]>
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 | |
65dff759 LZ |
148 | static struct cgroup_name root_cgroup_name = { .name = "/" }; |
149 | ||
794611a1 LZ |
150 | /* |
151 | * Assign a monotonically increasing serial number to cgroups. It | |
152 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
153 | * numbers. Also, as cgroups are always appended to the parent's | |
154 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
155 | * the ascending serial number order on the list. Protected by |
156 | * cgroup_mutex. | |
794611a1 | 157 | */ |
00356bd5 | 158 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 159 | |
ddbcc7e8 | 160 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
161 | * check for fork/exit handlers to call. This avoids us having to do |
162 | * extra work in the fork/exit path if none of the subsystems need to | |
163 | * be called. | |
ddbcc7e8 | 164 | */ |
8947f9d5 | 165 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 166 | |
628f7cd4 TH |
167 | static struct cftype cgroup_base_files[]; |
168 | ||
59f5296b | 169 | static void cgroup_put(struct cgroup *cgrp); |
f2e85d57 TH |
170 | static int rebind_subsystems(struct cgroupfs_root *root, |
171 | unsigned long added_mask, unsigned removed_mask); | |
f20104de | 172 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 173 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
2bb566cb TH |
174 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
175 | bool is_add); | |
b1a21367 | 176 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 177 | |
95109b62 TH |
178 | /** |
179 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
180 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 181 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 182 | * |
ca8bdcaf TH |
183 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
184 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
185 | * the caller is responsible for pinning the returned css if it wants to | |
186 | * keep accessing it outside the said locks. This function may return | |
187 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
188 | */ |
189 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 190 | struct cgroup_subsys *ss) |
95109b62 | 191 | { |
ca8bdcaf | 192 | if (ss) |
aec25020 | 193 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 TH |
194 | lockdep_is_held(&cgroup_tree_mutex) || |
195 | lockdep_is_held(&cgroup_mutex)); | |
ca8bdcaf TH |
196 | else |
197 | return &cgrp->dummy_css; | |
95109b62 | 198 | } |
42809dd4 | 199 | |
ddbcc7e8 | 200 | /* convenient tests for these bits */ |
54766d4a | 201 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 202 | { |
54766d4a | 203 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
204 | } |
205 | ||
59f5296b TH |
206 | struct cgroup_subsys_state *seq_css(struct seq_file *seq) |
207 | { | |
2bd59d48 TH |
208 | struct kernfs_open_file *of = seq->private; |
209 | struct cgroup *cgrp = of->kn->parent->priv; | |
210 | struct cftype *cft = seq_cft(seq); | |
211 | ||
212 | /* | |
213 | * This is open and unprotected implementation of cgroup_css(). | |
214 | * seq_css() is only called from a kernfs file operation which has | |
215 | * an active reference on the file. Because all the subsystem | |
216 | * files are drained before a css is disassociated with a cgroup, | |
217 | * the matching css from the cgroup's subsys table is guaranteed to | |
218 | * be and stay valid until the enclosing operation is complete. | |
219 | */ | |
220 | if (cft->ss) | |
221 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
222 | else | |
223 | return &cgrp->dummy_css; | |
59f5296b TH |
224 | } |
225 | EXPORT_SYMBOL_GPL(seq_css); | |
226 | ||
78574cf9 LZ |
227 | /** |
228 | * cgroup_is_descendant - test ancestry | |
229 | * @cgrp: the cgroup to be tested | |
230 | * @ancestor: possible ancestor of @cgrp | |
231 | * | |
232 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
233 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
234 | * and @ancestor are accessible. | |
235 | */ | |
236 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
237 | { | |
238 | while (cgrp) { | |
239 | if (cgrp == ancestor) | |
240 | return true; | |
241 | cgrp = cgrp->parent; | |
242 | } | |
243 | return false; | |
244 | } | |
245 | EXPORT_SYMBOL_GPL(cgroup_is_descendant); | |
ddbcc7e8 | 246 | |
e9685a03 | 247 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
248 | { |
249 | const int bits = | |
bd89aabc PM |
250 | (1 << CGRP_RELEASABLE) | |
251 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
252 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
253 | } |
254 | ||
e9685a03 | 255 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 256 | { |
bd89aabc | 257 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
258 | } |
259 | ||
1c6727af TH |
260 | /** |
261 | * for_each_css - iterate all css's of a cgroup | |
262 | * @css: the iteration cursor | |
263 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
264 | * @cgrp: the target cgroup to iterate css's of | |
265 | * | |
266 | * Should be called under cgroup_mutex. | |
267 | */ | |
268 | #define for_each_css(css, ssid, cgrp) \ | |
269 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
270 | if (!((css) = rcu_dereference_check( \ | |
271 | (cgrp)->subsys[(ssid)], \ | |
ace2bee8 | 272 | lockdep_is_held(&cgroup_tree_mutex) || \ |
1c6727af TH |
273 | lockdep_is_held(&cgroup_mutex)))) { } \ |
274 | else | |
275 | ||
30159ec7 | 276 | /** |
3ed80a62 | 277 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 278 | * @ss: the iteration cursor |
780cd8b3 | 279 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 280 | */ |
780cd8b3 | 281 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
282 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
283 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 284 | |
5549c497 TH |
285 | /* iterate across the active hierarchies */ |
286 | #define for_each_active_root(root) \ | |
287 | list_for_each_entry((root), &cgroup_roots, root_list) | |
ddbcc7e8 | 288 | |
7ae1bad9 TH |
289 | /** |
290 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
291 | * @cgrp: the cgroup to be checked for liveness | |
292 | * | |
47cfcd09 TH |
293 | * On success, returns true; the mutex should be later unlocked. On |
294 | * failure returns false with no lock held. | |
7ae1bad9 | 295 | */ |
b9777cf8 | 296 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
297 | { |
298 | mutex_lock(&cgroup_mutex); | |
54766d4a | 299 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
300 | mutex_unlock(&cgroup_mutex); |
301 | return false; | |
302 | } | |
303 | return true; | |
304 | } | |
7ae1bad9 | 305 | |
81a6a5cd PM |
306 | /* the list of cgroups eligible for automatic release. Protected by |
307 | * release_list_lock */ | |
308 | static LIST_HEAD(release_list); | |
cdcc136f | 309 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
310 | static void cgroup_release_agent(struct work_struct *work); |
311 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 312 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 313 | |
69d0206c TH |
314 | /* |
315 | * A cgroup can be associated with multiple css_sets as different tasks may | |
316 | * belong to different cgroups on different hierarchies. In the other | |
317 | * direction, a css_set is naturally associated with multiple cgroups. | |
318 | * This M:N relationship is represented by the following link structure | |
319 | * which exists for each association and allows traversing the associations | |
320 | * from both sides. | |
321 | */ | |
322 | struct cgrp_cset_link { | |
323 | /* the cgroup and css_set this link associates */ | |
324 | struct cgroup *cgrp; | |
325 | struct css_set *cset; | |
326 | ||
327 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
328 | struct list_head cset_link; | |
329 | ||
330 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
331 | struct list_head cgrp_link; | |
817929ec PM |
332 | }; |
333 | ||
334 | /* The default css_set - used by init and its children prior to any | |
335 | * hierarchies being mounted. It contains a pointer to the root state | |
336 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
337 | * reference-counted, to improve performance when child cgroups | |
338 | * haven't been created. | |
339 | */ | |
340 | ||
341 | static struct css_set init_css_set; | |
69d0206c | 342 | static struct cgrp_cset_link init_cgrp_cset_link; |
817929ec | 343 | |
0942eeee TH |
344 | /* |
345 | * css_set_lock protects the list of css_set objects, and the chain of | |
346 | * tasks off each css_set. Nests outside task->alloc_lock due to | |
72ec7029 | 347 | * css_task_iter_start(). |
0942eeee | 348 | */ |
817929ec PM |
349 | static DEFINE_RWLOCK(css_set_lock); |
350 | static int css_set_count; | |
351 | ||
7717f7ba PM |
352 | /* |
353 | * hash table for cgroup groups. This improves the performance to find | |
354 | * an existing css_set. This hash doesn't (currently) take into | |
355 | * account cgroups in empty hierarchies. | |
356 | */ | |
472b1053 | 357 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 358 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 359 | |
0ac801fe | 360 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 361 | { |
0ac801fe | 362 | unsigned long key = 0UL; |
30159ec7 TH |
363 | struct cgroup_subsys *ss; |
364 | int i; | |
472b1053 | 365 | |
30159ec7 | 366 | for_each_subsys(ss, i) |
0ac801fe LZ |
367 | key += (unsigned long)css[i]; |
368 | key = (key >> 16) ^ key; | |
472b1053 | 369 | |
0ac801fe | 370 | return key; |
472b1053 LZ |
371 | } |
372 | ||
0942eeee TH |
373 | /* |
374 | * We don't maintain the lists running through each css_set to its task | |
72ec7029 TH |
375 | * until after the first call to css_task_iter_start(). This reduces the |
376 | * fork()/exit() overhead for people who have cgroups compiled into their | |
377 | * kernel but not actually in use. | |
0942eeee | 378 | */ |
8947f9d5 | 379 | static int use_task_css_set_links __read_mostly; |
817929ec | 380 | |
5abb8855 | 381 | static void __put_css_set(struct css_set *cset, int taskexit) |
b4f48b63 | 382 | { |
69d0206c | 383 | struct cgrp_cset_link *link, *tmp_link; |
5abb8855 | 384 | |
146aa1bd LJ |
385 | /* |
386 | * Ensure that the refcount doesn't hit zero while any readers | |
387 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
388 | * rwlock | |
389 | */ | |
5abb8855 | 390 | if (atomic_add_unless(&cset->refcount, -1, 1)) |
146aa1bd LJ |
391 | return; |
392 | write_lock(&css_set_lock); | |
5abb8855 | 393 | if (!atomic_dec_and_test(&cset->refcount)) { |
146aa1bd LJ |
394 | write_unlock(&css_set_lock); |
395 | return; | |
396 | } | |
81a6a5cd | 397 | |
2c6ab6d2 | 398 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
5abb8855 | 399 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
400 | css_set_count--; |
401 | ||
69d0206c | 402 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 403 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 404 | |
69d0206c TH |
405 | list_del(&link->cset_link); |
406 | list_del(&link->cgrp_link); | |
71b5707e | 407 | |
ddd69148 | 408 | /* @cgrp can't go away while we're holding css_set_lock */ |
6f3d828f | 409 | if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { |
81a6a5cd | 410 | if (taskexit) |
bd89aabc PM |
411 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
412 | check_for_release(cgrp); | |
81a6a5cd | 413 | } |
2c6ab6d2 PM |
414 | |
415 | kfree(link); | |
81a6a5cd | 416 | } |
2c6ab6d2 PM |
417 | |
418 | write_unlock(&css_set_lock); | |
5abb8855 | 419 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
420 | } |
421 | ||
817929ec PM |
422 | /* |
423 | * refcounted get/put for css_set objects | |
424 | */ | |
5abb8855 | 425 | static inline void get_css_set(struct css_set *cset) |
817929ec | 426 | { |
5abb8855 | 427 | atomic_inc(&cset->refcount); |
817929ec PM |
428 | } |
429 | ||
5abb8855 | 430 | static inline void put_css_set(struct css_set *cset) |
817929ec | 431 | { |
5abb8855 | 432 | __put_css_set(cset, 0); |
817929ec PM |
433 | } |
434 | ||
5abb8855 | 435 | static inline void put_css_set_taskexit(struct css_set *cset) |
81a6a5cd | 436 | { |
5abb8855 | 437 | __put_css_set(cset, 1); |
81a6a5cd PM |
438 | } |
439 | ||
b326f9d0 | 440 | /** |
7717f7ba | 441 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
442 | * @cset: candidate css_set being tested |
443 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
444 | * @new_cgrp: cgroup that's being entered by the task |
445 | * @template: desired set of css pointers in css_set (pre-calculated) | |
446 | * | |
6f4b7e63 | 447 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
448 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
449 | */ | |
5abb8855 TH |
450 | static bool compare_css_sets(struct css_set *cset, |
451 | struct css_set *old_cset, | |
7717f7ba PM |
452 | struct cgroup *new_cgrp, |
453 | struct cgroup_subsys_state *template[]) | |
454 | { | |
455 | struct list_head *l1, *l2; | |
456 | ||
5abb8855 | 457 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) { |
7717f7ba PM |
458 | /* Not all subsystems matched */ |
459 | return false; | |
460 | } | |
461 | ||
462 | /* | |
463 | * Compare cgroup pointers in order to distinguish between | |
464 | * different cgroups in heirarchies with no subsystems. We | |
465 | * could get by with just this check alone (and skip the | |
466 | * memcmp above) but on most setups the memcmp check will | |
467 | * avoid the need for this more expensive check on almost all | |
468 | * candidates. | |
469 | */ | |
470 | ||
69d0206c TH |
471 | l1 = &cset->cgrp_links; |
472 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 473 | while (1) { |
69d0206c | 474 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 475 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
476 | |
477 | l1 = l1->next; | |
478 | l2 = l2->next; | |
479 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
480 | if (l1 == &cset->cgrp_links) { |
481 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
482 | break; |
483 | } else { | |
69d0206c | 484 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
485 | } |
486 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
487 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
488 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
489 | cgrp1 = link1->cgrp; | |
490 | cgrp2 = link2->cgrp; | |
7717f7ba | 491 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 492 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
493 | |
494 | /* | |
495 | * If this hierarchy is the hierarchy of the cgroup | |
496 | * that's changing, then we need to check that this | |
497 | * css_set points to the new cgroup; if it's any other | |
498 | * hierarchy, then this css_set should point to the | |
499 | * same cgroup as the old css_set. | |
500 | */ | |
5abb8855 TH |
501 | if (cgrp1->root == new_cgrp->root) { |
502 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
503 | return false; |
504 | } else { | |
5abb8855 | 505 | if (cgrp1 != cgrp2) |
7717f7ba PM |
506 | return false; |
507 | } | |
508 | } | |
509 | return true; | |
510 | } | |
511 | ||
b326f9d0 TH |
512 | /** |
513 | * find_existing_css_set - init css array and find the matching css_set | |
514 | * @old_cset: the css_set that we're using before the cgroup transition | |
515 | * @cgrp: the cgroup that we're moving into | |
516 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 517 | */ |
5abb8855 TH |
518 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
519 | struct cgroup *cgrp, | |
520 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 521 | { |
bd89aabc | 522 | struct cgroupfs_root *root = cgrp->root; |
30159ec7 | 523 | struct cgroup_subsys *ss; |
5abb8855 | 524 | struct css_set *cset; |
0ac801fe | 525 | unsigned long key; |
b326f9d0 | 526 | int i; |
817929ec | 527 | |
aae8aab4 BB |
528 | /* |
529 | * Build the set of subsystem state objects that we want to see in the | |
530 | * new css_set. while subsystems can change globally, the entries here | |
531 | * won't change, so no need for locking. | |
532 | */ | |
30159ec7 | 533 | for_each_subsys(ss, i) { |
a1a71b45 | 534 | if (root->subsys_mask & (1UL << i)) { |
817929ec PM |
535 | /* Subsystem is in this hierarchy. So we want |
536 | * the subsystem state from the new | |
537 | * cgroup */ | |
ca8bdcaf | 538 | template[i] = cgroup_css(cgrp, ss); |
817929ec PM |
539 | } else { |
540 | /* Subsystem is not in this hierarchy, so we | |
541 | * don't want to change the subsystem state */ | |
5abb8855 | 542 | template[i] = old_cset->subsys[i]; |
817929ec PM |
543 | } |
544 | } | |
545 | ||
0ac801fe | 546 | key = css_set_hash(template); |
5abb8855 TH |
547 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
548 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
549 | continue; |
550 | ||
551 | /* This css_set matches what we need */ | |
5abb8855 | 552 | return cset; |
472b1053 | 553 | } |
817929ec PM |
554 | |
555 | /* No existing cgroup group matched */ | |
556 | return NULL; | |
557 | } | |
558 | ||
69d0206c | 559 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 560 | { |
69d0206c | 561 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 562 | |
69d0206c TH |
563 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
564 | list_del(&link->cset_link); | |
36553434 LZ |
565 | kfree(link); |
566 | } | |
567 | } | |
568 | ||
69d0206c TH |
569 | /** |
570 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
571 | * @count: the number of links to allocate | |
572 | * @tmp_links: list_head the allocated links are put on | |
573 | * | |
574 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
575 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 576 | */ |
69d0206c | 577 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 578 | { |
69d0206c | 579 | struct cgrp_cset_link *link; |
817929ec | 580 | int i; |
69d0206c TH |
581 | |
582 | INIT_LIST_HEAD(tmp_links); | |
583 | ||
817929ec | 584 | for (i = 0; i < count; i++) { |
f4f4be2b | 585 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 586 | if (!link) { |
69d0206c | 587 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
588 | return -ENOMEM; |
589 | } | |
69d0206c | 590 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
591 | } |
592 | return 0; | |
593 | } | |
594 | ||
c12f65d4 LZ |
595 | /** |
596 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 597 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 598 | * @cset: the css_set to be linked |
c12f65d4 LZ |
599 | * @cgrp: the destination cgroup |
600 | */ | |
69d0206c TH |
601 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
602 | struct cgroup *cgrp) | |
c12f65d4 | 603 | { |
69d0206c | 604 | struct cgrp_cset_link *link; |
c12f65d4 | 605 | |
69d0206c TH |
606 | BUG_ON(list_empty(tmp_links)); |
607 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); | |
608 | link->cset = cset; | |
7717f7ba | 609 | link->cgrp = cgrp; |
69d0206c | 610 | list_move(&link->cset_link, &cgrp->cset_links); |
7717f7ba PM |
611 | /* |
612 | * Always add links to the tail of the list so that the list | |
613 | * is sorted by order of hierarchy creation | |
614 | */ | |
69d0206c | 615 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
616 | } |
617 | ||
b326f9d0 TH |
618 | /** |
619 | * find_css_set - return a new css_set with one cgroup updated | |
620 | * @old_cset: the baseline css_set | |
621 | * @cgrp: the cgroup to be updated | |
622 | * | |
623 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
624 | * substituted into the appropriate hierarchy. | |
817929ec | 625 | */ |
5abb8855 TH |
626 | static struct css_set *find_css_set(struct css_set *old_cset, |
627 | struct cgroup *cgrp) | |
817929ec | 628 | { |
b326f9d0 | 629 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 630 | struct css_set *cset; |
69d0206c TH |
631 | struct list_head tmp_links; |
632 | struct cgrp_cset_link *link; | |
0ac801fe | 633 | unsigned long key; |
472b1053 | 634 | |
b326f9d0 TH |
635 | lockdep_assert_held(&cgroup_mutex); |
636 | ||
817929ec PM |
637 | /* First see if we already have a cgroup group that matches |
638 | * the desired set */ | |
7e9abd89 | 639 | read_lock(&css_set_lock); |
5abb8855 TH |
640 | cset = find_existing_css_set(old_cset, cgrp, template); |
641 | if (cset) | |
642 | get_css_set(cset); | |
7e9abd89 | 643 | read_unlock(&css_set_lock); |
817929ec | 644 | |
5abb8855 TH |
645 | if (cset) |
646 | return cset; | |
817929ec | 647 | |
f4f4be2b | 648 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 649 | if (!cset) |
817929ec PM |
650 | return NULL; |
651 | ||
69d0206c | 652 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 653 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 654 | kfree(cset); |
817929ec PM |
655 | return NULL; |
656 | } | |
657 | ||
5abb8855 | 658 | atomic_set(&cset->refcount, 1); |
69d0206c | 659 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 TH |
660 | INIT_LIST_HEAD(&cset->tasks); |
661 | INIT_HLIST_NODE(&cset->hlist); | |
817929ec PM |
662 | |
663 | /* Copy the set of subsystem state objects generated in | |
664 | * find_existing_css_set() */ | |
5abb8855 | 665 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec PM |
666 | |
667 | write_lock(&css_set_lock); | |
668 | /* Add reference counts and links from the new css_set. */ | |
69d0206c | 669 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 670 | struct cgroup *c = link->cgrp; |
69d0206c | 671 | |
7717f7ba PM |
672 | if (c->root == cgrp->root) |
673 | c = cgrp; | |
69d0206c | 674 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 675 | } |
817929ec | 676 | |
69d0206c | 677 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 678 | |
817929ec | 679 | css_set_count++; |
472b1053 LZ |
680 | |
681 | /* Add this cgroup group to the hash table */ | |
5abb8855 TH |
682 | key = css_set_hash(cset->subsys); |
683 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 684 | |
817929ec PM |
685 | write_unlock(&css_set_lock); |
686 | ||
5abb8855 | 687 | return cset; |
b4f48b63 PM |
688 | } |
689 | ||
2bd59d48 TH |
690 | static struct cgroupfs_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
691 | { | |
692 | struct cgroup *top_cgrp = kf_root->kn->priv; | |
693 | ||
694 | return top_cgrp->root; | |
695 | } | |
696 | ||
f2e85d57 TH |
697 | static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end) |
698 | { | |
699 | int id; | |
700 | ||
701 | lockdep_assert_held(&cgroup_mutex); | |
702 | ||
703 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end, | |
704 | GFP_KERNEL); | |
705 | if (id < 0) | |
706 | return id; | |
707 | ||
708 | root->hierarchy_id = id; | |
709 | return 0; | |
710 | } | |
711 | ||
712 | static void cgroup_exit_root_id(struct cgroupfs_root *root) | |
713 | { | |
714 | lockdep_assert_held(&cgroup_mutex); | |
715 | ||
716 | if (root->hierarchy_id) { | |
717 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
718 | root->hierarchy_id = 0; | |
719 | } | |
720 | } | |
721 | ||
722 | static void cgroup_free_root(struct cgroupfs_root *root) | |
723 | { | |
724 | if (root) { | |
725 | /* hierarhcy ID shoulid already have been released */ | |
726 | WARN_ON_ONCE(root->hierarchy_id); | |
727 | ||
728 | idr_destroy(&root->cgroup_idr); | |
729 | kfree(root); | |
730 | } | |
731 | } | |
732 | ||
59f5296b TH |
733 | static void cgroup_get_root(struct cgroupfs_root *root) |
734 | { | |
2bd59d48 TH |
735 | /* |
736 | * The caller must ensure that @root is alive, which can be | |
737 | * achieved by holding a ref on one of the member cgroups or | |
738 | * following a registered reference to @root while holding | |
739 | * cgroup_tree_mutex. | |
740 | */ | |
741 | WARN_ON_ONCE(atomic_read(&root->refcnt) <= 0); | |
742 | atomic_inc(&root->refcnt); | |
59f5296b TH |
743 | } |
744 | ||
745 | static void cgroup_put_root(struct cgroupfs_root *root) | |
746 | { | |
f2e85d57 TH |
747 | struct cgroup *cgrp = &root->top_cgroup; |
748 | struct cgrp_cset_link *link, *tmp_link; | |
749 | int ret; | |
750 | ||
2bd59d48 TH |
751 | /* |
752 | * @root's refcnt reaching zero and its deregistration should be | |
753 | * atomic w.r.t. cgroup_tree_mutex. This ensures that | |
754 | * cgroup_get_root() is safe to invoke if @root is registered. | |
755 | */ | |
756 | mutex_lock(&cgroup_tree_mutex); | |
757 | if (!atomic_dec_and_test(&root->refcnt)) { | |
758 | mutex_unlock(&cgroup_tree_mutex); | |
759 | return; | |
760 | } | |
761 | mutex_lock(&cgroup_mutex); | |
f2e85d57 TH |
762 | |
763 | BUG_ON(root->number_of_cgroups != 1); | |
764 | BUG_ON(!list_empty(&cgrp->children)); | |
765 | ||
f2e85d57 TH |
766 | /* Rebind all subsystems back to the default hierarchy */ |
767 | if (root->flags & CGRP_ROOT_SUBSYS_BOUND) { | |
768 | ret = rebind_subsystems(root, 0, root->subsys_mask); | |
769 | /* Shouldn't be able to fail ... */ | |
770 | BUG_ON(ret); | |
771 | } | |
772 | ||
773 | /* | |
774 | * Release all the links from cset_links to this hierarchy's | |
775 | * root cgroup | |
776 | */ | |
777 | write_lock(&css_set_lock); | |
778 | ||
779 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
780 | list_del(&link->cset_link); | |
781 | list_del(&link->cgrp_link); | |
782 | kfree(link); | |
783 | } | |
784 | write_unlock(&css_set_lock); | |
785 | ||
786 | if (!list_empty(&root->root_list)) { | |
787 | list_del(&root->root_list); | |
788 | cgroup_root_count--; | |
789 | } | |
790 | ||
791 | cgroup_exit_root_id(root); | |
792 | ||
793 | mutex_unlock(&cgroup_mutex); | |
794 | mutex_unlock(&cgroup_tree_mutex); | |
f2e85d57 | 795 | |
2bd59d48 | 796 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
797 | cgroup_free_root(root); |
798 | } | |
799 | ||
7717f7ba PM |
800 | /* |
801 | * Return the cgroup for "task" from the given hierarchy. Must be | |
802 | * called with cgroup_mutex held. | |
803 | */ | |
804 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
805 | struct cgroupfs_root *root) | |
806 | { | |
5abb8855 | 807 | struct css_set *cset; |
7717f7ba PM |
808 | struct cgroup *res = NULL; |
809 | ||
810 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); | |
811 | read_lock(&css_set_lock); | |
812 | /* | |
813 | * No need to lock the task - since we hold cgroup_mutex the | |
814 | * task can't change groups, so the only thing that can happen | |
815 | * is that it exits and its css is set back to init_css_set. | |
816 | */ | |
a8ad805c | 817 | cset = task_css_set(task); |
5abb8855 | 818 | if (cset == &init_css_set) { |
7717f7ba PM |
819 | res = &root->top_cgroup; |
820 | } else { | |
69d0206c TH |
821 | struct cgrp_cset_link *link; |
822 | ||
823 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 824 | struct cgroup *c = link->cgrp; |
69d0206c | 825 | |
7717f7ba PM |
826 | if (c->root == root) { |
827 | res = c; | |
828 | break; | |
829 | } | |
830 | } | |
831 | } | |
832 | read_unlock(&css_set_lock); | |
833 | BUG_ON(!res); | |
834 | return res; | |
835 | } | |
836 | ||
ddbcc7e8 PM |
837 | /* |
838 | * There is one global cgroup mutex. We also require taking | |
839 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
840 | * See "The task_lock() exception", at the end of this comment. | |
841 | * | |
842 | * A task must hold cgroup_mutex to modify cgroups. | |
843 | * | |
844 | * Any task can increment and decrement the count field without lock. | |
845 | * So in general, code holding cgroup_mutex can't rely on the count | |
846 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 847 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
848 | * means that no tasks are currently attached, therefore there is no |
849 | * way a task attached to that cgroup can fork (the other way to | |
850 | * increment the count). So code holding cgroup_mutex can safely | |
851 | * assume that if the count is zero, it will stay zero. Similarly, if | |
852 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
853 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
854 | * needs that mutex. | |
855 | * | |
ddbcc7e8 PM |
856 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
857 | * (usually) take cgroup_mutex. These are the two most performance | |
858 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
859 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
860 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
861 | * to the release agent with the name of the cgroup (path relative to |
862 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
863 | * |
864 | * A cgroup can only be deleted if both its 'count' of using tasks | |
865 | * is zero, and its list of 'children' cgroups is empty. Since all | |
866 | * tasks in the system use _some_ cgroup, and since there is always at | |
867 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
868 | * always has either children cgroups and/or using tasks. So we don't | |
869 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
870 | * | |
871 | * The task_lock() exception | |
872 | * | |
873 | * The need for this exception arises from the action of | |
d0b2fdd2 | 874 | * cgroup_attach_task(), which overwrites one task's cgroup pointer with |
a043e3b2 | 875 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
876 | * several performance critical places that need to reference |
877 | * task->cgroup without the expense of grabbing a system global | |
878 | * mutex. Therefore except as noted below, when dereferencing or, as | |
d0b2fdd2 | 879 | * in cgroup_attach_task(), modifying a task's cgroup pointer we use |
ddbcc7e8 PM |
880 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
881 | * the task_struct routinely used for such matters. | |
882 | * | |
883 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 884 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
885 | */ |
886 | ||
628f7cd4 | 887 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
2bd59d48 | 888 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 889 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 890 | |
8d7e6fb0 | 891 | static struct cgroup_name *cgroup_alloc_name(const char *name_str) |
65dff759 LZ |
892 | { |
893 | struct cgroup_name *name; | |
894 | ||
8d7e6fb0 | 895 | name = kmalloc(sizeof(*name) + strlen(name_str) + 1, GFP_KERNEL); |
65dff759 LZ |
896 | if (!name) |
897 | return NULL; | |
8d7e6fb0 | 898 | strcpy(name->name, name_str); |
65dff759 LZ |
899 | return name; |
900 | } | |
901 | ||
8d7e6fb0 TH |
902 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
903 | char *buf) | |
904 | { | |
905 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && | |
906 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
907 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
908 | cft->ss->name, cft->name); | |
909 | else | |
910 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
911 | return buf; | |
912 | } | |
913 | ||
f2e85d57 TH |
914 | /** |
915 | * cgroup_file_mode - deduce file mode of a control file | |
916 | * @cft: the control file in question | |
917 | * | |
918 | * returns cft->mode if ->mode is not 0 | |
919 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
920 | * returns S_IRUGO if it has only a read handler | |
921 | * returns S_IWUSR if it has only a write hander | |
922 | */ | |
923 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
924 | { | |
925 | umode_t mode = 0; | |
926 | ||
927 | if (cft->mode) | |
928 | return cft->mode; | |
929 | ||
930 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
931 | mode |= S_IRUGO; | |
932 | ||
933 | if (cft->write_u64 || cft->write_s64 || cft->write_string || | |
934 | cft->trigger) | |
935 | mode |= S_IWUSR; | |
936 | ||
937 | return mode; | |
938 | } | |
939 | ||
be445626 LZ |
940 | static void cgroup_free_fn(struct work_struct *work) |
941 | { | |
ea15f8cc | 942 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 LZ |
943 | |
944 | mutex_lock(&cgroup_mutex); | |
be445626 LZ |
945 | cgrp->root->number_of_cgroups--; |
946 | mutex_unlock(&cgroup_mutex); | |
947 | ||
415cf07a | 948 | /* |
59f5296b TH |
949 | * We get a ref to the parent, and put the ref when this cgroup is |
950 | * being freed, so it's guaranteed that the parent won't be | |
951 | * destroyed before its children. | |
415cf07a | 952 | */ |
59f5296b | 953 | cgroup_put(cgrp->parent); |
415cf07a | 954 | |
59f5296b TH |
955 | /* put the root reference that we took when we created the cgroup */ |
956 | cgroup_put_root(cgrp->root); | |
be445626 | 957 | |
b1a21367 | 958 | cgroup_pidlist_destroy_all(cgrp); |
be445626 | 959 | |
65dff759 | 960 | kfree(rcu_dereference_raw(cgrp->name)); |
be445626 LZ |
961 | kfree(cgrp); |
962 | } | |
963 | ||
964 | static void cgroup_free_rcu(struct rcu_head *head) | |
965 | { | |
966 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
967 | ||
ea15f8cc | 968 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 969 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
970 | } |
971 | ||
59f5296b TH |
972 | static void cgroup_get(struct cgroup *cgrp) |
973 | { | |
2bd59d48 TH |
974 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
975 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
976 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
977 | } |
978 | ||
59f5296b TH |
979 | static void cgroup_put(struct cgroup *cgrp) |
980 | { | |
2bd59d48 TH |
981 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
982 | return; | |
983 | if (WARN_ON_ONCE(!cgroup_is_dead(cgrp))) | |
984 | return; | |
59f5296b | 985 | |
2bd59d48 TH |
986 | /* |
987 | * XXX: cgrp->id is only used to look up css's. As cgroup and | |
988 | * css's lifetimes will be decoupled, it should be made | |
989 | * per-subsystem and moved to css->id so that lookups are | |
990 | * successful until the target css is released. | |
991 | */ | |
992 | mutex_lock(&cgroup_mutex); | |
993 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
994 | mutex_unlock(&cgroup_mutex); | |
995 | cgrp->id = -1; | |
ddbcc7e8 | 996 | |
2bd59d48 | 997 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 PM |
998 | } |
999 | ||
2739d3cc | 1000 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1001 | { |
2bd59d48 | 1002 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1003 | |
ace2bee8 | 1004 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 1005 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1006 | } |
1007 | ||
13af07df | 1008 | /** |
628f7cd4 | 1009 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 1010 | * @cgrp: target cgroup |
13af07df AR |
1011 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
1012 | */ | |
628f7cd4 | 1013 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 1014 | { |
13af07df | 1015 | struct cgroup_subsys *ss; |
b420ba7d | 1016 | int i; |
05ef1d7c | 1017 | |
b420ba7d | 1018 | for_each_subsys(ss, i) { |
13af07df | 1019 | struct cftype_set *set; |
b420ba7d TH |
1020 | |
1021 | if (!test_bit(i, &subsys_mask)) | |
13af07df AR |
1022 | continue; |
1023 | list_for_each_entry(set, &ss->cftsets, node) | |
2bb566cb | 1024 | cgroup_addrm_files(cgrp, set->cfts, false); |
13af07df | 1025 | } |
ddbcc7e8 PM |
1026 | } |
1027 | ||
ddbcc7e8 | 1028 | static int rebind_subsystems(struct cgroupfs_root *root, |
a8a648c4 | 1029 | unsigned long added_mask, unsigned removed_mask) |
ddbcc7e8 | 1030 | { |
bd89aabc | 1031 | struct cgroup *cgrp = &root->top_cgroup; |
30159ec7 | 1032 | struct cgroup_subsys *ss; |
3126121f | 1033 | int i, ret; |
ddbcc7e8 | 1034 | |
ace2bee8 TH |
1035 | lockdep_assert_held(&cgroup_tree_mutex); |
1036 | lockdep_assert_held(&cgroup_mutex); | |
aae8aab4 | 1037 | |
ddbcc7e8 | 1038 | /* Check that any added subsystems are currently free */ |
3ed80a62 TH |
1039 | for_each_subsys(ss, i) |
1040 | if ((added_mask & (1 << i)) && ss->root != &cgroup_dummy_root) | |
1041 | return -EBUSY; | |
ddbcc7e8 | 1042 | |
3126121f TH |
1043 | ret = cgroup_populate_dir(cgrp, added_mask); |
1044 | if (ret) | |
3ed80a62 | 1045 | return ret; |
3126121f TH |
1046 | |
1047 | /* | |
1048 | * Nothing can fail from this point on. Remove files for the | |
1049 | * removed subsystems and rebind each subsystem. | |
1050 | */ | |
4ac06017 | 1051 | mutex_unlock(&cgroup_mutex); |
3126121f | 1052 | cgroup_clear_dir(cgrp, removed_mask); |
4ac06017 | 1053 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1054 | |
30159ec7 | 1055 | for_each_subsys(ss, i) { |
ddbcc7e8 | 1056 | unsigned long bit = 1UL << i; |
30159ec7 | 1057 | |
a1a71b45 | 1058 | if (bit & added_mask) { |
ddbcc7e8 | 1059 | /* We're binding this subsystem to this hierarchy */ |
ca8bdcaf TH |
1060 | BUG_ON(cgroup_css(cgrp, ss)); |
1061 | BUG_ON(!cgroup_css(cgroup_dummy_top, ss)); | |
1062 | BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top); | |
a8a648c4 | 1063 | |
73e80ed8 | 1064 | rcu_assign_pointer(cgrp->subsys[i], |
ca8bdcaf TH |
1065 | cgroup_css(cgroup_dummy_top, ss)); |
1066 | cgroup_css(cgrp, ss)->cgroup = cgrp; | |
a8a648c4 | 1067 | |
b2aa30f7 | 1068 | ss->root = root; |
ddbcc7e8 | 1069 | if (ss->bind) |
ca8bdcaf | 1070 | ss->bind(cgroup_css(cgrp, ss)); |
a8a648c4 | 1071 | |
cf5d5941 | 1072 | /* refcount was already taken, and we're keeping it */ |
a8a648c4 | 1073 | root->subsys_mask |= bit; |
a1a71b45 | 1074 | } else if (bit & removed_mask) { |
ddbcc7e8 | 1075 | /* We're removing this subsystem */ |
ca8bdcaf TH |
1076 | BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss)); |
1077 | BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp); | |
a8a648c4 | 1078 | |
ddbcc7e8 | 1079 | if (ss->bind) |
ca8bdcaf | 1080 | ss->bind(cgroup_css(cgroup_dummy_top, ss)); |
73e80ed8 | 1081 | |
ca8bdcaf | 1082 | cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top; |
73e80ed8 TH |
1083 | RCU_INIT_POINTER(cgrp->subsys[i], NULL); |
1084 | ||
9871bf95 | 1085 | cgroup_subsys[i]->root = &cgroup_dummy_root; |
a8a648c4 | 1086 | root->subsys_mask &= ~bit; |
ddbcc7e8 PM |
1087 | } |
1088 | } | |
ddbcc7e8 | 1089 | |
1672d040 TH |
1090 | /* |
1091 | * Mark @root has finished binding subsystems. @root->subsys_mask | |
1092 | * now matches the bound subsystems. | |
1093 | */ | |
1094 | root->flags |= CGRP_ROOT_SUBSYS_BOUND; | |
2bd59d48 | 1095 | kernfs_activate(cgrp->kn); |
1672d040 | 1096 | |
ddbcc7e8 PM |
1097 | return 0; |
1098 | } | |
1099 | ||
2bd59d48 TH |
1100 | static int cgroup_show_options(struct seq_file *seq, |
1101 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1102 | { |
2bd59d48 | 1103 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1104 | struct cgroup_subsys *ss; |
b85d2040 | 1105 | int ssid; |
ddbcc7e8 | 1106 | |
b85d2040 TH |
1107 | for_each_subsys(ss, ssid) |
1108 | if (root->subsys_mask & (1 << ssid)) | |
1109 | seq_printf(seq, ",%s", ss->name); | |
873fe09e TH |
1110 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1111 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1112 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1113 | seq_puts(seq, ",noprefix"); |
93438629 | 1114 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1115 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1116 | |
1117 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1118 | if (strlen(root->release_agent_path)) |
1119 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1120 | spin_unlock(&release_agent_path_lock); |
1121 | ||
2260e7fc | 1122 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags)) |
97978e6d | 1123 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1124 | if (strlen(root->name)) |
1125 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1126 | return 0; |
1127 | } | |
1128 | ||
1129 | struct cgroup_sb_opts { | |
a1a71b45 | 1130 | unsigned long subsys_mask; |
ddbcc7e8 | 1131 | unsigned long flags; |
81a6a5cd | 1132 | char *release_agent; |
2260e7fc | 1133 | bool cpuset_clone_children; |
c6d57f33 | 1134 | char *name; |
2c6ab6d2 PM |
1135 | /* User explicitly requested empty subsystem */ |
1136 | bool none; | |
ddbcc7e8 PM |
1137 | }; |
1138 | ||
aae8aab4 | 1139 | /* |
9871bf95 TH |
1140 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1141 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1142 | * array. This function takes refcounts on subsystems to be used, unless it | |
1143 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1144 | */ |
cf5d5941 | 1145 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1146 | { |
32a8cf23 DL |
1147 | char *token, *o = data; |
1148 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1149 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1150 | struct cgroup_subsys *ss; |
1151 | int i; | |
f9ab5b5b | 1152 | |
aae8aab4 BB |
1153 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1154 | ||
f9ab5b5b | 1155 | #ifdef CONFIG_CPUSETS |
073219e9 | 1156 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1157 | #endif |
ddbcc7e8 | 1158 | |
c6d57f33 | 1159 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1160 | |
1161 | while ((token = strsep(&o, ",")) != NULL) { | |
1162 | if (!*token) | |
1163 | return -EINVAL; | |
32a8cf23 | 1164 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1165 | /* Explicitly have no subsystems */ |
1166 | opts->none = true; | |
32a8cf23 DL |
1167 | continue; |
1168 | } | |
1169 | if (!strcmp(token, "all")) { | |
1170 | /* Mutually exclusive option 'all' + subsystem name */ | |
1171 | if (one_ss) | |
1172 | return -EINVAL; | |
1173 | all_ss = true; | |
1174 | continue; | |
1175 | } | |
873fe09e TH |
1176 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1177 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1178 | continue; | |
1179 | } | |
32a8cf23 | 1180 | if (!strcmp(token, "noprefix")) { |
93438629 | 1181 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1182 | continue; |
1183 | } | |
1184 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1185 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1186 | continue; |
1187 | } | |
03b1cde6 | 1188 | if (!strcmp(token, "xattr")) { |
93438629 | 1189 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1190 | continue; |
1191 | } | |
32a8cf23 | 1192 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1193 | /* Specifying two release agents is forbidden */ |
1194 | if (opts->release_agent) | |
1195 | return -EINVAL; | |
c6d57f33 | 1196 | opts->release_agent = |
e400c285 | 1197 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1198 | if (!opts->release_agent) |
1199 | return -ENOMEM; | |
32a8cf23 DL |
1200 | continue; |
1201 | } | |
1202 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1203 | const char *name = token + 5; |
1204 | /* Can't specify an empty name */ | |
1205 | if (!strlen(name)) | |
1206 | return -EINVAL; | |
1207 | /* Must match [\w.-]+ */ | |
1208 | for (i = 0; i < strlen(name); i++) { | |
1209 | char c = name[i]; | |
1210 | if (isalnum(c)) | |
1211 | continue; | |
1212 | if ((c == '.') || (c == '-') || (c == '_')) | |
1213 | continue; | |
1214 | return -EINVAL; | |
1215 | } | |
1216 | /* Specifying two names is forbidden */ | |
1217 | if (opts->name) | |
1218 | return -EINVAL; | |
1219 | opts->name = kstrndup(name, | |
e400c285 | 1220 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1221 | GFP_KERNEL); |
1222 | if (!opts->name) | |
1223 | return -ENOMEM; | |
32a8cf23 DL |
1224 | |
1225 | continue; | |
1226 | } | |
1227 | ||
30159ec7 | 1228 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1229 | if (strcmp(token, ss->name)) |
1230 | continue; | |
1231 | if (ss->disabled) | |
1232 | continue; | |
1233 | ||
1234 | /* Mutually exclusive option 'all' + subsystem name */ | |
1235 | if (all_ss) | |
1236 | return -EINVAL; | |
a1a71b45 | 1237 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1238 | one_ss = true; |
1239 | ||
1240 | break; | |
1241 | } | |
1242 | if (i == CGROUP_SUBSYS_COUNT) | |
1243 | return -ENOENT; | |
1244 | } | |
1245 | ||
1246 | /* | |
1247 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1248 | * otherwise if 'none', 'name=' and a subsystem name options |
1249 | * were not specified, let's default to 'all' | |
32a8cf23 | 1250 | */ |
30159ec7 TH |
1251 | if (all_ss || (!one_ss && !opts->none && !opts->name)) |
1252 | for_each_subsys(ss, i) | |
1253 | if (!ss->disabled) | |
1254 | set_bit(i, &opts->subsys_mask); | |
ddbcc7e8 | 1255 | |
2c6ab6d2 PM |
1256 | /* Consistency checks */ |
1257 | ||
873fe09e TH |
1258 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1259 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1260 | ||
1261 | if (opts->flags & CGRP_ROOT_NOPREFIX) { | |
1262 | pr_err("cgroup: sane_behavior: noprefix is not allowed\n"); | |
1263 | return -EINVAL; | |
1264 | } | |
1265 | ||
1266 | if (opts->cpuset_clone_children) { | |
1267 | pr_err("cgroup: sane_behavior: clone_children is not allowed\n"); | |
1268 | return -EINVAL; | |
1269 | } | |
86bf4b68 TH |
1270 | |
1271 | if (opts->flags & CGRP_ROOT_XATTR) | |
1272 | pr_warning("cgroup: sane_behavior: xattr is always available, flag unnecessary\n"); | |
873fe09e TH |
1273 | } |
1274 | ||
f9ab5b5b LZ |
1275 | /* |
1276 | * Option noprefix was introduced just for backward compatibility | |
1277 | * with the old cpuset, so we allow noprefix only if mounting just | |
1278 | * the cpuset subsystem. | |
1279 | */ | |
93438629 | 1280 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1281 | return -EINVAL; |
1282 | ||
2c6ab6d2 PM |
1283 | |
1284 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1285 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1286 | return -EINVAL; |
1287 | ||
1288 | /* | |
1289 | * We either have to specify by name or by subsystems. (So all | |
1290 | * empty hierarchies must have a name). | |
1291 | */ | |
a1a71b45 | 1292 | if (!opts->subsys_mask && !opts->name) |
ddbcc7e8 PM |
1293 | return -EINVAL; |
1294 | ||
1295 | return 0; | |
1296 | } | |
1297 | ||
2bd59d48 | 1298 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1299 | { |
1300 | int ret = 0; | |
2bd59d48 | 1301 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1302 | struct cgroup_sb_opts opts; |
a1a71b45 | 1303 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1304 | |
873fe09e TH |
1305 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1306 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1307 | return -EINVAL; | |
1308 | } | |
1309 | ||
ace2bee8 | 1310 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1311 | mutex_lock(&cgroup_mutex); |
1312 | ||
1313 | /* See what subsystems are wanted */ | |
1314 | ret = parse_cgroupfs_options(data, &opts); | |
1315 | if (ret) | |
1316 | goto out_unlock; | |
1317 | ||
a8a648c4 | 1318 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1319 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1320 | task_tgid_nr(current), current->comm); | |
1321 | ||
a1a71b45 AR |
1322 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1323 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1324 | |
cf5d5941 | 1325 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1326 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1327 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1328 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1329 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1330 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1331 | ret = -EINVAL; |
1332 | goto out_unlock; | |
1333 | } | |
1334 | ||
f172e67c TH |
1335 | /* remounting is not allowed for populated hierarchies */ |
1336 | if (root->number_of_cgroups > 1) { | |
1337 | ret = -EBUSY; | |
0670e08b | 1338 | goto out_unlock; |
cf5d5941 | 1339 | } |
ddbcc7e8 | 1340 | |
a8a648c4 | 1341 | ret = rebind_subsystems(root, added_mask, removed_mask); |
3126121f | 1342 | if (ret) |
0670e08b | 1343 | goto out_unlock; |
ddbcc7e8 | 1344 | |
69e943b7 TH |
1345 | if (opts.release_agent) { |
1346 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1347 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1348 | spin_unlock(&release_agent_path_lock); |
1349 | } | |
ddbcc7e8 | 1350 | out_unlock: |
66bdc9cf | 1351 | kfree(opts.release_agent); |
c6d57f33 | 1352 | kfree(opts.name); |
ddbcc7e8 | 1353 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1354 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1355 | return ret; |
1356 | } | |
1357 | ||
cc31edce PM |
1358 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1359 | { | |
2bd59d48 | 1360 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1361 | INIT_LIST_HEAD(&cgrp->sibling); |
1362 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1363 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1364 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1365 | INIT_LIST_HEAD(&cgrp->pidlists); |
1366 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1367 | cgrp->dummy_css.cgroup = cgrp; |
cc31edce | 1368 | } |
c6d57f33 | 1369 | |
ddbcc7e8 PM |
1370 | static void init_cgroup_root(struct cgroupfs_root *root) |
1371 | { | |
bd89aabc | 1372 | struct cgroup *cgrp = &root->top_cgroup; |
b0ca5a84 | 1373 | |
2bd59d48 | 1374 | atomic_set(&root->refcnt, 1); |
ddbcc7e8 PM |
1375 | INIT_LIST_HEAD(&root->root_list); |
1376 | root->number_of_cgroups = 1; | |
bd89aabc | 1377 | cgrp->root = root; |
a4ea1cc9 | 1378 | RCU_INIT_POINTER(cgrp->name, &root_cgroup_name); |
cc31edce | 1379 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1380 | idr_init(&root->cgroup_idr); |
ddbcc7e8 PM |
1381 | } |
1382 | ||
c6d57f33 PM |
1383 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1384 | { | |
1385 | struct cgroupfs_root *root; | |
1386 | ||
a1a71b45 | 1387 | if (!opts->subsys_mask && !opts->none) |
2bd59d48 | 1388 | return ERR_PTR(-EINVAL); |
c6d57f33 PM |
1389 | |
1390 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1391 | if (!root) | |
1392 | return ERR_PTR(-ENOMEM); | |
1393 | ||
1394 | init_cgroup_root(root); | |
2c6ab6d2 | 1395 | |
1672d040 TH |
1396 | /* |
1397 | * We need to set @root->subsys_mask now so that @root can be | |
1398 | * matched by cgroup_test_super() before it finishes | |
1399 | * initialization; otherwise, competing mounts with the same | |
1400 | * options may try to bind the same subsystems instead of waiting | |
1401 | * for the first one leading to unexpected mount errors. | |
1402 | * SUBSYS_BOUND will be set once actual binding is complete. | |
1403 | */ | |
a1a71b45 | 1404 | root->subsys_mask = opts->subsys_mask; |
c6d57f33 PM |
1405 | root->flags = opts->flags; |
1406 | if (opts->release_agent) | |
1407 | strcpy(root->release_agent_path, opts->release_agent); | |
1408 | if (opts->name) | |
1409 | strcpy(root->name, opts->name); | |
2260e7fc TH |
1410 | if (opts->cpuset_clone_children) |
1411 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1412 | return root; |
1413 | } | |
1414 | ||
d427dfeb TH |
1415 | static int cgroup_setup_root(struct cgroupfs_root *root) |
1416 | { | |
1417 | LIST_HEAD(tmp_links); | |
d427dfeb | 1418 | struct cgroup *root_cgrp = &root->top_cgroup; |
d427dfeb | 1419 | struct css_set *cset; |
d427dfeb TH |
1420 | int i, ret; |
1421 | ||
1422 | lockdep_assert_held(&cgroup_tree_mutex); | |
1423 | lockdep_assert_held(&cgroup_mutex); | |
d427dfeb TH |
1424 | |
1425 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); | |
1426 | if (ret < 0) | |
2bd59d48 | 1427 | goto out; |
d427dfeb TH |
1428 | root_cgrp->id = ret; |
1429 | ||
d427dfeb TH |
1430 | /* |
1431 | * We're accessing css_set_count without locking css_set_lock here, | |
1432 | * but that's OK - it can only be increased by someone holding | |
1433 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1434 | * have some link structures left over | |
1435 | */ | |
1436 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1437 | if (ret) | |
2bd59d48 | 1438 | goto out; |
d427dfeb TH |
1439 | |
1440 | /* ID 0 is reserved for dummy root, 1 for unified hierarchy */ | |
1441 | ret = cgroup_init_root_id(root, 2, 0); | |
1442 | if (ret) | |
2bd59d48 | 1443 | goto out; |
d427dfeb | 1444 | |
2bd59d48 TH |
1445 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1446 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1447 | root_cgrp); | |
1448 | if (IS_ERR(root->kf_root)) { | |
1449 | ret = PTR_ERR(root->kf_root); | |
1450 | goto exit_root_id; | |
1451 | } | |
1452 | root_cgrp->kn = root->kf_root->kn; | |
d427dfeb TH |
1453 | |
1454 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); | |
1455 | if (ret) | |
2bd59d48 | 1456 | goto destroy_root; |
d427dfeb TH |
1457 | |
1458 | ret = rebind_subsystems(root, root->subsys_mask, 0); | |
1459 | if (ret) | |
2bd59d48 | 1460 | goto destroy_root; |
d427dfeb TH |
1461 | |
1462 | /* | |
1463 | * There must be no failure case after here, since rebinding takes | |
1464 | * care of subsystems' refcounts, which are explicitly dropped in | |
1465 | * the failure exit path. | |
1466 | */ | |
1467 | list_add(&root->root_list, &cgroup_roots); | |
1468 | cgroup_root_count++; | |
1469 | ||
1470 | /* | |
1471 | * Link the top cgroup in this hierarchy into all the css_set | |
1472 | * objects. | |
1473 | */ | |
1474 | write_lock(&css_set_lock); | |
1475 | hash_for_each(css_set_table, i, cset, hlist) | |
1476 | link_css_set(&tmp_links, cset, root_cgrp); | |
1477 | write_unlock(&css_set_lock); | |
1478 | ||
1479 | BUG_ON(!list_empty(&root_cgrp->children)); | |
1480 | BUG_ON(root->number_of_cgroups != 1); | |
1481 | ||
2bd59d48 | 1482 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1483 | ret = 0; |
2bd59d48 | 1484 | goto out; |
d427dfeb | 1485 | |
2bd59d48 TH |
1486 | destroy_root: |
1487 | kernfs_destroy_root(root->kf_root); | |
1488 | root->kf_root = NULL; | |
1489 | exit_root_id: | |
d427dfeb | 1490 | cgroup_exit_root_id(root); |
2bd59d48 | 1491 | out: |
d427dfeb TH |
1492 | free_cgrp_cset_links(&tmp_links); |
1493 | return ret; | |
1494 | } | |
1495 | ||
f7e83571 | 1496 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1497 | int flags, const char *unused_dev_name, |
f7e83571 | 1498 | void *data) |
ddbcc7e8 | 1499 | { |
2bd59d48 | 1500 | struct cgroupfs_root *root; |
ddbcc7e8 | 1501 | struct cgroup_sb_opts opts; |
2bd59d48 | 1502 | struct dentry *dentry; |
8e30e2b8 | 1503 | int ret; |
ddbcc7e8 | 1504 | |
8e30e2b8 | 1505 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1506 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1507 | |
1508 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1509 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1510 | if (ret) |
8e30e2b8 | 1511 | goto out_unlock; |
ddbcc7e8 | 1512 | |
2bd59d48 TH |
1513 | /* look for a matching existing root */ |
1514 | for_each_active_root(root) { | |
1515 | bool name_match = false; | |
ddbcc7e8 | 1516 | |
2bd59d48 TH |
1517 | /* |
1518 | * If we asked for a name then it must match. Also, if | |
1519 | * name matches but sybsys_mask doesn't, we should fail. | |
1520 | * Remember whether name matched. | |
1521 | */ | |
1522 | if (opts.name) { | |
1523 | if (strcmp(opts.name, root->name)) | |
1524 | continue; | |
1525 | name_match = true; | |
1526 | } | |
ddbcc7e8 | 1527 | |
c6d57f33 | 1528 | /* |
2bd59d48 TH |
1529 | * If we asked for subsystems (or explicitly for no |
1530 | * subsystems) then they must match. | |
c6d57f33 | 1531 | */ |
2bd59d48 TH |
1532 | if ((opts.subsys_mask || opts.none) && |
1533 | (opts.subsys_mask != root->subsys_mask)) { | |
1534 | if (!name_match) | |
1535 | continue; | |
1536 | ret = -EBUSY; | |
1537 | goto out_unlock; | |
1538 | } | |
873fe09e | 1539 | |
c7ba8287 | 1540 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1541 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1542 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1543 | ret = -EINVAL; | |
8e30e2b8 | 1544 | goto out_unlock; |
2a0ff3fb JL |
1545 | } else { |
1546 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1547 | } | |
873fe09e | 1548 | } |
2bd59d48 TH |
1549 | |
1550 | cgroup_get_root(root); | |
1551 | goto out_unlock; | |
ddbcc7e8 PM |
1552 | } |
1553 | ||
2bd59d48 TH |
1554 | /* no such thing, create a new one */ |
1555 | root = cgroup_root_from_opts(&opts); | |
1556 | if (IS_ERR(root)) { | |
1557 | ret = PTR_ERR(root); | |
1558 | goto out_unlock; | |
1559 | } | |
1560 | ||
1561 | ret = cgroup_setup_root(root); | |
1562 | if (ret) | |
1563 | cgroup_free_root(root); | |
1564 | ||
8e30e2b8 | 1565 | out_unlock: |
e25e2cbb | 1566 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1567 | mutex_unlock(&cgroup_tree_mutex); |
8e30e2b8 | 1568 | |
c6d57f33 PM |
1569 | kfree(opts.release_agent); |
1570 | kfree(opts.name); | |
8e30e2b8 | 1571 | |
2bd59d48 | 1572 | if (ret) |
8e30e2b8 | 1573 | return ERR_PTR(ret); |
2bd59d48 TH |
1574 | |
1575 | dentry = kernfs_mount(fs_type, flags, root->kf_root); | |
1576 | if (IS_ERR(dentry)) | |
1577 | cgroup_put_root(root); | |
1578 | return dentry; | |
1579 | } | |
1580 | ||
1581 | static void cgroup_kill_sb(struct super_block *sb) | |
1582 | { | |
1583 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
1584 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); | |
1585 | ||
1586 | cgroup_put_root(root); | |
1587 | kernfs_kill_sb(sb); | |
ddbcc7e8 PM |
1588 | } |
1589 | ||
ddbcc7e8 PM |
1590 | static struct file_system_type cgroup_fs_type = { |
1591 | .name = "cgroup", | |
f7e83571 | 1592 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1593 | .kill_sb = cgroup_kill_sb, |
1594 | }; | |
1595 | ||
676db4af GKH |
1596 | static struct kobject *cgroup_kobj; |
1597 | ||
a043e3b2 LZ |
1598 | /** |
1599 | * cgroup_path - generate the path of a cgroup | |
1600 | * @cgrp: the cgroup in question | |
1601 | * @buf: the buffer to write the path into | |
1602 | * @buflen: the length of the buffer | |
1603 | * | |
65dff759 LZ |
1604 | * Writes path of cgroup into buf. Returns 0 on success, -errno on error. |
1605 | * | |
1606 | * We can't generate cgroup path using dentry->d_name, as accessing | |
1607 | * dentry->name must be protected by irq-unsafe dentry->d_lock or parent | |
1608 | * inode's i_mutex, while on the other hand cgroup_path() can be called | |
1609 | * with some irq-safe spinlocks held. | |
ddbcc7e8 | 1610 | */ |
bd89aabc | 1611 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
ddbcc7e8 | 1612 | { |
65dff759 | 1613 | int ret = -ENAMETOOLONG; |
ddbcc7e8 | 1614 | char *start; |
febfcef6 | 1615 | |
da1f296f TH |
1616 | if (!cgrp->parent) { |
1617 | if (strlcpy(buf, "/", buflen) >= buflen) | |
1618 | return -ENAMETOOLONG; | |
ddbcc7e8 PM |
1619 | return 0; |
1620 | } | |
1621 | ||
316eb661 | 1622 | start = buf + buflen - 1; |
316eb661 | 1623 | *start = '\0'; |
9a9686b6 | 1624 | |
65dff759 | 1625 | rcu_read_lock(); |
da1f296f | 1626 | do { |
65dff759 LZ |
1627 | const char *name = cgroup_name(cgrp); |
1628 | int len; | |
1629 | ||
1630 | len = strlen(name); | |
ddbcc7e8 | 1631 | if ((start -= len) < buf) |
65dff759 LZ |
1632 | goto out; |
1633 | memcpy(start, name, len); | |
9a9686b6 | 1634 | |
ddbcc7e8 | 1635 | if (--start < buf) |
65dff759 | 1636 | goto out; |
ddbcc7e8 | 1637 | *start = '/'; |
65dff759 LZ |
1638 | |
1639 | cgrp = cgrp->parent; | |
da1f296f | 1640 | } while (cgrp->parent); |
65dff759 | 1641 | ret = 0; |
ddbcc7e8 | 1642 | memmove(buf, start, buf + buflen - start); |
65dff759 LZ |
1643 | out: |
1644 | rcu_read_unlock(); | |
1645 | return ret; | |
ddbcc7e8 | 1646 | } |
67523c48 | 1647 | EXPORT_SYMBOL_GPL(cgroup_path); |
ddbcc7e8 | 1648 | |
857a2beb | 1649 | /** |
913ffdb5 | 1650 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1651 | * @task: target task |
857a2beb TH |
1652 | * @buf: the buffer to write the path into |
1653 | * @buflen: the length of the buffer | |
1654 | * | |
913ffdb5 TH |
1655 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1656 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1657 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1658 | * cgroup controller callbacks. | |
1659 | * | |
1660 | * Returns 0 on success, fails with -%ENAMETOOLONG if @buflen is too short. | |
857a2beb | 1661 | */ |
913ffdb5 | 1662 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb TH |
1663 | { |
1664 | struct cgroupfs_root *root; | |
913ffdb5 TH |
1665 | struct cgroup *cgrp; |
1666 | int hierarchy_id = 1, ret = 0; | |
1667 | ||
1668 | if (buflen < 2) | |
1669 | return -ENAMETOOLONG; | |
857a2beb TH |
1670 | |
1671 | mutex_lock(&cgroup_mutex); | |
1672 | ||
913ffdb5 TH |
1673 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1674 | ||
857a2beb TH |
1675 | if (root) { |
1676 | cgrp = task_cgroup_from_root(task, root); | |
1677 | ret = cgroup_path(cgrp, buf, buflen); | |
913ffdb5 TH |
1678 | } else { |
1679 | /* if no hierarchy exists, everyone is in "/" */ | |
1680 | memcpy(buf, "/", 2); | |
857a2beb TH |
1681 | } |
1682 | ||
1683 | mutex_unlock(&cgroup_mutex); | |
857a2beb TH |
1684 | return ret; |
1685 | } | |
913ffdb5 | 1686 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1687 | |
2f7ee569 TH |
1688 | /* |
1689 | * Control Group taskset | |
1690 | */ | |
134d3373 TH |
1691 | struct task_and_cgroup { |
1692 | struct task_struct *task; | |
1693 | struct cgroup *cgrp; | |
6f4b7e63 | 1694 | struct css_set *cset; |
134d3373 TH |
1695 | }; |
1696 | ||
2f7ee569 TH |
1697 | struct cgroup_taskset { |
1698 | struct task_and_cgroup single; | |
1699 | struct flex_array *tc_array; | |
1700 | int tc_array_len; | |
1701 | int idx; | |
1702 | struct cgroup *cur_cgrp; | |
1703 | }; | |
1704 | ||
1705 | /** | |
1706 | * cgroup_taskset_first - reset taskset and return the first task | |
1707 | * @tset: taskset of interest | |
1708 | * | |
1709 | * @tset iteration is initialized and the first task is returned. | |
1710 | */ | |
1711 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1712 | { | |
1713 | if (tset->tc_array) { | |
1714 | tset->idx = 0; | |
1715 | return cgroup_taskset_next(tset); | |
1716 | } else { | |
1717 | tset->cur_cgrp = tset->single.cgrp; | |
1718 | return tset->single.task; | |
1719 | } | |
1720 | } | |
1721 | EXPORT_SYMBOL_GPL(cgroup_taskset_first); | |
1722 | ||
1723 | /** | |
1724 | * cgroup_taskset_next - iterate to the next task in taskset | |
1725 | * @tset: taskset of interest | |
1726 | * | |
1727 | * Return the next task in @tset. Iteration must have been initialized | |
1728 | * with cgroup_taskset_first(). | |
1729 | */ | |
1730 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1731 | { | |
1732 | struct task_and_cgroup *tc; | |
1733 | ||
1734 | if (!tset->tc_array || tset->idx >= tset->tc_array_len) | |
1735 | return NULL; | |
1736 | ||
1737 | tc = flex_array_get(tset->tc_array, tset->idx++); | |
1738 | tset->cur_cgrp = tc->cgrp; | |
1739 | return tc->task; | |
1740 | } | |
1741 | EXPORT_SYMBOL_GPL(cgroup_taskset_next); | |
1742 | ||
1743 | /** | |
d99c8727 | 1744 | * cgroup_taskset_cur_css - return the matching css for the current task |
2f7ee569 | 1745 | * @tset: taskset of interest |
d99c8727 | 1746 | * @subsys_id: the ID of the target subsystem |
2f7ee569 | 1747 | * |
d99c8727 TH |
1748 | * Return the css for the current (last returned) task of @tset for |
1749 | * subsystem specified by @subsys_id. This function must be preceded by | |
1750 | * either cgroup_taskset_first() or cgroup_taskset_next(). | |
2f7ee569 | 1751 | */ |
d99c8727 TH |
1752 | struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset, |
1753 | int subsys_id) | |
2f7ee569 | 1754 | { |
ca8bdcaf | 1755 | return cgroup_css(tset->cur_cgrp, cgroup_subsys[subsys_id]); |
2f7ee569 | 1756 | } |
d99c8727 | 1757 | EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css); |
2f7ee569 TH |
1758 | |
1759 | /** | |
1760 | * cgroup_taskset_size - return the number of tasks in taskset | |
1761 | * @tset: taskset of interest | |
1762 | */ | |
1763 | int cgroup_taskset_size(struct cgroup_taskset *tset) | |
1764 | { | |
1765 | return tset->tc_array ? tset->tc_array_len : 1; | |
1766 | } | |
1767 | EXPORT_SYMBOL_GPL(cgroup_taskset_size); | |
1768 | ||
1769 | ||
74a1166d BB |
1770 | /* |
1771 | * cgroup_task_migrate - move a task from one cgroup to another. | |
1772 | * | |
d0b2fdd2 | 1773 | * Must be called with cgroup_mutex and threadgroup locked. |
74a1166d | 1774 | */ |
5abb8855 TH |
1775 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1776 | struct task_struct *tsk, | |
1777 | struct css_set *new_cset) | |
74a1166d | 1778 | { |
5abb8855 | 1779 | struct css_set *old_cset; |
74a1166d BB |
1780 | |
1781 | /* | |
026085ef MSB |
1782 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1783 | * setting such that we can't race against cgroup_exit() changing the | |
1784 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1785 | */ |
c84cdf75 | 1786 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1787 | old_cset = task_css_set(tsk); |
74a1166d | 1788 | |
74a1166d | 1789 | task_lock(tsk); |
5abb8855 | 1790 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d BB |
1791 | task_unlock(tsk); |
1792 | ||
1793 | /* Update the css_set linked lists if we're using them */ | |
1794 | write_lock(&css_set_lock); | |
1795 | if (!list_empty(&tsk->cg_list)) | |
5abb8855 | 1796 | list_move(&tsk->cg_list, &new_cset->tasks); |
74a1166d BB |
1797 | write_unlock(&css_set_lock); |
1798 | ||
1799 | /* | |
5abb8855 TH |
1800 | * We just gained a reference on old_cset by taking it from the |
1801 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1802 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1803 | */ |
5abb8855 TH |
1804 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
1805 | put_css_set(old_cset); | |
74a1166d BB |
1806 | } |
1807 | ||
a043e3b2 | 1808 | /** |
081aa458 | 1809 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup |
74a1166d | 1810 | * @cgrp: the cgroup to attach to |
081aa458 LZ |
1811 | * @tsk: the task or the leader of the threadgroup to be attached |
1812 | * @threadgroup: attach the whole threadgroup? | |
74a1166d | 1813 | * |
257058ae | 1814 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
081aa458 | 1815 | * task_lock of @tsk or each thread in the threadgroup individually in turn. |
74a1166d | 1816 | */ |
47cfcd09 TH |
1817 | static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk, |
1818 | bool threadgroup) | |
74a1166d BB |
1819 | { |
1820 | int retval, i, group_size; | |
74a1166d | 1821 | struct cgroupfs_root *root = cgrp->root; |
1c6727af | 1822 | struct cgroup_subsys_state *css, *failed_css = NULL; |
74a1166d | 1823 | /* threadgroup list cursor and array */ |
081aa458 | 1824 | struct task_struct *leader = tsk; |
134d3373 | 1825 | struct task_and_cgroup *tc; |
d846687d | 1826 | struct flex_array *group; |
2f7ee569 | 1827 | struct cgroup_taskset tset = { }; |
74a1166d BB |
1828 | |
1829 | /* | |
1830 | * step 0: in order to do expensive, possibly blocking operations for | |
1831 | * every thread, we cannot iterate the thread group list, since it needs | |
1832 | * rcu or tasklist locked. instead, build an array of all threads in the | |
257058ae TH |
1833 | * group - group_rwsem prevents new threads from appearing, and if |
1834 | * threads exit, this will just be an over-estimate. | |
74a1166d | 1835 | */ |
081aa458 LZ |
1836 | if (threadgroup) |
1837 | group_size = get_nr_threads(tsk); | |
1838 | else | |
1839 | group_size = 1; | |
d846687d | 1840 | /* flex_array supports very large thread-groups better than kmalloc. */ |
134d3373 | 1841 | group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL); |
74a1166d BB |
1842 | if (!group) |
1843 | return -ENOMEM; | |
d846687d | 1844 | /* pre-allocate to guarantee space while iterating in rcu read-side. */ |
3ac1707a | 1845 | retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL); |
d846687d BB |
1846 | if (retval) |
1847 | goto out_free_group_list; | |
74a1166d | 1848 | |
74a1166d | 1849 | i = 0; |
fb5d2b4c MSB |
1850 | /* |
1851 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1852 | * already PF_EXITING could be freed from underneath us unless we | |
1853 | * take an rcu_read_lock. | |
1854 | */ | |
1855 | rcu_read_lock(); | |
74a1166d | 1856 | do { |
134d3373 TH |
1857 | struct task_and_cgroup ent; |
1858 | ||
cd3d0952 TH |
1859 | /* @tsk either already exited or can't exit until the end */ |
1860 | if (tsk->flags & PF_EXITING) | |
ea84753c | 1861 | goto next; |
cd3d0952 | 1862 | |
74a1166d BB |
1863 | /* as per above, nr_threads may decrease, but not increase. */ |
1864 | BUG_ON(i >= group_size); | |
134d3373 TH |
1865 | ent.task = tsk; |
1866 | ent.cgrp = task_cgroup_from_root(tsk, root); | |
892a2b90 MSB |
1867 | /* nothing to do if this task is already in the cgroup */ |
1868 | if (ent.cgrp == cgrp) | |
ea84753c | 1869 | goto next; |
61d1d219 MSB |
1870 | /* |
1871 | * saying GFP_ATOMIC has no effect here because we did prealloc | |
1872 | * earlier, but it's good form to communicate our expectations. | |
1873 | */ | |
134d3373 | 1874 | retval = flex_array_put(group, i, &ent, GFP_ATOMIC); |
d846687d | 1875 | BUG_ON(retval != 0); |
74a1166d | 1876 | i++; |
ea84753c | 1877 | next: |
081aa458 LZ |
1878 | if (!threadgroup) |
1879 | break; | |
74a1166d | 1880 | } while_each_thread(leader, tsk); |
fb5d2b4c | 1881 | rcu_read_unlock(); |
74a1166d BB |
1882 | /* remember the number of threads in the array for later. */ |
1883 | group_size = i; | |
2f7ee569 TH |
1884 | tset.tc_array = group; |
1885 | tset.tc_array_len = group_size; | |
74a1166d | 1886 | |
134d3373 TH |
1887 | /* methods shouldn't be called if no task is actually migrating */ |
1888 | retval = 0; | |
892a2b90 | 1889 | if (!group_size) |
b07ef774 | 1890 | goto out_free_group_list; |
134d3373 | 1891 | |
74a1166d BB |
1892 | /* |
1893 | * step 1: check that we can legitimately attach to the cgroup. | |
1894 | */ | |
1c6727af TH |
1895 | for_each_css(css, i, cgrp) { |
1896 | if (css->ss->can_attach) { | |
1897 | retval = css->ss->can_attach(css, &tset); | |
74a1166d | 1898 | if (retval) { |
1c6727af | 1899 | failed_css = css; |
74a1166d BB |
1900 | goto out_cancel_attach; |
1901 | } | |
1902 | } | |
74a1166d BB |
1903 | } |
1904 | ||
1905 | /* | |
1906 | * step 2: make sure css_sets exist for all threads to be migrated. | |
1907 | * we use find_css_set, which allocates a new one if necessary. | |
1908 | */ | |
74a1166d | 1909 | for (i = 0; i < group_size; i++) { |
a8ad805c TH |
1910 | struct css_set *old_cset; |
1911 | ||
134d3373 | 1912 | tc = flex_array_get(group, i); |
a8ad805c | 1913 | old_cset = task_css_set(tc->task); |
6f4b7e63 LZ |
1914 | tc->cset = find_css_set(old_cset, cgrp); |
1915 | if (!tc->cset) { | |
61d1d219 MSB |
1916 | retval = -ENOMEM; |
1917 | goto out_put_css_set_refs; | |
74a1166d BB |
1918 | } |
1919 | } | |
1920 | ||
1921 | /* | |
494c167c TH |
1922 | * step 3: now that we're guaranteed success wrt the css_sets, |
1923 | * proceed to move all tasks to the new cgroup. There are no | |
1924 | * failure cases after here, so this is the commit point. | |
74a1166d | 1925 | */ |
74a1166d | 1926 | for (i = 0; i < group_size; i++) { |
134d3373 | 1927 | tc = flex_array_get(group, i); |
6f4b7e63 | 1928 | cgroup_task_migrate(tc->cgrp, tc->task, tc->cset); |
74a1166d BB |
1929 | } |
1930 | /* nothing is sensitive to fork() after this point. */ | |
1931 | ||
1932 | /* | |
494c167c | 1933 | * step 4: do subsystem attach callbacks. |
74a1166d | 1934 | */ |
1c6727af TH |
1935 | for_each_css(css, i, cgrp) |
1936 | if (css->ss->attach) | |
1937 | css->ss->attach(css, &tset); | |
74a1166d BB |
1938 | |
1939 | /* | |
1940 | * step 5: success! and cleanup | |
1941 | */ | |
74a1166d | 1942 | retval = 0; |
61d1d219 MSB |
1943 | out_put_css_set_refs: |
1944 | if (retval) { | |
1945 | for (i = 0; i < group_size; i++) { | |
1946 | tc = flex_array_get(group, i); | |
6f4b7e63 | 1947 | if (!tc->cset) |
61d1d219 | 1948 | break; |
6f4b7e63 | 1949 | put_css_set(tc->cset); |
61d1d219 | 1950 | } |
74a1166d BB |
1951 | } |
1952 | out_cancel_attach: | |
74a1166d | 1953 | if (retval) { |
1c6727af TH |
1954 | for_each_css(css, i, cgrp) { |
1955 | if (css == failed_css) | |
74a1166d | 1956 | break; |
1c6727af TH |
1957 | if (css->ss->cancel_attach) |
1958 | css->ss->cancel_attach(css, &tset); | |
74a1166d BB |
1959 | } |
1960 | } | |
74a1166d | 1961 | out_free_group_list: |
d846687d | 1962 | flex_array_free(group); |
74a1166d BB |
1963 | return retval; |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
1968 | * function to attach either it or all tasks in its threadgroup. Will lock |
1969 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 1970 | */ |
74a1166d | 1971 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 1972 | { |
bbcb81d0 | 1973 | struct task_struct *tsk; |
c69e8d9c | 1974 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
1975 | int ret; |
1976 | ||
74a1166d BB |
1977 | if (!cgroup_lock_live_group(cgrp)) |
1978 | return -ENODEV; | |
1979 | ||
b78949eb MSB |
1980 | retry_find_task: |
1981 | rcu_read_lock(); | |
bbcb81d0 | 1982 | if (pid) { |
73507f33 | 1983 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
1984 | if (!tsk) { |
1985 | rcu_read_unlock(); | |
dd4b0a46 | 1986 | ret = -ESRCH; |
b78949eb | 1987 | goto out_unlock_cgroup; |
bbcb81d0 | 1988 | } |
74a1166d BB |
1989 | /* |
1990 | * even if we're attaching all tasks in the thread group, we | |
1991 | * only need to check permissions on one of them. | |
1992 | */ | |
c69e8d9c | 1993 | tcred = __task_cred(tsk); |
14a590c3 EB |
1994 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
1995 | !uid_eq(cred->euid, tcred->uid) && | |
1996 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 1997 | rcu_read_unlock(); |
b78949eb MSB |
1998 | ret = -EACCES; |
1999 | goto out_unlock_cgroup; | |
bbcb81d0 | 2000 | } |
b78949eb MSB |
2001 | } else |
2002 | tsk = current; | |
cd3d0952 TH |
2003 | |
2004 | if (threadgroup) | |
b78949eb | 2005 | tsk = tsk->group_leader; |
c4c27fbd MG |
2006 | |
2007 | /* | |
14a40ffc | 2008 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2009 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2010 | * with no rt_runtime allocated. Just say no. | |
2011 | */ | |
14a40ffc | 2012 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2013 | ret = -EINVAL; |
2014 | rcu_read_unlock(); | |
2015 | goto out_unlock_cgroup; | |
2016 | } | |
2017 | ||
b78949eb MSB |
2018 | get_task_struct(tsk); |
2019 | rcu_read_unlock(); | |
2020 | ||
2021 | threadgroup_lock(tsk); | |
2022 | if (threadgroup) { | |
2023 | if (!thread_group_leader(tsk)) { | |
2024 | /* | |
2025 | * a race with de_thread from another thread's exec() | |
2026 | * may strip us of our leadership, if this happens, | |
2027 | * there is no choice but to throw this task away and | |
2028 | * try again; this is | |
2029 | * "double-double-toil-and-trouble-check locking". | |
2030 | */ | |
2031 | threadgroup_unlock(tsk); | |
2032 | put_task_struct(tsk); | |
2033 | goto retry_find_task; | |
2034 | } | |
081aa458 LZ |
2035 | } |
2036 | ||
2037 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2038 | ||
cd3d0952 TH |
2039 | threadgroup_unlock(tsk); |
2040 | ||
bbcb81d0 | 2041 | put_task_struct(tsk); |
b78949eb | 2042 | out_unlock_cgroup: |
47cfcd09 | 2043 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
2044 | return ret; |
2045 | } | |
2046 | ||
7ae1bad9 TH |
2047 | /** |
2048 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2049 | * @from: attach to all cgroups of a given task | |
2050 | * @tsk: the task to be attached | |
2051 | */ | |
2052 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2053 | { | |
2054 | struct cgroupfs_root *root; | |
2055 | int retval = 0; | |
2056 | ||
47cfcd09 | 2057 | mutex_lock(&cgroup_mutex); |
7ae1bad9 | 2058 | for_each_active_root(root) { |
6f4b7e63 | 2059 | struct cgroup *from_cgrp = task_cgroup_from_root(from, root); |
7ae1bad9 | 2060 | |
6f4b7e63 | 2061 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2062 | if (retval) |
2063 | break; | |
2064 | } | |
47cfcd09 | 2065 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2066 | |
2067 | return retval; | |
2068 | } | |
2069 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2070 | ||
182446d0 TH |
2071 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2072 | struct cftype *cft, u64 pid) | |
74a1166d | 2073 | { |
182446d0 | 2074 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2075 | } |
2076 | ||
182446d0 TH |
2077 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2078 | struct cftype *cft, u64 tgid) | |
af351026 | 2079 | { |
182446d0 | 2080 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2081 | } |
2082 | ||
182446d0 TH |
2083 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
2084 | struct cftype *cft, const char *buffer) | |
e788e066 | 2085 | { |
5f469907 TH |
2086 | struct cgroupfs_root *root = css->cgroup->root; |
2087 | ||
2088 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
182446d0 | 2089 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2090 | return -ENODEV; |
69e943b7 | 2091 | spin_lock(&release_agent_path_lock); |
5f469907 TH |
2092 | strlcpy(root->release_agent_path, buffer, |
2093 | sizeof(root->release_agent_path)); | |
69e943b7 | 2094 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2095 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2096 | return 0; |
2097 | } | |
2098 | ||
2da8ca82 | 2099 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2100 | { |
2da8ca82 | 2101 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2102 | |
e788e066 PM |
2103 | if (!cgroup_lock_live_group(cgrp)) |
2104 | return -ENODEV; | |
2105 | seq_puts(seq, cgrp->root->release_agent_path); | |
2106 | seq_putc(seq, '\n'); | |
47cfcd09 | 2107 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2108 | return 0; |
2109 | } | |
2110 | ||
2da8ca82 | 2111 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2112 | { |
2da8ca82 TH |
2113 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2114 | ||
2115 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2116 | return 0; |
2117 | } | |
2118 | ||
2bd59d48 TH |
2119 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2120 | size_t nbytes, loff_t off) | |
355e0c48 | 2121 | { |
2bd59d48 TH |
2122 | struct cgroup *cgrp = of->kn->parent->priv; |
2123 | struct cftype *cft = of->kn->priv; | |
2124 | struct cgroup_subsys_state *css; | |
a742c59d | 2125 | int ret; |
355e0c48 | 2126 | |
2bd59d48 TH |
2127 | /* |
2128 | * kernfs guarantees that a file isn't deleted with operations in | |
2129 | * flight, which means that the matching css is and stays alive and | |
2130 | * doesn't need to be pinned. The RCU locking is not necessary | |
2131 | * either. It's just for the convenience of using cgroup_css(). | |
2132 | */ | |
2133 | rcu_read_lock(); | |
2134 | css = cgroup_css(cgrp, cft->ss); | |
2135 | rcu_read_unlock(); | |
a742c59d TH |
2136 | |
2137 | if (cft->write_string) { | |
2138 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2139 | } else if (cft->write_u64) { | |
2140 | unsigned long long v; | |
2141 | ret = kstrtoull(buf, 0, &v); | |
2142 | if (!ret) | |
2143 | ret = cft->write_u64(css, cft, v); | |
2144 | } else if (cft->write_s64) { | |
2145 | long long v; | |
2146 | ret = kstrtoll(buf, 0, &v); | |
2147 | if (!ret) | |
2148 | ret = cft->write_s64(css, cft, v); | |
2149 | } else if (cft->trigger) { | |
2150 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2151 | } else { |
a742c59d | 2152 | ret = -EINVAL; |
e73d2c61 | 2153 | } |
2bd59d48 | 2154 | |
a742c59d | 2155 | return ret ?: nbytes; |
355e0c48 PM |
2156 | } |
2157 | ||
6612f05b | 2158 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2159 | { |
2bd59d48 | 2160 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2161 | } |
2162 | ||
6612f05b | 2163 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2164 | { |
2bd59d48 | 2165 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2166 | } |
2167 | ||
6612f05b | 2168 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2169 | { |
2bd59d48 | 2170 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2171 | } |
2172 | ||
91796569 | 2173 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2174 | { |
7da11279 TH |
2175 | struct cftype *cft = seq_cft(m); |
2176 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2177 | |
2da8ca82 TH |
2178 | if (cft->seq_show) |
2179 | return cft->seq_show(m, arg); | |
e73d2c61 | 2180 | |
f4c753b7 | 2181 | if (cft->read_u64) |
896f5199 TH |
2182 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2183 | else if (cft->read_s64) | |
2184 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2185 | else | |
2186 | return -EINVAL; | |
2187 | return 0; | |
91796569 PM |
2188 | } |
2189 | ||
2bd59d48 TH |
2190 | static struct kernfs_ops cgroup_kf_single_ops = { |
2191 | .atomic_write_len = PAGE_SIZE, | |
2192 | .write = cgroup_file_write, | |
2193 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2194 | }; |
2195 | ||
2bd59d48 TH |
2196 | static struct kernfs_ops cgroup_kf_ops = { |
2197 | .atomic_write_len = PAGE_SIZE, | |
2198 | .write = cgroup_file_write, | |
2199 | .seq_start = cgroup_seqfile_start, | |
2200 | .seq_next = cgroup_seqfile_next, | |
2201 | .seq_stop = cgroup_seqfile_stop, | |
2202 | .seq_show = cgroup_seqfile_show, | |
2203 | }; | |
ddbcc7e8 PM |
2204 | |
2205 | /* | |
2206 | * cgroup_rename - Only allow simple rename of directories in place. | |
2207 | */ | |
2bd59d48 TH |
2208 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2209 | const char *new_name_str) | |
ddbcc7e8 | 2210 | { |
2bd59d48 | 2211 | struct cgroup *cgrp = kn->priv; |
65dff759 | 2212 | struct cgroup_name *name, *old_name; |
2bd59d48 | 2213 | int ret; |
65dff759 | 2214 | |
2bd59d48 | 2215 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2216 | return -ENOTDIR; |
2bd59d48 | 2217 | if (kn->parent != new_parent) |
ddbcc7e8 | 2218 | return -EIO; |
65dff759 | 2219 | |
6db8e85c TH |
2220 | /* |
2221 | * This isn't a proper migration and its usefulness is very | |
2222 | * limited. Disallow if sane_behavior. | |
2223 | */ | |
2224 | if (cgroup_sane_behavior(cgrp)) | |
2225 | return -EPERM; | |
2226 | ||
2bd59d48 | 2227 | name = cgroup_alloc_name(new_name_str); |
65dff759 LZ |
2228 | if (!name) |
2229 | return -ENOMEM; | |
2230 | ||
2bd59d48 TH |
2231 | mutex_lock(&cgroup_tree_mutex); |
2232 | mutex_lock(&cgroup_mutex); | |
2233 | ||
2234 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
2235 | if (!ret) { | |
2236 | old_name = rcu_dereference_protected(cgrp->name, true); | |
2237 | rcu_assign_pointer(cgrp->name, name); | |
2238 | } else { | |
2239 | old_name = name; | |
65dff759 LZ |
2240 | } |
2241 | ||
2bd59d48 TH |
2242 | mutex_unlock(&cgroup_mutex); |
2243 | mutex_unlock(&cgroup_tree_mutex); | |
65dff759 LZ |
2244 | |
2245 | kfree_rcu(old_name, rcu_head); | |
2bd59d48 | 2246 | return ret; |
ddbcc7e8 PM |
2247 | } |
2248 | ||
2bb566cb | 2249 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2250 | { |
8d7e6fb0 | 2251 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2252 | struct kernfs_node *kn; |
2253 | struct lock_class_key *key = NULL; | |
05ef1d7c | 2254 | |
2bd59d48 TH |
2255 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2256 | key = &cft->lockdep_key; | |
2257 | #endif | |
2258 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2259 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2260 | NULL, false, key); | |
2261 | if (IS_ERR(kn)) | |
2262 | return PTR_ERR(kn); | |
2263 | return 0; | |
ddbcc7e8 PM |
2264 | } |
2265 | ||
b1f28d31 TH |
2266 | /** |
2267 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2268 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2269 | * @cfts: array of cftypes to be added |
2270 | * @is_add: whether to add or remove | |
2271 | * | |
2272 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2273 | * For removals, this function never fails. If addition fails, this |
2274 | * function doesn't remove files already added. The caller is responsible | |
2275 | * for cleaning up. | |
b1f28d31 | 2276 | */ |
2bb566cb TH |
2277 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2278 | bool is_add) | |
ddbcc7e8 | 2279 | { |
03b1cde6 | 2280 | struct cftype *cft; |
b1f28d31 TH |
2281 | int ret; |
2282 | ||
ace2bee8 | 2283 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2284 | |
2285 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2286 | /* does cft->flags tell us to skip this file on @cgrp? */ |
873fe09e TH |
2287 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2288 | continue; | |
f33fddc2 G |
2289 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2290 | continue; | |
2291 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2292 | continue; | |
2293 | ||
2739d3cc | 2294 | if (is_add) { |
2bb566cb | 2295 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2296 | if (ret) { |
2739d3cc | 2297 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2298 | cft->name, ret); |
2299 | return ret; | |
2300 | } | |
2739d3cc LZ |
2301 | } else { |
2302 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2303 | } |
ddbcc7e8 | 2304 | } |
b1f28d31 | 2305 | return 0; |
ddbcc7e8 PM |
2306 | } |
2307 | ||
8e3f6541 | 2308 | static void cgroup_cfts_prepare(void) |
e8c82d20 | 2309 | __acquires(&cgroup_mutex) |
8e3f6541 TH |
2310 | { |
2311 | /* | |
2312 | * Thanks to the entanglement with vfs inode locking, we can't walk | |
2313 | * the existing cgroups under cgroup_mutex and create files. | |
492eb21b TH |
2314 | * Instead, we use css_for_each_descendant_pre() and drop RCU read |
2315 | * lock before calling cgroup_addrm_files(). | |
8e3f6541 | 2316 | */ |
ace2bee8 | 2317 | mutex_lock(&cgroup_tree_mutex); |
8e3f6541 TH |
2318 | mutex_lock(&cgroup_mutex); |
2319 | } | |
2320 | ||
2bb566cb | 2321 | static int cgroup_cfts_commit(struct cftype *cfts, bool is_add) |
e8c82d20 | 2322 | __releases(&cgroup_mutex) |
8e3f6541 TH |
2323 | { |
2324 | LIST_HEAD(pending); | |
2bb566cb | 2325 | struct cgroup_subsys *ss = cfts[0].ss; |
492eb21b | 2326 | struct cgroup *root = &ss->root->top_cgroup; |
59f5296b | 2327 | struct cgroup *prev = NULL; |
492eb21b | 2328 | struct cgroup_subsys_state *css; |
00356bd5 | 2329 | u64 update_before; |
9ccece80 | 2330 | int ret = 0; |
8e3f6541 | 2331 | |
4ac06017 TH |
2332 | mutex_unlock(&cgroup_mutex); |
2333 | ||
8e3f6541 | 2334 | /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */ |
2bd59d48 | 2335 | if (!cfts || ss->root == &cgroup_dummy_root) { |
ace2bee8 | 2336 | mutex_unlock(&cgroup_tree_mutex); |
9ccece80 | 2337 | return 0; |
8e3f6541 TH |
2338 | } |
2339 | ||
2bd59d48 TH |
2340 | cgroup_get_root(ss->root); |
2341 | ||
8e3f6541 | 2342 | /* |
e8c82d20 LZ |
2343 | * All cgroups which are created after we drop cgroup_mutex will |
2344 | * have the updated set of files, so we only need to update the | |
00356bd5 | 2345 | * cgroups created before the current @cgroup_serial_nr_next. |
8e3f6541 | 2346 | */ |
00356bd5 | 2347 | update_before = cgroup_serial_nr_next; |
e8c82d20 | 2348 | |
e8c82d20 | 2349 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2350 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2351 | struct cgroup *cgrp = css->cgroup; |
2352 | ||
e8c82d20 LZ |
2353 | if (cgroup_is_dead(cgrp)) |
2354 | continue; | |
2355 | ||
59f5296b TH |
2356 | cgroup_get(cgrp); |
2357 | if (prev) | |
2358 | cgroup_put(prev); | |
2359 | prev = cgrp; | |
8e3f6541 | 2360 | |
2bd59d48 | 2361 | if (cgrp->serial_nr < update_before && !cgroup_is_dead(cgrp)) { |
2bb566cb | 2362 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
2bd59d48 TH |
2363 | if (is_add) |
2364 | kernfs_activate(cgrp->kn); | |
2365 | } | |
9ccece80 TH |
2366 | if (ret) |
2367 | break; | |
8e3f6541 | 2368 | } |
ace2bee8 | 2369 | mutex_unlock(&cgroup_tree_mutex); |
59f5296b TH |
2370 | cgroup_put(prev); |
2371 | cgroup_put_root(ss->root); | |
9ccece80 | 2372 | return ret; |
8e3f6541 TH |
2373 | } |
2374 | ||
2da440a2 TH |
2375 | static void cgroup_exit_cftypes(struct cftype *cfts) |
2376 | { | |
2377 | struct cftype *cft; | |
2378 | ||
2bd59d48 TH |
2379 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2380 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2381 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2382 | kfree(cft->kf_ops); | |
2383 | cft->kf_ops = NULL; | |
2da440a2 | 2384 | cft->ss = NULL; |
2bd59d48 | 2385 | } |
2da440a2 TH |
2386 | } |
2387 | ||
2bd59d48 | 2388 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2389 | { |
2390 | struct cftype *cft; | |
2391 | ||
2bd59d48 TH |
2392 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2393 | struct kernfs_ops *kf_ops; | |
2394 | ||
2395 | if (cft->seq_start) | |
2396 | kf_ops = &cgroup_kf_ops; | |
2397 | else | |
2398 | kf_ops = &cgroup_kf_single_ops; | |
2399 | ||
2400 | /* | |
2401 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2402 | * make a copy of kf_ops to set its atomic_write_len. | |
2403 | */ | |
2404 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2405 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2406 | if (!kf_ops) { | |
2407 | cgroup_exit_cftypes(cfts); | |
2408 | return -ENOMEM; | |
2409 | } | |
2410 | kf_ops->atomic_write_len = cft->max_write_len; | |
2411 | } | |
2412 | ||
2413 | cft->kf_ops = kf_ops; | |
2da440a2 | 2414 | cft->ss = ss; |
2bd59d48 TH |
2415 | } |
2416 | ||
2417 | return 0; | |
2da440a2 TH |
2418 | } |
2419 | ||
80b13586 TH |
2420 | /** |
2421 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
2422 | * @cfts: zero-length name terminated array of cftypes | |
2423 | * | |
2424 | * Unregister @cfts. Files described by @cfts are removed from all | |
2425 | * existing cgroups and all future cgroups won't have them either. This | |
2426 | * function can be called anytime whether @cfts' subsys is attached or not. | |
2427 | * | |
2428 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2429 | * registered. | |
2430 | */ | |
2431 | int cgroup_rm_cftypes(struct cftype *cfts) | |
2432 | { | |
2433 | struct cftype *found = NULL; | |
2434 | struct cftype_set *set; | |
2435 | ||
2436 | if (!cfts || !cfts[0].ss) | |
2437 | return -ENOENT; | |
2438 | ||
2439 | cgroup_cfts_prepare(); | |
2440 | ||
2441 | list_for_each_entry(set, &cfts[0].ss->cftsets, node) { | |
2442 | if (set->cfts == cfts) { | |
2443 | list_del(&set->node); | |
2444 | kfree(set); | |
2445 | found = cfts; | |
2446 | break; | |
2447 | } | |
2448 | } | |
2449 | ||
2450 | cgroup_cfts_commit(found, false); | |
2451 | cgroup_exit_cftypes(cfts); | |
2452 | return found ? 0 : -ENOENT; | |
2453 | } | |
2454 | ||
8e3f6541 TH |
2455 | /** |
2456 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2457 | * @ss: target cgroup subsystem | |
2458 | * @cfts: zero-length name terminated array of cftypes | |
2459 | * | |
2460 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2461 | * existing cgroups to which @ss is attached and all future cgroups will | |
2462 | * have them too. This function can be called anytime whether @ss is | |
2463 | * attached or not. | |
2464 | * | |
2465 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2466 | * function currently returns 0 as long as @cfts registration is successful | |
2467 | * even if some file creation attempts on existing cgroups fail. | |
2468 | */ | |
03b1cde6 | 2469 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 TH |
2470 | { |
2471 | struct cftype_set *set; | |
9ccece80 | 2472 | int ret; |
8e3f6541 TH |
2473 | |
2474 | set = kzalloc(sizeof(*set), GFP_KERNEL); | |
2475 | if (!set) | |
2476 | return -ENOMEM; | |
2477 | ||
2bd59d48 TH |
2478 | ret = cgroup_init_cftypes(ss, cfts); |
2479 | if (ret) | |
2480 | return ret; | |
2bb566cb | 2481 | |
8e3f6541 TH |
2482 | cgroup_cfts_prepare(); |
2483 | set->cfts = cfts; | |
2484 | list_add_tail(&set->node, &ss->cftsets); | |
2bb566cb | 2485 | ret = cgroup_cfts_commit(cfts, true); |
9ccece80 | 2486 | if (ret) |
2bb566cb | 2487 | cgroup_rm_cftypes(cfts); |
9ccece80 | 2488 | return ret; |
8e3f6541 TH |
2489 | } |
2490 | EXPORT_SYMBOL_GPL(cgroup_add_cftypes); | |
2491 | ||
a043e3b2 LZ |
2492 | /** |
2493 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2494 | * @cgrp: the cgroup in question | |
2495 | * | |
2496 | * Return the number of tasks in the cgroup. | |
2497 | */ | |
bd89aabc | 2498 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2499 | { |
2500 | int count = 0; | |
69d0206c | 2501 | struct cgrp_cset_link *link; |
817929ec PM |
2502 | |
2503 | read_lock(&css_set_lock); | |
69d0206c TH |
2504 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2505 | count += atomic_read(&link->cset->refcount); | |
817929ec | 2506 | read_unlock(&css_set_lock); |
bbcb81d0 PM |
2507 | return count; |
2508 | } | |
2509 | ||
817929ec | 2510 | /* |
0942eeee TH |
2511 | * To reduce the fork() overhead for systems that are not actually using |
2512 | * their cgroups capability, we don't maintain the lists running through | |
2513 | * each css_set to its tasks until we see the list actually used - in other | |
72ec7029 | 2514 | * words after the first call to css_task_iter_start(). |
31a7df01 | 2515 | */ |
3df91fe3 | 2516 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2517 | { |
2518 | struct task_struct *p, *g; | |
2519 | write_lock(&css_set_lock); | |
2520 | use_task_css_set_links = 1; | |
3ce3230a FW |
2521 | /* |
2522 | * We need tasklist_lock because RCU is not safe against | |
2523 | * while_each_thread(). Besides, a forking task that has passed | |
2524 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
2525 | * is not guaranteed to have its child immediately visible in the | |
2526 | * tasklist if we walk through it with RCU. | |
2527 | */ | |
2528 | read_lock(&tasklist_lock); | |
31a7df01 CW |
2529 | do_each_thread(g, p) { |
2530 | task_lock(p); | |
0e04388f LZ |
2531 | /* |
2532 | * We should check if the process is exiting, otherwise | |
2533 | * it will race with cgroup_exit() in that the list | |
2534 | * entry won't be deleted though the process has exited. | |
2535 | */ | |
2536 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
a8ad805c | 2537 | list_add(&p->cg_list, &task_css_set(p)->tasks); |
31a7df01 CW |
2538 | task_unlock(p); |
2539 | } while_each_thread(g, p); | |
3ce3230a | 2540 | read_unlock(&tasklist_lock); |
31a7df01 CW |
2541 | write_unlock(&css_set_lock); |
2542 | } | |
2543 | ||
53fa5261 | 2544 | /** |
492eb21b TH |
2545 | * css_next_child - find the next child of a given css |
2546 | * @pos_css: the current position (%NULL to initiate traversal) | |
2547 | * @parent_css: css whose children to walk | |
53fa5261 | 2548 | * |
492eb21b | 2549 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
2550 | * under either cgroup_mutex or RCU read lock. The only requirement is |
2551 | * that @parent_css and @pos_css are accessible. The next sibling is | |
2552 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 2553 | */ |
492eb21b TH |
2554 | struct cgroup_subsys_state * |
2555 | css_next_child(struct cgroup_subsys_state *pos_css, | |
2556 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 2557 | { |
492eb21b TH |
2558 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
2559 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
2560 | struct cgroup *next; |
2561 | ||
ace2bee8 | 2562 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
2563 | |
2564 | /* | |
2565 | * @pos could already have been removed. Once a cgroup is removed, | |
2566 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
2567 | * changes. As CGRP_DEAD assertion is serialized and happens |
2568 | * before the cgroup is taken off the ->sibling list, if we see it | |
2569 | * unasserted, it's guaranteed that the next sibling hasn't | |
2570 | * finished its grace period even if it's already removed, and thus | |
2571 | * safe to dereference from this RCU critical section. If | |
2572 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
2573 | * to be visible as %true here. | |
3b287a50 TH |
2574 | * |
2575 | * If @pos is dead, its next pointer can't be dereferenced; | |
2576 | * however, as each cgroup is given a monotonically increasing | |
2577 | * unique serial number and always appended to the sibling list, | |
2578 | * the next one can be found by walking the parent's children until | |
2579 | * we see a cgroup with higher serial number than @pos's. While | |
2580 | * this path can be slower, it's taken only when either the current | |
2581 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 2582 | */ |
3b287a50 TH |
2583 | if (!pos) { |
2584 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
2585 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 2586 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
2587 | } else { |
2588 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
2589 | if (next->serial_nr > pos->serial_nr) | |
2590 | break; | |
53fa5261 TH |
2591 | } |
2592 | ||
492eb21b TH |
2593 | if (&next->sibling == &cgrp->children) |
2594 | return NULL; | |
2595 | ||
ca8bdcaf | 2596 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 2597 | } |
492eb21b | 2598 | EXPORT_SYMBOL_GPL(css_next_child); |
53fa5261 | 2599 | |
574bd9f7 | 2600 | /** |
492eb21b | 2601 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 2602 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2603 | * @root: css whose descendants to walk |
574bd9f7 | 2604 | * |
492eb21b | 2605 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
2606 | * to visit for pre-order traversal of @root's descendants. @root is |
2607 | * included in the iteration and the first node to be visited. | |
75501a6d | 2608 | * |
87fb54f1 TH |
2609 | * While this function requires cgroup_mutex or RCU read locking, it |
2610 | * doesn't require the whole traversal to be contained in a single critical | |
2611 | * section. This function will return the correct next descendant as long | |
2612 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 2613 | */ |
492eb21b TH |
2614 | struct cgroup_subsys_state * |
2615 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
2616 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2617 | { |
492eb21b | 2618 | struct cgroup_subsys_state *next; |
574bd9f7 | 2619 | |
ace2bee8 | 2620 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2621 | |
bd8815a6 | 2622 | /* if first iteration, visit @root */ |
7805d000 | 2623 | if (!pos) |
bd8815a6 | 2624 | return root; |
574bd9f7 TH |
2625 | |
2626 | /* visit the first child if exists */ | |
492eb21b | 2627 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
2628 | if (next) |
2629 | return next; | |
2630 | ||
2631 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
2632 | while (pos != root) { |
2633 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 2634 | if (next) |
574bd9f7 | 2635 | return next; |
492eb21b | 2636 | pos = css_parent(pos); |
7805d000 | 2637 | } |
574bd9f7 TH |
2638 | |
2639 | return NULL; | |
2640 | } | |
492eb21b | 2641 | EXPORT_SYMBOL_GPL(css_next_descendant_pre); |
574bd9f7 | 2642 | |
12a9d2fe | 2643 | /** |
492eb21b TH |
2644 | * css_rightmost_descendant - return the rightmost descendant of a css |
2645 | * @pos: css of interest | |
12a9d2fe | 2646 | * |
492eb21b TH |
2647 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
2648 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 2649 | * subtree of @pos. |
75501a6d | 2650 | * |
87fb54f1 TH |
2651 | * While this function requires cgroup_mutex or RCU read locking, it |
2652 | * doesn't require the whole traversal to be contained in a single critical | |
2653 | * section. This function will return the correct rightmost descendant as | |
2654 | * long as @pos is accessible. | |
12a9d2fe | 2655 | */ |
492eb21b TH |
2656 | struct cgroup_subsys_state * |
2657 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 2658 | { |
492eb21b | 2659 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 2660 | |
ace2bee8 | 2661 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
2662 | |
2663 | do { | |
2664 | last = pos; | |
2665 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
2666 | pos = NULL; | |
492eb21b | 2667 | css_for_each_child(tmp, last) |
12a9d2fe TH |
2668 | pos = tmp; |
2669 | } while (pos); | |
2670 | ||
2671 | return last; | |
2672 | } | |
492eb21b | 2673 | EXPORT_SYMBOL_GPL(css_rightmost_descendant); |
12a9d2fe | 2674 | |
492eb21b TH |
2675 | static struct cgroup_subsys_state * |
2676 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 2677 | { |
492eb21b | 2678 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
2679 | |
2680 | do { | |
2681 | last = pos; | |
492eb21b | 2682 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
2683 | } while (pos); |
2684 | ||
2685 | return last; | |
2686 | } | |
2687 | ||
2688 | /** | |
492eb21b | 2689 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 2690 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2691 | * @root: css whose descendants to walk |
574bd9f7 | 2692 | * |
492eb21b | 2693 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
2694 | * to visit for post-order traversal of @root's descendants. @root is |
2695 | * included in the iteration and the last node to be visited. | |
75501a6d | 2696 | * |
87fb54f1 TH |
2697 | * While this function requires cgroup_mutex or RCU read locking, it |
2698 | * doesn't require the whole traversal to be contained in a single critical | |
2699 | * section. This function will return the correct next descendant as long | |
2700 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
2701 | * @cgroup. | |
574bd9f7 | 2702 | */ |
492eb21b TH |
2703 | struct cgroup_subsys_state * |
2704 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
2705 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2706 | { |
492eb21b | 2707 | struct cgroup_subsys_state *next; |
574bd9f7 | 2708 | |
ace2bee8 | 2709 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2710 | |
58b79a91 TH |
2711 | /* if first iteration, visit leftmost descendant which may be @root */ |
2712 | if (!pos) | |
2713 | return css_leftmost_descendant(root); | |
574bd9f7 | 2714 | |
bd8815a6 TH |
2715 | /* if we visited @root, we're done */ |
2716 | if (pos == root) | |
2717 | return NULL; | |
2718 | ||
574bd9f7 | 2719 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 2720 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 2721 | if (next) |
492eb21b | 2722 | return css_leftmost_descendant(next); |
574bd9f7 TH |
2723 | |
2724 | /* no sibling left, visit parent */ | |
bd8815a6 | 2725 | return css_parent(pos); |
574bd9f7 | 2726 | } |
492eb21b | 2727 | EXPORT_SYMBOL_GPL(css_next_descendant_post); |
574bd9f7 | 2728 | |
0942eeee | 2729 | /** |
72ec7029 | 2730 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
2731 | * @it: the iterator to advance |
2732 | * | |
2733 | * Advance @it to the next css_set to walk. | |
d515876e | 2734 | */ |
72ec7029 | 2735 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
2736 | { |
2737 | struct list_head *l = it->cset_link; | |
2738 | struct cgrp_cset_link *link; | |
2739 | struct css_set *cset; | |
2740 | ||
2741 | /* Advance to the next non-empty css_set */ | |
2742 | do { | |
2743 | l = l->next; | |
72ec7029 | 2744 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
2745 | it->cset_link = NULL; |
2746 | return; | |
2747 | } | |
2748 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
2749 | cset = link->cset; | |
2750 | } while (list_empty(&cset->tasks)); | |
2751 | it->cset_link = l; | |
2752 | it->task = cset->tasks.next; | |
2753 | } | |
2754 | ||
0942eeee | 2755 | /** |
72ec7029 TH |
2756 | * css_task_iter_start - initiate task iteration |
2757 | * @css: the css to walk tasks of | |
0942eeee TH |
2758 | * @it: the task iterator to use |
2759 | * | |
72ec7029 TH |
2760 | * Initiate iteration through the tasks of @css. The caller can call |
2761 | * css_task_iter_next() to walk through the tasks until the function | |
2762 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
2763 | * called. | |
0942eeee TH |
2764 | * |
2765 | * Note that this function acquires a lock which is released when the | |
2766 | * iteration finishes. The caller can't sleep while iteration is in | |
2767 | * progress. | |
2768 | */ | |
72ec7029 TH |
2769 | void css_task_iter_start(struct cgroup_subsys_state *css, |
2770 | struct css_task_iter *it) | |
c6ca5750 | 2771 | __acquires(css_set_lock) |
817929ec PM |
2772 | { |
2773 | /* | |
72ec7029 TH |
2774 | * The first time anyone tries to iterate across a css, we need to |
2775 | * enable the list linking each css_set to its tasks, and fix up | |
2776 | * all existing tasks. | |
817929ec | 2777 | */ |
31a7df01 CW |
2778 | if (!use_task_css_set_links) |
2779 | cgroup_enable_task_cg_lists(); | |
2780 | ||
817929ec | 2781 | read_lock(&css_set_lock); |
c59cd3d8 | 2782 | |
72ec7029 TH |
2783 | it->origin_css = css; |
2784 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 2785 | |
72ec7029 | 2786 | css_advance_task_iter(it); |
817929ec PM |
2787 | } |
2788 | ||
0942eeee | 2789 | /** |
72ec7029 | 2790 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
2791 | * @it: the task iterator being iterated |
2792 | * | |
2793 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
2794 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
2795 | * reaches the end. | |
0942eeee | 2796 | */ |
72ec7029 | 2797 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
2798 | { |
2799 | struct task_struct *res; | |
2800 | struct list_head *l = it->task; | |
69d0206c | 2801 | struct cgrp_cset_link *link; |
817929ec PM |
2802 | |
2803 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 2804 | if (!it->cset_link) |
817929ec PM |
2805 | return NULL; |
2806 | res = list_entry(l, struct task_struct, cg_list); | |
2807 | /* Advance iterator to find next entry */ | |
2808 | l = l->next; | |
69d0206c TH |
2809 | link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link); |
2810 | if (l == &link->cset->tasks) { | |
0942eeee TH |
2811 | /* |
2812 | * We reached the end of this task list - move on to the | |
2813 | * next cgrp_cset_link. | |
2814 | */ | |
72ec7029 | 2815 | css_advance_task_iter(it); |
817929ec PM |
2816 | } else { |
2817 | it->task = l; | |
2818 | } | |
2819 | return res; | |
2820 | } | |
2821 | ||
0942eeee | 2822 | /** |
72ec7029 | 2823 | * css_task_iter_end - finish task iteration |
0942eeee TH |
2824 | * @it: the task iterator to finish |
2825 | * | |
72ec7029 | 2826 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 2827 | */ |
72ec7029 | 2828 | void css_task_iter_end(struct css_task_iter *it) |
c6ca5750 | 2829 | __releases(css_set_lock) |
817929ec PM |
2830 | { |
2831 | read_unlock(&css_set_lock); | |
2832 | } | |
2833 | ||
31a7df01 CW |
2834 | static inline int started_after_time(struct task_struct *t1, |
2835 | struct timespec *time, | |
2836 | struct task_struct *t2) | |
2837 | { | |
2838 | int start_diff = timespec_compare(&t1->start_time, time); | |
2839 | if (start_diff > 0) { | |
2840 | return 1; | |
2841 | } else if (start_diff < 0) { | |
2842 | return 0; | |
2843 | } else { | |
2844 | /* | |
2845 | * Arbitrarily, if two processes started at the same | |
2846 | * time, we'll say that the lower pointer value | |
2847 | * started first. Note that t2 may have exited by now | |
2848 | * so this may not be a valid pointer any longer, but | |
2849 | * that's fine - it still serves to distinguish | |
2850 | * between two tasks started (effectively) simultaneously. | |
2851 | */ | |
2852 | return t1 > t2; | |
2853 | } | |
2854 | } | |
2855 | ||
2856 | /* | |
2857 | * This function is a callback from heap_insert() and is used to order | |
2858 | * the heap. | |
2859 | * In this case we order the heap in descending task start time. | |
2860 | */ | |
2861 | static inline int started_after(void *p1, void *p2) | |
2862 | { | |
2863 | struct task_struct *t1 = p1; | |
2864 | struct task_struct *t2 = p2; | |
2865 | return started_after_time(t1, &t2->start_time, t2); | |
2866 | } | |
2867 | ||
2868 | /** | |
72ec7029 TH |
2869 | * css_scan_tasks - iterate though all the tasks in a css |
2870 | * @css: the css to iterate tasks of | |
e535837b TH |
2871 | * @test: optional test callback |
2872 | * @process: process callback | |
2873 | * @data: data passed to @test and @process | |
2874 | * @heap: optional pre-allocated heap used for task iteration | |
31a7df01 | 2875 | * |
72ec7029 TH |
2876 | * Iterate through all the tasks in @css, calling @test for each, and if it |
2877 | * returns %true, call @process for it also. | |
31a7df01 | 2878 | * |
e535837b | 2879 | * @test may be NULL, meaning always true (select all tasks), which |
72ec7029 | 2880 | * effectively duplicates css_task_iter_{start,next,end}() but does not |
e535837b TH |
2881 | * lock css_set_lock for the call to @process. |
2882 | * | |
2883 | * It is guaranteed that @process will act on every task that is a member | |
72ec7029 TH |
2884 | * of @css for the duration of this call. This function may or may not |
2885 | * call @process for tasks that exit or move to a different css during the | |
2886 | * call, or are forked or move into the css during the call. | |
31a7df01 | 2887 | * |
e535837b TH |
2888 | * Note that @test may be called with locks held, and may in some |
2889 | * situations be called multiple times for the same task, so it should be | |
2890 | * cheap. | |
31a7df01 | 2891 | * |
e535837b TH |
2892 | * If @heap is non-NULL, a heap has been pre-allocated and will be used for |
2893 | * heap operations (and its "gt" member will be overwritten), else a | |
2894 | * temporary heap will be used (allocation of which may cause this function | |
2895 | * to fail). | |
31a7df01 | 2896 | */ |
72ec7029 TH |
2897 | int css_scan_tasks(struct cgroup_subsys_state *css, |
2898 | bool (*test)(struct task_struct *, void *), | |
2899 | void (*process)(struct task_struct *, void *), | |
2900 | void *data, struct ptr_heap *heap) | |
31a7df01 CW |
2901 | { |
2902 | int retval, i; | |
72ec7029 | 2903 | struct css_task_iter it; |
31a7df01 CW |
2904 | struct task_struct *p, *dropped; |
2905 | /* Never dereference latest_task, since it's not refcounted */ | |
2906 | struct task_struct *latest_task = NULL; | |
2907 | struct ptr_heap tmp_heap; | |
31a7df01 CW |
2908 | struct timespec latest_time = { 0, 0 }; |
2909 | ||
e535837b | 2910 | if (heap) { |
31a7df01 | 2911 | /* The caller supplied our heap and pre-allocated its memory */ |
31a7df01 CW |
2912 | heap->gt = &started_after; |
2913 | } else { | |
2914 | /* We need to allocate our own heap memory */ | |
2915 | heap = &tmp_heap; | |
2916 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
2917 | if (retval) | |
2918 | /* cannot allocate the heap */ | |
2919 | return retval; | |
2920 | } | |
2921 | ||
2922 | again: | |
2923 | /* | |
72ec7029 | 2924 | * Scan tasks in the css, using the @test callback to determine |
e535837b TH |
2925 | * which are of interest, and invoking @process callback on the |
2926 | * ones which need an update. Since we don't want to hold any | |
2927 | * locks during the task updates, gather tasks to be processed in a | |
2928 | * heap structure. The heap is sorted by descending task start | |
2929 | * time. If the statically-sized heap fills up, we overflow tasks | |
2930 | * that started later, and in future iterations only consider tasks | |
2931 | * that started after the latest task in the previous pass. This | |
31a7df01 CW |
2932 | * guarantees forward progress and that we don't miss any tasks. |
2933 | */ | |
2934 | heap->size = 0; | |
72ec7029 TH |
2935 | css_task_iter_start(css, &it); |
2936 | while ((p = css_task_iter_next(&it))) { | |
31a7df01 CW |
2937 | /* |
2938 | * Only affect tasks that qualify per the caller's callback, | |
2939 | * if he provided one | |
2940 | */ | |
e535837b | 2941 | if (test && !test(p, data)) |
31a7df01 CW |
2942 | continue; |
2943 | /* | |
2944 | * Only process tasks that started after the last task | |
2945 | * we processed | |
2946 | */ | |
2947 | if (!started_after_time(p, &latest_time, latest_task)) | |
2948 | continue; | |
2949 | dropped = heap_insert(heap, p); | |
2950 | if (dropped == NULL) { | |
2951 | /* | |
2952 | * The new task was inserted; the heap wasn't | |
2953 | * previously full | |
2954 | */ | |
2955 | get_task_struct(p); | |
2956 | } else if (dropped != p) { | |
2957 | /* | |
2958 | * The new task was inserted, and pushed out a | |
2959 | * different task | |
2960 | */ | |
2961 | get_task_struct(p); | |
2962 | put_task_struct(dropped); | |
2963 | } | |
2964 | /* | |
2965 | * Else the new task was newer than anything already in | |
2966 | * the heap and wasn't inserted | |
2967 | */ | |
2968 | } | |
72ec7029 | 2969 | css_task_iter_end(&it); |
31a7df01 CW |
2970 | |
2971 | if (heap->size) { | |
2972 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 2973 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 2974 | if (i == 0) { |
4fe91d51 PJ |
2975 | latest_time = q->start_time; |
2976 | latest_task = q; | |
31a7df01 CW |
2977 | } |
2978 | /* Process the task per the caller's callback */ | |
e535837b | 2979 | process(q, data); |
4fe91d51 | 2980 | put_task_struct(q); |
31a7df01 CW |
2981 | } |
2982 | /* | |
2983 | * If we had to process any tasks at all, scan again | |
2984 | * in case some of them were in the middle of forking | |
2985 | * children that didn't get processed. | |
2986 | * Not the most efficient way to do it, but it avoids | |
2987 | * having to take callback_mutex in the fork path | |
2988 | */ | |
2989 | goto again; | |
2990 | } | |
2991 | if (heap == &tmp_heap) | |
2992 | heap_free(&tmp_heap); | |
2993 | return 0; | |
2994 | } | |
2995 | ||
e535837b | 2996 | static void cgroup_transfer_one_task(struct task_struct *task, void *data) |
8cc99345 | 2997 | { |
e535837b | 2998 | struct cgroup *new_cgroup = data; |
8cc99345 | 2999 | |
47cfcd09 | 3000 | mutex_lock(&cgroup_mutex); |
8cc99345 | 3001 | cgroup_attach_task(new_cgroup, task, false); |
47cfcd09 | 3002 | mutex_unlock(&cgroup_mutex); |
8cc99345 TH |
3003 | } |
3004 | ||
3005 | /** | |
3006 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
3007 | * @to: cgroup to which the tasks will be moved | |
3008 | * @from: cgroup in which the tasks currently reside | |
3009 | */ | |
3010 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
3011 | { | |
72ec7029 TH |
3012 | return css_scan_tasks(&from->dummy_css, NULL, cgroup_transfer_one_task, |
3013 | to, NULL); | |
8cc99345 TH |
3014 | } |
3015 | ||
bbcb81d0 | 3016 | /* |
102a775e | 3017 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3018 | * |
3019 | * Reading this file can return large amounts of data if a cgroup has | |
3020 | * *lots* of attached tasks. So it may need several calls to read(), | |
3021 | * but we cannot guarantee that the information we produce is correct | |
3022 | * unless we produce it entirely atomically. | |
3023 | * | |
bbcb81d0 | 3024 | */ |
bbcb81d0 | 3025 | |
24528255 LZ |
3026 | /* which pidlist file are we talking about? */ |
3027 | enum cgroup_filetype { | |
3028 | CGROUP_FILE_PROCS, | |
3029 | CGROUP_FILE_TASKS, | |
3030 | }; | |
3031 | ||
3032 | /* | |
3033 | * A pidlist is a list of pids that virtually represents the contents of one | |
3034 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3035 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3036 | * to the cgroup. | |
3037 | */ | |
3038 | struct cgroup_pidlist { | |
3039 | /* | |
3040 | * used to find which pidlist is wanted. doesn't change as long as | |
3041 | * this particular list stays in the list. | |
3042 | */ | |
3043 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3044 | /* array of xids */ | |
3045 | pid_t *list; | |
3046 | /* how many elements the above list has */ | |
3047 | int length; | |
24528255 LZ |
3048 | /* each of these stored in a list by its cgroup */ |
3049 | struct list_head links; | |
3050 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3051 | struct cgroup *owner; | |
b1a21367 TH |
3052 | /* for delayed destruction */ |
3053 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3054 | }; |
3055 | ||
d1d9fd33 BB |
3056 | /* |
3057 | * The following two functions "fix" the issue where there are more pids | |
3058 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3059 | * TODO: replace with a kernel-wide solution to this problem | |
3060 | */ | |
3061 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3062 | static void *pidlist_allocate(int count) | |
3063 | { | |
3064 | if (PIDLIST_TOO_LARGE(count)) | |
3065 | return vmalloc(count * sizeof(pid_t)); | |
3066 | else | |
3067 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3068 | } | |
b1a21367 | 3069 | |
d1d9fd33 BB |
3070 | static void pidlist_free(void *p) |
3071 | { | |
3072 | if (is_vmalloc_addr(p)) | |
3073 | vfree(p); | |
3074 | else | |
3075 | kfree(p); | |
3076 | } | |
d1d9fd33 | 3077 | |
b1a21367 TH |
3078 | /* |
3079 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3080 | * should be left afterwards. | |
3081 | */ | |
3082 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3083 | { | |
3084 | struct cgroup_pidlist *l, *tmp_l; | |
3085 | ||
3086 | mutex_lock(&cgrp->pidlist_mutex); | |
3087 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3088 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3089 | mutex_unlock(&cgrp->pidlist_mutex); | |
3090 | ||
3091 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3092 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3093 | } | |
3094 | ||
3095 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3096 | { | |
3097 | struct delayed_work *dwork = to_delayed_work(work); | |
3098 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3099 | destroy_dwork); | |
3100 | struct cgroup_pidlist *tofree = NULL; | |
3101 | ||
3102 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3103 | |
3104 | /* | |
04502365 TH |
3105 | * Destroy iff we didn't get queued again. The state won't change |
3106 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3107 | */ |
04502365 | 3108 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3109 | list_del(&l->links); |
3110 | pidlist_free(l->list); | |
3111 | put_pid_ns(l->key.ns); | |
3112 | tofree = l; | |
3113 | } | |
3114 | ||
b1a21367 TH |
3115 | mutex_unlock(&l->owner->pidlist_mutex); |
3116 | kfree(tofree); | |
3117 | } | |
3118 | ||
bbcb81d0 | 3119 | /* |
102a775e | 3120 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3121 | * Returns the number of unique elements. |
bbcb81d0 | 3122 | */ |
6ee211ad | 3123 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3124 | { |
102a775e | 3125 | int src, dest = 1; |
102a775e BB |
3126 | |
3127 | /* | |
3128 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3129 | * edge cases first; no work needs to be done for either | |
3130 | */ | |
3131 | if (length == 0 || length == 1) | |
3132 | return length; | |
3133 | /* src and dest walk down the list; dest counts unique elements */ | |
3134 | for (src = 1; src < length; src++) { | |
3135 | /* find next unique element */ | |
3136 | while (list[src] == list[src-1]) { | |
3137 | src++; | |
3138 | if (src == length) | |
3139 | goto after; | |
3140 | } | |
3141 | /* dest always points to where the next unique element goes */ | |
3142 | list[dest] = list[src]; | |
3143 | dest++; | |
3144 | } | |
3145 | after: | |
102a775e BB |
3146 | return dest; |
3147 | } | |
3148 | ||
afb2bc14 TH |
3149 | /* |
3150 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3151 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3152 | * different per namespace, each namespace needs differently sorted list, | |
3153 | * making it impossible to use, for example, single rbtree of member tasks | |
3154 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3155 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3156 | * pidlists keyed by cgroup and namespace. | |
3157 | * | |
3158 | * All this extra complexity was caused by the original implementation | |
3159 | * committing to an entirely unnecessary property. In the long term, we | |
3160 | * want to do away with it. Explicitly scramble sort order if | |
3161 | * sane_behavior so that no such expectation exists in the new interface. | |
3162 | * | |
3163 | * Scrambling is done by swapping every two consecutive bits, which is | |
3164 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3165 | */ | |
3166 | static pid_t pid_fry(pid_t pid) | |
3167 | { | |
3168 | unsigned a = pid & 0x55555555; | |
3169 | unsigned b = pid & 0xAAAAAAAA; | |
3170 | ||
3171 | return (a << 1) | (b >> 1); | |
3172 | } | |
3173 | ||
3174 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3175 | { | |
3176 | if (cgroup_sane_behavior(cgrp)) | |
3177 | return pid_fry(pid); | |
3178 | else | |
3179 | return pid; | |
3180 | } | |
3181 | ||
102a775e BB |
3182 | static int cmppid(const void *a, const void *b) |
3183 | { | |
3184 | return *(pid_t *)a - *(pid_t *)b; | |
3185 | } | |
3186 | ||
afb2bc14 TH |
3187 | static int fried_cmppid(const void *a, const void *b) |
3188 | { | |
3189 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3190 | } | |
3191 | ||
e6b81710 TH |
3192 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3193 | enum cgroup_filetype type) | |
3194 | { | |
3195 | struct cgroup_pidlist *l; | |
3196 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3197 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3198 | ||
3199 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3200 | ||
3201 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3202 | if (l->key.type == type && l->key.ns == ns) | |
3203 | return l; | |
3204 | return NULL; | |
3205 | } | |
3206 | ||
72a8cb30 BB |
3207 | /* |
3208 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3209 | * returns with the lock on that pidlist already held, and takes care | |
3210 | * of the use count, or returns NULL with no locks held if we're out of | |
3211 | * memory. | |
3212 | */ | |
e6b81710 TH |
3213 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3214 | enum cgroup_filetype type) | |
72a8cb30 BB |
3215 | { |
3216 | struct cgroup_pidlist *l; | |
b70cc5fd | 3217 | |
e6b81710 TH |
3218 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3219 | ||
3220 | l = cgroup_pidlist_find(cgrp, type); | |
3221 | if (l) | |
3222 | return l; | |
3223 | ||
72a8cb30 | 3224 | /* entry not found; create a new one */ |
f4f4be2b | 3225 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3226 | if (!l) |
72a8cb30 | 3227 | return l; |
e6b81710 | 3228 | |
b1a21367 | 3229 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3230 | l->key.type = type; |
e6b81710 TH |
3231 | /* don't need task_nsproxy() if we're looking at ourself */ |
3232 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3233 | l->owner = cgrp; |
3234 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3235 | return l; |
3236 | } | |
3237 | ||
102a775e BB |
3238 | /* |
3239 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3240 | */ | |
72a8cb30 BB |
3241 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3242 | struct cgroup_pidlist **lp) | |
102a775e BB |
3243 | { |
3244 | pid_t *array; | |
3245 | int length; | |
3246 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3247 | struct css_task_iter it; |
817929ec | 3248 | struct task_struct *tsk; |
102a775e BB |
3249 | struct cgroup_pidlist *l; |
3250 | ||
4bac00d1 TH |
3251 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3252 | ||
102a775e BB |
3253 | /* |
3254 | * If cgroup gets more users after we read count, we won't have | |
3255 | * enough space - tough. This race is indistinguishable to the | |
3256 | * caller from the case that the additional cgroup users didn't | |
3257 | * show up until sometime later on. | |
3258 | */ | |
3259 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3260 | array = pidlist_allocate(length); |
102a775e BB |
3261 | if (!array) |
3262 | return -ENOMEM; | |
3263 | /* now, populate the array */ | |
72ec7029 TH |
3264 | css_task_iter_start(&cgrp->dummy_css, &it); |
3265 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3266 | if (unlikely(n == length)) |
817929ec | 3267 | break; |
102a775e | 3268 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3269 | if (type == CGROUP_FILE_PROCS) |
3270 | pid = task_tgid_vnr(tsk); | |
3271 | else | |
3272 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3273 | if (pid > 0) /* make sure to only use valid results */ |
3274 | array[n++] = pid; | |
817929ec | 3275 | } |
72ec7029 | 3276 | css_task_iter_end(&it); |
102a775e BB |
3277 | length = n; |
3278 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3279 | if (cgroup_sane_behavior(cgrp)) |
3280 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3281 | else | |
3282 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3283 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3284 | length = pidlist_uniq(array, length); |
e6b81710 | 3285 | |
e6b81710 | 3286 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3287 | if (!l) { |
e6b81710 | 3288 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3289 | pidlist_free(array); |
72a8cb30 | 3290 | return -ENOMEM; |
102a775e | 3291 | } |
e6b81710 TH |
3292 | |
3293 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3294 | pidlist_free(l->list); |
102a775e BB |
3295 | l->list = array; |
3296 | l->length = length; | |
72a8cb30 | 3297 | *lp = l; |
102a775e | 3298 | return 0; |
bbcb81d0 PM |
3299 | } |
3300 | ||
846c7bb0 | 3301 | /** |
a043e3b2 | 3302 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3303 | * @stats: cgroupstats to fill information into |
3304 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3305 | * been requested. | |
a043e3b2 LZ |
3306 | * |
3307 | * Build and fill cgroupstats so that taskstats can export it to user | |
3308 | * space. | |
846c7bb0 BS |
3309 | */ |
3310 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3311 | { | |
2bd59d48 | 3312 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3313 | struct cgroup *cgrp; |
72ec7029 | 3314 | struct css_task_iter it; |
846c7bb0 | 3315 | struct task_struct *tsk; |
33d283be | 3316 | |
2bd59d48 TH |
3317 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3318 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3319 | kernfs_type(kn) != KERNFS_DIR) | |
3320 | return -EINVAL; | |
3321 | ||
846c7bb0 | 3322 | /* |
2bd59d48 TH |
3323 | * We aren't being called from kernfs and there's no guarantee on |
3324 | * @kn->priv's validity. For this and css_tryget_from_dir(), | |
3325 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
846c7bb0 | 3326 | */ |
2bd59d48 TH |
3327 | rcu_read_lock(); |
3328 | cgrp = rcu_dereference(kn->priv); | |
3329 | if (!cgrp) { | |
3330 | rcu_read_unlock(); | |
3331 | return -ENOENT; | |
3332 | } | |
846c7bb0 | 3333 | |
72ec7029 TH |
3334 | css_task_iter_start(&cgrp->dummy_css, &it); |
3335 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3336 | switch (tsk->state) { |
3337 | case TASK_RUNNING: | |
3338 | stats->nr_running++; | |
3339 | break; | |
3340 | case TASK_INTERRUPTIBLE: | |
3341 | stats->nr_sleeping++; | |
3342 | break; | |
3343 | case TASK_UNINTERRUPTIBLE: | |
3344 | stats->nr_uninterruptible++; | |
3345 | break; | |
3346 | case TASK_STOPPED: | |
3347 | stats->nr_stopped++; | |
3348 | break; | |
3349 | default: | |
3350 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3351 | stats->nr_io_wait++; | |
3352 | break; | |
3353 | } | |
3354 | } | |
72ec7029 | 3355 | css_task_iter_end(&it); |
846c7bb0 | 3356 | |
2bd59d48 TH |
3357 | rcu_read_unlock(); |
3358 | return 0; | |
846c7bb0 BS |
3359 | } |
3360 | ||
8f3ff208 | 3361 | |
bbcb81d0 | 3362 | /* |
102a775e | 3363 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3364 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3365 | * in the cgroup->l->list array. |
bbcb81d0 | 3366 | */ |
cc31edce | 3367 | |
102a775e | 3368 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3369 | { |
cc31edce PM |
3370 | /* |
3371 | * Initially we receive a position value that corresponds to | |
3372 | * one more than the last pid shown (or 0 on the first call or | |
3373 | * after a seek to the start). Use a binary-search to find the | |
3374 | * next pid to display, if any | |
3375 | */ | |
2bd59d48 | 3376 | struct kernfs_open_file *of = s->private; |
7da11279 | 3377 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3378 | struct cgroup_pidlist *l; |
7da11279 | 3379 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3380 | int index = 0, pid = *pos; |
4bac00d1 TH |
3381 | int *iter, ret; |
3382 | ||
3383 | mutex_lock(&cgrp->pidlist_mutex); | |
3384 | ||
3385 | /* | |
5d22444f | 3386 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3387 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3388 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3389 | * could already have been destroyed. |
3390 | */ | |
5d22444f TH |
3391 | if (of->priv) |
3392 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3393 | |
3394 | /* | |
3395 | * Either this is the first start() after open or the matching | |
3396 | * pidlist has been destroyed inbetween. Create a new one. | |
3397 | */ | |
5d22444f TH |
3398 | if (!of->priv) { |
3399 | ret = pidlist_array_load(cgrp, type, | |
3400 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3401 | if (ret) |
3402 | return ERR_PTR(ret); | |
3403 | } | |
5d22444f | 3404 | l = of->priv; |
cc31edce | 3405 | |
cc31edce | 3406 | if (pid) { |
102a775e | 3407 | int end = l->length; |
20777766 | 3408 | |
cc31edce PM |
3409 | while (index < end) { |
3410 | int mid = (index + end) / 2; | |
afb2bc14 | 3411 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3412 | index = mid; |
3413 | break; | |
afb2bc14 | 3414 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3415 | index = mid + 1; |
3416 | else | |
3417 | end = mid; | |
3418 | } | |
3419 | } | |
3420 | /* If we're off the end of the array, we're done */ | |
102a775e | 3421 | if (index >= l->length) |
cc31edce PM |
3422 | return NULL; |
3423 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3424 | iter = l->list + index; |
afb2bc14 | 3425 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3426 | return iter; |
3427 | } | |
3428 | ||
102a775e | 3429 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3430 | { |
2bd59d48 | 3431 | struct kernfs_open_file *of = s->private; |
5d22444f | 3432 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3433 | |
5d22444f TH |
3434 | if (l) |
3435 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3436 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3437 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3438 | } |
3439 | ||
102a775e | 3440 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3441 | { |
2bd59d48 | 3442 | struct kernfs_open_file *of = s->private; |
5d22444f | 3443 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3444 | pid_t *p = v; |
3445 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3446 | /* |
3447 | * Advance to the next pid in the array. If this goes off the | |
3448 | * end, we're done | |
3449 | */ | |
3450 | p++; | |
3451 | if (p >= end) { | |
3452 | return NULL; | |
3453 | } else { | |
7da11279 | 3454 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3455 | return p; |
3456 | } | |
3457 | } | |
3458 | ||
102a775e | 3459 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3460 | { |
3461 | return seq_printf(s, "%d\n", *(int *)v); | |
3462 | } | |
bbcb81d0 | 3463 | |
102a775e BB |
3464 | /* |
3465 | * seq_operations functions for iterating on pidlists through seq_file - | |
3466 | * independent of whether it's tasks or procs | |
3467 | */ | |
3468 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3469 | .start = cgroup_pidlist_start, | |
3470 | .stop = cgroup_pidlist_stop, | |
3471 | .next = cgroup_pidlist_next, | |
3472 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3473 | }; |
3474 | ||
182446d0 TH |
3475 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3476 | struct cftype *cft) | |
81a6a5cd | 3477 | { |
182446d0 | 3478 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3479 | } |
3480 | ||
182446d0 TH |
3481 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3482 | struct cftype *cft, u64 val) | |
6379c106 | 3483 | { |
182446d0 | 3484 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3485 | if (val) |
182446d0 | 3486 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3487 | else |
182446d0 | 3488 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3489 | return 0; |
3490 | } | |
3491 | ||
182446d0 TH |
3492 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3493 | struct cftype *cft) | |
97978e6d | 3494 | { |
182446d0 | 3495 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3496 | } |
3497 | ||
182446d0 TH |
3498 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3499 | struct cftype *cft, u64 val) | |
97978e6d DL |
3500 | { |
3501 | if (val) | |
182446d0 | 3502 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3503 | else |
182446d0 | 3504 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3505 | return 0; |
3506 | } | |
3507 | ||
d5c56ced | 3508 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3509 | { |
d5c56ced | 3510 | .name = "cgroup.procs", |
6612f05b TH |
3511 | .seq_start = cgroup_pidlist_start, |
3512 | .seq_next = cgroup_pidlist_next, | |
3513 | .seq_stop = cgroup_pidlist_stop, | |
3514 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3515 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3516 | .write_u64 = cgroup_procs_write, |
74a1166d | 3517 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3518 | }, |
97978e6d DL |
3519 | { |
3520 | .name = "cgroup.clone_children", | |
873fe09e | 3521 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3522 | .read_u64 = cgroup_clone_children_read, |
3523 | .write_u64 = cgroup_clone_children_write, | |
3524 | }, | |
873fe09e TH |
3525 | { |
3526 | .name = "cgroup.sane_behavior", | |
3527 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3528 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3529 | }, |
d5c56ced TH |
3530 | |
3531 | /* | |
3532 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3533 | * don't exist if sane_behavior. If you're depending on these, be | |
3534 | * prepared to be burned. | |
3535 | */ | |
3536 | { | |
3537 | .name = "tasks", | |
3538 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3539 | .seq_start = cgroup_pidlist_start, |
3540 | .seq_next = cgroup_pidlist_next, | |
3541 | .seq_stop = cgroup_pidlist_stop, | |
3542 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3543 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3544 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3545 | .mode = S_IRUGO | S_IWUSR, |
3546 | }, | |
3547 | { | |
3548 | .name = "notify_on_release", | |
3549 | .flags = CFTYPE_INSANE, | |
3550 | .read_u64 = cgroup_read_notify_on_release, | |
3551 | .write_u64 = cgroup_write_notify_on_release, | |
3552 | }, | |
6e6ff25b TH |
3553 | { |
3554 | .name = "release_agent", | |
cc5943a7 | 3555 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3556 | .seq_show = cgroup_release_agent_show, |
6e6ff25b | 3557 | .write_string = cgroup_release_agent_write, |
5f469907 | 3558 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 3559 | }, |
db0416b6 | 3560 | { } /* terminate */ |
bbcb81d0 PM |
3561 | }; |
3562 | ||
13af07df | 3563 | /** |
628f7cd4 | 3564 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3565 | * @cgrp: target cgroup |
13af07df | 3566 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3567 | * |
3568 | * On failure, no file is added. | |
13af07df | 3569 | */ |
628f7cd4 | 3570 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3571 | { |
ddbcc7e8 | 3572 | struct cgroup_subsys *ss; |
b420ba7d | 3573 | int i, ret = 0; |
bbcb81d0 | 3574 | |
8e3f6541 | 3575 | /* process cftsets of each subsystem */ |
b420ba7d | 3576 | for_each_subsys(ss, i) { |
8e3f6541 | 3577 | struct cftype_set *set; |
b420ba7d TH |
3578 | |
3579 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3580 | continue; |
8e3f6541 | 3581 | |
bee55099 | 3582 | list_for_each_entry(set, &ss->cftsets, node) { |
2bb566cb | 3583 | ret = cgroup_addrm_files(cgrp, set->cfts, true); |
bee55099 TH |
3584 | if (ret < 0) |
3585 | goto err; | |
3586 | } | |
ddbcc7e8 | 3587 | } |
ddbcc7e8 | 3588 | return 0; |
bee55099 TH |
3589 | err: |
3590 | cgroup_clear_dir(cgrp, subsys_mask); | |
3591 | return ret; | |
ddbcc7e8 PM |
3592 | } |
3593 | ||
0c21ead1 TH |
3594 | /* |
3595 | * css destruction is four-stage process. | |
3596 | * | |
3597 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3598 | * Implemented in kill_css(). | |
3599 | * | |
3600 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3601 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3602 | * by invoking offline_css(). After offlining, the base ref is put. | |
3603 | * Implemented in css_killed_work_fn(). | |
3604 | * | |
3605 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3606 | * accessors are inside RCU read sections. css_release() schedules the | |
3607 | * RCU callback. | |
3608 | * | |
3609 | * 4. After the grace period, the css can be freed. Implemented in | |
3610 | * css_free_work_fn(). | |
3611 | * | |
3612 | * It is actually hairier because both step 2 and 4 require process context | |
3613 | * and thus involve punting to css->destroy_work adding two additional | |
3614 | * steps to the already complex sequence. | |
3615 | */ | |
35ef10da | 3616 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3617 | { |
3618 | struct cgroup_subsys_state *css = | |
35ef10da | 3619 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 3620 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3621 | |
0ae78e0b TH |
3622 | if (css->parent) |
3623 | css_put(css->parent); | |
3624 | ||
0c21ead1 | 3625 | css->ss->css_free(css); |
2bd59d48 | 3626 | cgroup_put(cgrp); |
48ddbe19 TH |
3627 | } |
3628 | ||
0c21ead1 | 3629 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3630 | { |
3631 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3632 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3633 | |
35ef10da | 3634 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 3635 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
3636 | } |
3637 | ||
d3daf28d TH |
3638 | static void css_release(struct percpu_ref *ref) |
3639 | { | |
3640 | struct cgroup_subsys_state *css = | |
3641 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3642 | ||
aec25020 | 3643 | rcu_assign_pointer(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 3644 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
3645 | } |
3646 | ||
623f926b TH |
3647 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
3648 | struct cgroup *cgrp) | |
ddbcc7e8 | 3649 | { |
bd89aabc | 3650 | css->cgroup = cgrp; |
72c97e54 | 3651 | css->ss = ss; |
ddbcc7e8 | 3652 | css->flags = 0; |
0ae78e0b TH |
3653 | |
3654 | if (cgrp->parent) | |
ca8bdcaf | 3655 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 3656 | else |
38b53aba | 3657 | css->flags |= CSS_ROOT; |
48ddbe19 | 3658 | |
ca8bdcaf | 3659 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
3660 | } |
3661 | ||
2a4ac633 | 3662 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 3663 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3664 | { |
623f926b | 3665 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
3666 | int ret = 0; |
3667 | ||
ace2bee8 | 3668 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3669 | lockdep_assert_held(&cgroup_mutex); |
3670 | ||
92fb9748 | 3671 | if (ss->css_online) |
eb95419b | 3672 | ret = ss->css_online(css); |
ae7f164a | 3673 | if (!ret) { |
eb95419b | 3674 | css->flags |= CSS_ONLINE; |
f20104de | 3675 | css->cgroup->nr_css++; |
aec25020 | 3676 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 3677 | } |
b1929db4 | 3678 | return ret; |
a31f2d3f TH |
3679 | } |
3680 | ||
2a4ac633 | 3681 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 3682 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3683 | { |
623f926b | 3684 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 3685 | |
ace2bee8 | 3686 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3687 | lockdep_assert_held(&cgroup_mutex); |
3688 | ||
3689 | if (!(css->flags & CSS_ONLINE)) | |
3690 | return; | |
3691 | ||
d7eeac19 | 3692 | if (ss->css_offline) |
eb95419b | 3693 | ss->css_offline(css); |
a31f2d3f | 3694 | |
eb95419b | 3695 | css->flags &= ~CSS_ONLINE; |
09a503ea | 3696 | css->cgroup->nr_css--; |
aec25020 | 3697 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); |
a31f2d3f TH |
3698 | } |
3699 | ||
c81c925a TH |
3700 | /** |
3701 | * create_css - create a cgroup_subsys_state | |
3702 | * @cgrp: the cgroup new css will be associated with | |
3703 | * @ss: the subsys of new css | |
3704 | * | |
3705 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
3706 | * css is online and installed in @cgrp with all interface files created. | |
3707 | * Returns 0 on success, -errno on failure. | |
3708 | */ | |
3709 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
3710 | { | |
3711 | struct cgroup *parent = cgrp->parent; | |
3712 | struct cgroup_subsys_state *css; | |
3713 | int err; | |
3714 | ||
c81c925a TH |
3715 | lockdep_assert_held(&cgroup_mutex); |
3716 | ||
3717 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
3718 | if (IS_ERR(css)) | |
3719 | return PTR_ERR(css); | |
3720 | ||
3721 | err = percpu_ref_init(&css->refcnt, css_release); | |
3722 | if (err) | |
3723 | goto err_free; | |
3724 | ||
3725 | init_css(css, ss, cgrp); | |
3726 | ||
aec25020 | 3727 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a TH |
3728 | if (err) |
3729 | goto err_free; | |
3730 | ||
3731 | err = online_css(css); | |
3732 | if (err) | |
3733 | goto err_free; | |
3734 | ||
59f5296b | 3735 | cgroup_get(cgrp); |
c81c925a TH |
3736 | css_get(css->parent); |
3737 | ||
3738 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
3739 | parent->parent) { | |
3740 | 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", | |
3741 | current->comm, current->pid, ss->name); | |
3742 | if (!strcmp(ss->name, "memory")) | |
3743 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
3744 | ss->warned_broken_hierarchy = true; | |
3745 | } | |
3746 | ||
3747 | return 0; | |
3748 | ||
3749 | err_free: | |
3750 | percpu_ref_cancel_init(&css->refcnt); | |
3751 | ss->css_free(css); | |
3752 | return err; | |
3753 | } | |
3754 | ||
2bd59d48 | 3755 | /** |
a043e3b2 LZ |
3756 | * cgroup_create - create a cgroup |
3757 | * @parent: cgroup that will be parent of the new cgroup | |
2bd59d48 TH |
3758 | * @name_str: name of the new cgroup |
3759 | * @mode: mode to set on new cgroup | |
ddbcc7e8 | 3760 | */ |
2bd59d48 TH |
3761 | static long cgroup_create(struct cgroup *parent, const char *name_str, |
3762 | umode_t mode) | |
ddbcc7e8 | 3763 | { |
bd89aabc | 3764 | struct cgroup *cgrp; |
65dff759 | 3765 | struct cgroup_name *name; |
ddbcc7e8 | 3766 | struct cgroupfs_root *root = parent->root; |
b58c8998 | 3767 | int ssid, err; |
ddbcc7e8 | 3768 | struct cgroup_subsys *ss; |
2bd59d48 | 3769 | struct kernfs_node *kn; |
ddbcc7e8 | 3770 | |
0a950f65 | 3771 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
3772 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
3773 | if (!cgrp) | |
ddbcc7e8 PM |
3774 | return -ENOMEM; |
3775 | ||
2bd59d48 | 3776 | name = cgroup_alloc_name(name_str); |
b58c8998 TH |
3777 | if (!name) { |
3778 | err = -ENOMEM; | |
65dff759 | 3779 | goto err_free_cgrp; |
b58c8998 | 3780 | } |
65dff759 LZ |
3781 | rcu_assign_pointer(cgrp->name, name); |
3782 | ||
ace2bee8 TH |
3783 | mutex_lock(&cgroup_tree_mutex); |
3784 | ||
976c06bc TH |
3785 | /* |
3786 | * Only live parents can have children. Note that the liveliness | |
3787 | * check isn't strictly necessary because cgroup_mkdir() and | |
3788 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
3789 | * anyway so that locking is contained inside cgroup proper and we | |
3790 | * don't get nasty surprises if we ever grow another caller. | |
3791 | */ | |
3792 | if (!cgroup_lock_live_group(parent)) { | |
3793 | err = -ENODEV; | |
ace2bee8 | 3794 | goto err_unlock_tree; |
0ab02ca8 LZ |
3795 | } |
3796 | ||
3797 | /* | |
3798 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
3799 | * a half-baked cgroup. | |
3800 | */ | |
3801 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
3802 | if (cgrp->id < 0) { | |
3803 | err = -ENOMEM; | |
3804 | goto err_unlock; | |
976c06bc TH |
3805 | } |
3806 | ||
cc31edce | 3807 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3808 | |
bd89aabc | 3809 | cgrp->parent = parent; |
0ae78e0b | 3810 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 3811 | cgrp->root = parent->root; |
ddbcc7e8 | 3812 | |
b6abdb0e LZ |
3813 | if (notify_on_release(parent)) |
3814 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3815 | ||
2260e7fc TH |
3816 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
3817 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 3818 | |
2bd59d48 TH |
3819 | /* create the directory */ |
3820 | kn = kernfs_create_dir(parent->kn, name->name, mode, cgrp); | |
3821 | if (IS_ERR(kn)) { | |
3822 | err = PTR_ERR(kn); | |
0ab02ca8 | 3823 | goto err_free_id; |
2bd59d48 TH |
3824 | } |
3825 | cgrp->kn = kn; | |
ddbcc7e8 | 3826 | |
00356bd5 | 3827 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 3828 | |
4e139afc | 3829 | /* allocation complete, commit to creation */ |
4e139afc TH |
3830 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3831 | root->number_of_cgroups++; | |
28fd6f30 | 3832 | |
2bd59d48 TH |
3833 | /* |
3834 | * Grab a reference on the root and parent so that they don't get | |
3835 | * deleted while there are child cgroups. | |
3836 | */ | |
3837 | cgroup_get_root(root); | |
59f5296b | 3838 | cgroup_get(parent); |
415cf07a | 3839 | |
0d80255e TH |
3840 | /* |
3841 | * @cgrp is now fully operational. If something fails after this | |
3842 | * point, it'll be released via the normal destruction path. | |
3843 | */ | |
4e96ee8e LZ |
3844 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
3845 | ||
2bb566cb | 3846 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
3847 | if (err) |
3848 | goto err_destroy; | |
3849 | ||
9d403e99 | 3850 | /* let's create and online css's */ |
b85d2040 TH |
3851 | for_each_subsys(ss, ssid) { |
3852 | if (root->subsys_mask & (1 << ssid)) { | |
3853 | err = create_css(cgrp, ss); | |
3854 | if (err) | |
3855 | goto err_destroy; | |
3856 | } | |
a8638030 | 3857 | } |
ddbcc7e8 | 3858 | |
2bd59d48 TH |
3859 | kernfs_activate(kn); |
3860 | ||
ddbcc7e8 | 3861 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3862 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
3863 | |
3864 | return 0; | |
3865 | ||
0a950f65 | 3866 | err_free_id: |
4e96ee8e | 3867 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
3868 | err_unlock: |
3869 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
3870 | err_unlock_tree: |
3871 | mutex_unlock(&cgroup_tree_mutex); | |
65dff759 | 3872 | kfree(rcu_dereference_raw(cgrp->name)); |
4b8b47eb | 3873 | err_free_cgrp: |
bd89aabc | 3874 | kfree(cgrp); |
ddbcc7e8 | 3875 | return err; |
4b8b47eb TH |
3876 | |
3877 | err_destroy: | |
3878 | cgroup_destroy_locked(cgrp); | |
3879 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3880 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 3881 | return err; |
ddbcc7e8 PM |
3882 | } |
3883 | ||
2bd59d48 TH |
3884 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
3885 | umode_t mode) | |
ddbcc7e8 | 3886 | { |
2bd59d48 | 3887 | struct cgroup *parent = parent_kn->priv; |
ddbcc7e8 | 3888 | |
2bd59d48 | 3889 | return cgroup_create(parent, name, mode); |
ddbcc7e8 PM |
3890 | } |
3891 | ||
223dbc38 TH |
3892 | /* |
3893 | * This is called when the refcnt of a css is confirmed to be killed. | |
3894 | * css_tryget() is now guaranteed to fail. | |
3895 | */ | |
3896 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 3897 | { |
223dbc38 TH |
3898 | struct cgroup_subsys_state *css = |
3899 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
3900 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 3901 | |
ace2bee8 | 3902 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
3903 | mutex_lock(&cgroup_mutex); |
3904 | ||
09a503ea TH |
3905 | /* |
3906 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
3907 | * initate destruction. | |
3908 | */ | |
3909 | offline_css(css); | |
3910 | ||
f20104de TH |
3911 | /* |
3912 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
3913 | * be disabled before proceeding to the second phase of cgroup | |
3914 | * destruction. If we are the last one, kick it off. | |
3915 | */ | |
09a503ea | 3916 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
3917 | cgroup_destroy_css_killed(cgrp); |
3918 | ||
3919 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3920 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
3921 | |
3922 | /* | |
3923 | * Put the css refs from kill_css(). Each css holds an extra | |
3924 | * reference to the cgroup's dentry and cgroup removal proceeds | |
3925 | * regardless of css refs. On the last put of each css, whenever | |
3926 | * that may be, the extra dentry ref is put so that dentry | |
3927 | * destruction happens only after all css's are released. | |
3928 | */ | |
3929 | css_put(css); | |
d3daf28d TH |
3930 | } |
3931 | ||
223dbc38 TH |
3932 | /* css kill confirmation processing requires process context, bounce */ |
3933 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
3934 | { |
3935 | struct cgroup_subsys_state *css = | |
3936 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3937 | ||
223dbc38 | 3938 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 3939 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
3940 | } |
3941 | ||
edae0c33 TH |
3942 | /** |
3943 | * kill_css - destroy a css | |
3944 | * @css: css to destroy | |
3945 | * | |
3c14f8b4 TH |
3946 | * This function initiates destruction of @css by removing cgroup interface |
3947 | * files and putting its base reference. ->css_offline() will be invoked | |
3948 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
3949 | * reference count reaches zero, @css will be released. | |
edae0c33 TH |
3950 | */ |
3951 | static void kill_css(struct cgroup_subsys_state *css) | |
3952 | { | |
2bd59d48 TH |
3953 | /* |
3954 | * This must happen before css is disassociated with its cgroup. | |
3955 | * See seq_css() for details. | |
3956 | */ | |
aec25020 | 3957 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 3958 | |
edae0c33 TH |
3959 | /* |
3960 | * Killing would put the base ref, but we need to keep it alive | |
3961 | * until after ->css_offline(). | |
3962 | */ | |
3963 | css_get(css); | |
3964 | ||
3965 | /* | |
3966 | * cgroup core guarantees that, by the time ->css_offline() is | |
3967 | * invoked, no new css reference will be given out via | |
3968 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
3969 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
3970 | * guarantee that the ref is seen as killed on all CPUs on return. | |
3971 | * | |
3972 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
3973 | * css is confirmed to be seen as killed on all CPUs. | |
3974 | */ | |
3975 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
3976 | } |
3977 | ||
3978 | /** | |
3979 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
3980 | * @cgrp: cgroup to be destroyed | |
3981 | * | |
3982 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
3983 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
3984 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
3985 | * invoked. To satisfy all the requirements, destruction is implemented in | |
3986 | * the following two steps. | |
3987 | * | |
3988 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
3989 | * userland visible parts and start killing the percpu refcnts of | |
3990 | * css's. Set up so that the next stage will be kicked off once all | |
3991 | * the percpu refcnts are confirmed to be killed. | |
3992 | * | |
3993 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
3994 | * rest of destruction. Once all cgroup references are gone, the | |
3995 | * cgroup is RCU-freed. | |
3996 | * | |
3997 | * This function implements s1. After this step, @cgrp is gone as far as | |
3998 | * the userland is concerned and a new cgroup with the same name may be | |
3999 | * created. As cgroup doesn't care about the names internally, this | |
4000 | * doesn't cause any problem. | |
4001 | */ | |
42809dd4 TH |
4002 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4003 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4004 | { |
bb78a92f | 4005 | struct cgroup *child; |
2bd59d48 TH |
4006 | struct cgroup_subsys_state *css; |
4007 | struct kernfs_node *kn; | |
ddd69148 | 4008 | bool empty; |
1c6727af | 4009 | int ssid; |
ddbcc7e8 | 4010 | |
ace2bee8 | 4011 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
4012 | lockdep_assert_held(&cgroup_mutex); |
4013 | ||
ddd69148 | 4014 | /* |
6f3d828f TH |
4015 | * css_set_lock synchronizes access to ->cset_links and prevents |
4016 | * @cgrp from being removed while __put_css_set() is in progress. | |
ddd69148 TH |
4017 | */ |
4018 | read_lock(&css_set_lock); | |
bb78a92f | 4019 | empty = list_empty(&cgrp->cset_links); |
ddd69148 TH |
4020 | read_unlock(&css_set_lock); |
4021 | if (!empty) | |
ddbcc7e8 | 4022 | return -EBUSY; |
a043e3b2 | 4023 | |
bb78a92f HD |
4024 | /* |
4025 | * Make sure there's no live children. We can't test ->children | |
4026 | * emptiness as dead children linger on it while being destroyed; | |
4027 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
4028 | */ | |
4029 | empty = true; | |
4030 | rcu_read_lock(); | |
4031 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
4032 | empty = cgroup_is_dead(child); | |
4033 | if (!empty) | |
4034 | break; | |
4035 | } | |
4036 | rcu_read_unlock(); | |
4037 | if (!empty) | |
4038 | return -EBUSY; | |
4039 | ||
88703267 | 4040 | /* |
edae0c33 TH |
4041 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
4042 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
4043 | * percpu refs of all css's are confirmed to be killed. This |
4044 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 4045 | */ |
4ac06017 | 4046 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
4047 | for_each_css(css, ssid, cgrp) |
4048 | kill_css(css); | |
4ac06017 | 4049 | mutex_lock(&cgroup_mutex); |
455050d2 TH |
4050 | |
4051 | /* | |
4052 | * Mark @cgrp dead. This prevents further task migration and child | |
4053 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 4054 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 4055 | * resume iteration after dropping RCU read lock. See |
492eb21b | 4056 | * css_next_child() for details. |
455050d2 | 4057 | */ |
54766d4a | 4058 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 4059 | |
455050d2 TH |
4060 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
4061 | raw_spin_lock(&release_list_lock); | |
4062 | if (!list_empty(&cgrp->release_list)) | |
4063 | list_del_init(&cgrp->release_list); | |
4064 | raw_spin_unlock(&release_list_lock); | |
4065 | ||
4066 | /* | |
f20104de TH |
4067 | * If @cgrp has css's attached, the second stage of cgroup |
4068 | * destruction is kicked off from css_killed_work_fn() after the | |
4069 | * refs of all attached css's are killed. If @cgrp doesn't have | |
4070 | * any css, we kick it off here. | |
4071 | */ | |
4072 | if (!cgrp->nr_css) | |
4073 | cgroup_destroy_css_killed(cgrp); | |
4074 | ||
2bd59d48 TH |
4075 | /* remove @cgrp directory along with the base files */ |
4076 | mutex_unlock(&cgroup_mutex); | |
4077 | ||
455050d2 | 4078 | /* |
2bd59d48 TH |
4079 | * There are two control paths which try to determine cgroup from |
4080 | * dentry without going through kernfs - cgroupstats_build() and | |
4081 | * css_tryget_from_dir(). Those are supported by RCU protecting | |
4082 | * clearing of cgrp->kn->priv backpointer, which should happen | |
4083 | * after all files under it have been removed. | |
455050d2 | 4084 | */ |
2bd59d48 TH |
4085 | kn = cgrp->kn; |
4086 | kernfs_get(kn); | |
4087 | ||
4088 | kernfs_remove(cgrp->kn); | |
4089 | ||
4090 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); | |
4091 | kernfs_put(kn); | |
4092 | ||
4ac06017 | 4093 | mutex_lock(&cgroup_mutex); |
455050d2 | 4094 | |
ea15f8cc TH |
4095 | return 0; |
4096 | }; | |
4097 | ||
d3daf28d | 4098 | /** |
f20104de | 4099 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
4100 | * @work: cgroup->destroy_free_work |
4101 | * | |
4102 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
4103 | * destroyed after all css's are offlined and performs the rest of |
4104 | * destruction. This is the second step of destruction described in the | |
4105 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 4106 | */ |
f20104de | 4107 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 4108 | { |
ea15f8cc | 4109 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 4110 | |
ace2bee8 | 4111 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 4112 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 4113 | |
999cd8a4 | 4114 | /* delete this cgroup from parent->children */ |
eb6fd504 | 4115 | list_del_rcu(&cgrp->sibling); |
ed957793 | 4116 | |
59f5296b | 4117 | cgroup_put(cgrp); |
ddbcc7e8 | 4118 | |
bd89aabc | 4119 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 4120 | check_for_release(parent); |
ddbcc7e8 PM |
4121 | } |
4122 | ||
2bd59d48 | 4123 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4124 | { |
2bd59d48 TH |
4125 | struct cgroup *cgrp = kn->priv; |
4126 | int ret = 0; | |
4127 | ||
4128 | /* | |
4129 | * This is self-destruction but @kn can't be removed while this | |
4130 | * callback is in progress. Let's break active protection. Once | |
4131 | * the protection is broken, @cgrp can be destroyed at any point. | |
4132 | * Pin it so that it stays accessible. | |
4133 | */ | |
4134 | cgroup_get(cgrp); | |
4135 | kernfs_break_active_protection(kn); | |
42809dd4 | 4136 | |
ace2bee8 | 4137 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 4138 | mutex_lock(&cgroup_mutex); |
2bd59d48 TH |
4139 | |
4140 | /* | |
4141 | * @cgrp might already have been destroyed while we're trying to | |
4142 | * grab the mutexes. | |
4143 | */ | |
4144 | if (!cgroup_is_dead(cgrp)) | |
4145 | ret = cgroup_destroy_locked(cgrp); | |
4146 | ||
42809dd4 | 4147 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4148 | mutex_unlock(&cgroup_tree_mutex); |
42809dd4 | 4149 | |
2bd59d48 TH |
4150 | kernfs_unbreak_active_protection(kn); |
4151 | cgroup_put(cgrp); | |
42809dd4 TH |
4152 | return ret; |
4153 | } | |
4154 | ||
2bd59d48 TH |
4155 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4156 | .remount_fs = cgroup_remount, | |
4157 | .show_options = cgroup_show_options, | |
4158 | .mkdir = cgroup_mkdir, | |
4159 | .rmdir = cgroup_rmdir, | |
4160 | .rename = cgroup_rename, | |
4161 | }; | |
4162 | ||
06a11920 | 4163 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4164 | { |
ddbcc7e8 | 4165 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4166 | |
4167 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4168 | |
ace2bee8 | 4169 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
4170 | mutex_lock(&cgroup_mutex); |
4171 | ||
de00ffa5 | 4172 | INIT_LIST_HEAD(&ss->cftsets); |
8e3f6541 | 4173 | |
ddbcc7e8 | 4174 | /* Create the top cgroup state for this subsystem */ |
9871bf95 | 4175 | ss->root = &cgroup_dummy_root; |
ca8bdcaf | 4176 | css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss)); |
ddbcc7e8 PM |
4177 | /* We don't handle early failures gracefully */ |
4178 | BUG_ON(IS_ERR(css)); | |
623f926b | 4179 | init_css(css, ss, cgroup_dummy_top); |
ddbcc7e8 | 4180 | |
e8d55fde | 4181 | /* Update the init_css_set to contain a subsys |
817929ec | 4182 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
4183 | * newly registered, all tasks and hence the |
4184 | * init_css_set is in the subsystem's top cgroup. */ | |
aec25020 | 4185 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4186 | |
4187 | need_forkexit_callback |= ss->fork || ss->exit; | |
4188 | ||
e8d55fde LZ |
4189 | /* At system boot, before all subsystems have been |
4190 | * registered, no tasks have been forked, so we don't | |
4191 | * need to invoke fork callbacks here. */ | |
4192 | BUG_ON(!list_empty(&init_task.tasks)); | |
4193 | ||
ae7f164a | 4194 | BUG_ON(online_css(css)); |
a8638030 | 4195 | |
648bb56d | 4196 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4197 | mutex_unlock(&cgroup_tree_mutex); |
e6a1105b BB |
4198 | } |
4199 | ||
ddbcc7e8 | 4200 | /** |
a043e3b2 LZ |
4201 | * cgroup_init_early - cgroup initialization at system boot |
4202 | * | |
4203 | * Initialize cgroups at system boot, and initialize any | |
4204 | * subsystems that request early init. | |
ddbcc7e8 PM |
4205 | */ |
4206 | int __init cgroup_init_early(void) | |
4207 | { | |
30159ec7 | 4208 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4209 | int i; |
30159ec7 | 4210 | |
146aa1bd | 4211 | atomic_set(&init_css_set.refcount, 1); |
69d0206c | 4212 | INIT_LIST_HEAD(&init_css_set.cgrp_links); |
817929ec | 4213 | INIT_LIST_HEAD(&init_css_set.tasks); |
472b1053 | 4214 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 4215 | css_set_count = 1; |
9871bf95 TH |
4216 | init_cgroup_root(&cgroup_dummy_root); |
4217 | cgroup_root_count = 1; | |
a4ea1cc9 | 4218 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4219 | |
69d0206c | 4220 | init_cgrp_cset_link.cset = &init_css_set; |
9871bf95 TH |
4221 | init_cgrp_cset_link.cgrp = cgroup_dummy_top; |
4222 | list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links); | |
69d0206c | 4223 | list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links); |
ddbcc7e8 | 4224 | |
3ed80a62 | 4225 | for_each_subsys(ss, i) { |
aec25020 | 4226 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4227 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4228 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4229 | ss->id, ss->name); |
073219e9 TH |
4230 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4231 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4232 | ||
aec25020 | 4233 | ss->id = i; |
073219e9 | 4234 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4235 | |
4236 | if (ss->early_init) | |
4237 | cgroup_init_subsys(ss); | |
4238 | } | |
4239 | return 0; | |
4240 | } | |
4241 | ||
4242 | /** | |
a043e3b2 LZ |
4243 | * cgroup_init - cgroup initialization |
4244 | * | |
4245 | * Register cgroup filesystem and /proc file, and initialize | |
4246 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4247 | */ |
4248 | int __init cgroup_init(void) | |
4249 | { | |
30159ec7 | 4250 | struct cgroup_subsys *ss; |
0ac801fe | 4251 | unsigned long key; |
30159ec7 | 4252 | int i, err; |
a424316c | 4253 | |
2bd59d48 | 4254 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
2da440a2 | 4255 | |
3ed80a62 | 4256 | for_each_subsys(ss, i) { |
ddbcc7e8 PM |
4257 | if (!ss->early_init) |
4258 | cgroup_init_subsys(ss); | |
de00ffa5 TH |
4259 | |
4260 | /* | |
4261 | * cftype registration needs kmalloc and can't be done | |
4262 | * during early_init. Register base cftypes separately. | |
4263 | */ | |
4264 | if (ss->base_cftypes) | |
4265 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
ddbcc7e8 PM |
4266 | } |
4267 | ||
fa3ca07e | 4268 | /* allocate id for the dummy hierarchy */ |
54e7b4eb | 4269 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4270 | |
82fe9b0d TH |
4271 | /* Add init_css_set to the hash table */ |
4272 | key = css_set_hash(init_css_set.subsys); | |
4273 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4274 | ||
fc76df70 | 4275 | BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1)); |
676db4af | 4276 | |
4e96ee8e LZ |
4277 | err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top, |
4278 | 0, 1, GFP_KERNEL); | |
4279 | BUG_ON(err < 0); | |
4280 | ||
54e7b4eb TH |
4281 | mutex_unlock(&cgroup_mutex); |
4282 | ||
676db4af | 4283 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4284 | if (!cgroup_kobj) |
4285 | return -ENOMEM; | |
676db4af | 4286 | |
ddbcc7e8 | 4287 | err = register_filesystem(&cgroup_fs_type); |
676db4af GKH |
4288 | if (err < 0) { |
4289 | kobject_put(cgroup_kobj); | |
2bd59d48 | 4290 | return err; |
676db4af | 4291 | } |
ddbcc7e8 | 4292 | |
46ae220b | 4293 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 4294 | return 0; |
ddbcc7e8 | 4295 | } |
b4f48b63 | 4296 | |
e5fca243 TH |
4297 | static int __init cgroup_wq_init(void) |
4298 | { | |
4299 | /* | |
4300 | * There isn't much point in executing destruction path in | |
4301 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
ab3f5faa HD |
4302 | * |
4303 | * XXX: Must be ordered to make sure parent is offlined after | |
4304 | * children. The ordering requirement is for memcg where a | |
4305 | * parent's offline may wait for a child's leading to deadlock. In | |
4306 | * the long term, this should be fixed from memcg side. | |
e5fca243 TH |
4307 | * |
4308 | * We would prefer to do this in cgroup_init() above, but that | |
4309 | * is called before init_workqueues(): so leave this until after. | |
4310 | */ | |
ab3f5faa | 4311 | cgroup_destroy_wq = alloc_ordered_workqueue("cgroup_destroy", 0); |
e5fca243 | 4312 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4313 | |
4314 | /* | |
4315 | * Used to destroy pidlists and separate to serve as flush domain. | |
4316 | * Cap @max_active to 1 too. | |
4317 | */ | |
4318 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4319 | 0, 1); | |
4320 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4321 | ||
e5fca243 TH |
4322 | return 0; |
4323 | } | |
4324 | core_initcall(cgroup_wq_init); | |
4325 | ||
a424316c PM |
4326 | /* |
4327 | * proc_cgroup_show() | |
4328 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4329 | * - Used for /proc/<pid>/cgroup. | |
4330 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4331 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4332 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4333 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4334 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4335 | * cgroup to top_cgroup. | |
4336 | */ | |
4337 | ||
4338 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4339 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4340 | { |
4341 | struct pid *pid; | |
4342 | struct task_struct *tsk; | |
4343 | char *buf; | |
4344 | int retval; | |
4345 | struct cgroupfs_root *root; | |
4346 | ||
4347 | retval = -ENOMEM; | |
4348 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4349 | if (!buf) | |
4350 | goto out; | |
4351 | ||
4352 | retval = -ESRCH; | |
4353 | pid = m->private; | |
4354 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4355 | if (!tsk) | |
4356 | goto out_free; | |
4357 | ||
4358 | retval = 0; | |
4359 | ||
4360 | mutex_lock(&cgroup_mutex); | |
4361 | ||
e5f6a860 | 4362 | for_each_active_root(root) { |
a424316c | 4363 | struct cgroup_subsys *ss; |
bd89aabc | 4364 | struct cgroup *cgrp; |
b85d2040 | 4365 | int ssid, count = 0; |
a424316c | 4366 | |
2c6ab6d2 | 4367 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 TH |
4368 | for_each_subsys(ss, ssid) |
4369 | if (root->subsys_mask & (1 << ssid)) | |
4370 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4371 | if (strlen(root->name)) |
4372 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4373 | root->name); | |
a424316c | 4374 | seq_putc(m, ':'); |
7717f7ba | 4375 | cgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 4376 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); |
a424316c PM |
4377 | if (retval < 0) |
4378 | goto out_unlock; | |
4379 | seq_puts(m, buf); | |
4380 | seq_putc(m, '\n'); | |
4381 | } | |
4382 | ||
4383 | out_unlock: | |
4384 | mutex_unlock(&cgroup_mutex); | |
4385 | put_task_struct(tsk); | |
4386 | out_free: | |
4387 | kfree(buf); | |
4388 | out: | |
4389 | return retval; | |
4390 | } | |
4391 | ||
a424316c PM |
4392 | /* Display information about each subsystem and each hierarchy */ |
4393 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4394 | { | |
30159ec7 | 4395 | struct cgroup_subsys *ss; |
a424316c | 4396 | int i; |
a424316c | 4397 | |
8bab8dde | 4398 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4399 | /* |
4400 | * ideally we don't want subsystems moving around while we do this. | |
4401 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4402 | * subsys/hierarchy state. | |
4403 | */ | |
a424316c | 4404 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4405 | |
4406 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4407 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4408 | ss->name, ss->root->hierarchy_id, | |
8bab8dde | 4409 | ss->root->number_of_cgroups, !ss->disabled); |
30159ec7 | 4410 | |
a424316c PM |
4411 | mutex_unlock(&cgroup_mutex); |
4412 | return 0; | |
4413 | } | |
4414 | ||
4415 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4416 | { | |
9dce07f1 | 4417 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4418 | } |
4419 | ||
828c0950 | 4420 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4421 | .open = cgroupstats_open, |
4422 | .read = seq_read, | |
4423 | .llseek = seq_lseek, | |
4424 | .release = single_release, | |
4425 | }; | |
4426 | ||
b4f48b63 PM |
4427 | /** |
4428 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4429 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4430 | * |
4431 | * Description: A task inherits its parent's cgroup at fork(). | |
4432 | * | |
4433 | * A pointer to the shared css_set was automatically copied in | |
4434 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
9bb71308 TH |
4435 | * it was not made under the protection of RCU or cgroup_mutex, so |
4436 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
4437 | * have already changed current->cgroups, allowing the previously | |
4438 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
4439 | * |
4440 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4441 | * task, and the passed argument 'child' points to the child task. | |
4442 | */ | |
4443 | void cgroup_fork(struct task_struct *child) | |
4444 | { | |
9bb71308 | 4445 | task_lock(current); |
a8ad805c | 4446 | get_css_set(task_css_set(current)); |
817929ec | 4447 | child->cgroups = current->cgroups; |
9bb71308 | 4448 | task_unlock(current); |
817929ec | 4449 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4450 | } |
4451 | ||
817929ec | 4452 | /** |
a043e3b2 LZ |
4453 | * cgroup_post_fork - called on a new task after adding it to the task list |
4454 | * @child: the task in question | |
4455 | * | |
5edee61e TH |
4456 | * Adds the task to the list running through its css_set if necessary and |
4457 | * call the subsystem fork() callbacks. Has to be after the task is | |
4458 | * visible on the task list in case we race with the first call to | |
0942eeee | 4459 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4460 | * list. |
a043e3b2 | 4461 | */ |
817929ec PM |
4462 | void cgroup_post_fork(struct task_struct *child) |
4463 | { | |
30159ec7 | 4464 | struct cgroup_subsys *ss; |
5edee61e TH |
4465 | int i; |
4466 | ||
3ce3230a FW |
4467 | /* |
4468 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
4469 | * under the tasklist_lock and we read it here after we added the child | |
4470 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
4471 | * yet in the tasklist when we walked through it from | |
4472 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
4473 | * should be visible now due to the paired locking and barriers implied | |
4474 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
4475 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
4476 | * lock on fork. | |
4477 | */ | |
817929ec PM |
4478 | if (use_task_css_set_links) { |
4479 | write_lock(&css_set_lock); | |
d8783832 TH |
4480 | task_lock(child); |
4481 | if (list_empty(&child->cg_list)) | |
a8ad805c | 4482 | list_add(&child->cg_list, &task_css_set(child)->tasks); |
d8783832 | 4483 | task_unlock(child); |
817929ec PM |
4484 | write_unlock(&css_set_lock); |
4485 | } | |
5edee61e TH |
4486 | |
4487 | /* | |
4488 | * Call ss->fork(). This must happen after @child is linked on | |
4489 | * css_set; otherwise, @child might change state between ->fork() | |
4490 | * and addition to css_set. | |
4491 | */ | |
4492 | if (need_forkexit_callback) { | |
3ed80a62 | 4493 | for_each_subsys(ss, i) |
5edee61e TH |
4494 | if (ss->fork) |
4495 | ss->fork(child); | |
5edee61e | 4496 | } |
817929ec | 4497 | } |
5edee61e | 4498 | |
b4f48b63 PM |
4499 | /** |
4500 | * cgroup_exit - detach cgroup from exiting task | |
4501 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4502 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4503 | * |
4504 | * Description: Detach cgroup from @tsk and release it. | |
4505 | * | |
4506 | * Note that cgroups marked notify_on_release force every task in | |
4507 | * them to take the global cgroup_mutex mutex when exiting. | |
4508 | * This could impact scaling on very large systems. Be reluctant to | |
4509 | * use notify_on_release cgroups where very high task exit scaling | |
4510 | * is required on large systems. | |
4511 | * | |
4512 | * the_top_cgroup_hack: | |
4513 | * | |
4514 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4515 | * | |
4516 | * We call cgroup_exit() while the task is still competent to | |
4517 | * handle notify_on_release(), then leave the task attached to the | |
4518 | * root cgroup in each hierarchy for the remainder of its exit. | |
4519 | * | |
4520 | * To do this properly, we would increment the reference count on | |
4521 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4522 | * code we would add a second cgroup function call, to drop that | |
4523 | * reference. This would just create an unnecessary hot spot on | |
4524 | * the top_cgroup reference count, to no avail. | |
4525 | * | |
4526 | * Normally, holding a reference to a cgroup without bumping its | |
4527 | * count is unsafe. The cgroup could go away, or someone could | |
4528 | * attach us to a different cgroup, decrementing the count on | |
4529 | * the first cgroup that we never incremented. But in this case, | |
4530 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4531 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4532 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4533 | */ |
4534 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4535 | { | |
30159ec7 | 4536 | struct cgroup_subsys *ss; |
5abb8855 | 4537 | struct css_set *cset; |
d41d5a01 | 4538 | int i; |
817929ec PM |
4539 | |
4540 | /* | |
4541 | * Unlink from the css_set task list if necessary. | |
4542 | * Optimistically check cg_list before taking | |
4543 | * css_set_lock | |
4544 | */ | |
4545 | if (!list_empty(&tsk->cg_list)) { | |
4546 | write_lock(&css_set_lock); | |
4547 | if (!list_empty(&tsk->cg_list)) | |
8d258797 | 4548 | list_del_init(&tsk->cg_list); |
817929ec PM |
4549 | write_unlock(&css_set_lock); |
4550 | } | |
4551 | ||
b4f48b63 PM |
4552 | /* Reassign the task to the init_css_set. */ |
4553 | task_lock(tsk); | |
a8ad805c TH |
4554 | cset = task_css_set(tsk); |
4555 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 PZ |
4556 | |
4557 | if (run_callbacks && need_forkexit_callback) { | |
3ed80a62 TH |
4558 | /* see cgroup_post_fork() for details */ |
4559 | for_each_subsys(ss, i) { | |
d41d5a01 | 4560 | if (ss->exit) { |
eb95419b TH |
4561 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4562 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4563 | |
eb95419b | 4564 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4565 | } |
4566 | } | |
4567 | } | |
b4f48b63 | 4568 | task_unlock(tsk); |
d41d5a01 | 4569 | |
5abb8855 | 4570 | put_css_set_taskexit(cset); |
b4f48b63 | 4571 | } |
697f4161 | 4572 | |
bd89aabc | 4573 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4574 | { |
f50daa70 | 4575 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4576 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4577 | /* |
4578 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4579 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4580 | * it now |
4581 | */ | |
81a6a5cd | 4582 | int need_schedule_work = 0; |
f50daa70 | 4583 | |
cdcc136f | 4584 | raw_spin_lock(&release_list_lock); |
54766d4a | 4585 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4586 | list_empty(&cgrp->release_list)) { |
4587 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4588 | need_schedule_work = 1; |
4589 | } | |
cdcc136f | 4590 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4591 | if (need_schedule_work) |
4592 | schedule_work(&release_agent_work); | |
4593 | } | |
4594 | } | |
4595 | ||
81a6a5cd PM |
4596 | /* |
4597 | * Notify userspace when a cgroup is released, by running the | |
4598 | * configured release agent with the name of the cgroup (path | |
4599 | * relative to the root of cgroup file system) as the argument. | |
4600 | * | |
4601 | * Most likely, this user command will try to rmdir this cgroup. | |
4602 | * | |
4603 | * This races with the possibility that some other task will be | |
4604 | * attached to this cgroup before it is removed, or that some other | |
4605 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4606 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4607 | * unused, and this cgroup will be reprieved from its death sentence, | |
4608 | * to continue to serve a useful existence. Next time it's released, | |
4609 | * we will get notified again, if it still has 'notify_on_release' set. | |
4610 | * | |
4611 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4612 | * means only wait until the task is successfully execve()'d. The | |
4613 | * separate release agent task is forked by call_usermodehelper(), | |
4614 | * then control in this thread returns here, without waiting for the | |
4615 | * release agent task. We don't bother to wait because the caller of | |
4616 | * this routine has no use for the exit status of the release agent | |
4617 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4618 | */ |
81a6a5cd PM |
4619 | static void cgroup_release_agent(struct work_struct *work) |
4620 | { | |
4621 | BUG_ON(work != &release_agent_work); | |
4622 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4623 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4624 | while (!list_empty(&release_list)) { |
4625 | char *argv[3], *envp[3]; | |
4626 | int i; | |
e788e066 | 4627 | char *pathbuf = NULL, *agentbuf = NULL; |
bd89aabc | 4628 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4629 | struct cgroup, |
4630 | release_list); | |
bd89aabc | 4631 | list_del_init(&cgrp->release_list); |
cdcc136f | 4632 | raw_spin_unlock(&release_list_lock); |
81a6a5cd | 4633 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
e788e066 PM |
4634 | if (!pathbuf) |
4635 | goto continue_free; | |
4636 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
4637 | goto continue_free; | |
4638 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4639 | if (!agentbuf) | |
4640 | goto continue_free; | |
81a6a5cd PM |
4641 | |
4642 | i = 0; | |
e788e066 PM |
4643 | argv[i++] = agentbuf; |
4644 | argv[i++] = pathbuf; | |
81a6a5cd PM |
4645 | argv[i] = NULL; |
4646 | ||
4647 | i = 0; | |
4648 | /* minimal command environment */ | |
4649 | envp[i++] = "HOME=/"; | |
4650 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4651 | envp[i] = NULL; | |
4652 | ||
4653 | /* Drop the lock while we invoke the usermode helper, | |
4654 | * since the exec could involve hitting disk and hence | |
4655 | * be a slow process */ | |
4656 | mutex_unlock(&cgroup_mutex); | |
4657 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4658 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4659 | continue_free: |
4660 | kfree(pathbuf); | |
4661 | kfree(agentbuf); | |
cdcc136f | 4662 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4663 | } |
cdcc136f | 4664 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4665 | mutex_unlock(&cgroup_mutex); |
4666 | } | |
8bab8dde PM |
4667 | |
4668 | static int __init cgroup_disable(char *str) | |
4669 | { | |
30159ec7 | 4670 | struct cgroup_subsys *ss; |
8bab8dde | 4671 | char *token; |
30159ec7 | 4672 | int i; |
8bab8dde PM |
4673 | |
4674 | while ((token = strsep(&str, ",")) != NULL) { | |
4675 | if (!*token) | |
4676 | continue; | |
be45c900 | 4677 | |
3ed80a62 | 4678 | for_each_subsys(ss, i) { |
8bab8dde PM |
4679 | if (!strcmp(token, ss->name)) { |
4680 | ss->disabled = 1; | |
4681 | printk(KERN_INFO "Disabling %s control group" | |
4682 | " subsystem\n", ss->name); | |
4683 | break; | |
4684 | } | |
4685 | } | |
4686 | } | |
4687 | return 1; | |
4688 | } | |
4689 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 4690 | |
b77d7b60 | 4691 | /** |
5a17f543 | 4692 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
4693 | * @dentry: directory dentry of interest |
4694 | * @ss: subsystem of interest | |
b77d7b60 | 4695 | * |
5a17f543 TH |
4696 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
4697 | * to get the corresponding css and return it. If such css doesn't exist | |
4698 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 4699 | */ |
5a17f543 TH |
4700 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
4701 | struct cgroup_subsys *ss) | |
e5d1367f | 4702 | { |
2bd59d48 TH |
4703 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
4704 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 4705 | struct cgroup *cgrp; |
b77d7b60 | 4706 | |
35cf0836 | 4707 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
4708 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
4709 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
4710 | return ERR_PTR(-EBADF); |
4711 | ||
5a17f543 TH |
4712 | rcu_read_lock(); |
4713 | ||
2bd59d48 TH |
4714 | /* |
4715 | * This path doesn't originate from kernfs and @kn could already | |
4716 | * have been or be removed at any point. @kn->priv is RCU | |
4717 | * protected for this access. See destroy_locked() for details. | |
4718 | */ | |
4719 | cgrp = rcu_dereference(kn->priv); | |
4720 | if (cgrp) | |
4721 | css = cgroup_css(cgrp, ss); | |
5a17f543 TH |
4722 | |
4723 | if (!css || !css_tryget(css)) | |
4724 | css = ERR_PTR(-ENOENT); | |
4725 | ||
4726 | rcu_read_unlock(); | |
4727 | return css; | |
e5d1367f | 4728 | } |
e5d1367f | 4729 | |
1cb650b9 LZ |
4730 | /** |
4731 | * css_from_id - lookup css by id | |
4732 | * @id: the cgroup id | |
4733 | * @ss: cgroup subsys to be looked into | |
4734 | * | |
4735 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
4736 | * Should be called under rcu_read_lock(). | |
4737 | */ | |
4738 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
4739 | { | |
4740 | struct cgroup *cgrp; | |
4741 | ||
ace2bee8 | 4742 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
4743 | |
4744 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
4745 | if (cgrp) | |
d1625964 | 4746 | return cgroup_css(cgrp, ss); |
1cb650b9 | 4747 | return NULL; |
e5d1367f SE |
4748 | } |
4749 | ||
fe693435 | 4750 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
4751 | static struct cgroup_subsys_state * |
4752 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
4753 | { |
4754 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
4755 | ||
4756 | if (!css) | |
4757 | return ERR_PTR(-ENOMEM); | |
4758 | ||
4759 | return css; | |
4760 | } | |
4761 | ||
eb95419b | 4762 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 4763 | { |
eb95419b | 4764 | kfree(css); |
fe693435 PM |
4765 | } |
4766 | ||
182446d0 TH |
4767 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
4768 | struct cftype *cft) | |
fe693435 | 4769 | { |
182446d0 | 4770 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
4771 | } |
4772 | ||
182446d0 TH |
4773 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
4774 | struct cftype *cft) | |
fe693435 PM |
4775 | { |
4776 | return (u64)(unsigned long)current->cgroups; | |
4777 | } | |
4778 | ||
182446d0 | 4779 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 4780 | struct cftype *cft) |
fe693435 PM |
4781 | { |
4782 | u64 count; | |
4783 | ||
4784 | rcu_read_lock(); | |
a8ad805c | 4785 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
4786 | rcu_read_unlock(); |
4787 | return count; | |
4788 | } | |
4789 | ||
2da8ca82 | 4790 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4791 | { |
69d0206c | 4792 | struct cgrp_cset_link *link; |
5abb8855 | 4793 | struct css_set *cset; |
7717f7ba PM |
4794 | |
4795 | read_lock(&css_set_lock); | |
4796 | rcu_read_lock(); | |
5abb8855 | 4797 | cset = rcu_dereference(current->cgroups); |
69d0206c | 4798 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 4799 | struct cgroup *c = link->cgrp; |
59f5296b TH |
4800 | const char *name = "?"; |
4801 | ||
4802 | if (c != cgroup_dummy_top) | |
4803 | name = cgroup_name(c); | |
7717f7ba | 4804 | |
2c6ab6d2 PM |
4805 | seq_printf(seq, "Root %d group %s\n", |
4806 | c->root->hierarchy_id, name); | |
7717f7ba PM |
4807 | } |
4808 | rcu_read_unlock(); | |
4809 | read_unlock(&css_set_lock); | |
4810 | return 0; | |
4811 | } | |
4812 | ||
4813 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 4814 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4815 | { |
2da8ca82 | 4816 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 4817 | struct cgrp_cset_link *link; |
7717f7ba PM |
4818 | |
4819 | read_lock(&css_set_lock); | |
182446d0 | 4820 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 4821 | struct css_set *cset = link->cset; |
7717f7ba PM |
4822 | struct task_struct *task; |
4823 | int count = 0; | |
5abb8855 TH |
4824 | seq_printf(seq, "css_set %p\n", cset); |
4825 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
7717f7ba PM |
4826 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { |
4827 | seq_puts(seq, " ...\n"); | |
4828 | break; | |
4829 | } else { | |
4830 | seq_printf(seq, " task %d\n", | |
4831 | task_pid_vnr(task)); | |
4832 | } | |
4833 | } | |
4834 | } | |
4835 | read_unlock(&css_set_lock); | |
4836 | return 0; | |
4837 | } | |
4838 | ||
182446d0 | 4839 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 4840 | { |
182446d0 | 4841 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
4842 | } |
4843 | ||
4844 | static struct cftype debug_files[] = { | |
fe693435 PM |
4845 | { |
4846 | .name = "taskcount", | |
4847 | .read_u64 = debug_taskcount_read, | |
4848 | }, | |
4849 | ||
4850 | { | |
4851 | .name = "current_css_set", | |
4852 | .read_u64 = current_css_set_read, | |
4853 | }, | |
4854 | ||
4855 | { | |
4856 | .name = "current_css_set_refcount", | |
4857 | .read_u64 = current_css_set_refcount_read, | |
4858 | }, | |
4859 | ||
7717f7ba PM |
4860 | { |
4861 | .name = "current_css_set_cg_links", | |
2da8ca82 | 4862 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
4863 | }, |
4864 | ||
4865 | { | |
4866 | .name = "cgroup_css_links", | |
2da8ca82 | 4867 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
4868 | }, |
4869 | ||
fe693435 PM |
4870 | { |
4871 | .name = "releasable", | |
4872 | .read_u64 = releasable_read, | |
4873 | }, | |
fe693435 | 4874 | |
4baf6e33 TH |
4875 | { } /* terminate */ |
4876 | }; | |
fe693435 | 4877 | |
073219e9 | 4878 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
4879 | .css_alloc = debug_css_alloc, |
4880 | .css_free = debug_css_free, | |
4baf6e33 | 4881 | .base_cftypes = debug_files, |
fe693435 PM |
4882 | }; |
4883 | #endif /* CONFIG_CGROUP_DEBUG */ |