that @test may be called with locks held, and may in some |
2770 | * situations be called multiple times for the same task, so it should be | |
2771 | * cheap. | |
31a7df01 | 2772 | * |
e535837b TH |
2773 | * If @heap is non-NULL, a heap has been pre-allocated and will be used for |
2774 | * heap operations (and its "gt" member will be overwritten), else a | |
2775 | * temporary heap will be used (allocation of which may cause this function | |
2776 | * to fail). | |
31a7df01 | 2777 | */ |
72ec7029 TH |
2778 | int css_scan_tasks(struct cgroup_subsys_state *css, |
2779 | bool (*test)(struct task_struct *, void *), | |
2780 | void (*process)(struct task_struct *, void *), | |
2781 | void *data, struct ptr_heap *heap) | |
31a7df01 CW |
2782 | { |
2783 | int retval, i; | |
72ec7029 | 2784 | struct css_task_iter it; |
31a7df01 CW |
2785 | struct task_struct *p, *dropped; |
2786 | /* Never dereference latest_task, since it's not refcounted */ | |
2787 | struct task_struct *latest_task = NULL; | |
2788 | struct ptr_heap tmp_heap; | |
31a7df01 CW |
2789 | struct timespec latest_time = { 0, 0 }; |
2790 | ||
e535837b | 2791 | if (heap) { |
31a7df01 | 2792 | /* The caller supplied our heap and pre-allocated its memory */ |
31a7df01 CW |
2793 | heap->gt = &started_after; |
2794 | } else { | |
2795 | /* We need to allocate our own heap memory */ | |
2796 | heap = &tmp_heap; | |
2797 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
2798 | if (retval) | |
2799 | /* cannot allocate the heap */ | |
2800 | return retval; | |
2801 | } | |
2802 | ||
2803 | again: | |
2804 | /* | |
72ec7029 | 2805 | * Scan tasks in the css, using the @test callback to determine |
e535837b TH |
2806 | * which are of interest, and invoking @process callback on the |
2807 | * ones which need an update. Since we don't want to hold any | |
2808 | * locks during the task updates, gather tasks to be processed in a | |
2809 | * heap structure. The heap is sorted by descending task start | |
2810 | * time. If the statically-sized heap fills up, we overflow tasks | |
2811 | * that started later, and in future iterations only consider tasks | |
2812 | * that started after the latest task in the previous pass. This | |
31a7df01 CW |
2813 | * guarantees forward progress and that we don't miss any tasks. |
2814 | */ | |
2815 | heap->size = 0; | |
72ec7029 TH |
2816 | css_task_iter_start(css, &it); |
2817 | while ((p = css_task_iter_next(&it))) { | |
31a7df01 CW |
2818 | /* |
2819 | * Only affect tasks that qualify per the caller's callback, | |
2820 | * if he provided one | |
2821 | */ | |
e535837b | 2822 | if (test && !test(p, data)) |
31a7df01 CW |
2823 | continue; |
2824 | /* | |
2825 | * Only process tasks that started after the last task | |
2826 | * we processed | |
2827 | */ | |
2828 | if (!started_after_time(p, &latest_time, latest_task)) | |
2829 | continue; | |
2830 | dropped = heap_insert(heap, p); | |
2831 | if (dropped == NULL) { | |
2832 | /* | |
2833 | * The new task was inserted; the heap wasn't | |
2834 | * previously full | |
2835 | */ | |
2836 | get_task_struct(p); | |
2837 | } else if (dropped != p) { | |
2838 | /* | |
2839 | * The new task was inserted, and pushed out a | |
2840 | * different task | |
2841 | */ | |
2842 | get_task_struct(p); | |
2843 | put_task_struct(dropped); | |
2844 | } | |
2845 | /* | |
2846 | * Else the new task was newer than anything already in | |
2847 | * the heap and wasn't inserted | |
2848 | */ | |
2849 | } | |
72ec7029 | 2850 | css_task_iter_end(&it); |
31a7df01 CW |
2851 | |
2852 | if (heap->size) { | |
2853 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 2854 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 2855 | if (i == 0) { |
4fe91d51 PJ |
2856 | latest_time = q->start_time; |
2857 | latest_task = q; | |
31a7df01 CW |
2858 | } |
2859 | /* Process the task per the caller's callback */ | |
e535837b | 2860 | process(q, data); |
4fe91d51 | 2861 | put_task_struct(q); |
31a7df01 CW |
2862 | } |
2863 | /* | |
2864 | * If we had to process any tasks at all, scan again | |
2865 | * in case some of them were in the middle of forking | |
2866 | * children that didn't get processed. | |
2867 | * Not the most efficient way to do it, but it avoids | |
2868 | * having to take callback_mutex in the fork path | |
2869 | */ | |
2870 | goto again; | |
2871 | } | |
2872 | if (heap == &tmp_heap) | |
2873 | heap_free(&tmp_heap); | |
2874 | return 0; | |
2875 | } | |
2876 | ||
e535837b | 2877 | static void cgroup_transfer_one_task(struct task_struct *task, void *data) |
8cc99345 | 2878 | { |
e535837b | 2879 | struct cgroup *new_cgroup = data; |
8cc99345 | 2880 | |
47cfcd09 | 2881 | mutex_lock(&cgroup_mutex); |
8cc99345 | 2882 | cgroup_attach_task(new_cgroup, task, false); |
47cfcd09 | 2883 | mutex_unlock(&cgroup_mutex); |
8cc99345 TH |
2884 | } |
2885 | ||
2886 | /** | |
2887 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
2888 | * @to: cgroup to which the tasks will be moved | |
2889 | * @from: cgroup in which the tasks currently reside | |
2890 | */ | |
2891 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
2892 | { | |
72ec7029 TH |
2893 | return css_scan_tasks(&from->dummy_css, NULL, cgroup_transfer_one_task, |
2894 | to, NULL); | |
8cc99345 TH |
2895 | } |
2896 | ||
bbcb81d0 | 2897 | /* |
102a775e | 2898 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
2899 | * |
2900 | * Reading this file can return large amounts of data if a cgroup has | |
2901 | * *lots* of attached tasks. So it may need several calls to read(), | |
2902 | * but we cannot guarantee that the information we produce is correct | |
2903 | * unless we produce it entirely atomically. | |
2904 | * | |
bbcb81d0 | 2905 | */ |
bbcb81d0 | 2906 | |
24528255 LZ |
2907 | /* which pidlist file are we talking about? */ |
2908 | enum cgroup_filetype { | |
2909 | CGROUP_FILE_PROCS, | |
2910 | CGROUP_FILE_TASKS, | |
2911 | }; | |
2912 | ||
2913 | /* | |
2914 | * A pidlist is a list of pids that virtually represents the contents of one | |
2915 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
2916 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
2917 | * to the cgroup. | |
2918 | */ | |
2919 | struct cgroup_pidlist { | |
2920 | /* | |
2921 | * used to find which pidlist is wanted. doesn't change as long as | |
2922 | * this particular list stays in the list. | |
2923 | */ | |
2924 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
2925 | /* array of xids */ | |
2926 | pid_t *list; | |
2927 | /* how many elements the above list has */ | |
2928 | int length; | |
24528255 LZ |
2929 | /* each of these stored in a list by its cgroup */ |
2930 | struct list_head links; | |
2931 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
2932 | struct cgroup *owner; | |
b1a21367 TH |
2933 | /* for delayed destruction */ |
2934 | struct delayed_work destroy_dwork; | |
24528255 LZ |
2935 | }; |
2936 | ||
d1d9fd33 BB |
2937 | /* |
2938 | * The following two functions "fix" the issue where there are more pids | |
2939 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
2940 | * TODO: replace with a kernel-wide solution to this problem | |
2941 | */ | |
2942 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
2943 | static void *pidlist_allocate(int count) | |
2944 | { | |
2945 | if (PIDLIST_TOO_LARGE(count)) | |
2946 | return vmalloc(count * sizeof(pid_t)); | |
2947 | else | |
2948 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
2949 | } | |
b1a21367 | 2950 | |
d1d9fd33 BB |
2951 | static void pidlist_free(void *p) |
2952 | { | |
2953 | if (is_vmalloc_addr(p)) | |
2954 | vfree(p); | |
2955 | else | |
2956 | kfree(p); | |
2957 | } | |
d1d9fd33 | 2958 | |
b1a21367 TH |
2959 | /* |
2960 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
2961 | * should be left afterwards. | |
2962 | */ | |
2963 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
2964 | { | |
2965 | struct cgroup_pidlist *l, *tmp_l; | |
2966 | ||
2967 | mutex_lock(&cgrp->pidlist_mutex); | |
2968 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
2969 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
2970 | mutex_unlock(&cgrp->pidlist_mutex); | |
2971 | ||
2972 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
2973 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
2974 | } | |
2975 | ||
2976 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
2977 | { | |
2978 | struct delayed_work *dwork = to_delayed_work(work); | |
2979 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
2980 | destroy_dwork); | |
2981 | struct cgroup_pidlist *tofree = NULL; | |
2982 | ||
2983 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
2984 | |
2985 | /* | |
04502365 TH |
2986 | * Destroy iff we didn't get queued again. The state won't change |
2987 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 2988 | */ |
04502365 | 2989 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
2990 | list_del(&l->links); |
2991 | pidlist_free(l->list); | |
2992 | put_pid_ns(l->key.ns); | |
2993 | tofree = l; | |
2994 | } | |
2995 | ||
b1a21367 TH |
2996 | mutex_unlock(&l->owner->pidlist_mutex); |
2997 | kfree(tofree); | |
2998 | } | |
2999 | ||
bbcb81d0 | 3000 | /* |
102a775e | 3001 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3002 | * Returns the number of unique elements. |
bbcb81d0 | 3003 | */ |
6ee211ad | 3004 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3005 | { |
102a775e | 3006 | int src, dest = 1; |
102a775e BB |
3007 | |
3008 | /* | |
3009 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3010 | * edge cases first; no work needs to be done for either | |
3011 | */ | |
3012 | if (length == 0 || length == 1) | |
3013 | return length; | |
3014 | /* src and dest walk down the list; dest counts unique elements */ | |
3015 | for (src = 1; src < length; src++) { | |
3016 | /* find next unique element */ | |
3017 | while (list[src] == list[src-1]) { | |
3018 | src++; | |
3019 | if (src == length) | |
3020 | goto after; | |
3021 | } | |
3022 | /* dest always points to where the next unique element goes */ | |
3023 | list[dest] = list[src]; | |
3024 | dest++; | |
3025 | } | |
3026 | after: | |
102a775e BB |
3027 | return dest; |
3028 | } | |
3029 | ||
afb2bc14 TH |
3030 | /* |
3031 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3032 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3033 | * different per namespace, each namespace needs differently sorted list, | |
3034 | * making it impossible to use, for example, single rbtree of member tasks | |
3035 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3036 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3037 | * pidlists keyed by cgroup and namespace. | |
3038 | * | |
3039 | * All this extra complexity was caused by the original implementation | |
3040 | * committing to an entirely unnecessary property. In the long term, we | |
3041 | * want to do away with it. Explicitly scramble sort order if | |
3042 | * sane_behavior so that no such expectation exists in the new interface. | |
3043 | * | |
3044 | * Scrambling is done by swapping every two consecutive bits, which is | |
3045 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3046 | */ | |
3047 | static pid_t pid_fry(pid_t pid) | |
3048 | { | |
3049 | unsigned a = pid & 0x55555555; | |
3050 | unsigned b = pid & 0xAAAAAAAA; | |
3051 | ||
3052 | return (a << 1) | (b >> 1); | |
3053 | } | |
3054 | ||
3055 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3056 | { | |
3057 | if (cgroup_sane_behavior(cgrp)) | |
3058 | return pid_fry(pid); | |
3059 | else | |
3060 | return pid; | |
3061 | } | |
3062 | ||
102a775e BB |
3063 | static int cmppid(const void *a, const void *b) |
3064 | { | |
3065 | return *(pid_t *)a - *(pid_t *)b; | |
3066 | } | |
3067 | ||
afb2bc14 TH |
3068 | static int fried_cmppid(const void *a, const void *b) |
3069 | { | |
3070 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3071 | } | |
3072 | ||
e6b81710 TH |
3073 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3074 | enum cgroup_filetype type) | |
3075 | { | |
3076 | struct cgroup_pidlist *l; | |
3077 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3078 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3079 | ||
3080 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3081 | ||
3082 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3083 | if (l->key.type == type && l->key.ns == ns) | |
3084 | return l; | |
3085 | return NULL; | |
3086 | } | |
3087 | ||
72a8cb30 BB |
3088 | /* |
3089 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3090 | * returns with the lock on that pidlist already held, and takes care | |
3091 | * of the use count, or returns NULL with no locks held if we're out of | |
3092 | * memory. | |
3093 | */ | |
e6b81710 TH |
3094 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3095 | enum cgroup_filetype type) | |
72a8cb30 BB |
3096 | { |
3097 | struct cgroup_pidlist *l; | |
b70cc5fd | 3098 | |
e6b81710 TH |
3099 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3100 | ||
3101 | l = cgroup_pidlist_find(cgrp, type); | |
3102 | if (l) | |
3103 | return l; | |
3104 | ||
72a8cb30 | 3105 | /* entry not found; create a new one */ |
f4f4be2b | 3106 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3107 | if (!l) |
72a8cb30 | 3108 | return l; |
e6b81710 | 3109 | |
b1a21367 | 3110 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3111 | l->key.type = type; |
e6b81710 TH |
3112 | /* don't need task_nsproxy() if we're looking at ourself */ |
3113 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3114 | l->owner = cgrp; |
3115 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3116 | return l; |
3117 | } | |
3118 | ||
102a775e BB |
3119 | /* |
3120 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3121 | */ | |
72a8cb30 BB |
3122 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3123 | struct cgroup_pidlist **lp) | |
102a775e BB |
3124 | { |
3125 | pid_t *array; | |
3126 | int length; | |
3127 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3128 | struct css_task_iter it; |
817929ec | 3129 | struct task_struct *tsk; |
102a775e BB |
3130 | struct cgroup_pidlist *l; |
3131 | ||
4bac00d1 TH |
3132 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3133 | ||
102a775e BB |
3134 | /* |
3135 | * If cgroup gets more users after we read count, we won't have | |
3136 | * enough space - tough. This race is indistinguishable to the | |
3137 | * caller from the case that the additional cgroup users didn't | |
3138 | * show up until sometime later on. | |
3139 | */ | |
3140 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3141 | array = pidlist_allocate(length); |
102a775e BB |
3142 | if (!array) |
3143 | return -ENOMEM; | |
3144 | /* now, populate the array */ | |
72ec7029 TH |
3145 | css_task_iter_start(&cgrp->dummy_css, &it); |
3146 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3147 | if (unlikely(n == length)) |
817929ec | 3148 | break; |
102a775e | 3149 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3150 | if (type == CGROUP_FILE_PROCS) |
3151 | pid = task_tgid_vnr(tsk); | |
3152 | else | |
3153 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3154 | if (pid > 0) /* make sure to only use valid results */ |
3155 | array[n++] = pid; | |
817929ec | 3156 | } |
72ec7029 | 3157 | css_task_iter_end(&it); |
102a775e BB |
3158 | length = n; |
3159 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3160 | if (cgroup_sane_behavior(cgrp)) |
3161 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3162 | else | |
3163 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3164 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3165 | length = pidlist_uniq(array, length); |
e6b81710 | 3166 | |
e6b81710 | 3167 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3168 | if (!l) { |
e6b81710 | 3169 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3170 | pidlist_free(array); |
72a8cb30 | 3171 | return -ENOMEM; |
102a775e | 3172 | } |
e6b81710 TH |
3173 | |
3174 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3175 | pidlist_free(l->list); |
102a775e BB |
3176 | l->list = array; |
3177 | l->length = length; | |
72a8cb30 | 3178 | *lp = l; |
102a775e | 3179 | return 0; |
bbcb81d0 PM |
3180 | } |
3181 | ||
846c7bb0 | 3182 | /** |
a043e3b2 | 3183 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3184 | * @stats: cgroupstats to fill information into |
3185 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3186 | * been requested. | |
a043e3b2 LZ |
3187 | * |
3188 | * Build and fill cgroupstats so that taskstats can export it to user | |
3189 | * space. | |
846c7bb0 BS |
3190 | */ |
3191 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3192 | { | |
2bd59d48 | 3193 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3194 | struct cgroup *cgrp; |
72ec7029 | 3195 | struct css_task_iter it; |
846c7bb0 | 3196 | struct task_struct *tsk; |
33d283be | 3197 | |
2bd59d48 TH |
3198 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3199 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3200 | kernfs_type(kn) != KERNFS_DIR) | |
3201 | return -EINVAL; | |
3202 | ||
846c7bb0 | 3203 | /* |
2bd59d48 TH |
3204 | * We aren't being called from kernfs and there's no guarantee on |
3205 | * @kn->priv's validity. For this and css_tryget_from_dir(), | |
3206 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
846c7bb0 | 3207 | */ |
2bd59d48 TH |
3208 | rcu_read_lock(); |
3209 | cgrp = rcu_dereference(kn->priv); | |
3210 | if (!cgrp) { | |
3211 | rcu_read_unlock(); | |
3212 | return -ENOENT; | |
3213 | } | |
846c7bb0 | 3214 | |
72ec7029 TH |
3215 | css_task_iter_start(&cgrp->dummy_css, &it); |
3216 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3217 | switch (tsk->state) { |
3218 | case TASK_RUNNING: | |
3219 | stats->nr_running++; | |
3220 | break; | |
3221 | case TASK_INTERRUPTIBLE: | |
3222 | stats->nr_sleeping++; | |
3223 | break; | |
3224 | case TASK_UNINTERRUPTIBLE: | |
3225 | stats->nr_uninterruptible++; | |
3226 | break; | |
3227 | case TASK_STOPPED: | |
3228 | stats->nr_stopped++; | |
3229 | break; | |
3230 | default: | |
3231 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3232 | stats->nr_io_wait++; | |
3233 | break; | |
3234 | } | |
3235 | } | |
72ec7029 | 3236 | css_task_iter_end(&it); |
846c7bb0 | 3237 | |
2bd59d48 TH |
3238 | rcu_read_unlock(); |
3239 | return 0; | |
846c7bb0 BS |
3240 | } |
3241 | ||
8f3ff208 | 3242 | |
bbcb81d0 | 3243 | /* |
102a775e | 3244 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3245 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3246 | * in the cgroup->l->list array. |
bbcb81d0 | 3247 | */ |
cc31edce | 3248 | |
102a775e | 3249 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3250 | { |
cc31edce PM |
3251 | /* |
3252 | * Initially we receive a position value that corresponds to | |
3253 | * one more than the last pid shown (or 0 on the first call or | |
3254 | * after a seek to the start). Use a binary-search to find the | |
3255 | * next pid to display, if any | |
3256 | */ | |
2bd59d48 | 3257 | struct kernfs_open_file *of = s->private; |
7da11279 | 3258 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3259 | struct cgroup_pidlist *l; |
7da11279 | 3260 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3261 | int index = 0, pid = *pos; |
4bac00d1 TH |
3262 | int *iter, ret; |
3263 | ||
3264 | mutex_lock(&cgrp->pidlist_mutex); | |
3265 | ||
3266 | /* | |
5d22444f | 3267 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3268 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3269 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3270 | * could already have been destroyed. |
3271 | */ | |
5d22444f TH |
3272 | if (of->priv) |
3273 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3274 | |
3275 | /* | |
3276 | * Either this is the first start() after open or the matching | |
3277 | * pidlist has been destroyed inbetween. Create a new one. | |
3278 | */ | |
5d22444f TH |
3279 | if (!of->priv) { |
3280 | ret = pidlist_array_load(cgrp, type, | |
3281 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3282 | if (ret) |
3283 | return ERR_PTR(ret); | |
3284 | } | |
5d22444f | 3285 | l = of->priv; |
cc31edce | 3286 | |
cc31edce | 3287 | if (pid) { |
102a775e | 3288 | int end = l->length; |
20777766 | 3289 | |
cc31edce PM |
3290 | while (index < end) { |
3291 | int mid = (index + end) / 2; | |
afb2bc14 | 3292 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3293 | index = mid; |
3294 | break; | |
afb2bc14 | 3295 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3296 | index = mid + 1; |
3297 | else | |
3298 | end = mid; | |
3299 | } | |
3300 | } | |
3301 | /* If we're off the end of the array, we're done */ | |
102a775e | 3302 | if (index >= l->length) |
cc31edce PM |
3303 | return NULL; |
3304 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3305 | iter = l->list + index; |
afb2bc14 | 3306 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3307 | return iter; |
3308 | } | |
3309 | ||
102a775e | 3310 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3311 | { |
2bd59d48 | 3312 | struct kernfs_open_file *of = s->private; |
5d22444f | 3313 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3314 | |
5d22444f TH |
3315 | if (l) |
3316 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3317 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3318 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3319 | } |
3320 | ||
102a775e | 3321 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3322 | { |
2bd59d48 | 3323 | struct kernfs_open_file *of = s->private; |
5d22444f | 3324 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3325 | pid_t *p = v; |
3326 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3327 | /* |
3328 | * Advance to the next pid in the array. If this goes off the | |
3329 | * end, we're done | |
3330 | */ | |
3331 | p++; | |
3332 | if (p >= end) { | |
3333 | return NULL; | |
3334 | } else { | |
7da11279 | 3335 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3336 | return p; |
3337 | } | |
3338 | } | |
3339 | ||
102a775e | 3340 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3341 | { |
3342 | return seq_printf(s, "%d\n", *(int *)v); | |
3343 | } | |
bbcb81d0 | 3344 | |
102a775e BB |
3345 | /* |
3346 | * seq_operations functions for iterating on pidlists through seq_file - | |
3347 | * independent of whether it's tasks or procs | |
3348 | */ | |
3349 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3350 | .start = cgroup_pidlist_start, | |
3351 | .stop = cgroup_pidlist_stop, | |
3352 | .next = cgroup_pidlist_next, | |
3353 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3354 | }; |
3355 | ||
182446d0 TH |
3356 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3357 | struct cftype *cft) | |
81a6a5cd | 3358 | { |
182446d0 | 3359 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3360 | } |
3361 | ||
182446d0 TH |
3362 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3363 | struct cftype *cft, u64 val) | |
6379c106 | 3364 | { |
182446d0 | 3365 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3366 | if (val) |
182446d0 | 3367 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3368 | else |
182446d0 | 3369 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3370 | return 0; |
3371 | } | |
3372 | ||
182446d0 TH |
3373 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3374 | struct cftype *cft) | |
97978e6d | 3375 | { |
182446d0 | 3376 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3377 | } |
3378 | ||
182446d0 TH |
3379 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3380 | struct cftype *cft, u64 val) | |
97978e6d DL |
3381 | { |
3382 | if (val) | |
182446d0 | 3383 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3384 | else |
182446d0 | 3385 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3386 | return 0; |
3387 | } | |
3388 | ||
d5c56ced | 3389 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3390 | { |
d5c56ced | 3391 | .name = "cgroup.procs", |
6612f05b TH |
3392 | .seq_start = cgroup_pidlist_start, |
3393 | .seq_next = cgroup_pidlist_next, | |
3394 | .seq_stop = cgroup_pidlist_stop, | |
3395 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3396 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3397 | .write_u64 = cgroup_procs_write, |
74a1166d | 3398 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3399 | }, |
97978e6d DL |
3400 | { |
3401 | .name = "cgroup.clone_children", | |
873fe09e | 3402 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3403 | .read_u64 = cgroup_clone_children_read, |
3404 | .write_u64 = cgroup_clone_children_write, | |
3405 | }, | |
873fe09e TH |
3406 | { |
3407 | .name = "cgroup.sane_behavior", | |
3408 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3409 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3410 | }, |
d5c56ced TH |
3411 | |
3412 | /* | |
3413 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3414 | * don't exist if sane_behavior. If you're depending on these, be | |
3415 | * prepared to be burned. | |
3416 | */ | |
3417 | { | |
3418 | .name = "tasks", | |
3419 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3420 | .seq_start = cgroup_pidlist_start, |
3421 | .seq_next = cgroup_pidlist_next, | |
3422 | .seq_stop = cgroup_pidlist_stop, | |
3423 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3424 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3425 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3426 | .mode = S_IRUGO | S_IWUSR, |
3427 | }, | |
3428 | { | |
3429 | .name = "notify_on_release", | |
3430 | .flags = CFTYPE_INSANE, | |
3431 | .read_u64 = cgroup_read_notify_on_release, | |
3432 | .write_u64 = cgroup_write_notify_on_release, | |
3433 | }, | |
6e6ff25b TH |
3434 | { |
3435 | .name = "release_agent", | |
cc5943a7 | 3436 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3437 | .seq_show = cgroup_release_agent_show, |
6e6ff25b | 3438 | .write_string = cgroup_release_agent_write, |
5f469907 | 3439 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 3440 | }, |
db0416b6 | 3441 | { } /* terminate */ |
bbcb81d0 PM |
3442 | }; |
3443 | ||
13af07df | 3444 | /** |
628f7cd4 | 3445 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3446 | * @cgrp: target cgroup |
13af07df | 3447 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3448 | * |
3449 | * On failure, no file is added. | |
13af07df | 3450 | */ |
628f7cd4 | 3451 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3452 | { |
ddbcc7e8 | 3453 | struct cgroup_subsys *ss; |
b420ba7d | 3454 | int i, ret = 0; |
bbcb81d0 | 3455 | |
8e3f6541 | 3456 | /* process cftsets of each subsystem */ |
b420ba7d | 3457 | for_each_subsys(ss, i) { |
0adb0704 | 3458 | struct cftype *cfts; |
b420ba7d TH |
3459 | |
3460 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3461 | continue; |
8e3f6541 | 3462 | |
0adb0704 TH |
3463 | list_for_each_entry(cfts, &ss->cfts, node) { |
3464 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
3465 | if (ret < 0) |
3466 | goto err; | |
3467 | } | |
ddbcc7e8 | 3468 | } |
ddbcc7e8 | 3469 | return 0; |
bee55099 TH |
3470 | err: |
3471 | cgroup_clear_dir(cgrp, subsys_mask); | |
3472 | return ret; | |
ddbcc7e8 PM |
3473 | } |
3474 | ||
0c21ead1 TH |
3475 | /* |
3476 | * css destruction is four-stage process. | |
3477 | * | |
3478 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3479 | * Implemented in kill_css(). | |
3480 | * | |
3481 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3482 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3483 | * by invoking offline_css(). After offlining, the base ref is put. | |
3484 | * Implemented in css_killed_work_fn(). | |
3485 | * | |
3486 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3487 | * accessors are inside RCU read sections. css_release() schedules the | |
3488 | * RCU callback. | |
3489 | * | |
3490 | * 4. After the grace period, the css can be freed. Implemented in | |
3491 | * css_free_work_fn(). | |
3492 | * | |
3493 | * It is actually hairier because both step 2 and 4 require process context | |
3494 | * and thus involve punting to css->destroy_work adding two additional | |
3495 | * steps to the already complex sequence. | |
3496 | */ | |
35ef10da | 3497 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3498 | { |
3499 | struct cgroup_subsys_state *css = | |
35ef10da | 3500 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 3501 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3502 | |
0ae78e0b TH |
3503 | if (css->parent) |
3504 | css_put(css->parent); | |
3505 | ||
0c21ead1 | 3506 | css->ss->css_free(css); |
2bd59d48 | 3507 | cgroup_put(cgrp); |
48ddbe19 TH |
3508 | } |
3509 | ||
0c21ead1 | 3510 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3511 | { |
3512 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3513 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3514 | |
35ef10da | 3515 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 3516 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
3517 | } |
3518 | ||
d3daf28d TH |
3519 | static void css_release(struct percpu_ref *ref) |
3520 | { | |
3521 | struct cgroup_subsys_state *css = | |
3522 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3523 | ||
aec25020 | 3524 | rcu_assign_pointer(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 3525 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
3526 | } |
3527 | ||
623f926b TH |
3528 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
3529 | struct cgroup *cgrp) | |
ddbcc7e8 | 3530 | { |
bd89aabc | 3531 | css->cgroup = cgrp; |
72c97e54 | 3532 | css->ss = ss; |
ddbcc7e8 | 3533 | css->flags = 0; |
0ae78e0b TH |
3534 | |
3535 | if (cgrp->parent) | |
ca8bdcaf | 3536 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 3537 | else |
38b53aba | 3538 | css->flags |= CSS_ROOT; |
48ddbe19 | 3539 | |
ca8bdcaf | 3540 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
3541 | } |
3542 | ||
2a4ac633 | 3543 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 3544 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3545 | { |
623f926b | 3546 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
3547 | int ret = 0; |
3548 | ||
ace2bee8 | 3549 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3550 | lockdep_assert_held(&cgroup_mutex); |
3551 | ||
92fb9748 | 3552 | if (ss->css_online) |
eb95419b | 3553 | ret = ss->css_online(css); |
ae7f164a | 3554 | if (!ret) { |
eb95419b | 3555 | css->flags |= CSS_ONLINE; |
f20104de | 3556 | css->cgroup->nr_css++; |
aec25020 | 3557 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 3558 | } |
b1929db4 | 3559 | return ret; |
a31f2d3f TH |
3560 | } |
3561 | ||
2a4ac633 | 3562 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 3563 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3564 | { |
623f926b | 3565 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 3566 | |
ace2bee8 | 3567 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3568 | lockdep_assert_held(&cgroup_mutex); |
3569 | ||
3570 | if (!(css->flags & CSS_ONLINE)) | |
3571 | return; | |
3572 | ||
d7eeac19 | 3573 | if (ss->css_offline) |
eb95419b | 3574 | ss->css_offline(css); |
a31f2d3f | 3575 | |
eb95419b | 3576 | css->flags &= ~CSS_ONLINE; |
09a503ea | 3577 | css->cgroup->nr_css--; |
aec25020 | 3578 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); |
a31f2d3f TH |
3579 | } |
3580 | ||
c81c925a TH |
3581 | /** |
3582 | * create_css - create a cgroup_subsys_state | |
3583 | * @cgrp: the cgroup new css will be associated with | |
3584 | * @ss: the subsys of new css | |
3585 | * | |
3586 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
3587 | * css is online and installed in @cgrp with all interface files created. | |
3588 | * Returns 0 on success, -errno on failure. | |
3589 | */ | |
3590 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
3591 | { | |
3592 | struct cgroup *parent = cgrp->parent; | |
3593 | struct cgroup_subsys_state *css; | |
3594 | int err; | |
3595 | ||
c81c925a TH |
3596 | lockdep_assert_held(&cgroup_mutex); |
3597 | ||
3598 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
3599 | if (IS_ERR(css)) | |
3600 | return PTR_ERR(css); | |
3601 | ||
3602 | err = percpu_ref_init(&css->refcnt, css_release); | |
3603 | if (err) | |
3604 | goto err_free; | |
3605 | ||
3606 | init_css(css, ss, cgrp); | |
3607 | ||
aec25020 | 3608 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a TH |
3609 | if (err) |
3610 | goto err_free; | |
3611 | ||
3612 | err = online_css(css); | |
3613 | if (err) | |
3614 | goto err_free; | |
3615 | ||
59f5296b | 3616 | cgroup_get(cgrp); |
c81c925a TH |
3617 | css_get(css->parent); |
3618 | ||
3619 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
3620 | parent->parent) { | |
3621 | 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", | |
3622 | current->comm, current->pid, ss->name); | |
3623 | if (!strcmp(ss->name, "memory")) | |
3624 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
3625 | ss->warned_broken_hierarchy = true; | |
3626 | } | |
3627 | ||
3628 | return 0; | |
3629 | ||
3630 | err_free: | |
3631 | percpu_ref_cancel_init(&css->refcnt); | |
3632 | ss->css_free(css); | |
3633 | return err; | |
3634 | } | |
3635 | ||
2bd59d48 | 3636 | /** |
a043e3b2 LZ |
3637 | * cgroup_create - create a cgroup |
3638 | * @parent: cgroup that will be parent of the new cgroup | |
e61734c5 | 3639 | * @name: name of the new cgroup |
2bd59d48 | 3640 | * @mode: mode to set on new cgroup |
ddbcc7e8 | 3641 | */ |
e61734c5 | 3642 | static long cgroup_create(struct cgroup *parent, const char *name, |
2bd59d48 | 3643 | umode_t mode) |
ddbcc7e8 | 3644 | { |
bd89aabc | 3645 | struct cgroup *cgrp; |
ddbcc7e8 | 3646 | struct cgroupfs_root *root = parent->root; |
b58c8998 | 3647 | int ssid, err; |
ddbcc7e8 | 3648 | struct cgroup_subsys *ss; |
2bd59d48 | 3649 | struct kernfs_node *kn; |
ddbcc7e8 | 3650 | |
0a950f65 | 3651 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
3652 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
3653 | if (!cgrp) | |
ddbcc7e8 PM |
3654 | return -ENOMEM; |
3655 | ||
ace2bee8 TH |
3656 | mutex_lock(&cgroup_tree_mutex); |
3657 | ||
976c06bc TH |
3658 | /* |
3659 | * Only live parents can have children. Note that the liveliness | |
3660 | * check isn't strictly necessary because cgroup_mkdir() and | |
3661 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
3662 | * anyway so that locking is contained inside cgroup proper and we | |
3663 | * don't get nasty surprises if we ever grow another caller. | |
3664 | */ | |
3665 | if (!cgroup_lock_live_group(parent)) { | |
3666 | err = -ENODEV; | |
ace2bee8 | 3667 | goto err_unlock_tree; |
0ab02ca8 LZ |
3668 | } |
3669 | ||
3670 | /* | |
3671 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
3672 | * a half-baked cgroup. | |
3673 | */ | |
3674 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
3675 | if (cgrp->id < 0) { | |
3676 | err = -ENOMEM; | |
3677 | goto err_unlock; | |
976c06bc TH |
3678 | } |
3679 | ||
cc31edce | 3680 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3681 | |
bd89aabc | 3682 | cgrp->parent = parent; |
0ae78e0b | 3683 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 3684 | cgrp->root = parent->root; |
ddbcc7e8 | 3685 | |
b6abdb0e LZ |
3686 | if (notify_on_release(parent)) |
3687 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3688 | ||
2260e7fc TH |
3689 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
3690 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 3691 | |
2bd59d48 | 3692 | /* create the directory */ |
e61734c5 | 3693 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 TH |
3694 | if (IS_ERR(kn)) { |
3695 | err = PTR_ERR(kn); | |
0ab02ca8 | 3696 | goto err_free_id; |
2bd59d48 TH |
3697 | } |
3698 | cgrp->kn = kn; | |
ddbcc7e8 | 3699 | |
6f30558f TH |
3700 | /* |
3701 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
3702 | * that @cgrp->kn is always accessible. | |
3703 | */ | |
3704 | kernfs_get(kn); | |
3705 | ||
00356bd5 | 3706 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 3707 | |
4e139afc | 3708 | /* allocation complete, commit to creation */ |
4e139afc | 3709 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3c9c825b | 3710 | atomic_inc(&root->nr_cgrps); |
28fd6f30 | 3711 | |
2bd59d48 TH |
3712 | /* |
3713 | * Grab a reference on the root and parent so that they don't get | |
3714 | * deleted while there are child cgroups. | |
3715 | */ | |
3716 | cgroup_get_root(root); | |
59f5296b | 3717 | cgroup_get(parent); |
415cf07a | 3718 | |
0d80255e TH |
3719 | /* |
3720 | * @cgrp is now fully operational. If something fails after this | |
3721 | * point, it'll be released via the normal destruction path. | |
3722 | */ | |
4e96ee8e LZ |
3723 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
3724 | ||
2bb566cb | 3725 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
3726 | if (err) |
3727 | goto err_destroy; | |
3728 | ||
9d403e99 | 3729 | /* let's create and online css's */ |
b85d2040 TH |
3730 | for_each_subsys(ss, ssid) { |
3731 | if (root->subsys_mask & (1 << ssid)) { | |
3732 | err = create_css(cgrp, ss); | |
3733 | if (err) | |
3734 | goto err_destroy; | |
3735 | } | |
a8638030 | 3736 | } |
ddbcc7e8 | 3737 | |
2bd59d48 TH |
3738 | kernfs_activate(kn); |
3739 | ||
ddbcc7e8 | 3740 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3741 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
3742 | |
3743 | return 0; | |
3744 | ||
0a950f65 | 3745 | err_free_id: |
4e96ee8e | 3746 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
3747 | err_unlock: |
3748 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
3749 | err_unlock_tree: |
3750 | mutex_unlock(&cgroup_tree_mutex); | |
bd89aabc | 3751 | kfree(cgrp); |
ddbcc7e8 | 3752 | return err; |
4b8b47eb TH |
3753 | |
3754 | err_destroy: | |
3755 | cgroup_destroy_locked(cgrp); | |
3756 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3757 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 3758 | return err; |
ddbcc7e8 PM |
3759 | } |
3760 | ||
2bd59d48 TH |
3761 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
3762 | umode_t mode) | |
ddbcc7e8 | 3763 | { |
2bd59d48 | 3764 | struct cgroup *parent = parent_kn->priv; |
ddbcc7e8 | 3765 | |
2bd59d48 | 3766 | return cgroup_create(parent, name, mode); |
ddbcc7e8 PM |
3767 | } |
3768 | ||
223dbc38 TH |
3769 | /* |
3770 | * This is called when the refcnt of a css is confirmed to be killed. | |
3771 | * css_tryget() is now guaranteed to fail. | |
3772 | */ | |
3773 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 3774 | { |
223dbc38 TH |
3775 | struct cgroup_subsys_state *css = |
3776 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
3777 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 3778 | |
ace2bee8 | 3779 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
3780 | mutex_lock(&cgroup_mutex); |
3781 | ||
09a503ea TH |
3782 | /* |
3783 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
3784 | * initate destruction. | |
3785 | */ | |
3786 | offline_css(css); | |
3787 | ||
f20104de TH |
3788 | /* |
3789 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
3790 | * be disabled before proceeding to the second phase of cgroup | |
3791 | * destruction. If we are the last one, kick it off. | |
3792 | */ | |
09a503ea | 3793 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
3794 | cgroup_destroy_css_killed(cgrp); |
3795 | ||
3796 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3797 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
3798 | |
3799 | /* | |
3800 | * Put the css refs from kill_css(). Each css holds an extra | |
3801 | * reference to the cgroup's dentry and cgroup removal proceeds | |
3802 | * regardless of css refs. On the last put of each css, whenever | |
3803 | * that may be, the extra dentry ref is put so that dentry | |
3804 | * destruction happens only after all css's are released. | |
3805 | */ | |
3806 | css_put(css); | |
d3daf28d TH |
3807 | } |
3808 | ||
223dbc38 TH |
3809 | /* css kill confirmation processing requires process context, bounce */ |
3810 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
3811 | { |
3812 | struct cgroup_subsys_state *css = | |
3813 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3814 | ||
223dbc38 | 3815 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 3816 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
3817 | } |
3818 | ||
edae0c33 TH |
3819 | /** |
3820 | * kill_css - destroy a css | |
3821 | * @css: css to destroy | |
3822 | * | |
3c14f8b4 TH |
3823 | * This function initiates destruction of @css by removing cgroup interface |
3824 | * files and putting its base reference. ->css_offline() will be invoked | |
3825 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
3826 | * reference count reaches zero, @css will be released. | |
edae0c33 TH |
3827 | */ |
3828 | static void kill_css(struct cgroup_subsys_state *css) | |
3829 | { | |
2bd59d48 TH |
3830 | /* |
3831 | * This must happen before css is disassociated with its cgroup. | |
3832 | * See seq_css() for details. | |
3833 | */ | |
aec25020 | 3834 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 3835 | |
edae0c33 TH |
3836 | /* |
3837 | * Killing would put the base ref, but we need to keep it alive | |
3838 | * until after ->css_offline(). | |
3839 | */ | |
3840 | css_get(css); | |
3841 | ||
3842 | /* | |
3843 | * cgroup core guarantees that, by the time ->css_offline() is | |
3844 | * invoked, no new css reference will be given out via | |
3845 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
3846 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
3847 | * guarantee that the ref is seen as killed on all CPUs on return. | |
3848 | * | |
3849 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
3850 | * css is confirmed to be seen as killed on all CPUs. | |
3851 | */ | |
3852 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
3853 | } |
3854 | ||
3855 | /** | |
3856 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
3857 | * @cgrp: cgroup to be destroyed | |
3858 | * | |
3859 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
3860 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
3861 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
3862 | * invoked. To satisfy all the requirements, destruction is implemented in | |
3863 | * the following two steps. | |
3864 | * | |
3865 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
3866 | * userland visible parts and start killing the percpu refcnts of | |
3867 | * css's. Set up so that the next stage will be kicked off once all | |
3868 | * the percpu refcnts are confirmed to be killed. | |
3869 | * | |
3870 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
3871 | * rest of destruction. Once all cgroup references are gone, the | |
3872 | * cgroup is RCU-freed. | |
3873 | * | |
3874 | * This function implements s1. After this step, @cgrp is gone as far as | |
3875 | * the userland is concerned and a new cgroup with the same name may be | |
3876 | * created. As cgroup doesn't care about the names internally, this | |
3877 | * doesn't cause any problem. | |
3878 | */ | |
42809dd4 TH |
3879 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
3880 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 3881 | { |
bb78a92f | 3882 | struct cgroup *child; |
2bd59d48 | 3883 | struct cgroup_subsys_state *css; |
ddd69148 | 3884 | bool empty; |
1c6727af | 3885 | int ssid; |
ddbcc7e8 | 3886 | |
ace2bee8 | 3887 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
3888 | lockdep_assert_held(&cgroup_mutex); |
3889 | ||
ddd69148 | 3890 | /* |
6f3d828f TH |
3891 | * css_set_lock synchronizes access to ->cset_links and prevents |
3892 | * @cgrp from being removed while __put_css_set() is in progress. | |
ddd69148 TH |
3893 | */ |
3894 | read_lock(&css_set_lock); | |
bb78a92f | 3895 | empty = list_empty(&cgrp->cset_links); |
ddd69148 TH |
3896 | read_unlock(&css_set_lock); |
3897 | if (!empty) | |
ddbcc7e8 | 3898 | return -EBUSY; |
a043e3b2 | 3899 | |
bb78a92f HD |
3900 | /* |
3901 | * Make sure there's no live children. We can't test ->children | |
3902 | * emptiness as dead children linger on it while being destroyed; | |
3903 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
3904 | */ | |
3905 | empty = true; | |
3906 | rcu_read_lock(); | |
3907 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
3908 | empty = cgroup_is_dead(child); | |
3909 | if (!empty) | |
3910 | break; | |
3911 | } | |
3912 | rcu_read_unlock(); | |
3913 | if (!empty) | |
3914 | return -EBUSY; | |
3915 | ||
88703267 | 3916 | /* |
edae0c33 TH |
3917 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
3918 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
3919 | * percpu refs of all css's are confirmed to be killed. This |
3920 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 3921 | */ |
4ac06017 | 3922 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
3923 | for_each_css(css, ssid, cgrp) |
3924 | kill_css(css); | |
4ac06017 | 3925 | mutex_lock(&cgroup_mutex); |
455050d2 TH |
3926 | |
3927 | /* | |
3928 | * Mark @cgrp dead. This prevents further task migration and child | |
3929 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 3930 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 3931 | * resume iteration after dropping RCU read lock. See |
492eb21b | 3932 | * css_next_child() for details. |
455050d2 | 3933 | */ |
54766d4a | 3934 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 3935 | |
455050d2 TH |
3936 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
3937 | raw_spin_lock(&release_list_lock); | |
3938 | if (!list_empty(&cgrp->release_list)) | |
3939 | list_del_init(&cgrp->release_list); | |
3940 | raw_spin_unlock(&release_list_lock); | |
3941 | ||
3942 | /* | |
f20104de TH |
3943 | * If @cgrp has css's attached, the second stage of cgroup |
3944 | * destruction is kicked off from css_killed_work_fn() after the | |
3945 | * refs of all attached css's are killed. If @cgrp doesn't have | |
3946 | * any css, we kick it off here. | |
3947 | */ | |
3948 | if (!cgrp->nr_css) | |
3949 | cgroup_destroy_css_killed(cgrp); | |
3950 | ||
2bd59d48 TH |
3951 | /* remove @cgrp directory along with the base files */ |
3952 | mutex_unlock(&cgroup_mutex); | |
3953 | ||
455050d2 | 3954 | /* |
2bd59d48 TH |
3955 | * There are two control paths which try to determine cgroup from |
3956 | * dentry without going through kernfs - cgroupstats_build() and | |
3957 | * css_tryget_from_dir(). Those are supported by RCU protecting | |
3958 | * clearing of cgrp->kn->priv backpointer, which should happen | |
3959 | * after all files under it have been removed. | |
455050d2 | 3960 | */ |
6f30558f | 3961 | kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ |
2bd59d48 | 3962 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); |
2bd59d48 | 3963 | |
4ac06017 | 3964 | mutex_lock(&cgroup_mutex); |
455050d2 | 3965 | |
ea15f8cc TH |
3966 | return 0; |
3967 | }; | |
3968 | ||
d3daf28d | 3969 | /** |
f20104de | 3970 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
3971 | * @work: cgroup->destroy_free_work |
3972 | * | |
3973 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
3974 | * destroyed after all css's are offlined and performs the rest of |
3975 | * destruction. This is the second step of destruction described in the | |
3976 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 3977 | */ |
f20104de | 3978 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 3979 | { |
ea15f8cc | 3980 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 3981 | |
ace2bee8 | 3982 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 3983 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 3984 | |
999cd8a4 | 3985 | /* delete this cgroup from parent->children */ |
eb6fd504 | 3986 | list_del_rcu(&cgrp->sibling); |
ed957793 | 3987 | |
59f5296b | 3988 | cgroup_put(cgrp); |
ddbcc7e8 | 3989 | |
bd89aabc | 3990 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 3991 | check_for_release(parent); |
ddbcc7e8 PM |
3992 | } |
3993 | ||
2bd59d48 | 3994 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 3995 | { |
2bd59d48 TH |
3996 | struct cgroup *cgrp = kn->priv; |
3997 | int ret = 0; | |
3998 | ||
3999 | /* | |
4000 | * This is self-destruction but @kn can't be removed while this | |
4001 | * callback is in progress. Let's break active protection. Once | |
4002 | * the protection is broken, @cgrp can be destroyed at any point. | |
4003 | * Pin it so that it stays accessible. | |
4004 | */ | |
4005 | cgroup_get(cgrp); | |
4006 | kernfs_break_active_protection(kn); | |
42809dd4 | 4007 | |
ace2bee8 | 4008 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 4009 | mutex_lock(&cgroup_mutex); |
2bd59d48 TH |
4010 | |
4011 | /* | |
4012 | * @cgrp might already have been destroyed while we're trying to | |
4013 | * grab the mutexes. | |
4014 | */ | |
4015 | if (!cgroup_is_dead(cgrp)) | |
4016 | ret = cgroup_destroy_locked(cgrp); | |
4017 | ||
42809dd4 | 4018 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4019 | mutex_unlock(&cgroup_tree_mutex); |
42809dd4 | 4020 | |
2bd59d48 TH |
4021 | kernfs_unbreak_active_protection(kn); |
4022 | cgroup_put(cgrp); | |
42809dd4 TH |
4023 | return ret; |
4024 | } | |
4025 | ||
2bd59d48 TH |
4026 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4027 | .remount_fs = cgroup_remount, | |
4028 | .show_options = cgroup_show_options, | |
4029 | .mkdir = cgroup_mkdir, | |
4030 | .rmdir = cgroup_rmdir, | |
4031 | .rename = cgroup_rename, | |
4032 | }; | |
4033 | ||
06a11920 | 4034 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4035 | { |
ddbcc7e8 | 4036 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4037 | |
4038 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4039 | |
ace2bee8 | 4040 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
4041 | mutex_lock(&cgroup_mutex); |
4042 | ||
0adb0704 | 4043 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4044 | |
ddbcc7e8 | 4045 | /* Create the top cgroup state for this subsystem */ |
9871bf95 | 4046 | ss->root = &cgroup_dummy_root; |
ca8bdcaf | 4047 | css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss)); |
ddbcc7e8 PM |
4048 | /* We don't handle early failures gracefully */ |
4049 | BUG_ON(IS_ERR(css)); | |
623f926b | 4050 | init_css(css, ss, cgroup_dummy_top); |
ddbcc7e8 | 4051 | |
e8d55fde | 4052 | /* Update the init_css_set to contain a subsys |
817929ec | 4053 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
4054 | * newly registered, all tasks and hence the |
4055 | * init_css_set is in the subsystem's top cgroup. */ | |
aec25020 | 4056 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4057 | |
4058 | need_forkexit_callback |= ss->fork || ss->exit; | |
4059 | ||
e8d55fde LZ |
4060 | /* At system boot, before all subsystems have been |
4061 | * registered, no tasks have been forked, so we don't | |
4062 | * need to invoke fork callbacks here. */ | |
4063 | BUG_ON(!list_empty(&init_task.tasks)); | |
4064 | ||
ae7f164a | 4065 | BUG_ON(online_css(css)); |
a8638030 | 4066 | |
648bb56d | 4067 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4068 | mutex_unlock(&cgroup_tree_mutex); |
e6a1105b BB |
4069 | } |
4070 | ||
ddbcc7e8 | 4071 | /** |
a043e3b2 LZ |
4072 | * cgroup_init_early - cgroup initialization at system boot |
4073 | * | |
4074 | * Initialize cgroups at system boot, and initialize any | |
4075 | * subsystems that request early init. | |
ddbcc7e8 PM |
4076 | */ |
4077 | int __init cgroup_init_early(void) | |
4078 | { | |
30159ec7 | 4079 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4080 | int i; |
30159ec7 | 4081 | |
146aa1bd | 4082 | atomic_set(&init_css_set.refcount, 1); |
69d0206c | 4083 | INIT_LIST_HEAD(&init_css_set.cgrp_links); |
817929ec | 4084 | INIT_LIST_HEAD(&init_css_set.tasks); |
472b1053 | 4085 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 4086 | css_set_count = 1; |
9871bf95 TH |
4087 | init_cgroup_root(&cgroup_dummy_root); |
4088 | cgroup_root_count = 1; | |
a4ea1cc9 | 4089 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4090 | |
69d0206c | 4091 | init_cgrp_cset_link.cset = &init_css_set; |
9871bf95 TH |
4092 | init_cgrp_cset_link.cgrp = cgroup_dummy_top; |
4093 | list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links); | |
69d0206c | 4094 | list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links); |
ddbcc7e8 | 4095 | |
3ed80a62 | 4096 | for_each_subsys(ss, i) { |
aec25020 | 4097 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4098 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4099 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4100 | ss->id, ss->name); |
073219e9 TH |
4101 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4102 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4103 | ||
aec25020 | 4104 | ss->id = i; |
073219e9 | 4105 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4106 | |
4107 | if (ss->early_init) | |
4108 | cgroup_init_subsys(ss); | |
4109 | } | |
4110 | return 0; | |
4111 | } | |
4112 | ||
4113 | /** | |
a043e3b2 LZ |
4114 | * cgroup_init - cgroup initialization |
4115 | * | |
4116 | * Register cgroup filesystem and /proc file, and initialize | |
4117 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4118 | */ |
4119 | int __init cgroup_init(void) | |
4120 | { | |
30159ec7 | 4121 | struct cgroup_subsys *ss; |
0ac801fe | 4122 | unsigned long key; |
30159ec7 | 4123 | int i, err; |
a424316c | 4124 | |
2bd59d48 | 4125 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
2da440a2 | 4126 | |
3ed80a62 | 4127 | for_each_subsys(ss, i) { |
ddbcc7e8 PM |
4128 | if (!ss->early_init) |
4129 | cgroup_init_subsys(ss); | |
de00ffa5 TH |
4130 | |
4131 | /* | |
4132 | * cftype registration needs kmalloc and can't be done | |
4133 | * during early_init. Register base cftypes separately. | |
4134 | */ | |
4135 | if (ss->base_cftypes) | |
4136 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
ddbcc7e8 PM |
4137 | } |
4138 | ||
fa3ca07e | 4139 | /* allocate id for the dummy hierarchy */ |
54e7b4eb | 4140 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4141 | |
82fe9b0d TH |
4142 | /* Add init_css_set to the hash table */ |
4143 | key = css_set_hash(init_css_set.subsys); | |
4144 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4145 | ||
fc76df70 | 4146 | BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1)); |
676db4af | 4147 | |
4e96ee8e LZ |
4148 | err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top, |
4149 | 0, 1, GFP_KERNEL); | |
4150 | BUG_ON(err < 0); | |
4151 | ||
54e7b4eb TH |
4152 | mutex_unlock(&cgroup_mutex); |
4153 | ||
676db4af | 4154 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4155 | if (!cgroup_kobj) |
4156 | return -ENOMEM; | |
676db4af | 4157 | |
ddbcc7e8 | 4158 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
4159 | if (err < 0) { |
4160 | kobject_put(cgroup_kobj); | |
2bd59d48 | 4161 | return err; |
676db4af | 4162 | } |
ddbcc7e8 | 4163 | |
46ae220b | 4164 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 4165 | return 0; |
ddbcc7e8 | 4166 | } |
b4f48b63 | 4167 | |
e5fca243 TH |
4168 | static int __init cgroup_wq_init(void) |
4169 | { | |
4170 | /* | |
4171 | * There isn't much point in executing destruction path in | |
4172 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
ab3f5faa HD |
4173 | * |
4174 | * XXX: Must be ordered to make sure parent is offlined after | |
4175 | * children. The ordering requirement is for memcg where a | |
4176 | * parent's offline may wait for a child's leading to deadlock. In | |
4177 | * the long term, this should be fixed from memcg side. | |
e5fca243 TH |
4178 | * |
4179 | * We would prefer to do this in cgroup_init() above, but that | |
4180 | * is called before init_workqueues(): so leave this until after. | |
4181 | */ | |
ab3f5faa | 4182 | cgroup_destroy_wq = alloc_ordered_workqueue("cgroup_destroy", 0); |
e5fca243 | 4183 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4184 | |
4185 | /* | |
4186 | * Used to destroy pidlists and separate to serve as flush domain. | |
4187 | * Cap @max_active to 1 too. | |
4188 | */ | |
4189 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4190 | 0, 1); | |
4191 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4192 | ||
e5fca243 TH |
4193 | return 0; |
4194 | } | |
4195 | core_initcall(cgroup_wq_init); | |
4196 | ||
a424316c PM |
4197 | /* |
4198 | * proc_cgroup_show() | |
4199 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4200 | * - Used for /proc/<pid>/cgroup. | |
4201 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4202 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4203 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4204 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4205 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4206 | * cgroup to top_cgroup. | |
4207 | */ | |
4208 | ||
4209 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4210 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4211 | { |
4212 | struct pid *pid; | |
4213 | struct task_struct *tsk; | |
e61734c5 | 4214 | char *buf, *path; |
a424316c PM |
4215 | int retval; |
4216 | struct cgroupfs_root *root; | |
4217 | ||
4218 | retval = -ENOMEM; | |
e61734c5 | 4219 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4220 | if (!buf) |
4221 | goto out; | |
4222 | ||
4223 | retval = -ESRCH; | |
4224 | pid = m->private; | |
4225 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4226 | if (!tsk) | |
4227 | goto out_free; | |
4228 | ||
4229 | retval = 0; | |
4230 | ||
4231 | mutex_lock(&cgroup_mutex); | |
4232 | ||
e5f6a860 | 4233 | for_each_active_root(root) { |
a424316c | 4234 | struct cgroup_subsys *ss; |
bd89aabc | 4235 | struct cgroup *cgrp; |
b85d2040 | 4236 | int ssid, count = 0; |
a424316c | 4237 | |
2c6ab6d2 | 4238 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 TH |
4239 | for_each_subsys(ss, ssid) |
4240 | if (root->subsys_mask & (1 << ssid)) | |
4241 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4242 | if (strlen(root->name)) |
4243 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4244 | root->name); | |
a424316c | 4245 | seq_putc(m, ':'); |
7717f7ba | 4246 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
4247 | path = cgroup_path(cgrp, buf, PATH_MAX); |
4248 | if (!path) { | |
4249 | retval = -ENAMETOOLONG; | |
a424316c | 4250 | goto out_unlock; |
e61734c5 TH |
4251 | } |
4252 | seq_puts(m, path); | |
a424316c PM |
4253 | seq_putc(m, '\n'); |
4254 | } | |
4255 | ||
4256 | out_unlock: | |
4257 | mutex_unlock(&cgroup_mutex); | |
4258 | put_task_struct(tsk); | |
4259 | out_free: | |
4260 | kfree(buf); | |
4261 | out: | |
4262 | return retval; | |
4263 | } | |
4264 | ||
a424316c PM |
4265 | /* Display information about each subsystem and each hierarchy */ |
4266 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4267 | { | |
30159ec7 | 4268 | struct cgroup_subsys *ss; |
a424316c | 4269 | int i; |
a424316c | 4270 | |
8bab8dde | 4271 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4272 | /* |
4273 | * ideally we don't want subsystems moving around while we do this. | |
4274 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4275 | * subsys/hierarchy state. | |
4276 | */ | |
a424316c | 4277 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4278 | |
4279 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4280 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4281 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 4282 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 4283 | |
a424316c PM |
4284 | mutex_unlock(&cgroup_mutex); |
4285 | return 0; | |
4286 | } | |
4287 | ||
4288 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4289 | { | |
9dce07f1 | 4290 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4291 | } |
4292 | ||
828c0950 | 4293 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4294 | .open = cgroupstats_open, |
4295 | .read = seq_read, | |
4296 | .llseek = seq_lseek, | |
4297 | .release = single_release, | |
4298 | }; | |
4299 | ||
b4f48b63 PM |
4300 | /** |
4301 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4302 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4303 | * |
4304 | * Description: A task inherits its parent's cgroup at fork(). | |
4305 | * | |
4306 | * A pointer to the shared css_set was automatically copied in | |
4307 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
9bb71308 TH |
4308 | * it was not made under the protection of RCU or cgroup_mutex, so |
4309 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
4310 | * have already changed current->cgroups, allowing the previously | |
4311 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
4312 | * |
4313 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4314 | * task, and the passed argument 'child' points to the child task. | |
4315 | */ | |
4316 | void cgroup_fork(struct task_struct *child) | |
4317 | { | |
9bb71308 | 4318 | task_lock(current); |
a8ad805c | 4319 | get_css_set(task_css_set(current)); |
817929ec | 4320 | child->cgroups = current->cgroups; |
9bb71308 | 4321 | task_unlock(current); |
817929ec | 4322 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4323 | } |
4324 | ||
817929ec | 4325 | /** |
a043e3b2 LZ |
4326 | * cgroup_post_fork - called on a new task after adding it to the task list |
4327 | * @child: the task in question | |
4328 | * | |
5edee61e TH |
4329 | * Adds the task to the list running through its css_set if necessary and |
4330 | * call the subsystem fork() callbacks. Has to be after the task is | |
4331 | * visible on the task list in case we race with the first call to | |
0942eeee | 4332 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4333 | * list. |
a043e3b2 | 4334 | */ |
817929ec PM |
4335 | void cgroup_post_fork(struct task_struct *child) |
4336 | { | |
30159ec7 | 4337 | struct cgroup_subsys *ss; |
5edee61e TH |
4338 | int i; |
4339 | ||
3ce3230a FW |
4340 | /* |
4341 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
4342 | * under the tasklist_lock and we read it here after we added the child | |
4343 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
4344 | * yet in the tasklist when we walked through it from | |
4345 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
4346 | * should be visible now due to the paired locking and barriers implied | |
4347 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
4348 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
4349 | * lock on fork. | |
4350 | */ | |
817929ec PM |
4351 | if (use_task_css_set_links) { |
4352 | write_lock(&css_set_lock); | |
d8783832 TH |
4353 | task_lock(child); |
4354 | if (list_empty(&child->cg_list)) | |
a8ad805c | 4355 | list_add(&child->cg_list, &task_css_set(child)->tasks); |
d8783832 | 4356 | task_unlock(child); |
817929ec PM |
4357 | write_unlock(&css_set_lock); |
4358 | } | |
5edee61e TH |
4359 | |
4360 | /* | |
4361 | * Call ss->fork(). This must happen after @child is linked on | |
4362 | * css_set; otherwise, @child might change state between ->fork() | |
4363 | * and addition to css_set. | |
4364 | */ | |
4365 | if (need_forkexit_callback) { | |
3ed80a62 | 4366 | for_each_subsys(ss, i) |
5edee61e TH |
4367 | if (ss->fork) |
4368 | ss->fork(child); | |
5edee61e | 4369 | } |
817929ec | 4370 | } |
5edee61e | 4371 | |
b4f48b63 PM |
4372 | /** |
4373 | * cgroup_exit - detach cgroup from exiting task | |
4374 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4375 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4376 | * |
4377 | * Description: Detach cgroup from @tsk and release it. | |
4378 | * | |
4379 | * Note that cgroups marked notify_on_release force every task in | |
4380 | * them to take the global cgroup_mutex mutex when exiting. | |
4381 | * This could impact scaling on very large systems. Be reluctant to | |
4382 | * use notify_on_release cgroups where very high task exit scaling | |
4383 | * is required on large systems. | |
4384 | * | |
4385 | * the_top_cgroup_hack: | |
4386 | * | |
4387 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4388 | * | |
4389 | * We call cgroup_exit() while the task is still competent to | |
4390 | * handle notify_on_release(), then leave the task attached to the | |
4391 | * root cgroup in each hierarchy for the remainder of its exit. | |
4392 | * | |
4393 | * To do this properly, we would increment the reference count on | |
4394 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4395 | * code we would add a second cgroup function call, to drop that | |
4396 | * reference. This would just create an unnecessary hot spot on | |
4397 | * the top_cgroup reference count, to no avail. | |
4398 | * | |
4399 | * Normally, holding a reference to a cgroup without bumping its | |
4400 | * count is unsafe. The cgroup could go away, or someone could | |
4401 | * attach us to a different cgroup, decrementing the count on | |
4402 | * the first cgroup that we never incremented. But in this case, | |
4403 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4404 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4405 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4406 | */ |
4407 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4408 | { | |
30159ec7 | 4409 | struct cgroup_subsys *ss; |
5abb8855 | 4410 | struct css_set *cset; |
d41d5a01 | 4411 | int i; |
817929ec PM |
4412 | |
4413 | /* | |
4414 | * Unlink from the css_set task list if necessary. | |
4415 | * Optimistically check cg_list before taking | |
4416 | * css_set_lock | |
4417 | */ | |
4418 | if (!list_empty(&tsk->cg_list)) { | |
4419 | write_lock(&css_set_lock); | |
4420 | if (!list_empty(&tsk->cg_list)) | |
8d258797 | 4421 | list_del_init(&tsk->cg_list); |
817929ec PM |
4422 | write_unlock(&css_set_lock); |
4423 | } | |
4424 | ||
b4f48b63 PM |
4425 | /* Reassign the task to the init_css_set. */ |
4426 | task_lock(tsk); | |
a8ad805c TH |
4427 | cset = task_css_set(tsk); |
4428 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 PZ |
4429 | |
4430 | if (run_callbacks && need_forkexit_callback) { | |
3ed80a62 TH |
4431 | /* see cgroup_post_fork() for details */ |
4432 | for_each_subsys(ss, i) { | |
d41d5a01 | 4433 | if (ss->exit) { |
eb95419b TH |
4434 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4435 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4436 | |
eb95419b | 4437 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4438 | } |
4439 | } | |
4440 | } | |
b4f48b63 | 4441 | task_unlock(tsk); |
d41d5a01 | 4442 | |
5abb8855 | 4443 | put_css_set_taskexit(cset); |
b4f48b63 | 4444 | } |
697f4161 | 4445 | |
bd89aabc | 4446 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4447 | { |
f50daa70 | 4448 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4449 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4450 | /* |
4451 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4452 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4453 | * it now |
4454 | */ | |
81a6a5cd | 4455 | int need_schedule_work = 0; |
f50daa70 | 4456 | |
cdcc136f | 4457 | raw_spin_lock(&release_list_lock); |
54766d4a | 4458 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4459 | list_empty(&cgrp->release_list)) { |
4460 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4461 | need_schedule_work = 1; |
4462 | } | |
cdcc136f | 4463 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4464 | if (need_schedule_work) |
4465 | schedule_work(&release_agent_work); | |
4466 | } | |
4467 | } | |
4468 | ||
81a6a5cd PM |
4469 | /* |
4470 | * Notify userspace when a cgroup is released, by running the | |
4471 | * configured release agent with the name of the cgroup (path | |
4472 | * relative to the root of cgroup file system) as the argument. | |
4473 | * | |
4474 | * Most likely, this user command will try to rmdir this cgroup. | |
4475 | * | |
4476 | * This races with the possibility that some other task will be | |
4477 | * attached to this cgroup before it is removed, or that some other | |
4478 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4479 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4480 | * unused, and this cgroup will be reprieved from its death sentence, | |
4481 | * to continue to serve a useful existence. Next time it's released, | |
4482 | * we will get notified again, if it still has 'notify_on_release' set. | |
4483 | * | |
4484 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4485 | * means only wait until the task is successfully execve()'d. The | |
4486 | * separate release agent task is forked by call_usermodehelper(), | |
4487 | * then control in this thread returns here, without waiting for the | |
4488 | * release agent task. We don't bother to wait because the caller of | |
4489 | * this routine has no use for the exit status of the release agent | |
4490 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4491 | */ |
81a6a5cd PM |
4492 | static void cgroup_release_agent(struct work_struct *work) |
4493 | { | |
4494 | BUG_ON(work != &release_agent_work); | |
4495 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4496 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4497 | while (!list_empty(&release_list)) { |
4498 | char *argv[3], *envp[3]; | |
4499 | int i; | |
e61734c5 | 4500 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 4501 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4502 | struct cgroup, |
4503 | release_list); | |
bd89aabc | 4504 | list_del_init(&cgrp->release_list); |
cdcc136f | 4505 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 4506 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
4507 | if (!pathbuf) |
4508 | goto continue_free; | |
e61734c5 TH |
4509 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
4510 | if (!path) | |
e788e066 PM |
4511 | goto continue_free; |
4512 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4513 | if (!agentbuf) | |
4514 | goto continue_free; | |
81a6a5cd PM |
4515 | |
4516 | i = 0; | |
e788e066 | 4517 | argv[i++] = agentbuf; |
e61734c5 | 4518 | argv[i++] = path; |
81a6a5cd PM |
4519 | argv[i] = NULL; |
4520 | ||
4521 | i = 0; | |
4522 | /* minimal command environment */ | |
4523 | envp[i++] = "HOME=/"; | |
4524 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4525 | envp[i] = NULL; | |
4526 | ||
4527 | /* Drop the lock while we invoke the usermode helper, | |
4528 | * since the exec could involve hitting disk and hence | |
4529 | * be a slow process */ | |
4530 | mutex_unlock(&cgroup_mutex); | |
4531 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4532 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4533 | continue_free: |
4534 | kfree(pathbuf); | |
4535 | kfree(agentbuf); | |
cdcc136f | 4536 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4537 | } |
cdcc136f | 4538 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4539 | mutex_unlock(&cgroup_mutex); |
4540 | } | |
8bab8dde PM |
4541 | |
4542 | static int __init cgroup_disable(char *str) | |
4543 | { | |
30159ec7 | 4544 | struct cgroup_subsys *ss; |
8bab8dde | 4545 | char *token; |
30159ec7 | 4546 | int i; |
8bab8dde PM |
4547 | |
4548 | while ((token = strsep(&str, ",")) != NULL) { | |
4549 | if (!*token) | |
4550 | continue; | |
be45c900 | 4551 | |
3ed80a62 | 4552 | for_each_subsys(ss, i) { |
8bab8dde PM |
4553 | if (!strcmp(token, ss->name)) { |
4554 | ss->disabled = 1; | |
4555 | printk(KERN_INFO "Disabling %s control group" | |
4556 | " subsystem\n", ss->name); | |
4557 | break; | |
4558 | } | |
4559 | } | |
4560 | } | |
4561 | return 1; | |
4562 | } | |
4563 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 4564 | |
b77d7b60 | 4565 | /** |
5a17f543 | 4566 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
4567 | * @dentry: directory dentry of interest |
4568 | * @ss: subsystem of interest | |
b77d7b60 | 4569 | * |
5a17f543 TH |
4570 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
4571 | * to get the corresponding css and return it. If such css doesn't exist | |
4572 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 4573 | */ |
5a17f543 TH |
4574 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
4575 | struct cgroup_subsys *ss) | |
e5d1367f | 4576 | { |
2bd59d48 TH |
4577 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
4578 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 4579 | struct cgroup *cgrp; |
b77d7b60 | 4580 | |
35cf0836 | 4581 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
4582 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
4583 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
4584 | return ERR_PTR(-EBADF); |
4585 | ||
5a17f543 TH |
4586 | rcu_read_lock(); |
4587 | ||
2bd59d48 TH |
4588 | /* |
4589 | * This path doesn't originate from kernfs and @kn could already | |
4590 | * have been or be removed at any point. @kn->priv is RCU | |
4591 | * protected for this access. See destroy_locked() for details. | |
4592 | */ | |
4593 | cgrp = rcu_dereference(kn->priv); | |
4594 | if (cgrp) | |
4595 | css = cgroup_css(cgrp, ss); | |
5a17f543 TH |
4596 | |
4597 | if (!css || !css_tryget(css)) | |
4598 | css = ERR_PTR(-ENOENT); | |
4599 | ||
4600 | rcu_read_unlock(); | |
4601 | return css; | |
e5d1367f | 4602 | } |
e5d1367f | 4603 | |
1cb650b9 LZ |
4604 | /** |
4605 | * css_from_id - lookup css by id | |
4606 | * @id: the cgroup id | |
4607 | * @ss: cgroup subsys to be looked into | |
4608 | * | |
4609 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
4610 | * Should be called under rcu_read_lock(). | |
4611 | */ | |
4612 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
4613 | { | |
4614 | struct cgroup *cgrp; | |
4615 | ||
ace2bee8 | 4616 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
4617 | |
4618 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
4619 | if (cgrp) | |
d1625964 | 4620 | return cgroup_css(cgrp, ss); |
1cb650b9 | 4621 | return NULL; |
e5d1367f SE |
4622 | } |
4623 | ||
fe693435 | 4624 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
4625 | static struct cgroup_subsys_state * |
4626 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
4627 | { |
4628 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
4629 | ||
4630 | if (!css) | |
4631 | return ERR_PTR(-ENOMEM); | |
4632 | ||
4633 | return css; | |
4634 | } | |
4635 | ||
eb95419b | 4636 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 4637 | { |
eb95419b | 4638 | kfree(css); |
fe693435 PM |
4639 | } |
4640 | ||
182446d0 TH |
4641 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
4642 | struct cftype *cft) | |
fe693435 | 4643 | { |
182446d0 | 4644 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
4645 | } |
4646 | ||
182446d0 TH |
4647 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
4648 | struct cftype *cft) | |
fe693435 PM |
4649 | { |
4650 | return (u64)(unsigned long)current->cgroups; | |
4651 | } | |
4652 | ||
182446d0 | 4653 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 4654 | struct cftype *cft) |
fe693435 PM |
4655 | { |
4656 | u64 count; | |
4657 | ||
4658 | rcu_read_lock(); | |
a8ad805c | 4659 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
4660 | rcu_read_unlock(); |
4661 | return count; | |
4662 | } | |
4663 | ||
2da8ca82 | 4664 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4665 | { |
69d0206c | 4666 | struct cgrp_cset_link *link; |
5abb8855 | 4667 | struct css_set *cset; |
e61734c5 TH |
4668 | char *name_buf; |
4669 | ||
4670 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
4671 | if (!name_buf) | |
4672 | return -ENOMEM; | |
7717f7ba PM |
4673 | |
4674 | read_lock(&css_set_lock); | |
4675 | rcu_read_lock(); | |
5abb8855 | 4676 | cset = rcu_dereference(current->cgroups); |
69d0206c | 4677 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 4678 | struct cgroup *c = link->cgrp; |
59f5296b TH |
4679 | const char *name = "?"; |
4680 | ||
e61734c5 TH |
4681 | if (c != cgroup_dummy_top) { |
4682 | cgroup_name(c, name_buf, NAME_MAX + 1); | |
4683 | name = name_buf; | |
4684 | } | |
7717f7ba | 4685 | |
2c6ab6d2 PM |
4686 | seq_printf(seq, "Root %d group %s\n", |
4687 | c->root->hierarchy_id, name); | |
7717f7ba PM |
4688 | } |
4689 | rcu_read_unlock(); | |
4690 | read_unlock(&css_set_lock); | |
e61734c5 | 4691 | kfree(name_buf); |
7717f7ba PM |
4692 | return 0; |
4693 | } | |
4694 | ||
4695 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 4696 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4697 | { |
2da8ca82 | 4698 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 4699 | struct cgrp_cset_link *link; |
7717f7ba PM |
4700 | |
4701 | read_lock(&css_set_lock); | |
182446d0 | 4702 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 4703 | struct css_set *cset = link->cset; |
7717f7ba PM |
4704 | struct task_struct *task; |
4705 | int count = 0; | |
5abb8855 TH |
4706 | seq_printf(seq, "css_set %p\n", cset); |
4707 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
7717f7ba PM |
4708 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { |
4709 | seq_puts(seq, " ...\n"); | |
4710 | break; | |
4711 | } else { | |
4712 | seq_printf(seq, " task %d\n", | |
4713 | task_pid_vnr(task)); | |
4714 | } | |
4715 | } | |
4716 | } | |
4717 | read_unlock(&css_set_lock); | |
4718 | return 0; | |
4719 | } | |
4720 | ||
182446d0 | 4721 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 4722 | { |
182446d0 | 4723 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
4724 | } |
4725 | ||
4726 | static struct cftype debug_files[] = { | |
fe693435 PM |
4727 | { |
4728 | .name = "taskcount", | |
4729 | .read_u64 = debug_taskcount_read, | |
4730 | }, | |
4731 | ||
4732 | { | |
4733 | .name = "current_css_set", | |
4734 | .read_u64 = current_css_set_read, | |
4735 | }, | |
4736 | ||
4737 | { | |
4738 | .name = "current_css_set_refcount", | |
4739 | .read_u64 = current_css_set_refcount_read, | |
4740 | }, | |
4741 | ||
7717f7ba PM |
4742 | { |
4743 | .name = "current_css_set_cg_links", | |
2da8ca82 | 4744 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
4745 | }, |
4746 | ||
4747 | { | |
4748 | .name = "cgroup_css_links", | |
2da8ca82 | 4749 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
4750 | }, |
4751 | ||
fe693435 PM |
4752 | { |
4753 | .name = "releasable", | |
4754 | .read_u64 = releasable_read, | |
4755 | }, | |
fe693435 | 4756 | |
4baf6e33 TH |
4757 | { } /* terminate */ |
4758 | }; | |
fe693435 | 4759 | |
073219e9 | 4760 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
4761 | .css_alloc = debug_css_alloc, |
4762 | .css_free = debug_css_free, | |
4baf6e33 | 4763 | .base_cftypes = debug_files, |
fe693435 PM |
4764 | }; |
4765 | #endif /* CONFIG_CGROUP_DEBUG */ |