]>
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 | |
6f30558f TH |
960 | kernfs_put(cgrp->kn); |
961 | ||
65dff759 | 962 | kfree(rcu_dereference_raw(cgrp->name)); |
be445626 LZ |
963 | kfree(cgrp); |
964 | } | |
965 | ||
966 | static void cgroup_free_rcu(struct rcu_head *head) | |
967 | { | |
968 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
969 | ||
ea15f8cc | 970 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 971 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
972 | } |
973 | ||
59f5296b TH |
974 | static void cgroup_get(struct cgroup *cgrp) |
975 | { | |
2bd59d48 TH |
976 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
977 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
978 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
979 | } |
980 | ||
59f5296b TH |
981 | static void cgroup_put(struct cgroup *cgrp) |
982 | { | |
2bd59d48 TH |
983 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
984 | return; | |
985 | if (WARN_ON_ONCE(!cgroup_is_dead(cgrp))) | |
986 | return; | |
59f5296b | 987 | |
2bd59d48 TH |
988 | /* |
989 | * XXX: cgrp->id is only used to look up css's. As cgroup and | |
990 | * css's lifetimes will be decoupled, it should be made | |
991 | * per-subsystem and moved to css->id so that lookups are | |
992 | * successful until the target css is released. | |
993 | */ | |
994 | mutex_lock(&cgroup_mutex); | |
995 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
996 | mutex_unlock(&cgroup_mutex); | |
997 | cgrp->id = -1; | |
ddbcc7e8 | 998 | |
2bd59d48 | 999 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 PM |
1000 | } |
1001 | ||
2739d3cc | 1002 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1003 | { |
2bd59d48 | 1004 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1005 | |
ace2bee8 | 1006 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 1007 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1008 | } |
1009 | ||
13af07df | 1010 | /** |
628f7cd4 | 1011 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 1012 | * @cgrp: target cgroup |
13af07df AR |
1013 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
1014 | */ | |
628f7cd4 | 1015 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 1016 | { |
13af07df | 1017 | struct cgroup_subsys *ss; |
b420ba7d | 1018 | int i; |
05ef1d7c | 1019 | |
b420ba7d | 1020 | for_each_subsys(ss, i) { |
0adb0704 | 1021 | struct cftype *cfts; |
b420ba7d TH |
1022 | |
1023 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 1024 | continue; |
0adb0704 TH |
1025 | list_for_each_entry(cfts, &ss->cfts, node) |
1026 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 1027 | } |
ddbcc7e8 PM |
1028 | } |
1029 | ||
ddbcc7e8 | 1030 | static int rebind_subsystems(struct cgroupfs_root *root, |
a8a648c4 | 1031 | unsigned long added_mask, unsigned removed_mask) |
ddbcc7e8 | 1032 | { |
bd89aabc | 1033 | struct cgroup *cgrp = &root->top_cgroup; |
30159ec7 | 1034 | struct cgroup_subsys *ss; |
3126121f | 1035 | int i, ret; |
ddbcc7e8 | 1036 | |
ace2bee8 TH |
1037 | lockdep_assert_held(&cgroup_tree_mutex); |
1038 | lockdep_assert_held(&cgroup_mutex); | |
aae8aab4 | 1039 | |
ddbcc7e8 | 1040 | /* Check that any added subsystems are currently free */ |
3ed80a62 TH |
1041 | for_each_subsys(ss, i) |
1042 | if ((added_mask & (1 << i)) && ss->root != &cgroup_dummy_root) | |
1043 | return -EBUSY; | |
ddbcc7e8 | 1044 | |
3126121f TH |
1045 | ret = cgroup_populate_dir(cgrp, added_mask); |
1046 | if (ret) | |
3ed80a62 | 1047 | return ret; |
3126121f TH |
1048 | |
1049 | /* | |
1050 | * Nothing can fail from this point on. Remove files for the | |
1051 | * removed subsystems and rebind each subsystem. | |
1052 | */ | |
4ac06017 | 1053 | mutex_unlock(&cgroup_mutex); |
3126121f | 1054 | cgroup_clear_dir(cgrp, removed_mask); |
4ac06017 | 1055 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1056 | |
30159ec7 | 1057 | for_each_subsys(ss, i) { |
ddbcc7e8 | 1058 | unsigned long bit = 1UL << i; |
30159ec7 | 1059 | |
a1a71b45 | 1060 | if (bit & added_mask) { |
ddbcc7e8 | 1061 | /* We're binding this subsystem to this hierarchy */ |
ca8bdcaf TH |
1062 | BUG_ON(cgroup_css(cgrp, ss)); |
1063 | BUG_ON(!cgroup_css(cgroup_dummy_top, ss)); | |
1064 | BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top); | |
a8a648c4 | 1065 | |
73e80ed8 | 1066 | rcu_assign_pointer(cgrp->subsys[i], |
ca8bdcaf TH |
1067 | cgroup_css(cgroup_dummy_top, ss)); |
1068 | cgroup_css(cgrp, ss)->cgroup = cgrp; | |
a8a648c4 | 1069 | |
b2aa30f7 | 1070 | ss->root = root; |
ddbcc7e8 | 1071 | if (ss->bind) |
ca8bdcaf | 1072 | ss->bind(cgroup_css(cgrp, ss)); |
a8a648c4 | 1073 | |
cf5d5941 | 1074 | /* refcount was already taken, and we're keeping it */ |
a8a648c4 | 1075 | root->subsys_mask |= bit; |
a1a71b45 | 1076 | } else if (bit & removed_mask) { |
ddbcc7e8 | 1077 | /* We're removing this subsystem */ |
ca8bdcaf TH |
1078 | BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss)); |
1079 | BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp); | |
a8a648c4 | 1080 | |
ddbcc7e8 | 1081 | if (ss->bind) |
ca8bdcaf | 1082 | ss->bind(cgroup_css(cgroup_dummy_top, ss)); |
73e80ed8 | 1083 | |
ca8bdcaf | 1084 | cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top; |
73e80ed8 TH |
1085 | RCU_INIT_POINTER(cgrp->subsys[i], NULL); |
1086 | ||
9871bf95 | 1087 | cgroup_subsys[i]->root = &cgroup_dummy_root; |
a8a648c4 | 1088 | root->subsys_mask &= ~bit; |
ddbcc7e8 PM |
1089 | } |
1090 | } | |
ddbcc7e8 | 1091 | |
1672d040 TH |
1092 | /* |
1093 | * Mark @root has finished binding subsystems. @root->subsys_mask | |
1094 | * now matches the bound subsystems. | |
1095 | */ | |
1096 | root->flags |= CGRP_ROOT_SUBSYS_BOUND; | |
2bd59d48 | 1097 | kernfs_activate(cgrp->kn); |
1672d040 | 1098 | |
ddbcc7e8 PM |
1099 | return 0; |
1100 | } | |
1101 | ||
2bd59d48 TH |
1102 | static int cgroup_show_options(struct seq_file *seq, |
1103 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1104 | { |
2bd59d48 | 1105 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1106 | struct cgroup_subsys *ss; |
b85d2040 | 1107 | int ssid; |
ddbcc7e8 | 1108 | |
b85d2040 TH |
1109 | for_each_subsys(ss, ssid) |
1110 | if (root->subsys_mask & (1 << ssid)) | |
1111 | seq_printf(seq, ",%s", ss->name); | |
873fe09e TH |
1112 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1113 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1114 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1115 | seq_puts(seq, ",noprefix"); |
93438629 | 1116 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1117 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1118 | |
1119 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1120 | if (strlen(root->release_agent_path)) |
1121 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1122 | spin_unlock(&release_agent_path_lock); |
1123 | ||
2260e7fc | 1124 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags)) |
97978e6d | 1125 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1126 | if (strlen(root->name)) |
1127 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1128 | return 0; |
1129 | } | |
1130 | ||
1131 | struct cgroup_sb_opts { | |
a1a71b45 | 1132 | unsigned long subsys_mask; |
ddbcc7e8 | 1133 | unsigned long flags; |
81a6a5cd | 1134 | char *release_agent; |
2260e7fc | 1135 | bool cpuset_clone_children; |
c6d57f33 | 1136 | char *name; |
2c6ab6d2 PM |
1137 | /* User explicitly requested empty subsystem */ |
1138 | bool none; | |
ddbcc7e8 PM |
1139 | }; |
1140 | ||
aae8aab4 | 1141 | /* |
9871bf95 TH |
1142 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1143 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1144 | * array. This function takes refcounts on subsystems to be used, unless it | |
1145 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1146 | */ |
cf5d5941 | 1147 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1148 | { |
32a8cf23 DL |
1149 | char *token, *o = data; |
1150 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1151 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1152 | struct cgroup_subsys *ss; |
1153 | int i; | |
f9ab5b5b | 1154 | |
aae8aab4 BB |
1155 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1156 | ||
f9ab5b5b | 1157 | #ifdef CONFIG_CPUSETS |
073219e9 | 1158 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1159 | #endif |
ddbcc7e8 | 1160 | |
c6d57f33 | 1161 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1162 | |
1163 | while ((token = strsep(&o, ",")) != NULL) { | |
1164 | if (!*token) | |
1165 | return -EINVAL; | |
32a8cf23 | 1166 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1167 | /* Explicitly have no subsystems */ |
1168 | opts->none = true; | |
32a8cf23 DL |
1169 | continue; |
1170 | } | |
1171 | if (!strcmp(token, "all")) { | |
1172 | /* Mutually exclusive option 'all' + subsystem name */ | |
1173 | if (one_ss) | |
1174 | return -EINVAL; | |
1175 | all_ss = true; | |
1176 | continue; | |
1177 | } | |
873fe09e TH |
1178 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1179 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1180 | continue; | |
1181 | } | |
32a8cf23 | 1182 | if (!strcmp(token, "noprefix")) { |
93438629 | 1183 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1184 | continue; |
1185 | } | |
1186 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1187 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1188 | continue; |
1189 | } | |
03b1cde6 | 1190 | if (!strcmp(token, "xattr")) { |
93438629 | 1191 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1192 | continue; |
1193 | } | |
32a8cf23 | 1194 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1195 | /* Specifying two release agents is forbidden */ |
1196 | if (opts->release_agent) | |
1197 | return -EINVAL; | |
c6d57f33 | 1198 | opts->release_agent = |
e400c285 | 1199 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1200 | if (!opts->release_agent) |
1201 | return -ENOMEM; | |
32a8cf23 DL |
1202 | continue; |
1203 | } | |
1204 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1205 | const char *name = token + 5; |
1206 | /* Can't specify an empty name */ | |
1207 | if (!strlen(name)) | |
1208 | return -EINVAL; | |
1209 | /* Must match [\w.-]+ */ | |
1210 | for (i = 0; i < strlen(name); i++) { | |
1211 | char c = name[i]; | |
1212 | if (isalnum(c)) | |
1213 | continue; | |
1214 | if ((c == '.') || (c == '-') || (c == '_')) | |
1215 | continue; | |
1216 | return -EINVAL; | |
1217 | } | |
1218 | /* Specifying two names is forbidden */ | |
1219 | if (opts->name) | |
1220 | return -EINVAL; | |
1221 | opts->name = kstrndup(name, | |
e400c285 | 1222 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1223 | GFP_KERNEL); |
1224 | if (!opts->name) | |
1225 | return -ENOMEM; | |
32a8cf23 DL |
1226 | |
1227 | continue; | |
1228 | } | |
1229 | ||
30159ec7 | 1230 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1231 | if (strcmp(token, ss->name)) |
1232 | continue; | |
1233 | if (ss->disabled) | |
1234 | continue; | |
1235 | ||
1236 | /* Mutually exclusive option 'all' + subsystem name */ | |
1237 | if (all_ss) | |
1238 | return -EINVAL; | |
a1a71b45 | 1239 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1240 | one_ss = true; |
1241 | ||
1242 | break; | |
1243 | } | |
1244 | if (i == CGROUP_SUBSYS_COUNT) | |
1245 | return -ENOENT; | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1250 | * otherwise if 'none', 'name=' and a subsystem name options |
1251 | * were not specified, let's default to 'all' | |
32a8cf23 | 1252 | */ |
30159ec7 TH |
1253 | if (all_ss || (!one_ss && !opts->none && !opts->name)) |
1254 | for_each_subsys(ss, i) | |
1255 | if (!ss->disabled) | |
1256 | set_bit(i, &opts->subsys_mask); | |
ddbcc7e8 | 1257 | |
2c6ab6d2 PM |
1258 | /* Consistency checks */ |
1259 | ||
873fe09e TH |
1260 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1261 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1262 | ||
1263 | if (opts->flags & CGRP_ROOT_NOPREFIX) { | |
1264 | pr_err("cgroup: sane_behavior: noprefix is not allowed\n"); | |
1265 | return -EINVAL; | |
1266 | } | |
1267 | ||
1268 | if (opts->cpuset_clone_children) { | |
1269 | pr_err("cgroup: sane_behavior: clone_children is not allowed\n"); | |
1270 | return -EINVAL; | |
1271 | } | |
86bf4b68 TH |
1272 | |
1273 | if (opts->flags & CGRP_ROOT_XATTR) | |
1274 | pr_warning("cgroup: sane_behavior: xattr is always available, flag unnecessary\n"); | |
873fe09e TH |
1275 | } |
1276 | ||
f9ab5b5b LZ |
1277 | /* |
1278 | * Option noprefix was introduced just for backward compatibility | |
1279 | * with the old cpuset, so we allow noprefix only if mounting just | |
1280 | * the cpuset subsystem. | |
1281 | */ | |
93438629 | 1282 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1283 | return -EINVAL; |
1284 | ||
2c6ab6d2 PM |
1285 | |
1286 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1287 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1288 | return -EINVAL; |
1289 | ||
1290 | /* | |
1291 | * We either have to specify by name or by subsystems. (So all | |
1292 | * empty hierarchies must have a name). | |
1293 | */ | |
a1a71b45 | 1294 | if (!opts->subsys_mask && !opts->name) |
ddbcc7e8 PM |
1295 | return -EINVAL; |
1296 | ||
1297 | return 0; | |
1298 | } | |
1299 | ||
2bd59d48 | 1300 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1301 | { |
1302 | int ret = 0; | |
2bd59d48 | 1303 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1304 | struct cgroup_sb_opts opts; |
a1a71b45 | 1305 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1306 | |
873fe09e TH |
1307 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1308 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1309 | return -EINVAL; | |
1310 | } | |
1311 | ||
ace2bee8 | 1312 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1313 | mutex_lock(&cgroup_mutex); |
1314 | ||
1315 | /* See what subsystems are wanted */ | |
1316 | ret = parse_cgroupfs_options(data, &opts); | |
1317 | if (ret) | |
1318 | goto out_unlock; | |
1319 | ||
a8a648c4 | 1320 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1321 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1322 | task_tgid_nr(current), current->comm); | |
1323 | ||
a1a71b45 AR |
1324 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1325 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1326 | |
cf5d5941 | 1327 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1328 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1329 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1330 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1331 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1332 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1333 | ret = -EINVAL; |
1334 | goto out_unlock; | |
1335 | } | |
1336 | ||
f172e67c TH |
1337 | /* remounting is not allowed for populated hierarchies */ |
1338 | if (root->number_of_cgroups > 1) { | |
1339 | ret = -EBUSY; | |
0670e08b | 1340 | goto out_unlock; |
cf5d5941 | 1341 | } |
ddbcc7e8 | 1342 | |
a8a648c4 | 1343 | ret = rebind_subsystems(root, added_mask, removed_mask); |
3126121f | 1344 | if (ret) |
0670e08b | 1345 | goto out_unlock; |
ddbcc7e8 | 1346 | |
69e943b7 TH |
1347 | if (opts.release_agent) { |
1348 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1349 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1350 | spin_unlock(&release_agent_path_lock); |
1351 | } | |
ddbcc7e8 | 1352 | out_unlock: |
66bdc9cf | 1353 | kfree(opts.release_agent); |
c6d57f33 | 1354 | kfree(opts.name); |
ddbcc7e8 | 1355 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1356 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1357 | return ret; |
1358 | } | |
1359 | ||
cc31edce PM |
1360 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1361 | { | |
2bd59d48 | 1362 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1363 | INIT_LIST_HEAD(&cgrp->sibling); |
1364 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1365 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1366 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1367 | INIT_LIST_HEAD(&cgrp->pidlists); |
1368 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1369 | cgrp->dummy_css.cgroup = cgrp; |
cc31edce | 1370 | } |
c6d57f33 | 1371 | |
ddbcc7e8 PM |
1372 | static void init_cgroup_root(struct cgroupfs_root *root) |
1373 | { | |
bd89aabc | 1374 | struct cgroup *cgrp = &root->top_cgroup; |
b0ca5a84 | 1375 | |
2bd59d48 | 1376 | atomic_set(&root->refcnt, 1); |
ddbcc7e8 PM |
1377 | INIT_LIST_HEAD(&root->root_list); |
1378 | root->number_of_cgroups = 1; | |
bd89aabc | 1379 | cgrp->root = root; |
a4ea1cc9 | 1380 | RCU_INIT_POINTER(cgrp->name, &root_cgroup_name); |
cc31edce | 1381 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1382 | idr_init(&root->cgroup_idr); |
ddbcc7e8 PM |
1383 | } |
1384 | ||
c6d57f33 PM |
1385 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1386 | { | |
1387 | struct cgroupfs_root *root; | |
1388 | ||
a1a71b45 | 1389 | if (!opts->subsys_mask && !opts->none) |
2bd59d48 | 1390 | return ERR_PTR(-EINVAL); |
c6d57f33 PM |
1391 | |
1392 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1393 | if (!root) | |
1394 | return ERR_PTR(-ENOMEM); | |
1395 | ||
1396 | init_cgroup_root(root); | |
2c6ab6d2 | 1397 | |
1672d040 TH |
1398 | /* |
1399 | * We need to set @root->subsys_mask now so that @root can be | |
1400 | * matched by cgroup_test_super() before it finishes | |
1401 | * initialization; otherwise, competing mounts with the same | |
1402 | * options may try to bind the same subsystems instead of waiting | |
1403 | * for the first one leading to unexpected mount errors. | |
1404 | * SUBSYS_BOUND will be set once actual binding is complete. | |
1405 | */ | |
a1a71b45 | 1406 | root->subsys_mask = opts->subsys_mask; |
c6d57f33 PM |
1407 | root->flags = opts->flags; |
1408 | if (opts->release_agent) | |
1409 | strcpy(root->release_agent_path, opts->release_agent); | |
1410 | if (opts->name) | |
1411 | strcpy(root->name, opts->name); | |
2260e7fc TH |
1412 | if (opts->cpuset_clone_children) |
1413 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1414 | return root; |
1415 | } | |
1416 | ||
d427dfeb TH |
1417 | static int cgroup_setup_root(struct cgroupfs_root *root) |
1418 | { | |
1419 | LIST_HEAD(tmp_links); | |
d427dfeb | 1420 | struct cgroup *root_cgrp = &root->top_cgroup; |
d427dfeb | 1421 | struct css_set *cset; |
d427dfeb TH |
1422 | int i, ret; |
1423 | ||
1424 | lockdep_assert_held(&cgroup_tree_mutex); | |
1425 | lockdep_assert_held(&cgroup_mutex); | |
d427dfeb TH |
1426 | |
1427 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); | |
1428 | if (ret < 0) | |
2bd59d48 | 1429 | goto out; |
d427dfeb TH |
1430 | root_cgrp->id = ret; |
1431 | ||
d427dfeb TH |
1432 | /* |
1433 | * We're accessing css_set_count without locking css_set_lock here, | |
1434 | * but that's OK - it can only be increased by someone holding | |
1435 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1436 | * have some link structures left over | |
1437 | */ | |
1438 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1439 | if (ret) | |
2bd59d48 | 1440 | goto out; |
d427dfeb TH |
1441 | |
1442 | /* ID 0 is reserved for dummy root, 1 for unified hierarchy */ | |
1443 | ret = cgroup_init_root_id(root, 2, 0); | |
1444 | if (ret) | |
2bd59d48 | 1445 | goto out; |
d427dfeb | 1446 | |
2bd59d48 TH |
1447 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1448 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1449 | root_cgrp); | |
1450 | if (IS_ERR(root->kf_root)) { | |
1451 | ret = PTR_ERR(root->kf_root); | |
1452 | goto exit_root_id; | |
1453 | } | |
1454 | root_cgrp->kn = root->kf_root->kn; | |
d427dfeb TH |
1455 | |
1456 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); | |
1457 | if (ret) | |
2bd59d48 | 1458 | goto destroy_root; |
d427dfeb TH |
1459 | |
1460 | ret = rebind_subsystems(root, root->subsys_mask, 0); | |
1461 | if (ret) | |
2bd59d48 | 1462 | goto destroy_root; |
d427dfeb TH |
1463 | |
1464 | /* | |
1465 | * There must be no failure case after here, since rebinding takes | |
1466 | * care of subsystems' refcounts, which are explicitly dropped in | |
1467 | * the failure exit path. | |
1468 | */ | |
1469 | list_add(&root->root_list, &cgroup_roots); | |
1470 | cgroup_root_count++; | |
1471 | ||
1472 | /* | |
1473 | * Link the top cgroup in this hierarchy into all the css_set | |
1474 | * objects. | |
1475 | */ | |
1476 | write_lock(&css_set_lock); | |
1477 | hash_for_each(css_set_table, i, cset, hlist) | |
1478 | link_css_set(&tmp_links, cset, root_cgrp); | |
1479 | write_unlock(&css_set_lock); | |
1480 | ||
1481 | BUG_ON(!list_empty(&root_cgrp->children)); | |
1482 | BUG_ON(root->number_of_cgroups != 1); | |
1483 | ||
2bd59d48 | 1484 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1485 | ret = 0; |
2bd59d48 | 1486 | goto out; |
d427dfeb | 1487 | |
2bd59d48 TH |
1488 | destroy_root: |
1489 | kernfs_destroy_root(root->kf_root); | |
1490 | root->kf_root = NULL; | |
1491 | exit_root_id: | |
d427dfeb | 1492 | cgroup_exit_root_id(root); |
2bd59d48 | 1493 | out: |
d427dfeb TH |
1494 | free_cgrp_cset_links(&tmp_links); |
1495 | return ret; | |
1496 | } | |
1497 | ||
f7e83571 | 1498 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1499 | int flags, const char *unused_dev_name, |
f7e83571 | 1500 | void *data) |
ddbcc7e8 | 1501 | { |
2bd59d48 | 1502 | struct cgroupfs_root *root; |
ddbcc7e8 | 1503 | struct cgroup_sb_opts opts; |
2bd59d48 | 1504 | struct dentry *dentry; |
8e30e2b8 | 1505 | int ret; |
ddbcc7e8 | 1506 | |
8e30e2b8 | 1507 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1508 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1509 | |
1510 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1511 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1512 | if (ret) |
8e30e2b8 | 1513 | goto out_unlock; |
ddbcc7e8 | 1514 | |
2bd59d48 TH |
1515 | /* look for a matching existing root */ |
1516 | for_each_active_root(root) { | |
1517 | bool name_match = false; | |
ddbcc7e8 | 1518 | |
2bd59d48 TH |
1519 | /* |
1520 | * If we asked for a name then it must match. Also, if | |
1521 | * name matches but sybsys_mask doesn't, we should fail. | |
1522 | * Remember whether name matched. | |
1523 | */ | |
1524 | if (opts.name) { | |
1525 | if (strcmp(opts.name, root->name)) | |
1526 | continue; | |
1527 | name_match = true; | |
1528 | } | |
ddbcc7e8 | 1529 | |
c6d57f33 | 1530 | /* |
2bd59d48 TH |
1531 | * If we asked for subsystems (or explicitly for no |
1532 | * subsystems) then they must match. | |
c6d57f33 | 1533 | */ |
2bd59d48 TH |
1534 | if ((opts.subsys_mask || opts.none) && |
1535 | (opts.subsys_mask != root->subsys_mask)) { | |
1536 | if (!name_match) | |
1537 | continue; | |
1538 | ret = -EBUSY; | |
1539 | goto out_unlock; | |
1540 | } | |
873fe09e | 1541 | |
c7ba8287 | 1542 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1543 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1544 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1545 | ret = -EINVAL; | |
8e30e2b8 | 1546 | goto out_unlock; |
2a0ff3fb JL |
1547 | } else { |
1548 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1549 | } | |
873fe09e | 1550 | } |
2bd59d48 TH |
1551 | |
1552 | cgroup_get_root(root); | |
1553 | goto out_unlock; | |
ddbcc7e8 PM |
1554 | } |
1555 | ||
2bd59d48 TH |
1556 | /* no such thing, create a new one */ |
1557 | root = cgroup_root_from_opts(&opts); | |
1558 | if (IS_ERR(root)) { | |
1559 | ret = PTR_ERR(root); | |
1560 | goto out_unlock; | |
1561 | } | |
1562 | ||
1563 | ret = cgroup_setup_root(root); | |
1564 | if (ret) | |
1565 | cgroup_free_root(root); | |
1566 | ||
8e30e2b8 | 1567 | out_unlock: |
e25e2cbb | 1568 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1569 | mutex_unlock(&cgroup_tree_mutex); |
8e30e2b8 | 1570 | |
c6d57f33 PM |
1571 | kfree(opts.release_agent); |
1572 | kfree(opts.name); | |
8e30e2b8 | 1573 | |
2bd59d48 | 1574 | if (ret) |
8e30e2b8 | 1575 | return ERR_PTR(ret); |
2bd59d48 TH |
1576 | |
1577 | dentry = kernfs_mount(fs_type, flags, root->kf_root); | |
1578 | if (IS_ERR(dentry)) | |
1579 | cgroup_put_root(root); | |
1580 | return dentry; | |
1581 | } | |
1582 | ||
1583 | static void cgroup_kill_sb(struct super_block *sb) | |
1584 | { | |
1585 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
1586 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); | |
1587 | ||
1588 | cgroup_put_root(root); | |
1589 | kernfs_kill_sb(sb); | |
ddbcc7e8 PM |
1590 | } |
1591 | ||
ddbcc7e8 PM |
1592 | static struct file_system_type cgroup_fs_type = { |
1593 | .name = "cgroup", | |
f7e83571 | 1594 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1595 | .kill_sb = cgroup_kill_sb, |
1596 | }; | |
1597 | ||
676db4af GKH |
1598 | static struct kobject *cgroup_kobj; |
1599 | ||
a043e3b2 LZ |
1600 | /** |
1601 | * cgroup_path - generate the path of a cgroup | |
1602 | * @cgrp: the cgroup in question | |
1603 | * @buf: the buffer to write the path into | |
1604 | * @buflen: the length of the buffer | |
1605 | * | |
65dff759 LZ |
1606 | * Writes path of cgroup into buf. Returns 0 on success, -errno on error. |
1607 | * | |
1608 | * We can't generate cgroup path using dentry->d_name, as accessing | |
1609 | * dentry->name must be protected by irq-unsafe dentry->d_lock or parent | |
1610 | * inode's i_mutex, while on the other hand cgroup_path() can be called | |
1611 | * with some irq-safe spinlocks held. | |
ddbcc7e8 | 1612 | */ |
bd89aabc | 1613 | int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen) |
ddbcc7e8 | 1614 | { |
65dff759 | 1615 | int ret = -ENAMETOOLONG; |
ddbcc7e8 | 1616 | char *start; |
febfcef6 | 1617 | |
da1f296f TH |
1618 | if (!cgrp->parent) { |
1619 | if (strlcpy(buf, "/", buflen) >= buflen) | |
1620 | return -ENAMETOOLONG; | |
ddbcc7e8 PM |
1621 | return 0; |
1622 | } | |
1623 | ||
316eb661 | 1624 | start = buf + buflen - 1; |
316eb661 | 1625 | *start = '\0'; |
9a9686b6 | 1626 | |
65dff759 | 1627 | rcu_read_lock(); |
da1f296f | 1628 | do { |
65dff759 LZ |
1629 | const char *name = cgroup_name(cgrp); |
1630 | int len; | |
1631 | ||
1632 | len = strlen(name); | |
ddbcc7e8 | 1633 | if ((start -= len) < buf) |
65dff759 LZ |
1634 | goto out; |
1635 | memcpy(start, name, len); | |
9a9686b6 | 1636 | |
ddbcc7e8 | 1637 | if (--start < buf) |
65dff759 | 1638 | goto out; |
ddbcc7e8 | 1639 | *start = '/'; |
65dff759 LZ |
1640 | |
1641 | cgrp = cgrp->parent; | |
da1f296f | 1642 | } while (cgrp->parent); |
65dff759 | 1643 | ret = 0; |
ddbcc7e8 | 1644 | memmove(buf, start, buf + buflen - start); |
65dff759 LZ |
1645 | out: |
1646 | rcu_read_unlock(); | |
1647 | return ret; | |
ddbcc7e8 | 1648 | } |
67523c48 | 1649 | EXPORT_SYMBOL_GPL(cgroup_path); |
ddbcc7e8 | 1650 | |
857a2beb | 1651 | /** |
913ffdb5 | 1652 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1653 | * @task: target task |
857a2beb TH |
1654 | * @buf: the buffer to write the path into |
1655 | * @buflen: the length of the buffer | |
1656 | * | |
913ffdb5 TH |
1657 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1658 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1659 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1660 | * cgroup controller callbacks. | |
1661 | * | |
1662 | * Returns 0 on success, fails with -%ENAMETOOLONG if @buflen is too short. | |
857a2beb | 1663 | */ |
913ffdb5 | 1664 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb TH |
1665 | { |
1666 | struct cgroupfs_root *root; | |
913ffdb5 TH |
1667 | struct cgroup *cgrp; |
1668 | int hierarchy_id = 1, ret = 0; | |
1669 | ||
1670 | if (buflen < 2) | |
1671 | return -ENAMETOOLONG; | |
857a2beb TH |
1672 | |
1673 | mutex_lock(&cgroup_mutex); | |
1674 | ||
913ffdb5 TH |
1675 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1676 | ||
857a2beb TH |
1677 | if (root) { |
1678 | cgrp = task_cgroup_from_root(task, root); | |
1679 | ret = cgroup_path(cgrp, buf, buflen); | |
913ffdb5 TH |
1680 | } else { |
1681 | /* if no hierarchy exists, everyone is in "/" */ | |
1682 | memcpy(buf, "/", 2); | |
857a2beb TH |
1683 | } |
1684 | ||
1685 | mutex_unlock(&cgroup_mutex); | |
857a2beb TH |
1686 | return ret; |
1687 | } | |
913ffdb5 | 1688 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1689 | |
2f7ee569 TH |
1690 | /* |
1691 | * Control Group taskset | |
1692 | */ | |
134d3373 TH |
1693 | struct task_and_cgroup { |
1694 | struct task_struct *task; | |
1695 | struct cgroup *cgrp; | |
6f4b7e63 | 1696 | struct css_set *cset; |
134d3373 TH |
1697 | }; |
1698 | ||
2f7ee569 TH |
1699 | struct cgroup_taskset { |
1700 | struct task_and_cgroup single; | |
1701 | struct flex_array *tc_array; | |
1702 | int tc_array_len; | |
1703 | int idx; | |
1704 | struct cgroup *cur_cgrp; | |
1705 | }; | |
1706 | ||
1707 | /** | |
1708 | * cgroup_taskset_first - reset taskset and return the first task | |
1709 | * @tset: taskset of interest | |
1710 | * | |
1711 | * @tset iteration is initialized and the first task is returned. | |
1712 | */ | |
1713 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1714 | { | |
1715 | if (tset->tc_array) { | |
1716 | tset->idx = 0; | |
1717 | return cgroup_taskset_next(tset); | |
1718 | } else { | |
1719 | tset->cur_cgrp = tset->single.cgrp; | |
1720 | return tset->single.task; | |
1721 | } | |
1722 | } | |
1723 | EXPORT_SYMBOL_GPL(cgroup_taskset_first); | |
1724 | ||
1725 | /** | |
1726 | * cgroup_taskset_next - iterate to the next task in taskset | |
1727 | * @tset: taskset of interest | |
1728 | * | |
1729 | * Return the next task in @tset. Iteration must have been initialized | |
1730 | * with cgroup_taskset_first(). | |
1731 | */ | |
1732 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1733 | { | |
1734 | struct task_and_cgroup *tc; | |
1735 | ||
1736 | if (!tset->tc_array || tset->idx >= tset->tc_array_len) | |
1737 | return NULL; | |
1738 | ||
1739 | tc = flex_array_get(tset->tc_array, tset->idx++); | |
1740 | tset->cur_cgrp = tc->cgrp; | |
1741 | return tc->task; | |
1742 | } | |
1743 | EXPORT_SYMBOL_GPL(cgroup_taskset_next); | |
1744 | ||
1745 | /** | |
d99c8727 | 1746 | * cgroup_taskset_cur_css - return the matching css for the current task |
2f7ee569 | 1747 | * @tset: taskset of interest |
d99c8727 | 1748 | * @subsys_id: the ID of the target subsystem |
2f7ee569 | 1749 | * |
d99c8727 TH |
1750 | * Return the css for the current (last returned) task of @tset for |
1751 | * subsystem specified by @subsys_id. This function must be preceded by | |
1752 | * either cgroup_taskset_first() or cgroup_taskset_next(). | |
2f7ee569 | 1753 | */ |
d99c8727 TH |
1754 | struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset, |
1755 | int subsys_id) | |
2f7ee569 | 1756 | { |
ca8bdcaf | 1757 | return cgroup_css(tset->cur_cgrp, cgroup_subsys[subsys_id]); |
2f7ee569 | 1758 | } |
d99c8727 | 1759 | EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css); |
2f7ee569 TH |
1760 | |
1761 | /** | |
1762 | * cgroup_taskset_size - return the number of tasks in taskset | |
1763 | * @tset: taskset of interest | |
1764 | */ | |
1765 | int cgroup_taskset_size(struct cgroup_taskset *tset) | |
1766 | { | |
1767 | return tset->tc_array ? tset->tc_array_len : 1; | |
1768 | } | |
1769 | EXPORT_SYMBOL_GPL(cgroup_taskset_size); | |
1770 | ||
1771 | ||
74a1166d BB |
1772 | /* |
1773 | * cgroup_task_migrate - move a task from one cgroup to another. | |
1774 | * | |
d0b2fdd2 | 1775 | * Must be called with cgroup_mutex and threadgroup locked. |
74a1166d | 1776 | */ |
5abb8855 TH |
1777 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1778 | struct task_struct *tsk, | |
1779 | struct css_set *new_cset) | |
74a1166d | 1780 | { |
5abb8855 | 1781 | struct css_set *old_cset; |
74a1166d BB |
1782 | |
1783 | /* | |
026085ef MSB |
1784 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1785 | * setting such that we can't race against cgroup_exit() changing the | |
1786 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1787 | */ |
c84cdf75 | 1788 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1789 | old_cset = task_css_set(tsk); |
74a1166d | 1790 | |
74a1166d | 1791 | task_lock(tsk); |
5abb8855 | 1792 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d BB |
1793 | task_unlock(tsk); |
1794 | ||
1795 | /* Update the css_set linked lists if we're using them */ | |
1796 | write_lock(&css_set_lock); | |
1797 | if (!list_empty(&tsk->cg_list)) | |
5abb8855 | 1798 | list_move(&tsk->cg_list, &new_cset->tasks); |
74a1166d BB |
1799 | write_unlock(&css_set_lock); |
1800 | ||
1801 | /* | |
5abb8855 TH |
1802 | * We just gained a reference on old_cset by taking it from the |
1803 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1804 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1805 | */ |
5abb8855 TH |
1806 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
1807 | put_css_set(old_cset); | |
74a1166d BB |
1808 | } |
1809 | ||
a043e3b2 | 1810 | /** |
081aa458 | 1811 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup |
74a1166d | 1812 | * @cgrp: the cgroup to attach to |
081aa458 LZ |
1813 | * @tsk: the task or the leader of the threadgroup to be attached |
1814 | * @threadgroup: attach the whole threadgroup? | |
74a1166d | 1815 | * |
257058ae | 1816 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
081aa458 | 1817 | * task_lock of @tsk or each thread in the threadgroup individually in turn. |
74a1166d | 1818 | */ |
47cfcd09 TH |
1819 | static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk, |
1820 | bool threadgroup) | |
74a1166d BB |
1821 | { |
1822 | int retval, i, group_size; | |
74a1166d | 1823 | struct cgroupfs_root *root = cgrp->root; |
1c6727af | 1824 | struct cgroup_subsys_state *css, *failed_css = NULL; |
74a1166d | 1825 | /* threadgroup list cursor and array */ |
081aa458 | 1826 | struct task_struct *leader = tsk; |
134d3373 | 1827 | struct task_and_cgroup *tc; |
d846687d | 1828 | struct flex_array *group; |
2f7ee569 | 1829 | struct cgroup_taskset tset = { }; |
74a1166d BB |
1830 | |
1831 | /* | |
1832 | * step 0: in order to do expensive, possibly blocking operations for | |
1833 | * every thread, we cannot iterate the thread group list, since it needs | |
1834 | * rcu or tasklist locked. instead, build an array of all threads in the | |
257058ae TH |
1835 | * group - group_rwsem prevents new threads from appearing, and if |
1836 | * threads exit, this will just be an over-estimate. | |
74a1166d | 1837 | */ |
081aa458 LZ |
1838 | if (threadgroup) |
1839 | group_size = get_nr_threads(tsk); | |
1840 | else | |
1841 | group_size = 1; | |
d846687d | 1842 | /* flex_array supports very large thread-groups better than kmalloc. */ |
134d3373 | 1843 | group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL); |
74a1166d BB |
1844 | if (!group) |
1845 | return -ENOMEM; | |
d846687d | 1846 | /* pre-allocate to guarantee space while iterating in rcu read-side. */ |
3ac1707a | 1847 | retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL); |
d846687d BB |
1848 | if (retval) |
1849 | goto out_free_group_list; | |
74a1166d | 1850 | |
74a1166d | 1851 | i = 0; |
fb5d2b4c MSB |
1852 | /* |
1853 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1854 | * already PF_EXITING could be freed from underneath us unless we | |
1855 | * take an rcu_read_lock. | |
1856 | */ | |
1857 | rcu_read_lock(); | |
74a1166d | 1858 | do { |
134d3373 TH |
1859 | struct task_and_cgroup ent; |
1860 | ||
cd3d0952 TH |
1861 | /* @tsk either already exited or can't exit until the end */ |
1862 | if (tsk->flags & PF_EXITING) | |
ea84753c | 1863 | goto next; |
cd3d0952 | 1864 | |
74a1166d BB |
1865 | /* as per above, nr_threads may decrease, but not increase. */ |
1866 | BUG_ON(i >= group_size); | |
134d3373 TH |
1867 | ent.task = tsk; |
1868 | ent.cgrp = task_cgroup_from_root(tsk, root); | |
892a2b90 MSB |
1869 | /* nothing to do if this task is already in the cgroup */ |
1870 | if (ent.cgrp == cgrp) | |
ea84753c | 1871 | goto next; |
61d1d219 MSB |
1872 | /* |
1873 | * saying GFP_ATOMIC has no effect here because we did prealloc | |
1874 | * earlier, but it's good form to communicate our expectations. | |
1875 | */ | |
134d3373 | 1876 | retval = flex_array_put(group, i, &ent, GFP_ATOMIC); |
d846687d | 1877 | BUG_ON(retval != 0); |
74a1166d | 1878 | i++; |
ea84753c | 1879 | next: |
081aa458 LZ |
1880 | if (!threadgroup) |
1881 | break; | |
74a1166d | 1882 | } while_each_thread(leader, tsk); |
fb5d2b4c | 1883 | rcu_read_unlock(); |
74a1166d BB |
1884 | /* remember the number of threads in the array for later. */ |
1885 | group_size = i; | |
2f7ee569 TH |
1886 | tset.tc_array = group; |
1887 | tset.tc_array_len = group_size; | |
74a1166d | 1888 | |
134d3373 TH |
1889 | /* methods shouldn't be called if no task is actually migrating */ |
1890 | retval = 0; | |
892a2b90 | 1891 | if (!group_size) |
b07ef774 | 1892 | goto out_free_group_list; |
134d3373 | 1893 | |
74a1166d BB |
1894 | /* |
1895 | * step 1: check that we can legitimately attach to the cgroup. | |
1896 | */ | |
1c6727af TH |
1897 | for_each_css(css, i, cgrp) { |
1898 | if (css->ss->can_attach) { | |
1899 | retval = css->ss->can_attach(css, &tset); | |
74a1166d | 1900 | if (retval) { |
1c6727af | 1901 | failed_css = css; |
74a1166d BB |
1902 | goto out_cancel_attach; |
1903 | } | |
1904 | } | |
74a1166d BB |
1905 | } |
1906 | ||
1907 | /* | |
1908 | * step 2: make sure css_sets exist for all threads to be migrated. | |
1909 | * we use find_css_set, which allocates a new one if necessary. | |
1910 | */ | |
74a1166d | 1911 | for (i = 0; i < group_size; i++) { |
a8ad805c TH |
1912 | struct css_set *old_cset; |
1913 | ||
134d3373 | 1914 | tc = flex_array_get(group, i); |
a8ad805c | 1915 | old_cset = task_css_set(tc->task); |
6f4b7e63 LZ |
1916 | tc->cset = find_css_set(old_cset, cgrp); |
1917 | if (!tc->cset) { | |
61d1d219 MSB |
1918 | retval = -ENOMEM; |
1919 | goto out_put_css_set_refs; | |
74a1166d BB |
1920 | } |
1921 | } | |
1922 | ||
1923 | /* | |
494c167c TH |
1924 | * step 3: now that we're guaranteed success wrt the css_sets, |
1925 | * proceed to move all tasks to the new cgroup. There are no | |
1926 | * failure cases after here, so this is the commit point. | |
74a1166d | 1927 | */ |
74a1166d | 1928 | for (i = 0; i < group_size; i++) { |
134d3373 | 1929 | tc = flex_array_get(group, i); |
6f4b7e63 | 1930 | cgroup_task_migrate(tc->cgrp, tc->task, tc->cset); |
74a1166d BB |
1931 | } |
1932 | /* nothing is sensitive to fork() after this point. */ | |
1933 | ||
1934 | /* | |
494c167c | 1935 | * step 4: do subsystem attach callbacks. |
74a1166d | 1936 | */ |
1c6727af TH |
1937 | for_each_css(css, i, cgrp) |
1938 | if (css->ss->attach) | |
1939 | css->ss->attach(css, &tset); | |
74a1166d BB |
1940 | |
1941 | /* | |
1942 | * step 5: success! and cleanup | |
1943 | */ | |
74a1166d | 1944 | retval = 0; |
61d1d219 MSB |
1945 | out_put_css_set_refs: |
1946 | if (retval) { | |
1947 | for (i = 0; i < group_size; i++) { | |
1948 | tc = flex_array_get(group, i); | |
6f4b7e63 | 1949 | if (!tc->cset) |
61d1d219 | 1950 | break; |
6f4b7e63 | 1951 | put_css_set(tc->cset); |
61d1d219 | 1952 | } |
74a1166d BB |
1953 | } |
1954 | out_cancel_attach: | |
74a1166d | 1955 | if (retval) { |
1c6727af TH |
1956 | for_each_css(css, i, cgrp) { |
1957 | if (css == failed_css) | |
74a1166d | 1958 | break; |
1c6727af TH |
1959 | if (css->ss->cancel_attach) |
1960 | css->ss->cancel_attach(css, &tset); | |
74a1166d BB |
1961 | } |
1962 | } | |
74a1166d | 1963 | out_free_group_list: |
d846687d | 1964 | flex_array_free(group); |
74a1166d BB |
1965 | return retval; |
1966 | } | |
1967 | ||
1968 | /* | |
1969 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
1970 | * function to attach either it or all tasks in its threadgroup. Will lock |
1971 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 1972 | */ |
74a1166d | 1973 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 1974 | { |
bbcb81d0 | 1975 | struct task_struct *tsk; |
c69e8d9c | 1976 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
1977 | int ret; |
1978 | ||
74a1166d BB |
1979 | if (!cgroup_lock_live_group(cgrp)) |
1980 | return -ENODEV; | |
1981 | ||
b78949eb MSB |
1982 | retry_find_task: |
1983 | rcu_read_lock(); | |
bbcb81d0 | 1984 | if (pid) { |
73507f33 | 1985 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
1986 | if (!tsk) { |
1987 | rcu_read_unlock(); | |
dd4b0a46 | 1988 | ret = -ESRCH; |
b78949eb | 1989 | goto out_unlock_cgroup; |
bbcb81d0 | 1990 | } |
74a1166d BB |
1991 | /* |
1992 | * even if we're attaching all tasks in the thread group, we | |
1993 | * only need to check permissions on one of them. | |
1994 | */ | |
c69e8d9c | 1995 | tcred = __task_cred(tsk); |
14a590c3 EB |
1996 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
1997 | !uid_eq(cred->euid, tcred->uid) && | |
1998 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 1999 | rcu_read_unlock(); |
b78949eb MSB |
2000 | ret = -EACCES; |
2001 | goto out_unlock_cgroup; | |
bbcb81d0 | 2002 | } |
b78949eb MSB |
2003 | } else |
2004 | tsk = current; | |
cd3d0952 TH |
2005 | |
2006 | if (threadgroup) | |
b78949eb | 2007 | tsk = tsk->group_leader; |
c4c27fbd MG |
2008 | |
2009 | /* | |
14a40ffc | 2010 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2011 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2012 | * with no rt_runtime allocated. Just say no. | |
2013 | */ | |
14a40ffc | 2014 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2015 | ret = -EINVAL; |
2016 | rcu_read_unlock(); | |
2017 | goto out_unlock_cgroup; | |
2018 | } | |
2019 | ||
b78949eb MSB |
2020 | get_task_struct(tsk); |
2021 | rcu_read_unlock(); | |
2022 | ||
2023 | threadgroup_lock(tsk); | |
2024 | if (threadgroup) { | |
2025 | if (!thread_group_leader(tsk)) { | |
2026 | /* | |
2027 | * a race with de_thread from another thread's exec() | |
2028 | * may strip us of our leadership, if this happens, | |
2029 | * there is no choice but to throw this task away and | |
2030 | * try again; this is | |
2031 | * "double-double-toil-and-trouble-check locking". | |
2032 | */ | |
2033 | threadgroup_unlock(tsk); | |
2034 | put_task_struct(tsk); | |
2035 | goto retry_find_task; | |
2036 | } | |
081aa458 LZ |
2037 | } |
2038 | ||
2039 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2040 | ||
cd3d0952 TH |
2041 | threadgroup_unlock(tsk); |
2042 | ||
bbcb81d0 | 2043 | put_task_struct(tsk); |
b78949eb | 2044 | out_unlock_cgroup: |
47cfcd09 | 2045 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
2046 | return ret; |
2047 | } | |
2048 | ||
7ae1bad9 TH |
2049 | /** |
2050 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2051 | * @from: attach to all cgroups of a given task | |
2052 | * @tsk: the task to be attached | |
2053 | */ | |
2054 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2055 | { | |
2056 | struct cgroupfs_root *root; | |
2057 | int retval = 0; | |
2058 | ||
47cfcd09 | 2059 | mutex_lock(&cgroup_mutex); |
7ae1bad9 | 2060 | for_each_active_root(root) { |
6f4b7e63 | 2061 | struct cgroup *from_cgrp = task_cgroup_from_root(from, root); |
7ae1bad9 | 2062 | |
6f4b7e63 | 2063 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2064 | if (retval) |
2065 | break; | |
2066 | } | |
47cfcd09 | 2067 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2068 | |
2069 | return retval; | |
2070 | } | |
2071 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2072 | ||
182446d0 TH |
2073 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2074 | struct cftype *cft, u64 pid) | |
74a1166d | 2075 | { |
182446d0 | 2076 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2077 | } |
2078 | ||
182446d0 TH |
2079 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2080 | struct cftype *cft, u64 tgid) | |
af351026 | 2081 | { |
182446d0 | 2082 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2083 | } |
2084 | ||
182446d0 TH |
2085 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
2086 | struct cftype *cft, const char *buffer) | |
e788e066 | 2087 | { |
5f469907 TH |
2088 | struct cgroupfs_root *root = css->cgroup->root; |
2089 | ||
2090 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
182446d0 | 2091 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2092 | return -ENODEV; |
69e943b7 | 2093 | spin_lock(&release_agent_path_lock); |
5f469907 TH |
2094 | strlcpy(root->release_agent_path, buffer, |
2095 | sizeof(root->release_agent_path)); | |
69e943b7 | 2096 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2097 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2098 | return 0; |
2099 | } | |
2100 | ||
2da8ca82 | 2101 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2102 | { |
2da8ca82 | 2103 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2104 | |
e788e066 PM |
2105 | if (!cgroup_lock_live_group(cgrp)) |
2106 | return -ENODEV; | |
2107 | seq_puts(seq, cgrp->root->release_agent_path); | |
2108 | seq_putc(seq, '\n'); | |
47cfcd09 | 2109 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2110 | return 0; |
2111 | } | |
2112 | ||
2da8ca82 | 2113 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2114 | { |
2da8ca82 TH |
2115 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2116 | ||
2117 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2118 | return 0; |
2119 | } | |
2120 | ||
2bd59d48 TH |
2121 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2122 | size_t nbytes, loff_t off) | |
355e0c48 | 2123 | { |
2bd59d48 TH |
2124 | struct cgroup *cgrp = of->kn->parent->priv; |
2125 | struct cftype *cft = of->kn->priv; | |
2126 | struct cgroup_subsys_state *css; | |
a742c59d | 2127 | int ret; |
355e0c48 | 2128 | |
2bd59d48 TH |
2129 | /* |
2130 | * kernfs guarantees that a file isn't deleted with operations in | |
2131 | * flight, which means that the matching css is and stays alive and | |
2132 | * doesn't need to be pinned. The RCU locking is not necessary | |
2133 | * either. It's just for the convenience of using cgroup_css(). | |
2134 | */ | |
2135 | rcu_read_lock(); | |
2136 | css = cgroup_css(cgrp, cft->ss); | |
2137 | rcu_read_unlock(); | |
a742c59d TH |
2138 | |
2139 | if (cft->write_string) { | |
2140 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2141 | } else if (cft->write_u64) { | |
2142 | unsigned long long v; | |
2143 | ret = kstrtoull(buf, 0, &v); | |
2144 | if (!ret) | |
2145 | ret = cft->write_u64(css, cft, v); | |
2146 | } else if (cft->write_s64) { | |
2147 | long long v; | |
2148 | ret = kstrtoll(buf, 0, &v); | |
2149 | if (!ret) | |
2150 | ret = cft->write_s64(css, cft, v); | |
2151 | } else if (cft->trigger) { | |
2152 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2153 | } else { |
a742c59d | 2154 | ret = -EINVAL; |
e73d2c61 | 2155 | } |
2bd59d48 | 2156 | |
a742c59d | 2157 | return ret ?: nbytes; |
355e0c48 PM |
2158 | } |
2159 | ||
6612f05b | 2160 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2161 | { |
2bd59d48 | 2162 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2163 | } |
2164 | ||
6612f05b | 2165 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2166 | { |
2bd59d48 | 2167 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2168 | } |
2169 | ||
6612f05b | 2170 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2171 | { |
2bd59d48 | 2172 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2173 | } |
2174 | ||
91796569 | 2175 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2176 | { |
7da11279 TH |
2177 | struct cftype *cft = seq_cft(m); |
2178 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2179 | |
2da8ca82 TH |
2180 | if (cft->seq_show) |
2181 | return cft->seq_show(m, arg); | |
e73d2c61 | 2182 | |
f4c753b7 | 2183 | if (cft->read_u64) |
896f5199 TH |
2184 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2185 | else if (cft->read_s64) | |
2186 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2187 | else | |
2188 | return -EINVAL; | |
2189 | return 0; | |
91796569 PM |
2190 | } |
2191 | ||
2bd59d48 TH |
2192 | static struct kernfs_ops cgroup_kf_single_ops = { |
2193 | .atomic_write_len = PAGE_SIZE, | |
2194 | .write = cgroup_file_write, | |
2195 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2196 | }; |
2197 | ||
2bd59d48 TH |
2198 | static struct kernfs_ops cgroup_kf_ops = { |
2199 | .atomic_write_len = PAGE_SIZE, | |
2200 | .write = cgroup_file_write, | |
2201 | .seq_start = cgroup_seqfile_start, | |
2202 | .seq_next = cgroup_seqfile_next, | |
2203 | .seq_stop = cgroup_seqfile_stop, | |
2204 | .seq_show = cgroup_seqfile_show, | |
2205 | }; | |
ddbcc7e8 PM |
2206 | |
2207 | /* | |
2208 | * cgroup_rename - Only allow simple rename of directories in place. | |
2209 | */ | |
2bd59d48 TH |
2210 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2211 | const char *new_name_str) | |
ddbcc7e8 | 2212 | { |
2bd59d48 | 2213 | struct cgroup *cgrp = kn->priv; |
65dff759 | 2214 | struct cgroup_name *name, *old_name; |
2bd59d48 | 2215 | int ret; |
65dff759 | 2216 | |
2bd59d48 | 2217 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2218 | return -ENOTDIR; |
2bd59d48 | 2219 | if (kn->parent != new_parent) |
ddbcc7e8 | 2220 | return -EIO; |
65dff759 | 2221 | |
6db8e85c TH |
2222 | /* |
2223 | * This isn't a proper migration and its usefulness is very | |
2224 | * limited. Disallow if sane_behavior. | |
2225 | */ | |
2226 | if (cgroup_sane_behavior(cgrp)) | |
2227 | return -EPERM; | |
2228 | ||
2bd59d48 | 2229 | name = cgroup_alloc_name(new_name_str); |
65dff759 LZ |
2230 | if (!name) |
2231 | return -ENOMEM; | |
2232 | ||
2bd59d48 TH |
2233 | mutex_lock(&cgroup_tree_mutex); |
2234 | mutex_lock(&cgroup_mutex); | |
2235 | ||
2236 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
2237 | if (!ret) { | |
2238 | old_name = rcu_dereference_protected(cgrp->name, true); | |
2239 | rcu_assign_pointer(cgrp->name, name); | |
2240 | } else { | |
2241 | old_name = name; | |
65dff759 LZ |
2242 | } |
2243 | ||
2bd59d48 TH |
2244 | mutex_unlock(&cgroup_mutex); |
2245 | mutex_unlock(&cgroup_tree_mutex); | |
65dff759 LZ |
2246 | |
2247 | kfree_rcu(old_name, rcu_head); | |
2bd59d48 | 2248 | return ret; |
ddbcc7e8 PM |
2249 | } |
2250 | ||
2bb566cb | 2251 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2252 | { |
8d7e6fb0 | 2253 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2254 | struct kernfs_node *kn; |
2255 | struct lock_class_key *key = NULL; | |
05ef1d7c | 2256 | |
2bd59d48 TH |
2257 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2258 | key = &cft->lockdep_key; | |
2259 | #endif | |
2260 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2261 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2262 | NULL, false, key); | |
2263 | if (IS_ERR(kn)) | |
2264 | return PTR_ERR(kn); | |
2265 | return 0; | |
ddbcc7e8 PM |
2266 | } |
2267 | ||
b1f28d31 TH |
2268 | /** |
2269 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2270 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2271 | * @cfts: array of cftypes to be added |
2272 | * @is_add: whether to add or remove | |
2273 | * | |
2274 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2275 | * For removals, this function never fails. If addition fails, this |
2276 | * function doesn't remove files already added. The caller is responsible | |
2277 | * for cleaning up. | |
b1f28d31 | 2278 | */ |
2bb566cb TH |
2279 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2280 | bool is_add) | |
ddbcc7e8 | 2281 | { |
03b1cde6 | 2282 | struct cftype *cft; |
b1f28d31 TH |
2283 | int ret; |
2284 | ||
ace2bee8 | 2285 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2286 | |
2287 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2288 | /* does cft->flags tell us to skip this file on @cgrp? */ |
873fe09e TH |
2289 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2290 | continue; | |
f33fddc2 G |
2291 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2292 | continue; | |
2293 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2294 | continue; | |
2295 | ||
2739d3cc | 2296 | if (is_add) { |
2bb566cb | 2297 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2298 | if (ret) { |
2739d3cc | 2299 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2300 | cft->name, ret); |
2301 | return ret; | |
2302 | } | |
2739d3cc LZ |
2303 | } else { |
2304 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2305 | } |
ddbcc7e8 | 2306 | } |
b1f28d31 | 2307 | return 0; |
ddbcc7e8 PM |
2308 | } |
2309 | ||
21a2d343 | 2310 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2311 | { |
2312 | LIST_HEAD(pending); | |
2bb566cb | 2313 | struct cgroup_subsys *ss = cfts[0].ss; |
492eb21b | 2314 | struct cgroup *root = &ss->root->top_cgroup; |
492eb21b | 2315 | struct cgroup_subsys_state *css; |
9ccece80 | 2316 | int ret = 0; |
8e3f6541 | 2317 | |
21a2d343 | 2318 | lockdep_assert_held(&cgroup_tree_mutex); |
4ac06017 | 2319 | |
21a2d343 TH |
2320 | /* don't bother if @ss isn't attached */ |
2321 | if (ss->root == &cgroup_dummy_root) | |
9ccece80 | 2322 | return 0; |
e8c82d20 | 2323 | |
e8c82d20 | 2324 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2325 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2326 | struct cgroup *cgrp = css->cgroup; |
2327 | ||
e8c82d20 LZ |
2328 | if (cgroup_is_dead(cgrp)) |
2329 | continue; | |
2330 | ||
21a2d343 | 2331 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2332 | if (ret) |
2333 | break; | |
8e3f6541 | 2334 | } |
21a2d343 TH |
2335 | |
2336 | if (is_add && !ret) | |
2337 | kernfs_activate(root->kn); | |
9ccece80 | 2338 | return ret; |
8e3f6541 TH |
2339 | } |
2340 | ||
2da440a2 TH |
2341 | static void cgroup_exit_cftypes(struct cftype *cfts) |
2342 | { | |
2343 | struct cftype *cft; | |
2344 | ||
2bd59d48 TH |
2345 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2346 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2347 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2348 | kfree(cft->kf_ops); | |
2349 | cft->kf_ops = NULL; | |
2da440a2 | 2350 | cft->ss = NULL; |
2bd59d48 | 2351 | } |
2da440a2 TH |
2352 | } |
2353 | ||
2bd59d48 | 2354 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2355 | { |
2356 | struct cftype *cft; | |
2357 | ||
2bd59d48 TH |
2358 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2359 | struct kernfs_ops *kf_ops; | |
2360 | ||
0adb0704 TH |
2361 | WARN_ON(cft->ss || cft->kf_ops); |
2362 | ||
2bd59d48 TH |
2363 | if (cft->seq_start) |
2364 | kf_ops = &cgroup_kf_ops; | |
2365 | else | |
2366 | kf_ops = &cgroup_kf_single_ops; | |
2367 | ||
2368 | /* | |
2369 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2370 | * make a copy of kf_ops to set its atomic_write_len. | |
2371 | */ | |
2372 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2373 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2374 | if (!kf_ops) { | |
2375 | cgroup_exit_cftypes(cfts); | |
2376 | return -ENOMEM; | |
2377 | } | |
2378 | kf_ops->atomic_write_len = cft->max_write_len; | |
2379 | } | |
2380 | ||
2381 | cft->kf_ops = kf_ops; | |
2da440a2 | 2382 | cft->ss = ss; |
2bd59d48 TH |
2383 | } |
2384 | ||
2385 | return 0; | |
2da440a2 TH |
2386 | } |
2387 | ||
21a2d343 TH |
2388 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
2389 | { | |
2390 | lockdep_assert_held(&cgroup_tree_mutex); | |
2391 | ||
2392 | if (!cfts || !cfts[0].ss) | |
2393 | return -ENOENT; | |
2394 | ||
2395 | list_del(&cfts->node); | |
2396 | cgroup_apply_cftypes(cfts, false); | |
2397 | cgroup_exit_cftypes(cfts); | |
2398 | return 0; | |
2399 | } | |
2400 | ||
80b13586 TH |
2401 | /** |
2402 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
2403 | * @cfts: zero-length name terminated array of cftypes | |
2404 | * | |
2405 | * Unregister @cfts. Files described by @cfts are removed from all | |
2406 | * existing cgroups and all future cgroups won't have them either. This | |
2407 | * function can be called anytime whether @cfts' subsys is attached or not. | |
2408 | * | |
2409 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2410 | * registered. | |
2411 | */ | |
2412 | int cgroup_rm_cftypes(struct cftype *cfts) | |
2413 | { | |
21a2d343 | 2414 | int ret; |
80b13586 | 2415 | |
21a2d343 TH |
2416 | mutex_lock(&cgroup_tree_mutex); |
2417 | ret = cgroup_rm_cftypes_locked(cfts); | |
2418 | mutex_unlock(&cgroup_tree_mutex); | |
2419 | return ret; | |
80b13586 TH |
2420 | } |
2421 | ||
8e3f6541 TH |
2422 | /** |
2423 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2424 | * @ss: target cgroup subsystem | |
2425 | * @cfts: zero-length name terminated array of cftypes | |
2426 | * | |
2427 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2428 | * existing cgroups to which @ss is attached and all future cgroups will | |
2429 | * have them too. This function can be called anytime whether @ss is | |
2430 | * attached or not. | |
2431 | * | |
2432 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2433 | * function currently returns 0 as long as @cfts registration is successful | |
2434 | * even if some file creation attempts on existing cgroups fail. | |
2435 | */ | |
03b1cde6 | 2436 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 2437 | { |
9ccece80 | 2438 | int ret; |
8e3f6541 | 2439 | |
2bd59d48 TH |
2440 | ret = cgroup_init_cftypes(ss, cfts); |
2441 | if (ret) | |
2442 | return ret; | |
2bb566cb | 2443 | |
21a2d343 TH |
2444 | mutex_lock(&cgroup_tree_mutex); |
2445 | ||
0adb0704 | 2446 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 2447 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 2448 | if (ret) |
21a2d343 TH |
2449 | cgroup_rm_cftypes_locked(cfts); |
2450 | ||
2451 | mutex_unlock(&cgroup_tree_mutex); | |
9ccece80 | 2452 | return ret; |
8e3f6541 TH |
2453 | } |
2454 | EXPORT_SYMBOL_GPL(cgroup_add_cftypes); | |
2455 | ||
a043e3b2 LZ |
2456 | /** |
2457 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2458 | * @cgrp: the cgroup in question | |
2459 | * | |
2460 | * Return the number of tasks in the cgroup. | |
2461 | */ | |
bd89aabc | 2462 | int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2463 | { |
2464 | int count = 0; | |
69d0206c | 2465 | struct cgrp_cset_link *link; |
817929ec PM |
2466 | |
2467 | read_lock(&css_set_lock); | |
69d0206c TH |
2468 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2469 | count += atomic_read(&link->cset->refcount); | |
817929ec | 2470 | read_unlock(&css_set_lock); |
bbcb81d0 PM |
2471 | return count; |
2472 | } | |
2473 | ||
817929ec | 2474 | /* |
0942eeee TH |
2475 | * To reduce the fork() overhead for systems that are not actually using |
2476 | * their cgroups capability, we don't maintain the lists running through | |
2477 | * each css_set to its tasks until we see the list actually used - in other | |
72ec7029 | 2478 | * words after the first call to css_task_iter_start(). |
31a7df01 | 2479 | */ |
3df91fe3 | 2480 | static void cgroup_enable_task_cg_lists(void) |
31a7df01 CW |
2481 | { |
2482 | struct task_struct *p, *g; | |
2483 | write_lock(&css_set_lock); | |
2484 | use_task_css_set_links = 1; | |
3ce3230a FW |
2485 | /* |
2486 | * We need tasklist_lock because RCU is not safe against | |
2487 | * while_each_thread(). Besides, a forking task that has passed | |
2488 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
2489 | * is not guaranteed to have its child immediately visible in the | |
2490 | * tasklist if we walk through it with RCU. | |
2491 | */ | |
2492 | read_lock(&tasklist_lock); | |
31a7df01 CW |
2493 | do_each_thread(g, p) { |
2494 | task_lock(p); | |
0e04388f LZ |
2495 | /* |
2496 | * We should check if the process is exiting, otherwise | |
2497 | * it will race with cgroup_exit() in that the list | |
2498 | * entry won't be deleted though the process has exited. | |
2499 | */ | |
2500 | if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list)) | |
a8ad805c | 2501 | list_add(&p->cg_list, &task_css_set(p)->tasks); |
31a7df01 CW |
2502 | task_unlock(p); |
2503 | } while_each_thread(g, p); | |
3ce3230a | 2504 | read_unlock(&tasklist_lock); |
31a7df01 CW |
2505 | write_unlock(&css_set_lock); |
2506 | } | |
2507 | ||
53fa5261 | 2508 | /** |
492eb21b TH |
2509 | * css_next_child - find the next child of a given css |
2510 | * @pos_css: the current position (%NULL to initiate traversal) | |
2511 | * @parent_css: css whose children to walk | |
53fa5261 | 2512 | * |
492eb21b | 2513 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
2514 | * under either cgroup_mutex or RCU read lock. The only requirement is |
2515 | * that @parent_css and @pos_css are accessible. The next sibling is | |
2516 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 2517 | */ |
492eb21b TH |
2518 | struct cgroup_subsys_state * |
2519 | css_next_child(struct cgroup_subsys_state *pos_css, | |
2520 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 2521 | { |
492eb21b TH |
2522 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
2523 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
2524 | struct cgroup *next; |
2525 | ||
ace2bee8 | 2526 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
2527 | |
2528 | /* | |
2529 | * @pos could already have been removed. Once a cgroup is removed, | |
2530 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
2531 | * changes. As CGRP_DEAD assertion is serialized and happens |
2532 | * before the cgroup is taken off the ->sibling list, if we see it | |
2533 | * unasserted, it's guaranteed that the next sibling hasn't | |
2534 | * finished its grace period even if it's already removed, and thus | |
2535 | * safe to dereference from this RCU critical section. If | |
2536 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
2537 | * to be visible as %true here. | |
3b287a50 TH |
2538 | * |
2539 | * If @pos is dead, its next pointer can't be dereferenced; | |
2540 | * however, as each cgroup is given a monotonically increasing | |
2541 | * unique serial number and always appended to the sibling list, | |
2542 | * the next one can be found by walking the parent's children until | |
2543 | * we see a cgroup with higher serial number than @pos's. While | |
2544 | * this path can be slower, it's taken only when either the current | |
2545 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 2546 | */ |
3b287a50 TH |
2547 | if (!pos) { |
2548 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
2549 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 2550 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
2551 | } else { |
2552 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
2553 | if (next->serial_nr > pos->serial_nr) | |
2554 | break; | |
53fa5261 TH |
2555 | } |
2556 | ||
492eb21b TH |
2557 | if (&next->sibling == &cgrp->children) |
2558 | return NULL; | |
2559 | ||
ca8bdcaf | 2560 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 2561 | } |
492eb21b | 2562 | EXPORT_SYMBOL_GPL(css_next_child); |
53fa5261 | 2563 | |
574bd9f7 | 2564 | /** |
492eb21b | 2565 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 2566 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2567 | * @root: css whose descendants to walk |
574bd9f7 | 2568 | * |
492eb21b | 2569 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
2570 | * to visit for pre-order traversal of @root's descendants. @root is |
2571 | * included in the iteration and the first node to be visited. | |
75501a6d | 2572 | * |
87fb54f1 TH |
2573 | * While this function requires cgroup_mutex or RCU read locking, it |
2574 | * doesn't require the whole traversal to be contained in a single critical | |
2575 | * section. This function will return the correct next descendant as long | |
2576 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 2577 | */ |
492eb21b TH |
2578 | struct cgroup_subsys_state * |
2579 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
2580 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2581 | { |
492eb21b | 2582 | struct cgroup_subsys_state *next; |
574bd9f7 | 2583 | |
ace2bee8 | 2584 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2585 | |
bd8815a6 | 2586 | /* if first iteration, visit @root */ |
7805d000 | 2587 | if (!pos) |
bd8815a6 | 2588 | return root; |
574bd9f7 TH |
2589 | |
2590 | /* visit the first child if exists */ | |
492eb21b | 2591 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
2592 | if (next) |
2593 | return next; | |
2594 | ||
2595 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
2596 | while (pos != root) { |
2597 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 2598 | if (next) |
574bd9f7 | 2599 | return next; |
492eb21b | 2600 | pos = css_parent(pos); |
7805d000 | 2601 | } |
574bd9f7 TH |
2602 | |
2603 | return NULL; | |
2604 | } | |
492eb21b | 2605 | EXPORT_SYMBOL_GPL(css_next_descendant_pre); |
574bd9f7 | 2606 | |
12a9d2fe | 2607 | /** |
492eb21b TH |
2608 | * css_rightmost_descendant - return the rightmost descendant of a css |
2609 | * @pos: css of interest | |
12a9d2fe | 2610 | * |
492eb21b TH |
2611 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
2612 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 2613 | * subtree of @pos. |
75501a6d | 2614 | * |
87fb54f1 TH |
2615 | * While this function requires cgroup_mutex or RCU read locking, it |
2616 | * doesn't require the whole traversal to be contained in a single critical | |
2617 | * section. This function will return the correct rightmost descendant as | |
2618 | * long as @pos is accessible. | |
12a9d2fe | 2619 | */ |
492eb21b TH |
2620 | struct cgroup_subsys_state * |
2621 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 2622 | { |
492eb21b | 2623 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 2624 | |
ace2bee8 | 2625 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
2626 | |
2627 | do { | |
2628 | last = pos; | |
2629 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
2630 | pos = NULL; | |
492eb21b | 2631 | css_for_each_child(tmp, last) |
12a9d2fe TH |
2632 | pos = tmp; |
2633 | } while (pos); | |
2634 | ||
2635 | return last; | |
2636 | } | |
492eb21b | 2637 | EXPORT_SYMBOL_GPL(css_rightmost_descendant); |
12a9d2fe | 2638 | |
492eb21b TH |
2639 | static struct cgroup_subsys_state * |
2640 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 2641 | { |
492eb21b | 2642 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
2643 | |
2644 | do { | |
2645 | last = pos; | |
492eb21b | 2646 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
2647 | } while (pos); |
2648 | ||
2649 | return last; | |
2650 | } | |
2651 | ||
2652 | /** | |
492eb21b | 2653 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 2654 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2655 | * @root: css whose descendants to walk |
574bd9f7 | 2656 | * |
492eb21b | 2657 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
2658 | * to visit for post-order traversal of @root's descendants. @root is |
2659 | * included in the iteration and the last node to be visited. | |
75501a6d | 2660 | * |
87fb54f1 TH |
2661 | * While this function requires cgroup_mutex or RCU read locking, it |
2662 | * doesn't require the whole traversal to be contained in a single critical | |
2663 | * section. This function will return the correct next descendant as long | |
2664 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
2665 | * @cgroup. | |
574bd9f7 | 2666 | */ |
492eb21b TH |
2667 | struct cgroup_subsys_state * |
2668 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
2669 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2670 | { |
492eb21b | 2671 | struct cgroup_subsys_state *next; |
574bd9f7 | 2672 | |
ace2bee8 | 2673 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2674 | |
58b79a91 TH |
2675 | /* if first iteration, visit leftmost descendant which may be @root */ |
2676 | if (!pos) | |
2677 | return css_leftmost_descendant(root); | |
574bd9f7 | 2678 | |
bd8815a6 TH |
2679 | /* if we visited @root, we're done */ |
2680 | if (pos == root) | |
2681 | return NULL; | |
2682 | ||
574bd9f7 | 2683 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 2684 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 2685 | if (next) |
492eb21b | 2686 | return css_leftmost_descendant(next); |
574bd9f7 TH |
2687 | |
2688 | /* no sibling left, visit parent */ | |
bd8815a6 | 2689 | return css_parent(pos); |
574bd9f7 | 2690 | } |
492eb21b | 2691 | EXPORT_SYMBOL_GPL(css_next_descendant_post); |
574bd9f7 | 2692 | |
0942eeee | 2693 | /** |
72ec7029 | 2694 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
2695 | * @it: the iterator to advance |
2696 | * | |
2697 | * Advance @it to the next css_set to walk. | |
d515876e | 2698 | */ |
72ec7029 | 2699 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
2700 | { |
2701 | struct list_head *l = it->cset_link; | |
2702 | struct cgrp_cset_link *link; | |
2703 | struct css_set *cset; | |
2704 | ||
2705 | /* Advance to the next non-empty css_set */ | |
2706 | do { | |
2707 | l = l->next; | |
72ec7029 | 2708 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
2709 | it->cset_link = NULL; |
2710 | return; | |
2711 | } | |
2712 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
2713 | cset = link->cset; | |
2714 | } while (list_empty(&cset->tasks)); | |
2715 | it->cset_link = l; | |
2716 | it->task = cset->tasks.next; | |
2717 | } | |
2718 | ||
0942eeee | 2719 | /** |
72ec7029 TH |
2720 | * css_task_iter_start - initiate task iteration |
2721 | * @css: the css to walk tasks of | |
0942eeee TH |
2722 | * @it: the task iterator to use |
2723 | * | |
72ec7029 TH |
2724 | * Initiate iteration through the tasks of @css. The caller can call |
2725 | * css_task_iter_next() to walk through the tasks until the function | |
2726 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
2727 | * called. | |
0942eeee TH |
2728 | * |
2729 | * Note that this function acquires a lock which is released when the | |
2730 | * iteration finishes. The caller can't sleep while iteration is in | |
2731 | * progress. | |
2732 | */ | |
72ec7029 TH |
2733 | void css_task_iter_start(struct cgroup_subsys_state *css, |
2734 | struct css_task_iter *it) | |
c6ca5750 | 2735 | __acquires(css_set_lock) |
817929ec PM |
2736 | { |
2737 | /* | |
72ec7029 TH |
2738 | * The first time anyone tries to iterate across a css, we need to |
2739 | * enable the list linking each css_set to its tasks, and fix up | |
2740 | * all existing tasks. | |
817929ec | 2741 | */ |
31a7df01 CW |
2742 | if (!use_task_css_set_links) |
2743 | cgroup_enable_task_cg_lists(); | |
2744 | ||
817929ec | 2745 | read_lock(&css_set_lock); |
c59cd3d8 | 2746 | |
72ec7029 TH |
2747 | it->origin_css = css; |
2748 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 2749 | |
72ec7029 | 2750 | css_advance_task_iter(it); |
817929ec PM |
2751 | } |
2752 | ||
0942eeee | 2753 | /** |
72ec7029 | 2754 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
2755 | * @it: the task iterator being iterated |
2756 | * | |
2757 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
2758 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
2759 | * reaches the end. | |
0942eeee | 2760 | */ |
72ec7029 | 2761 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
2762 | { |
2763 | struct task_struct *res; | |
2764 | struct list_head *l = it->task; | |
69d0206c | 2765 | struct cgrp_cset_link *link; |
817929ec PM |
2766 | |
2767 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 2768 | if (!it->cset_link) |
817929ec PM |
2769 | return NULL; |
2770 | res = list_entry(l, struct task_struct, cg_list); | |
2771 | /* Advance iterator to find next entry */ | |
2772 | l = l->next; | |
69d0206c TH |
2773 | link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link); |
2774 | if (l == &link->cset->tasks) { | |
0942eeee TH |
2775 | /* |
2776 | * We reached the end of this task list - move on to the | |
2777 | * next cgrp_cset_link. | |
2778 | */ | |
72ec7029 | 2779 | css_advance_task_iter(it); |
817929ec PM |
2780 | } else { |
2781 | it->task = l; | |
2782 | } | |
2783 | return res; | |
2784 | } | |
2785 | ||
0942eeee | 2786 | /** |
72ec7029 | 2787 | * css_task_iter_end - finish task iteration |
0942eeee TH |
2788 | * @it: the task iterator to finish |
2789 | * | |
72ec7029 | 2790 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 2791 | */ |
72ec7029 | 2792 | void css_task_iter_end(struct css_task_iter *it) |
c6ca5750 | 2793 | __releases(css_set_lock) |
817929ec PM |
2794 | { |
2795 | read_unlock(&css_set_lock); | |
2796 | } | |
2797 | ||
31a7df01 CW |
2798 | static inline int started_after_time(struct task_struct *t1, |
2799 | struct timespec *time, | |
2800 | struct task_struct *t2) | |
2801 | { | |
2802 | int start_diff = timespec_compare(&t1->start_time, time); | |
2803 | if (start_diff > 0) { | |
2804 | return 1; | |
2805 | } else if (start_diff < 0) { | |
2806 | return 0; | |
2807 | } else { | |
2808 | /* | |
2809 | * Arbitrarily, if two processes started at the same | |
2810 | * time, we'll say that the lower pointer value | |
2811 | * started first. Note that t2 may have exited by now | |
2812 | * so this may not be a valid pointer any longer, but | |
2813 | * that's fine - it still serves to distinguish | |
2814 | * between two tasks started (effectively) simultaneously. | |
2815 | */ | |
2816 | return t1 > t2; | |
2817 | } | |
2818 | } | |
2819 | ||
2820 | /* | |
2821 | * This function is a callback from heap_insert() and is used to order | |
2822 | * the heap. | |
2823 | * In this case we order the heap in descending task start time. | |
2824 | */ | |
2825 | static inline int started_after(void *p1, void *p2) | |
2826 | { | |
2827 | struct task_struct *t1 = p1; | |
2828 | struct task_struct *t2 = p2; | |
2829 | return started_after_time(t1, &t2->start_time, t2); | |
2830 | } | |
2831 | ||
2832 | /** | |
72ec7029 TH |
2833 | * css_scan_tasks - iterate though all the tasks in a css |
2834 | * @css: the css to iterate tasks of | |
e535837b TH |
2835 | * @test: optional test callback |
2836 | * @process: process callback | |
2837 | * @data: data passed to @test and @process | |
2838 | * @heap: optional pre-allocated heap used for task iteration | |
31a7df01 | 2839 | * |
72ec7029 TH |
2840 | * Iterate through all the tasks in @css, calling @test for each, and if it |
2841 | * returns %true, call @process for it also. | |
31a7df01 | 2842 | * |
e535837b | 2843 | * @test may be NULL, meaning always true (select all tasks), which |
72ec7029 | 2844 | * effectively duplicates css_task_iter_{start,next,end}() but does not |
e535837b TH |
2845 | * lock css_set_lock for the call to @process. |
2846 | * | |
2847 | * It is guaranteed that @process will act on every task that is a member | |
72ec7029 TH |
2848 | * of @css for the duration of this call. This function may or may not |
2849 | * call @process for tasks that exit or move to a different css during the | |
2850 | * call, or are forked or move into the css during the call. | |
31a7df01 | 2851 | * |
e535837b TH |
2852 | * Note that @test may be called with locks held, and may in some |
2853 | * situations be called multiple times for the same task, so it should be | |
2854 | * cheap. | |
31a7df01 | 2855 | * |
e535837b TH |
2856 | * If @heap is non-NULL, a heap has been pre-allocated and will be used for |
2857 | * heap operations (and its "gt" member will be overwritten), else a | |
2858 | * temporary heap will be used (allocation of which may cause this function | |
2859 | * to fail). | |
31a7df01 | 2860 | */ |
72ec7029 TH |
2861 | int css_scan_tasks(struct cgroup_subsys_state *css, |
2862 | bool (*test)(struct task_struct *, void *), | |
2863 | void (*process)(struct task_struct *, void *), | |
2864 | void *data, struct ptr_heap *heap) | |
31a7df01 CW |
2865 | { |
2866 | int retval, i; | |
72ec7029 | 2867 | struct css_task_iter it; |
31a7df01 CW |
2868 | struct task_struct *p, *dropped; |
2869 | /* Never dereference latest_task, since it's not refcounted */ | |
2870 | struct task_struct *latest_task = NULL; | |
2871 | struct ptr_heap tmp_heap; | |
31a7df01 CW |
2872 | struct timespec latest_time = { 0, 0 }; |
2873 | ||
e535837b | 2874 | if (heap) { |
31a7df01 | 2875 | /* The caller supplied our heap and pre-allocated its memory */ |
31a7df01 CW |
2876 | heap->gt = &started_after; |
2877 | } else { | |
2878 | /* We need to allocate our own heap memory */ | |
2879 | heap = &tmp_heap; | |
2880 | retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after); | |
2881 | if (retval) | |
2882 | /* cannot allocate the heap */ | |
2883 | return retval; | |
2884 | } | |
2885 | ||
2886 | again: | |
2887 | /* | |
72ec7029 | 2888 | * Scan tasks in the css, using the @test callback to determine |
e535837b TH |
2889 | * which are of interest, and invoking @process callback on the |
2890 | * ones which need an update. Since we don't want to hold any | |
2891 | * locks during the task updates, gather tasks to be processed in a | |
2892 | * heap structure. The heap is sorted by descending task start | |
2893 | * time. If the statically-sized heap fills up, we overflow tasks | |
2894 | * that started later, and in future iterations only consider tasks | |
2895 | * that started after the latest task in the previous pass. This | |
31a7df01 CW |
2896 | * guarantees forward progress and that we don't miss any tasks. |
2897 | */ | |
2898 | heap->size = 0; | |
72ec7029 TH |
2899 | css_task_iter_start(css, &it); |
2900 | while ((p = css_task_iter_next(&it))) { | |
31a7df01 CW |
2901 | /* |
2902 | * Only affect tasks that qualify per the caller's callback, | |
2903 | * if he provided one | |
2904 | */ | |
e535837b | 2905 | if (test && !test(p, data)) |
31a7df01 CW |
2906 | continue; |
2907 | /* | |
2908 | * Only process tasks that started after the last task | |
2909 | * we processed | |
2910 | */ | |
2911 | if (!started_after_time(p, &latest_time, latest_task)) | |
2912 | continue; | |
2913 | dropped = heap_insert(heap, p); | |
2914 | if (dropped == NULL) { | |
2915 | /* | |
2916 | * The new task was inserted; the heap wasn't | |
2917 | * previously full | |
2918 | */ | |
2919 | get_task_struct(p); | |
2920 | } else if (dropped != p) { | |
2921 | /* | |
2922 | * The new task was inserted, and pushed out a | |
2923 | * different task | |
2924 | */ | |
2925 | get_task_struct(p); | |
2926 | put_task_struct(dropped); | |
2927 | } | |
2928 | /* | |
2929 | * Else the new task was newer than anything already in | |
2930 | * the heap and wasn't inserted | |
2931 | */ | |
2932 | } | |
72ec7029 | 2933 | css_task_iter_end(&it); |
31a7df01 CW |
2934 | |
2935 | if (heap->size) { | |
2936 | for (i = 0; i < heap->size; i++) { | |
4fe91d51 | 2937 | struct task_struct *q = heap->ptrs[i]; |
31a7df01 | 2938 | if (i == 0) { |
4fe91d51 PJ |
2939 | latest_time = q->start_time; |
2940 | latest_task = q; | |
31a7df01 CW |
2941 | } |
2942 | /* Process the task per the caller's callback */ | |
e535837b | 2943 | process(q, data); |
4fe91d51 | 2944 | put_task_struct(q); |
31a7df01 CW |
2945 | } |
2946 | /* | |
2947 | * If we had to process any tasks at all, scan again | |
2948 | * in case some of them were in the middle of forking | |
2949 | * children that didn't get processed. | |
2950 | * Not the most efficient way to do it, but it avoids | |
2951 | * having to take callback_mutex in the fork path | |
2952 | */ | |
2953 | goto again; | |
2954 | } | |
2955 | if (heap == &tmp_heap) | |
2956 | heap_free(&tmp_heap); | |
2957 | return 0; | |
2958 | } | |
2959 | ||
e535837b | 2960 | static void cgroup_transfer_one_task(struct task_struct *task, void *data) |
8cc99345 | 2961 | { |
e535837b | 2962 | struct cgroup *new_cgroup = data; |
8cc99345 | 2963 | |
47cfcd09 | 2964 | mutex_lock(&cgroup_mutex); |
8cc99345 | 2965 | cgroup_attach_task(new_cgroup, task, false); |
47cfcd09 | 2966 | mutex_unlock(&cgroup_mutex); |
8cc99345 TH |
2967 | } |
2968 | ||
2969 | /** | |
2970 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
2971 | * @to: cgroup to which the tasks will be moved | |
2972 | * @from: cgroup in which the tasks currently reside | |
2973 | */ | |
2974 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
2975 | { | |
72ec7029 TH |
2976 | return css_scan_tasks(&from->dummy_css, NULL, cgroup_transfer_one_task, |
2977 | to, NULL); | |
8cc99345 TH |
2978 | } |
2979 | ||
bbcb81d0 | 2980 | /* |
102a775e | 2981 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
2982 | * |
2983 | * Reading this file can return large amounts of data if a cgroup has | |
2984 | * *lots* of attached tasks. So it may need several calls to read(), | |
2985 | * but we cannot guarantee that the information we produce is correct | |
2986 | * unless we produce it entirely atomically. | |
2987 | * | |
bbcb81d0 | 2988 | */ |
bbcb81d0 | 2989 | |
24528255 LZ |
2990 | /* which pidlist file are we talking about? */ |
2991 | enum cgroup_filetype { | |
2992 | CGROUP_FILE_PROCS, | |
2993 | CGROUP_FILE_TASKS, | |
2994 | }; | |
2995 | ||
2996 | /* | |
2997 | * A pidlist is a list of pids that virtually represents the contents of one | |
2998 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
2999 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3000 | * to the cgroup. | |
3001 | */ | |
3002 | struct cgroup_pidlist { | |
3003 | /* | |
3004 | * used to find which pidlist is wanted. doesn't change as long as | |
3005 | * this particular list stays in the list. | |
3006 | */ | |
3007 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3008 | /* array of xids */ | |
3009 | pid_t *list; | |
3010 | /* how many elements the above list has */ | |
3011 | int length; | |
24528255 LZ |
3012 | /* each of these stored in a list by its cgroup */ |
3013 | struct list_head links; | |
3014 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3015 | struct cgroup *owner; | |
b1a21367 TH |
3016 | /* for delayed destruction */ |
3017 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3018 | }; |
3019 | ||
d1d9fd33 BB |
3020 | /* |
3021 | * The following two functions "fix" the issue where there are more pids | |
3022 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3023 | * TODO: replace with a kernel-wide solution to this problem | |
3024 | */ | |
3025 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3026 | static void *pidlist_allocate(int count) | |
3027 | { | |
3028 | if (PIDLIST_TOO_LARGE(count)) | |
3029 | return vmalloc(count * sizeof(pid_t)); | |
3030 | else | |
3031 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3032 | } | |
b1a21367 | 3033 | |
d1d9fd33 BB |
3034 | static void pidlist_free(void *p) |
3035 | { | |
3036 | if (is_vmalloc_addr(p)) | |
3037 | vfree(p); | |
3038 | else | |
3039 | kfree(p); | |
3040 | } | |
d1d9fd33 | 3041 | |
b1a21367 TH |
3042 | /* |
3043 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3044 | * should be left afterwards. | |
3045 | */ | |
3046 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3047 | { | |
3048 | struct cgroup_pidlist *l, *tmp_l; | |
3049 | ||
3050 | mutex_lock(&cgrp->pidlist_mutex); | |
3051 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3052 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3053 | mutex_unlock(&cgrp->pidlist_mutex); | |
3054 | ||
3055 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3056 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3057 | } | |
3058 | ||
3059 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3060 | { | |
3061 | struct delayed_work *dwork = to_delayed_work(work); | |
3062 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3063 | destroy_dwork); | |
3064 | struct cgroup_pidlist *tofree = NULL; | |
3065 | ||
3066 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3067 | |
3068 | /* | |
04502365 TH |
3069 | * Destroy iff we didn't get queued again. The state won't change |
3070 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3071 | */ |
04502365 | 3072 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3073 | list_del(&l->links); |
3074 | pidlist_free(l->list); | |
3075 | put_pid_ns(l->key.ns); | |
3076 | tofree = l; | |
3077 | } | |
3078 | ||
b1a21367 TH |
3079 | mutex_unlock(&l->owner->pidlist_mutex); |
3080 | kfree(tofree); | |
3081 | } | |
3082 | ||
bbcb81d0 | 3083 | /* |
102a775e | 3084 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3085 | * Returns the number of unique elements. |
bbcb81d0 | 3086 | */ |
6ee211ad | 3087 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3088 | { |
102a775e | 3089 | int src, dest = 1; |
102a775e BB |
3090 | |
3091 | /* | |
3092 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3093 | * edge cases first; no work needs to be done for either | |
3094 | */ | |
3095 | if (length == 0 || length == 1) | |
3096 | return length; | |
3097 | /* src and dest walk down the list; dest counts unique elements */ | |
3098 | for (src = 1; src < length; src++) { | |
3099 | /* find next unique element */ | |
3100 | while (list[src] == list[src-1]) { | |
3101 | src++; | |
3102 | if (src == length) | |
3103 | goto after; | |
3104 | } | |
3105 | /* dest always points to where the next unique element goes */ | |
3106 | list[dest] = list[src]; | |
3107 | dest++; | |
3108 | } | |
3109 | after: | |
102a775e BB |
3110 | return dest; |
3111 | } | |
3112 | ||
afb2bc14 TH |
3113 | /* |
3114 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3115 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3116 | * different per namespace, each namespace needs differently sorted list, | |
3117 | * making it impossible to use, for example, single rbtree of member tasks | |
3118 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3119 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3120 | * pidlists keyed by cgroup and namespace. | |
3121 | * | |
3122 | * All this extra complexity was caused by the original implementation | |
3123 | * committing to an entirely unnecessary property. In the long term, we | |
3124 | * want to do away with it. Explicitly scramble sort order if | |
3125 | * sane_behavior so that no such expectation exists in the new interface. | |
3126 | * | |
3127 | * Scrambling is done by swapping every two consecutive bits, which is | |
3128 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3129 | */ | |
3130 | static pid_t pid_fry(pid_t pid) | |
3131 | { | |
3132 | unsigned a = pid & 0x55555555; | |
3133 | unsigned b = pid & 0xAAAAAAAA; | |
3134 | ||
3135 | return (a << 1) | (b >> 1); | |
3136 | } | |
3137 | ||
3138 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3139 | { | |
3140 | if (cgroup_sane_behavior(cgrp)) | |
3141 | return pid_fry(pid); | |
3142 | else | |
3143 | return pid; | |
3144 | } | |
3145 | ||
102a775e BB |
3146 | static int cmppid(const void *a, const void *b) |
3147 | { | |
3148 | return *(pid_t *)a - *(pid_t *)b; | |
3149 | } | |
3150 | ||
afb2bc14 TH |
3151 | static int fried_cmppid(const void *a, const void *b) |
3152 | { | |
3153 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3154 | } | |
3155 | ||
e6b81710 TH |
3156 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3157 | enum cgroup_filetype type) | |
3158 | { | |
3159 | struct cgroup_pidlist *l; | |
3160 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3161 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3162 | ||
3163 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3164 | ||
3165 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3166 | if (l->key.type == type && l->key.ns == ns) | |
3167 | return l; | |
3168 | return NULL; | |
3169 | } | |
3170 | ||
72a8cb30 BB |
3171 | /* |
3172 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3173 | * returns with the lock on that pidlist already held, and takes care | |
3174 | * of the use count, or returns NULL with no locks held if we're out of | |
3175 | * memory. | |
3176 | */ | |
e6b81710 TH |
3177 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3178 | enum cgroup_filetype type) | |
72a8cb30 BB |
3179 | { |
3180 | struct cgroup_pidlist *l; | |
b70cc5fd | 3181 | |
e6b81710 TH |
3182 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3183 | ||
3184 | l = cgroup_pidlist_find(cgrp, type); | |
3185 | if (l) | |
3186 | return l; | |
3187 | ||
72a8cb30 | 3188 | /* entry not found; create a new one */ |
f4f4be2b | 3189 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3190 | if (!l) |
72a8cb30 | 3191 | return l; |
e6b81710 | 3192 | |
b1a21367 | 3193 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3194 | l->key.type = type; |
e6b81710 TH |
3195 | /* don't need task_nsproxy() if we're looking at ourself */ |
3196 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3197 | l->owner = cgrp; |
3198 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3199 | return l; |
3200 | } | |
3201 | ||
102a775e BB |
3202 | /* |
3203 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3204 | */ | |
72a8cb30 BB |
3205 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3206 | struct cgroup_pidlist **lp) | |
102a775e BB |
3207 | { |
3208 | pid_t *array; | |
3209 | int length; | |
3210 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3211 | struct css_task_iter it; |
817929ec | 3212 | struct task_struct *tsk; |
102a775e BB |
3213 | struct cgroup_pidlist *l; |
3214 | ||
4bac00d1 TH |
3215 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3216 | ||
102a775e BB |
3217 | /* |
3218 | * If cgroup gets more users after we read count, we won't have | |
3219 | * enough space - tough. This race is indistinguishable to the | |
3220 | * caller from the case that the additional cgroup users didn't | |
3221 | * show up until sometime later on. | |
3222 | */ | |
3223 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3224 | array = pidlist_allocate(length); |
102a775e BB |
3225 | if (!array) |
3226 | return -ENOMEM; | |
3227 | /* now, populate the array */ | |
72ec7029 TH |
3228 | css_task_iter_start(&cgrp->dummy_css, &it); |
3229 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3230 | if (unlikely(n == length)) |
817929ec | 3231 | break; |
102a775e | 3232 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3233 | if (type == CGROUP_FILE_PROCS) |
3234 | pid = task_tgid_vnr(tsk); | |
3235 | else | |
3236 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3237 | if (pid > 0) /* make sure to only use valid results */ |
3238 | array[n++] = pid; | |
817929ec | 3239 | } |
72ec7029 | 3240 | css_task_iter_end(&it); |
102a775e BB |
3241 | length = n; |
3242 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3243 | if (cgroup_sane_behavior(cgrp)) |
3244 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3245 | else | |
3246 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3247 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3248 | length = pidlist_uniq(array, length); |
e6b81710 | 3249 | |
e6b81710 | 3250 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3251 | if (!l) { |
e6b81710 | 3252 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3253 | pidlist_free(array); |
72a8cb30 | 3254 | return -ENOMEM; |
102a775e | 3255 | } |
e6b81710 TH |
3256 | |
3257 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3258 | pidlist_free(l->list); |
102a775e BB |
3259 | l->list = array; |
3260 | l->length = length; | |
72a8cb30 | 3261 | *lp = l; |
102a775e | 3262 | return 0; |
bbcb81d0 PM |
3263 | } |
3264 | ||
846c7bb0 | 3265 | /** |
a043e3b2 | 3266 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3267 | * @stats: cgroupstats to fill information into |
3268 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3269 | * been requested. | |
a043e3b2 LZ |
3270 | * |
3271 | * Build and fill cgroupstats so that taskstats can export it to user | |
3272 | * space. | |
846c7bb0 BS |
3273 | */ |
3274 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3275 | { | |
2bd59d48 | 3276 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3277 | struct cgroup *cgrp; |
72ec7029 | 3278 | struct css_task_iter it; |
846c7bb0 | 3279 | struct task_struct *tsk; |
33d283be | 3280 | |
2bd59d48 TH |
3281 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3282 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3283 | kernfs_type(kn) != KERNFS_DIR) | |
3284 | return -EINVAL; | |
3285 | ||
846c7bb0 | 3286 | /* |
2bd59d48 TH |
3287 | * We aren't being called from kernfs and there's no guarantee on |
3288 | * @kn->priv's validity. For this and css_tryget_from_dir(), | |
3289 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
846c7bb0 | 3290 | */ |
2bd59d48 TH |
3291 | rcu_read_lock(); |
3292 | cgrp = rcu_dereference(kn->priv); | |
3293 | if (!cgrp) { | |
3294 | rcu_read_unlock(); | |
3295 | return -ENOENT; | |
3296 | } | |
846c7bb0 | 3297 | |
72ec7029 TH |
3298 | css_task_iter_start(&cgrp->dummy_css, &it); |
3299 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3300 | switch (tsk->state) { |
3301 | case TASK_RUNNING: | |
3302 | stats->nr_running++; | |
3303 | break; | |
3304 | case TASK_INTERRUPTIBLE: | |
3305 | stats->nr_sleeping++; | |
3306 | break; | |
3307 | case TASK_UNINTERRUPTIBLE: | |
3308 | stats->nr_uninterruptible++; | |
3309 | break; | |
3310 | case TASK_STOPPED: | |
3311 | stats->nr_stopped++; | |
3312 | break; | |
3313 | default: | |
3314 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3315 | stats->nr_io_wait++; | |
3316 | break; | |
3317 | } | |
3318 | } | |
72ec7029 | 3319 | css_task_iter_end(&it); |
846c7bb0 | 3320 | |
2bd59d48 TH |
3321 | rcu_read_unlock(); |
3322 | return 0; | |
846c7bb0 BS |
3323 | } |
3324 | ||
8f3ff208 | 3325 | |
bbcb81d0 | 3326 | /* |
102a775e | 3327 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3328 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3329 | * in the cgroup->l->list array. |
bbcb81d0 | 3330 | */ |
cc31edce | 3331 | |
102a775e | 3332 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3333 | { |
cc31edce PM |
3334 | /* |
3335 | * Initially we receive a position value that corresponds to | |
3336 | * one more than the last pid shown (or 0 on the first call or | |
3337 | * after a seek to the start). Use a binary-search to find the | |
3338 | * next pid to display, if any | |
3339 | */ | |
2bd59d48 | 3340 | struct kernfs_open_file *of = s->private; |
7da11279 | 3341 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3342 | struct cgroup_pidlist *l; |
7da11279 | 3343 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3344 | int index = 0, pid = *pos; |
4bac00d1 TH |
3345 | int *iter, ret; |
3346 | ||
3347 | mutex_lock(&cgrp->pidlist_mutex); | |
3348 | ||
3349 | /* | |
5d22444f | 3350 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3351 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3352 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3353 | * could already have been destroyed. |
3354 | */ | |
5d22444f TH |
3355 | if (of->priv) |
3356 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3357 | |
3358 | /* | |
3359 | * Either this is the first start() after open or the matching | |
3360 | * pidlist has been destroyed inbetween. Create a new one. | |
3361 | */ | |
5d22444f TH |
3362 | if (!of->priv) { |
3363 | ret = pidlist_array_load(cgrp, type, | |
3364 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3365 | if (ret) |
3366 | return ERR_PTR(ret); | |
3367 | } | |
5d22444f | 3368 | l = of->priv; |
cc31edce | 3369 | |
cc31edce | 3370 | if (pid) { |
102a775e | 3371 | int end = l->length; |
20777766 | 3372 | |
cc31edce PM |
3373 | while (index < end) { |
3374 | int mid = (index + end) / 2; | |
afb2bc14 | 3375 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3376 | index = mid; |
3377 | break; | |
afb2bc14 | 3378 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3379 | index = mid + 1; |
3380 | else | |
3381 | end = mid; | |
3382 | } | |
3383 | } | |
3384 | /* If we're off the end of the array, we're done */ | |
102a775e | 3385 | if (index >= l->length) |
cc31edce PM |
3386 | return NULL; |
3387 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3388 | iter = l->list + index; |
afb2bc14 | 3389 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3390 | return iter; |
3391 | } | |
3392 | ||
102a775e | 3393 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3394 | { |
2bd59d48 | 3395 | struct kernfs_open_file *of = s->private; |
5d22444f | 3396 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3397 | |
5d22444f TH |
3398 | if (l) |
3399 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3400 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3401 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3402 | } |
3403 | ||
102a775e | 3404 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3405 | { |
2bd59d48 | 3406 | struct kernfs_open_file *of = s->private; |
5d22444f | 3407 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3408 | pid_t *p = v; |
3409 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3410 | /* |
3411 | * Advance to the next pid in the array. If this goes off the | |
3412 | * end, we're done | |
3413 | */ | |
3414 | p++; | |
3415 | if (p >= end) { | |
3416 | return NULL; | |
3417 | } else { | |
7da11279 | 3418 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3419 | return p; |
3420 | } | |
3421 | } | |
3422 | ||
102a775e | 3423 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3424 | { |
3425 | return seq_printf(s, "%d\n", *(int *)v); | |
3426 | } | |
bbcb81d0 | 3427 | |
102a775e BB |
3428 | /* |
3429 | * seq_operations functions for iterating on pidlists through seq_file - | |
3430 | * independent of whether it's tasks or procs | |
3431 | */ | |
3432 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3433 | .start = cgroup_pidlist_start, | |
3434 | .stop = cgroup_pidlist_stop, | |
3435 | .next = cgroup_pidlist_next, | |
3436 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3437 | }; |
3438 | ||
182446d0 TH |
3439 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3440 | struct cftype *cft) | |
81a6a5cd | 3441 | { |
182446d0 | 3442 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3443 | } |
3444 | ||
182446d0 TH |
3445 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3446 | struct cftype *cft, u64 val) | |
6379c106 | 3447 | { |
182446d0 | 3448 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3449 | if (val) |
182446d0 | 3450 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3451 | else |
182446d0 | 3452 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3453 | return 0; |
3454 | } | |
3455 | ||
182446d0 TH |
3456 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3457 | struct cftype *cft) | |
97978e6d | 3458 | { |
182446d0 | 3459 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3460 | } |
3461 | ||
182446d0 TH |
3462 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3463 | struct cftype *cft, u64 val) | |
97978e6d DL |
3464 | { |
3465 | if (val) | |
182446d0 | 3466 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3467 | else |
182446d0 | 3468 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3469 | return 0; |
3470 | } | |
3471 | ||
d5c56ced | 3472 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3473 | { |
d5c56ced | 3474 | .name = "cgroup.procs", |
6612f05b TH |
3475 | .seq_start = cgroup_pidlist_start, |
3476 | .seq_next = cgroup_pidlist_next, | |
3477 | .seq_stop = cgroup_pidlist_stop, | |
3478 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3479 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3480 | .write_u64 = cgroup_procs_write, |
74a1166d | 3481 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3482 | }, |
97978e6d DL |
3483 | { |
3484 | .name = "cgroup.clone_children", | |
873fe09e | 3485 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3486 | .read_u64 = cgroup_clone_children_read, |
3487 | .write_u64 = cgroup_clone_children_write, | |
3488 | }, | |
873fe09e TH |
3489 | { |
3490 | .name = "cgroup.sane_behavior", | |
3491 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3492 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3493 | }, |
d5c56ced TH |
3494 | |
3495 | /* | |
3496 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3497 | * don't exist if sane_behavior. If you're depending on these, be | |
3498 | * prepared to be burned. | |
3499 | */ | |
3500 | { | |
3501 | .name = "tasks", | |
3502 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3503 | .seq_start = cgroup_pidlist_start, |
3504 | .seq_next = cgroup_pidlist_next, | |
3505 | .seq_stop = cgroup_pidlist_stop, | |
3506 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3507 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3508 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3509 | .mode = S_IRUGO | S_IWUSR, |
3510 | }, | |
3511 | { | |
3512 | .name = "notify_on_release", | |
3513 | .flags = CFTYPE_INSANE, | |
3514 | .read_u64 = cgroup_read_notify_on_release, | |
3515 | .write_u64 = cgroup_write_notify_on_release, | |
3516 | }, | |
6e6ff25b TH |
3517 | { |
3518 | .name = "release_agent", | |
cc5943a7 | 3519 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3520 | .seq_show = cgroup_release_agent_show, |
6e6ff25b | 3521 | .write_string = cgroup_release_agent_write, |
5f469907 | 3522 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 3523 | }, |
db0416b6 | 3524 | { } /* terminate */ |
bbcb81d0 PM |
3525 | }; |
3526 | ||
13af07df | 3527 | /** |
628f7cd4 | 3528 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3529 | * @cgrp: target cgroup |
13af07df | 3530 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3531 | * |
3532 | * On failure, no file is added. | |
13af07df | 3533 | */ |
628f7cd4 | 3534 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3535 | { |
ddbcc7e8 | 3536 | struct cgroup_subsys *ss; |
b420ba7d | 3537 | int i, ret = 0; |
bbcb81d0 | 3538 | |
8e3f6541 | 3539 | /* process cftsets of each subsystem */ |
b420ba7d | 3540 | for_each_subsys(ss, i) { |
0adb0704 | 3541 | struct cftype *cfts; |
b420ba7d TH |
3542 | |
3543 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3544 | continue; |
8e3f6541 | 3545 | |
0adb0704 TH |
3546 | list_for_each_entry(cfts, &ss->cfts, node) { |
3547 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
3548 | if (ret < 0) |
3549 | goto err; | |
3550 | } | |
ddbcc7e8 | 3551 | } |
ddbcc7e8 | 3552 | return 0; |
bee55099 TH |
3553 | err: |
3554 | cgroup_clear_dir(cgrp, subsys_mask); | |
3555 | return ret; | |
ddbcc7e8 PM |
3556 | } |
3557 | ||
0c21ead1 TH |
3558 | /* |
3559 | * css destruction is four-stage process. | |
3560 | * | |
3561 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3562 | * Implemented in kill_css(). | |
3563 | * | |
3564 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3565 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3566 | * by invoking offline_css(). After offlining, the base ref is put. | |
3567 | * Implemented in css_killed_work_fn(). | |
3568 | * | |
3569 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3570 | * accessors are inside RCU read sections. css_release() schedules the | |
3571 | * RCU callback. | |
3572 | * | |
3573 | * 4. After the grace period, the css can be freed. Implemented in | |
3574 | * css_free_work_fn(). | |
3575 | * | |
3576 | * It is actually hairier because both step 2 and 4 require process context | |
3577 | * and thus involve punting to css->destroy_work adding two additional | |
3578 | * steps to the already complex sequence. | |
3579 | */ | |
35ef10da | 3580 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3581 | { |
3582 | struct cgroup_subsys_state *css = | |
35ef10da | 3583 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 3584 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3585 | |
0ae78e0b TH |
3586 | if (css->parent) |
3587 | css_put(css->parent); | |
3588 | ||
0c21ead1 | 3589 | css->ss->css_free(css); |
2bd59d48 | 3590 | cgroup_put(cgrp); |
48ddbe19 TH |
3591 | } |
3592 | ||
0c21ead1 | 3593 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3594 | { |
3595 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3596 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3597 | |
35ef10da | 3598 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 3599 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
3600 | } |
3601 | ||
d3daf28d TH |
3602 | static void css_release(struct percpu_ref *ref) |
3603 | { | |
3604 | struct cgroup_subsys_state *css = | |
3605 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3606 | ||
aec25020 | 3607 | rcu_assign_pointer(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 3608 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
3609 | } |
3610 | ||
623f926b TH |
3611 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
3612 | struct cgroup *cgrp) | |
ddbcc7e8 | 3613 | { |
bd89aabc | 3614 | css->cgroup = cgrp; |
72c97e54 | 3615 | css->ss = ss; |
ddbcc7e8 | 3616 | css->flags = 0; |
0ae78e0b TH |
3617 | |
3618 | if (cgrp->parent) | |
ca8bdcaf | 3619 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 3620 | else |
38b53aba | 3621 | css->flags |= CSS_ROOT; |
48ddbe19 | 3622 | |
ca8bdcaf | 3623 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
3624 | } |
3625 | ||
2a4ac633 | 3626 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 3627 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3628 | { |
623f926b | 3629 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
3630 | int ret = 0; |
3631 | ||
ace2bee8 | 3632 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3633 | lockdep_assert_held(&cgroup_mutex); |
3634 | ||
92fb9748 | 3635 | if (ss->css_online) |
eb95419b | 3636 | ret = ss->css_online(css); |
ae7f164a | 3637 | if (!ret) { |
eb95419b | 3638 | css->flags |= CSS_ONLINE; |
f20104de | 3639 | css->cgroup->nr_css++; |
aec25020 | 3640 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 3641 | } |
b1929db4 | 3642 | return ret; |
a31f2d3f TH |
3643 | } |
3644 | ||
2a4ac633 | 3645 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 3646 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3647 | { |
623f926b | 3648 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 3649 | |
ace2bee8 | 3650 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3651 | lockdep_assert_held(&cgroup_mutex); |
3652 | ||
3653 | if (!(css->flags & CSS_ONLINE)) | |
3654 | return; | |
3655 | ||
d7eeac19 | 3656 | if (ss->css_offline) |
eb95419b | 3657 | ss->css_offline(css); |
a31f2d3f | 3658 | |
eb95419b | 3659 | css->flags &= ~CSS_ONLINE; |
09a503ea | 3660 | css->cgroup->nr_css--; |
aec25020 | 3661 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); |
a31f2d3f TH |
3662 | } |
3663 | ||
c81c925a TH |
3664 | /** |
3665 | * create_css - create a cgroup_subsys_state | |
3666 | * @cgrp: the cgroup new css will be associated with | |
3667 | * @ss: the subsys of new css | |
3668 | * | |
3669 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
3670 | * css is online and installed in @cgrp with all interface files created. | |
3671 | * Returns 0 on success, -errno on failure. | |
3672 | */ | |
3673 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
3674 | { | |
3675 | struct cgroup *parent = cgrp->parent; | |
3676 | struct cgroup_subsys_state *css; | |
3677 | int err; | |
3678 | ||
c81c925a TH |
3679 | lockdep_assert_held(&cgroup_mutex); |
3680 | ||
3681 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
3682 | if (IS_ERR(css)) | |
3683 | return PTR_ERR(css); | |
3684 | ||
3685 | err = percpu_ref_init(&css->refcnt, css_release); | |
3686 | if (err) | |
3687 | goto err_free; | |
3688 | ||
3689 | init_css(css, ss, cgrp); | |
3690 | ||
aec25020 | 3691 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a TH |
3692 | if (err) |
3693 | goto err_free; | |
3694 | ||
3695 | err = online_css(css); | |
3696 | if (err) | |
3697 | goto err_free; | |
3698 | ||
59f5296b | 3699 | cgroup_get(cgrp); |
c81c925a TH |
3700 | css_get(css->parent); |
3701 | ||
3702 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
3703 | parent->parent) { | |
3704 | 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", | |
3705 | current->comm, current->pid, ss->name); | |
3706 | if (!strcmp(ss->name, "memory")) | |
3707 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
3708 | ss->warned_broken_hierarchy = true; | |
3709 | } | |
3710 | ||
3711 | return 0; | |
3712 | ||
3713 | err_free: | |
3714 | percpu_ref_cancel_init(&css->refcnt); | |
3715 | ss->css_free(css); | |
3716 | return err; | |
3717 | } | |
3718 | ||
2bd59d48 | 3719 | /** |
a043e3b2 LZ |
3720 | * cgroup_create - create a cgroup |
3721 | * @parent: cgroup that will be parent of the new cgroup | |
2bd59d48 TH |
3722 | * @name_str: name of the new cgroup |
3723 | * @mode: mode to set on new cgroup | |
ddbcc7e8 | 3724 | */ |
2bd59d48 TH |
3725 | static long cgroup_create(struct cgroup *parent, const char *name_str, |
3726 | umode_t mode) | |
ddbcc7e8 | 3727 | { |
bd89aabc | 3728 | struct cgroup *cgrp; |
65dff759 | 3729 | struct cgroup_name *name; |
ddbcc7e8 | 3730 | struct cgroupfs_root *root = parent->root; |
b58c8998 | 3731 | int ssid, err; |
ddbcc7e8 | 3732 | struct cgroup_subsys *ss; |
2bd59d48 | 3733 | struct kernfs_node *kn; |
ddbcc7e8 | 3734 | |
0a950f65 | 3735 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
3736 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
3737 | if (!cgrp) | |
ddbcc7e8 PM |
3738 | return -ENOMEM; |
3739 | ||
2bd59d48 | 3740 | name = cgroup_alloc_name(name_str); |
b58c8998 TH |
3741 | if (!name) { |
3742 | err = -ENOMEM; | |
65dff759 | 3743 | goto err_free_cgrp; |
b58c8998 | 3744 | } |
65dff759 LZ |
3745 | rcu_assign_pointer(cgrp->name, name); |
3746 | ||
ace2bee8 TH |
3747 | mutex_lock(&cgroup_tree_mutex); |
3748 | ||
976c06bc TH |
3749 | /* |
3750 | * Only live parents can have children. Note that the liveliness | |
3751 | * check isn't strictly necessary because cgroup_mkdir() and | |
3752 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
3753 | * anyway so that locking is contained inside cgroup proper and we | |
3754 | * don't get nasty surprises if we ever grow another caller. | |
3755 | */ | |
3756 | if (!cgroup_lock_live_group(parent)) { | |
3757 | err = -ENODEV; | |
ace2bee8 | 3758 | goto err_unlock_tree; |
0ab02ca8 LZ |
3759 | } |
3760 | ||
3761 | /* | |
3762 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
3763 | * a half-baked cgroup. | |
3764 | */ | |
3765 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
3766 | if (cgrp->id < 0) { | |
3767 | err = -ENOMEM; | |
3768 | goto err_unlock; | |
976c06bc TH |
3769 | } |
3770 | ||
cc31edce | 3771 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3772 | |
bd89aabc | 3773 | cgrp->parent = parent; |
0ae78e0b | 3774 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 3775 | cgrp->root = parent->root; |
ddbcc7e8 | 3776 | |
b6abdb0e LZ |
3777 | if (notify_on_release(parent)) |
3778 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3779 | ||
2260e7fc TH |
3780 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
3781 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 3782 | |
2bd59d48 TH |
3783 | /* create the directory */ |
3784 | kn = kernfs_create_dir(parent->kn, name->name, mode, cgrp); | |
3785 | if (IS_ERR(kn)) { | |
3786 | err = PTR_ERR(kn); | |
0ab02ca8 | 3787 | goto err_free_id; |
2bd59d48 TH |
3788 | } |
3789 | cgrp->kn = kn; | |
ddbcc7e8 | 3790 | |
6f30558f TH |
3791 | /* |
3792 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
3793 | * that @cgrp->kn is always accessible. | |
3794 | */ | |
3795 | kernfs_get(kn); | |
3796 | ||
00356bd5 | 3797 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 3798 | |
4e139afc | 3799 | /* allocation complete, commit to creation */ |
4e139afc TH |
3800 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3801 | root->number_of_cgroups++; | |
28fd6f30 | 3802 | |
2bd59d48 TH |
3803 | /* |
3804 | * Grab a reference on the root and parent so that they don't get | |
3805 | * deleted while there are child cgroups. | |
3806 | */ | |
3807 | cgroup_get_root(root); | |
59f5296b | 3808 | cgroup_get(parent); |
415cf07a | 3809 | |
0d80255e TH |
3810 | /* |
3811 | * @cgrp is now fully operational. If something fails after this | |
3812 | * point, it'll be released via the normal destruction path. | |
3813 | */ | |
4e96ee8e LZ |
3814 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
3815 | ||
2bb566cb | 3816 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
3817 | if (err) |
3818 | goto err_destroy; | |
3819 | ||
9d403e99 | 3820 | /* let's create and online css's */ |
b85d2040 TH |
3821 | for_each_subsys(ss, ssid) { |
3822 | if (root->subsys_mask & (1 << ssid)) { | |
3823 | err = create_css(cgrp, ss); | |
3824 | if (err) | |
3825 | goto err_destroy; | |
3826 | } | |
a8638030 | 3827 | } |
ddbcc7e8 | 3828 | |
2bd59d48 TH |
3829 | kernfs_activate(kn); |
3830 | ||
ddbcc7e8 | 3831 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3832 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
3833 | |
3834 | return 0; | |
3835 | ||
0a950f65 | 3836 | err_free_id: |
4e96ee8e | 3837 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
3838 | err_unlock: |
3839 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
3840 | err_unlock_tree: |
3841 | mutex_unlock(&cgroup_tree_mutex); | |
65dff759 | 3842 | kfree(rcu_dereference_raw(cgrp->name)); |
4b8b47eb | 3843 | err_free_cgrp: |
bd89aabc | 3844 | kfree(cgrp); |
ddbcc7e8 | 3845 | return err; |
4b8b47eb TH |
3846 | |
3847 | err_destroy: | |
3848 | cgroup_destroy_locked(cgrp); | |
3849 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3850 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 3851 | return err; |
ddbcc7e8 PM |
3852 | } |
3853 | ||
2bd59d48 TH |
3854 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
3855 | umode_t mode) | |
ddbcc7e8 | 3856 | { |
2bd59d48 | 3857 | struct cgroup *parent = parent_kn->priv; |
ddbcc7e8 | 3858 | |
2bd59d48 | 3859 | return cgroup_create(parent, name, mode); |
ddbcc7e8 PM |
3860 | } |
3861 | ||
223dbc38 TH |
3862 | /* |
3863 | * This is called when the refcnt of a css is confirmed to be killed. | |
3864 | * css_tryget() is now guaranteed to fail. | |
3865 | */ | |
3866 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 3867 | { |
223dbc38 TH |
3868 | struct cgroup_subsys_state *css = |
3869 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
3870 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 3871 | |
ace2bee8 | 3872 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
3873 | mutex_lock(&cgroup_mutex); |
3874 | ||
09a503ea TH |
3875 | /* |
3876 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
3877 | * initate destruction. | |
3878 | */ | |
3879 | offline_css(css); | |
3880 | ||
f20104de TH |
3881 | /* |
3882 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
3883 | * be disabled before proceeding to the second phase of cgroup | |
3884 | * destruction. If we are the last one, kick it off. | |
3885 | */ | |
09a503ea | 3886 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
3887 | cgroup_destroy_css_killed(cgrp); |
3888 | ||
3889 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3890 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
3891 | |
3892 | /* | |
3893 | * Put the css refs from kill_css(). Each css holds an extra | |
3894 | * reference to the cgroup's dentry and cgroup removal proceeds | |
3895 | * regardless of css refs. On the last put of each css, whenever | |
3896 | * that may be, the extra dentry ref is put so that dentry | |
3897 | * destruction happens only after all css's are released. | |
3898 | */ | |
3899 | css_put(css); | |
d3daf28d TH |
3900 | } |
3901 | ||
223dbc38 TH |
3902 | /* css kill confirmation processing requires process context, bounce */ |
3903 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
3904 | { |
3905 | struct cgroup_subsys_state *css = | |
3906 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3907 | ||
223dbc38 | 3908 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 3909 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
3910 | } |
3911 | ||
edae0c33 TH |
3912 | /** |
3913 | * kill_css - destroy a css | |
3914 | * @css: css to destroy | |
3915 | * | |
3c14f8b4 TH |
3916 | * This function initiates destruction of @css by removing cgroup interface |
3917 | * files and putting its base reference. ->css_offline() will be invoked | |
3918 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
3919 | * reference count reaches zero, @css will be released. | |
edae0c33 TH |
3920 | */ |
3921 | static void kill_css(struct cgroup_subsys_state *css) | |
3922 | { | |
2bd59d48 TH |
3923 | /* |
3924 | * This must happen before css is disassociated with its cgroup. | |
3925 | * See seq_css() for details. | |
3926 | */ | |
aec25020 | 3927 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 3928 | |
edae0c33 TH |
3929 | /* |
3930 | * Killing would put the base ref, but we need to keep it alive | |
3931 | * until after ->css_offline(). | |
3932 | */ | |
3933 | css_get(css); | |
3934 | ||
3935 | /* | |
3936 | * cgroup core guarantees that, by the time ->css_offline() is | |
3937 | * invoked, no new css reference will be given out via | |
3938 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
3939 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
3940 | * guarantee that the ref is seen as killed on all CPUs on return. | |
3941 | * | |
3942 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
3943 | * css is confirmed to be seen as killed on all CPUs. | |
3944 | */ | |
3945 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
3946 | } |
3947 | ||
3948 | /** | |
3949 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
3950 | * @cgrp: cgroup to be destroyed | |
3951 | * | |
3952 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
3953 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
3954 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
3955 | * invoked. To satisfy all the requirements, destruction is implemented in | |
3956 | * the following two steps. | |
3957 | * | |
3958 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
3959 | * userland visible parts and start killing the percpu refcnts of | |
3960 | * css's. Set up so that the next stage will be kicked off once all | |
3961 | * the percpu refcnts are confirmed to be killed. | |
3962 | * | |
3963 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
3964 | * rest of destruction. Once all cgroup references are gone, the | |
3965 | * cgroup is RCU-freed. | |
3966 | * | |
3967 | * This function implements s1. After this step, @cgrp is gone as far as | |
3968 | * the userland is concerned and a new cgroup with the same name may be | |
3969 | * created. As cgroup doesn't care about the names internally, this | |
3970 | * doesn't cause any problem. | |
3971 | */ | |
42809dd4 TH |
3972 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
3973 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 3974 | { |
bb78a92f | 3975 | struct cgroup *child; |
2bd59d48 | 3976 | struct cgroup_subsys_state *css; |
ddd69148 | 3977 | bool empty; |
1c6727af | 3978 | int ssid; |
ddbcc7e8 | 3979 | |
ace2bee8 | 3980 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
3981 | lockdep_assert_held(&cgroup_mutex); |
3982 | ||
ddd69148 | 3983 | /* |
6f3d828f TH |
3984 | * css_set_lock synchronizes access to ->cset_links and prevents |
3985 | * @cgrp from being removed while __put_css_set() is in progress. | |
ddd69148 TH |
3986 | */ |
3987 | read_lock(&css_set_lock); | |
bb78a92f | 3988 | empty = list_empty(&cgrp->cset_links); |
ddd69148 TH |
3989 | read_unlock(&css_set_lock); |
3990 | if (!empty) | |
ddbcc7e8 | 3991 | return -EBUSY; |
a043e3b2 | 3992 | |
bb78a92f HD |
3993 | /* |
3994 | * Make sure there's no live children. We can't test ->children | |
3995 | * emptiness as dead children linger on it while being destroyed; | |
3996 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
3997 | */ | |
3998 | empty = true; | |
3999 | rcu_read_lock(); | |
4000 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
4001 | empty = cgroup_is_dead(child); | |
4002 | if (!empty) | |
4003 | break; | |
4004 | } | |
4005 | rcu_read_unlock(); | |
4006 | if (!empty) | |
4007 | return -EBUSY; | |
4008 | ||
88703267 | 4009 | /* |
edae0c33 TH |
4010 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
4011 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
4012 | * percpu refs of all css's are confirmed to be killed. This |
4013 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 4014 | */ |
4ac06017 | 4015 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
4016 | for_each_css(css, ssid, cgrp) |
4017 | kill_css(css); | |
4ac06017 | 4018 | mutex_lock(&cgroup_mutex); |
455050d2 TH |
4019 | |
4020 | /* | |
4021 | * Mark @cgrp dead. This prevents further task migration and child | |
4022 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 4023 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 4024 | * resume iteration after dropping RCU read lock. See |
492eb21b | 4025 | * css_next_child() for details. |
455050d2 | 4026 | */ |
54766d4a | 4027 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 4028 | |
455050d2 TH |
4029 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
4030 | raw_spin_lock(&release_list_lock); | |
4031 | if (!list_empty(&cgrp->release_list)) | |
4032 | list_del_init(&cgrp->release_list); | |
4033 | raw_spin_unlock(&release_list_lock); | |
4034 | ||
4035 | /* | |
f20104de TH |
4036 | * If @cgrp has css's attached, the second stage of cgroup |
4037 | * destruction is kicked off from css_killed_work_fn() after the | |
4038 | * refs of all attached css's are killed. If @cgrp doesn't have | |
4039 | * any css, we kick it off here. | |
4040 | */ | |
4041 | if (!cgrp->nr_css) | |
4042 | cgroup_destroy_css_killed(cgrp); | |
4043 | ||
2bd59d48 TH |
4044 | /* remove @cgrp directory along with the base files */ |
4045 | mutex_unlock(&cgroup_mutex); | |
4046 | ||
455050d2 | 4047 | /* |
2bd59d48 TH |
4048 | * There are two control paths which try to determine cgroup from |
4049 | * dentry without going through kernfs - cgroupstats_build() and | |
4050 | * css_tryget_from_dir(). Those are supported by RCU protecting | |
4051 | * clearing of cgrp->kn->priv backpointer, which should happen | |
4052 | * after all files under it have been removed. | |
455050d2 | 4053 | */ |
6f30558f | 4054 | kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ |
2bd59d48 | 4055 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); |
2bd59d48 | 4056 | |
4ac06017 | 4057 | mutex_lock(&cgroup_mutex); |
455050d2 | 4058 | |
ea15f8cc TH |
4059 | return 0; |
4060 | }; | |
4061 | ||
d3daf28d | 4062 | /** |
f20104de | 4063 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
4064 | * @work: cgroup->destroy_free_work |
4065 | * | |
4066 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
4067 | * destroyed after all css's are offlined and performs the rest of |
4068 | * destruction. This is the second step of destruction described in the | |
4069 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 4070 | */ |
f20104de | 4071 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 4072 | { |
ea15f8cc | 4073 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 4074 | |
ace2bee8 | 4075 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 4076 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 4077 | |
999cd8a4 | 4078 | /* delete this cgroup from parent->children */ |
eb6fd504 | 4079 | list_del_rcu(&cgrp->sibling); |
ed957793 | 4080 | |
59f5296b | 4081 | cgroup_put(cgrp); |
ddbcc7e8 | 4082 | |
bd89aabc | 4083 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 4084 | check_for_release(parent); |
ddbcc7e8 PM |
4085 | } |
4086 | ||
2bd59d48 | 4087 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4088 | { |
2bd59d48 TH |
4089 | struct cgroup *cgrp = kn->priv; |
4090 | int ret = 0; | |
4091 | ||
4092 | /* | |
4093 | * This is self-destruction but @kn can't be removed while this | |
4094 | * callback is in progress. Let's break active protection. Once | |
4095 | * the protection is broken, @cgrp can be destroyed at any point. | |
4096 | * Pin it so that it stays accessible. | |
4097 | */ | |
4098 | cgroup_get(cgrp); | |
4099 | kernfs_break_active_protection(kn); | |
42809dd4 | 4100 | |
ace2bee8 | 4101 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 4102 | mutex_lock(&cgroup_mutex); |
2bd59d48 TH |
4103 | |
4104 | /* | |
4105 | * @cgrp might already have been destroyed while we're trying to | |
4106 | * grab the mutexes. | |
4107 | */ | |
4108 | if (!cgroup_is_dead(cgrp)) | |
4109 | ret = cgroup_destroy_locked(cgrp); | |
4110 | ||
42809dd4 | 4111 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4112 | mutex_unlock(&cgroup_tree_mutex); |
42809dd4 | 4113 | |
2bd59d48 TH |
4114 | kernfs_unbreak_active_protection(kn); |
4115 | cgroup_put(cgrp); | |
42809dd4 TH |
4116 | return ret; |
4117 | } | |
4118 | ||
2bd59d48 TH |
4119 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4120 | .remount_fs = cgroup_remount, | |
4121 | .show_options = cgroup_show_options, | |
4122 | .mkdir = cgroup_mkdir, | |
4123 | .rmdir = cgroup_rmdir, | |
4124 | .rename = cgroup_rename, | |
4125 | }; | |
4126 | ||
06a11920 | 4127 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4128 | { |
ddbcc7e8 | 4129 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4130 | |
4131 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4132 | |
ace2bee8 | 4133 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
4134 | mutex_lock(&cgroup_mutex); |
4135 | ||
0adb0704 | 4136 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4137 | |
ddbcc7e8 | 4138 | /* Create the top cgroup state for this subsystem */ |
9871bf95 | 4139 | ss->root = &cgroup_dummy_root; |
ca8bdcaf | 4140 | css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss)); |
ddbcc7e8 PM |
4141 | /* We don't handle early failures gracefully */ |
4142 | BUG_ON(IS_ERR(css)); | |
623f926b | 4143 | init_css(css, ss, cgroup_dummy_top); |
ddbcc7e8 | 4144 | |
e8d55fde | 4145 | /* Update the init_css_set to contain a subsys |
817929ec | 4146 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
4147 | * newly registered, all tasks and hence the |
4148 | * init_css_set is in the subsystem's top cgroup. */ | |
aec25020 | 4149 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4150 | |
4151 | need_forkexit_callback |= ss->fork || ss->exit; | |
4152 | ||
e8d55fde LZ |
4153 | /* At system boot, before all subsystems have been |
4154 | * registered, no tasks have been forked, so we don't | |
4155 | * need to invoke fork callbacks here. */ | |
4156 | BUG_ON(!list_empty(&init_task.tasks)); | |
4157 | ||
ae7f164a | 4158 | BUG_ON(online_css(css)); |
a8638030 | 4159 | |
648bb56d | 4160 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4161 | mutex_unlock(&cgroup_tree_mutex); |
e6a1105b BB |
4162 | } |
4163 | ||
ddbcc7e8 | 4164 | /** |
a043e3b2 LZ |
4165 | * cgroup_init_early - cgroup initialization at system boot |
4166 | * | |
4167 | * Initialize cgroups at system boot, and initialize any | |
4168 | * subsystems that request early init. | |
ddbcc7e8 PM |
4169 | */ |
4170 | int __init cgroup_init_early(void) | |
4171 | { | |
30159ec7 | 4172 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4173 | int i; |
30159ec7 | 4174 | |
146aa1bd | 4175 | atomic_set(&init_css_set.refcount, 1); |
69d0206c | 4176 | INIT_LIST_HEAD(&init_css_set.cgrp_links); |
817929ec | 4177 | INIT_LIST_HEAD(&init_css_set.tasks); |
472b1053 | 4178 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 4179 | css_set_count = 1; |
9871bf95 TH |
4180 | init_cgroup_root(&cgroup_dummy_root); |
4181 | cgroup_root_count = 1; | |
a4ea1cc9 | 4182 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4183 | |
69d0206c | 4184 | init_cgrp_cset_link.cset = &init_css_set; |
9871bf95 TH |
4185 | init_cgrp_cset_link.cgrp = cgroup_dummy_top; |
4186 | list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links); | |
69d0206c | 4187 | list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links); |
ddbcc7e8 | 4188 | |
3ed80a62 | 4189 | for_each_subsys(ss, i) { |
aec25020 | 4190 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4191 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4192 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4193 | ss->id, ss->name); |
073219e9 TH |
4194 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4195 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4196 | ||
aec25020 | 4197 | ss->id = i; |
073219e9 | 4198 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4199 | |
4200 | if (ss->early_init) | |
4201 | cgroup_init_subsys(ss); | |
4202 | } | |
4203 | return 0; | |
4204 | } | |
4205 | ||
4206 | /** | |
a043e3b2 LZ |
4207 | * cgroup_init - cgroup initialization |
4208 | * | |
4209 | * Register cgroup filesystem and /proc file, and initialize | |
4210 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4211 | */ |
4212 | int __init cgroup_init(void) | |
4213 | { | |
30159ec7 | 4214 | struct cgroup_subsys *ss; |
0ac801fe | 4215 | unsigned long key; |
30159ec7 | 4216 | int i, err; |
a424316c | 4217 | |
2bd59d48 | 4218 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
2da440a2 | 4219 | |
3ed80a62 | 4220 | for_each_subsys(ss, i) { |
ddbcc7e8 PM |
4221 | if (!ss->early_init) |
4222 | cgroup_init_subsys(ss); | |
de00ffa5 TH |
4223 | |
4224 | /* | |
4225 | * cftype registration needs kmalloc and can't be done | |
4226 | * during early_init. Register base cftypes separately. | |
4227 | */ | |
4228 | if (ss->base_cftypes) | |
4229 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
ddbcc7e8 PM |
4230 | } |
4231 | ||
fa3ca07e | 4232 | /* allocate id for the dummy hierarchy */ |
54e7b4eb | 4233 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4234 | |
82fe9b0d TH |
4235 | /* Add init_css_set to the hash table */ |
4236 | key = css_set_hash(init_css_set.subsys); | |
4237 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4238 | ||
fc76df70 | 4239 | BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1)); |
676db4af | 4240 | |
4e96ee8e LZ |
4241 | err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top, |
4242 | 0, 1, GFP_KERNEL); | |
4243 | BUG_ON(err < 0); | |
4244 | ||
54e7b4eb TH |
4245 | mutex_unlock(&cgroup_mutex); |
4246 | ||
676db4af | 4247 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4248 | if (!cgroup_kobj) |
4249 | return -ENOMEM; | |
676db4af | 4250 | |
ddbcc7e8 | 4251 | err = register_filesystem(&cgroup_fs_type); |
676db4af GKH |
4252 | if (err < 0) { |
4253 | kobject_put(cgroup_kobj); | |
2bd59d48 | 4254 | return err; |
676db4af | 4255 | } |
ddbcc7e8 | 4256 | |
46ae220b | 4257 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 4258 | return 0; |
ddbcc7e8 | 4259 | } |
b4f48b63 | 4260 | |
e5fca243 TH |
4261 | static int __init cgroup_wq_init(void) |
4262 | { | |
4263 | /* | |
4264 | * There isn't much point in executing destruction path in | |
4265 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
ab3f5faa HD |
4266 | * |
4267 | * XXX: Must be ordered to make sure parent is offlined after | |
4268 | * children. The ordering requirement is for memcg where a | |
4269 | * parent's offline may wait for a child's leading to deadlock. In | |
4270 | * the long term, this should be fixed from memcg side. | |
e5fca243 TH |
4271 | * |
4272 | * We would prefer to do this in cgroup_init() above, but that | |
4273 | * is called before init_workqueues(): so leave this until after. | |
4274 | */ | |
ab3f5faa | 4275 | cgroup_destroy_wq = alloc_ordered_workqueue("cgroup_destroy", 0); |
e5fca243 | 4276 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4277 | |
4278 | /* | |
4279 | * Used to destroy pidlists and separate to serve as flush domain. | |
4280 | * Cap @max_active to 1 too. | |
4281 | */ | |
4282 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4283 | 0, 1); | |
4284 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4285 | ||
e5fca243 TH |
4286 | return 0; |
4287 | } | |
4288 | core_initcall(cgroup_wq_init); | |
4289 | ||
a424316c PM |
4290 | /* |
4291 | * proc_cgroup_show() | |
4292 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4293 | * - Used for /proc/<pid>/cgroup. | |
4294 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4295 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4296 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4297 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4298 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4299 | * cgroup to top_cgroup. | |
4300 | */ | |
4301 | ||
4302 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4303 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4304 | { |
4305 | struct pid *pid; | |
4306 | struct task_struct *tsk; | |
4307 | char *buf; | |
4308 | int retval; | |
4309 | struct cgroupfs_root *root; | |
4310 | ||
4311 | retval = -ENOMEM; | |
4312 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
4313 | if (!buf) | |
4314 | goto out; | |
4315 | ||
4316 | retval = -ESRCH; | |
4317 | pid = m->private; | |
4318 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4319 | if (!tsk) | |
4320 | goto out_free; | |
4321 | ||
4322 | retval = 0; | |
4323 | ||
4324 | mutex_lock(&cgroup_mutex); | |
4325 | ||
e5f6a860 | 4326 | for_each_active_root(root) { |
a424316c | 4327 | struct cgroup_subsys *ss; |
bd89aabc | 4328 | struct cgroup *cgrp; |
b85d2040 | 4329 | int ssid, count = 0; |
a424316c | 4330 | |
2c6ab6d2 | 4331 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 TH |
4332 | for_each_subsys(ss, ssid) |
4333 | if (root->subsys_mask & (1 << ssid)) | |
4334 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4335 | if (strlen(root->name)) |
4336 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4337 | root->name); | |
a424316c | 4338 | seq_putc(m, ':'); |
7717f7ba | 4339 | cgrp = task_cgroup_from_root(tsk, root); |
bd89aabc | 4340 | retval = cgroup_path(cgrp, buf, PAGE_SIZE); |
a424316c PM |
4341 | if (retval < 0) |
4342 | goto out_unlock; | |
4343 | seq_puts(m, buf); | |
4344 | seq_putc(m, '\n'); | |
4345 | } | |
4346 | ||
4347 | out_unlock: | |
4348 | mutex_unlock(&cgroup_mutex); | |
4349 | put_task_struct(tsk); | |
4350 | out_free: | |
4351 | kfree(buf); | |
4352 | out: | |
4353 | return retval; | |
4354 | } | |
4355 | ||
a424316c PM |
4356 | /* Display information about each subsystem and each hierarchy */ |
4357 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4358 | { | |
30159ec7 | 4359 | struct cgroup_subsys *ss; |
a424316c | 4360 | int i; |
a424316c | 4361 | |
8bab8dde | 4362 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4363 | /* |
4364 | * ideally we don't want subsystems moving around while we do this. | |
4365 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4366 | * subsys/hierarchy state. | |
4367 | */ | |
a424316c | 4368 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4369 | |
4370 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4371 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4372 | ss->name, ss->root->hierarchy_id, | |
8bab8dde | 4373 | ss->root->number_of_cgroups, !ss->disabled); |
30159ec7 | 4374 | |
a424316c PM |
4375 | mutex_unlock(&cgroup_mutex); |
4376 | return 0; | |
4377 | } | |
4378 | ||
4379 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4380 | { | |
9dce07f1 | 4381 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4382 | } |
4383 | ||
828c0950 | 4384 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4385 | .open = cgroupstats_open, |
4386 | .read = seq_read, | |
4387 | .llseek = seq_lseek, | |
4388 | .release = single_release, | |
4389 | }; | |
4390 | ||
b4f48b63 PM |
4391 | /** |
4392 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4393 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4394 | * |
4395 | * Description: A task inherits its parent's cgroup at fork(). | |
4396 | * | |
4397 | * A pointer to the shared css_set was automatically copied in | |
4398 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
9bb71308 TH |
4399 | * it was not made under the protection of RCU or cgroup_mutex, so |
4400 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
4401 | * have already changed current->cgroups, allowing the previously | |
4402 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
4403 | * |
4404 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4405 | * task, and the passed argument 'child' points to the child task. | |
4406 | */ | |
4407 | void cgroup_fork(struct task_struct *child) | |
4408 | { | |
9bb71308 | 4409 | task_lock(current); |
a8ad805c | 4410 | get_css_set(task_css_set(current)); |
817929ec | 4411 | child->cgroups = current->cgroups; |
9bb71308 | 4412 | task_unlock(current); |
817929ec | 4413 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4414 | } |
4415 | ||
817929ec | 4416 | /** |
a043e3b2 LZ |
4417 | * cgroup_post_fork - called on a new task after adding it to the task list |
4418 | * @child: the task in question | |
4419 | * | |
5edee61e TH |
4420 | * Adds the task to the list running through its css_set if necessary and |
4421 | * call the subsystem fork() callbacks. Has to be after the task is | |
4422 | * visible on the task list in case we race with the first call to | |
0942eeee | 4423 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4424 | * list. |
a043e3b2 | 4425 | */ |
817929ec PM |
4426 | void cgroup_post_fork(struct task_struct *child) |
4427 | { | |
30159ec7 | 4428 | struct cgroup_subsys *ss; |
5edee61e TH |
4429 | int i; |
4430 | ||
3ce3230a FW |
4431 | /* |
4432 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
4433 | * under the tasklist_lock and we read it here after we added the child | |
4434 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
4435 | * yet in the tasklist when we walked through it from | |
4436 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
4437 | * should be visible now due to the paired locking and barriers implied | |
4438 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
4439 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
4440 | * lock on fork. | |
4441 | */ | |
817929ec PM |
4442 | if (use_task_css_set_links) { |
4443 | write_lock(&css_set_lock); | |
d8783832 TH |
4444 | task_lock(child); |
4445 | if (list_empty(&child->cg_list)) | |
a8ad805c | 4446 | list_add(&child->cg_list, &task_css_set(child)->tasks); |
d8783832 | 4447 | task_unlock(child); |
817929ec PM |
4448 | write_unlock(&css_set_lock); |
4449 | } | |
5edee61e TH |
4450 | |
4451 | /* | |
4452 | * Call ss->fork(). This must happen after @child is linked on | |
4453 | * css_set; otherwise, @child might change state between ->fork() | |
4454 | * and addition to css_set. | |
4455 | */ | |
4456 | if (need_forkexit_callback) { | |
3ed80a62 | 4457 | for_each_subsys(ss, i) |
5edee61e TH |
4458 | if (ss->fork) |
4459 | ss->fork(child); | |
5edee61e | 4460 | } |
817929ec | 4461 | } |
5edee61e | 4462 | |
b4f48b63 PM |
4463 | /** |
4464 | * cgroup_exit - detach cgroup from exiting task | |
4465 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4466 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4467 | * |
4468 | * Description: Detach cgroup from @tsk and release it. | |
4469 | * | |
4470 | * Note that cgroups marked notify_on_release force every task in | |
4471 | * them to take the global cgroup_mutex mutex when exiting. | |
4472 | * This could impact scaling on very large systems. Be reluctant to | |
4473 | * use notify_on_release cgroups where very high task exit scaling | |
4474 | * is required on large systems. | |
4475 | * | |
4476 | * the_top_cgroup_hack: | |
4477 | * | |
4478 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4479 | * | |
4480 | * We call cgroup_exit() while the task is still competent to | |
4481 | * handle notify_on_release(), then leave the task attached to the | |
4482 | * root cgroup in each hierarchy for the remainder of its exit. | |
4483 | * | |
4484 | * To do this properly, we would increment the reference count on | |
4485 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4486 | * code we would add a second cgroup function call, to drop that | |
4487 | * reference. This would just create an unnecessary hot spot on | |
4488 | * the top_cgroup reference count, to no avail. | |
4489 | * | |
4490 | * Normally, holding a reference to a cgroup without bumping its | |
4491 | * count is unsafe. The cgroup could go away, or someone could | |
4492 | * attach us to a different cgroup, decrementing the count on | |
4493 | * the first cgroup that we never incremented. But in this case, | |
4494 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4495 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4496 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4497 | */ |
4498 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4499 | { | |
30159ec7 | 4500 | struct cgroup_subsys *ss; |
5abb8855 | 4501 | struct css_set *cset; |
d41d5a01 | 4502 | int i; |
817929ec PM |
4503 | |
4504 | /* | |
4505 | * Unlink from the css_set task list if necessary. | |
4506 | * Optimistically check cg_list before taking | |
4507 | * css_set_lock | |
4508 | */ | |
4509 | if (!list_empty(&tsk->cg_list)) { | |
4510 | write_lock(&css_set_lock); | |
4511 | if (!list_empty(&tsk->cg_list)) | |
8d258797 | 4512 | list_del_init(&tsk->cg_list); |
817929ec PM |
4513 | write_unlock(&css_set_lock); |
4514 | } | |
4515 | ||
b4f48b63 PM |
4516 | /* Reassign the task to the init_css_set. */ |
4517 | task_lock(tsk); | |
a8ad805c TH |
4518 | cset = task_css_set(tsk); |
4519 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 PZ |
4520 | |
4521 | if (run_callbacks && need_forkexit_callback) { | |
3ed80a62 TH |
4522 | /* see cgroup_post_fork() for details */ |
4523 | for_each_subsys(ss, i) { | |
d41d5a01 | 4524 | if (ss->exit) { |
eb95419b TH |
4525 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4526 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4527 | |
eb95419b | 4528 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4529 | } |
4530 | } | |
4531 | } | |
b4f48b63 | 4532 | task_unlock(tsk); |
d41d5a01 | 4533 | |
5abb8855 | 4534 | put_css_set_taskexit(cset); |
b4f48b63 | 4535 | } |
697f4161 | 4536 | |
bd89aabc | 4537 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4538 | { |
f50daa70 | 4539 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4540 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4541 | /* |
4542 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4543 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4544 | * it now |
4545 | */ | |
81a6a5cd | 4546 | int need_schedule_work = 0; |
f50daa70 | 4547 | |
cdcc136f | 4548 | raw_spin_lock(&release_list_lock); |
54766d4a | 4549 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4550 | list_empty(&cgrp->release_list)) { |
4551 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4552 | need_schedule_work = 1; |
4553 | } | |
cdcc136f | 4554 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4555 | if (need_schedule_work) |
4556 | schedule_work(&release_agent_work); | |
4557 | } | |
4558 | } | |
4559 | ||
81a6a5cd PM |
4560 | /* |
4561 | * Notify userspace when a cgroup is released, by running the | |
4562 | * configured release agent with the name of the cgroup (path | |
4563 | * relative to the root of cgroup file system) as the argument. | |
4564 | * | |
4565 | * Most likely, this user command will try to rmdir this cgroup. | |
4566 | * | |
4567 | * This races with the possibility that some other task will be | |
4568 | * attached to this cgroup before it is removed, or that some other | |
4569 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4570 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4571 | * unused, and this cgroup will be reprieved from its death sentence, | |
4572 | * to continue to serve a useful existence. Next time it's released, | |
4573 | * we will get notified again, if it still has 'notify_on_release' set. | |
4574 | * | |
4575 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4576 | * means only wait until the task is successfully execve()'d. The | |
4577 | * separate release agent task is forked by call_usermodehelper(), | |
4578 | * then control in this thread returns here, without waiting for the | |
4579 | * release agent task. We don't bother to wait because the caller of | |
4580 | * this routine has no use for the exit status of the release agent | |
4581 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4582 | */ |
81a6a5cd PM |
4583 | static void cgroup_release_agent(struct work_struct *work) |
4584 | { | |
4585 | BUG_ON(work != &release_agent_work); | |
4586 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4587 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4588 | while (!list_empty(&release_list)) { |
4589 | char *argv[3], *envp[3]; | |
4590 | int i; | |
e788e066 | 4591 | char *pathbuf = NULL, *agentbuf = NULL; |
bd89aabc | 4592 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4593 | struct cgroup, |
4594 | release_list); | |
bd89aabc | 4595 | list_del_init(&cgrp->release_list); |
cdcc136f | 4596 | raw_spin_unlock(&release_list_lock); |
81a6a5cd | 4597 | pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
e788e066 PM |
4598 | if (!pathbuf) |
4599 | goto continue_free; | |
4600 | if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) | |
4601 | goto continue_free; | |
4602 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4603 | if (!agentbuf) | |
4604 | goto continue_free; | |
81a6a5cd PM |
4605 | |
4606 | i = 0; | |
e788e066 PM |
4607 | argv[i++] = agentbuf; |
4608 | argv[i++] = pathbuf; | |
81a6a5cd PM |
4609 | argv[i] = NULL; |
4610 | ||
4611 | i = 0; | |
4612 | /* minimal command environment */ | |
4613 | envp[i++] = "HOME=/"; | |
4614 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4615 | envp[i] = NULL; | |
4616 | ||
4617 | /* Drop the lock while we invoke the usermode helper, | |
4618 | * since the exec could involve hitting disk and hence | |
4619 | * be a slow process */ | |
4620 | mutex_unlock(&cgroup_mutex); | |
4621 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4622 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4623 | continue_free: |
4624 | kfree(pathbuf); | |
4625 | kfree(agentbuf); | |
cdcc136f | 4626 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4627 | } |
cdcc136f | 4628 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4629 | mutex_unlock(&cgroup_mutex); |
4630 | } | |
8bab8dde PM |
4631 | |
4632 | static int __init cgroup_disable(char *str) | |
4633 | { | |
30159ec7 | 4634 | struct cgroup_subsys *ss; |
8bab8dde | 4635 | char *token; |
30159ec7 | 4636 | int i; |
8bab8dde PM |
4637 | |
4638 | while ((token = strsep(&str, ",")) != NULL) { | |
4639 | if (!*token) | |
4640 | continue; | |
be45c900 | 4641 | |
3ed80a62 | 4642 | for_each_subsys(ss, i) { |
8bab8dde PM |
4643 | if (!strcmp(token, ss->name)) { |
4644 | ss->disabled = 1; | |
4645 | printk(KERN_INFO "Disabling %s control group" | |
4646 | " subsystem\n", ss->name); | |
4647 | break; | |
4648 | } | |
4649 | } | |
4650 | } | |
4651 | return 1; | |
4652 | } | |
4653 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 4654 | |
b77d7b60 | 4655 | /** |
5a17f543 | 4656 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
4657 | * @dentry: directory dentry of interest |
4658 | * @ss: subsystem of interest | |
b77d7b60 | 4659 | * |
5a17f543 TH |
4660 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
4661 | * to get the corresponding css and return it. If such css doesn't exist | |
4662 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 4663 | */ |
5a17f543 TH |
4664 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
4665 | struct cgroup_subsys *ss) | |
e5d1367f | 4666 | { |
2bd59d48 TH |
4667 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
4668 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 4669 | struct cgroup *cgrp; |
b77d7b60 | 4670 | |
35cf0836 | 4671 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
4672 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
4673 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
4674 | return ERR_PTR(-EBADF); |
4675 | ||
5a17f543 TH |
4676 | rcu_read_lock(); |
4677 | ||
2bd59d48 TH |
4678 | /* |
4679 | * This path doesn't originate from kernfs and @kn could already | |
4680 | * have been or be removed at any point. @kn->priv is RCU | |
4681 | * protected for this access. See destroy_locked() for details. | |
4682 | */ | |
4683 | cgrp = rcu_dereference(kn->priv); | |
4684 | if (cgrp) | |
4685 | css = cgroup_css(cgrp, ss); | |
5a17f543 TH |
4686 | |
4687 | if (!css || !css_tryget(css)) | |
4688 | css = ERR_PTR(-ENOENT); | |
4689 | ||
4690 | rcu_read_unlock(); | |
4691 | return css; | |
e5d1367f | 4692 | } |
e5d1367f | 4693 | |
1cb650b9 LZ |
4694 | /** |
4695 | * css_from_id - lookup css by id | |
4696 | * @id: the cgroup id | |
4697 | * @ss: cgroup subsys to be looked into | |
4698 | * | |
4699 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
4700 | * Should be called under rcu_read_lock(). | |
4701 | */ | |
4702 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
4703 | { | |
4704 | struct cgroup *cgrp; | |
4705 | ||
ace2bee8 | 4706 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
4707 | |
4708 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
4709 | if (cgrp) | |
d1625964 | 4710 | return cgroup_css(cgrp, ss); |
1cb650b9 | 4711 | return NULL; |
e5d1367f SE |
4712 | } |
4713 | ||
fe693435 | 4714 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
4715 | static struct cgroup_subsys_state * |
4716 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
4717 | { |
4718 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
4719 | ||
4720 | if (!css) | |
4721 | return ERR_PTR(-ENOMEM); | |
4722 | ||
4723 | return css; | |
4724 | } | |
4725 | ||
eb95419b | 4726 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 4727 | { |
eb95419b | 4728 | kfree(css); |
fe693435 PM |
4729 | } |
4730 | ||
182446d0 TH |
4731 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
4732 | struct cftype *cft) | |
fe693435 | 4733 | { |
182446d0 | 4734 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
4735 | } |
4736 | ||
182446d0 TH |
4737 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
4738 | struct cftype *cft) | |
fe693435 PM |
4739 | { |
4740 | return (u64)(unsigned long)current->cgroups; | |
4741 | } | |
4742 | ||
182446d0 | 4743 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 4744 | struct cftype *cft) |
fe693435 PM |
4745 | { |
4746 | u64 count; | |
4747 | ||
4748 | rcu_read_lock(); | |
a8ad805c | 4749 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
4750 | rcu_read_unlock(); |
4751 | return count; | |
4752 | } | |
4753 | ||
2da8ca82 | 4754 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4755 | { |
69d0206c | 4756 | struct cgrp_cset_link *link; |
5abb8855 | 4757 | struct css_set *cset; |
7717f7ba PM |
4758 | |
4759 | read_lock(&css_set_lock); | |
4760 | rcu_read_lock(); | |
5abb8855 | 4761 | cset = rcu_dereference(current->cgroups); |
69d0206c | 4762 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 4763 | struct cgroup *c = link->cgrp; |
59f5296b TH |
4764 | const char *name = "?"; |
4765 | ||
4766 | if (c != cgroup_dummy_top) | |
4767 | name = cgroup_name(c); | |
7717f7ba | 4768 | |
2c6ab6d2 PM |
4769 | seq_printf(seq, "Root %d group %s\n", |
4770 | c->root->hierarchy_id, name); | |
7717f7ba PM |
4771 | } |
4772 | rcu_read_unlock(); | |
4773 | read_unlock(&css_set_lock); | |
4774 | return 0; | |
4775 | } | |
4776 | ||
4777 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 4778 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4779 | { |
2da8ca82 | 4780 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 4781 | struct cgrp_cset_link *link; |
7717f7ba PM |
4782 | |
4783 | read_lock(&css_set_lock); | |
182446d0 | 4784 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 4785 | struct css_set *cset = link->cset; |
7717f7ba PM |
4786 | struct task_struct *task; |
4787 | int count = 0; | |
5abb8855 TH |
4788 | seq_printf(seq, "css_set %p\n", cset); |
4789 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
7717f7ba PM |
4790 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { |
4791 | seq_puts(seq, " ...\n"); | |
4792 | break; | |
4793 | } else { | |
4794 | seq_printf(seq, " task %d\n", | |
4795 | task_pid_vnr(task)); | |
4796 | } | |
4797 | } | |
4798 | } | |
4799 | read_unlock(&css_set_lock); | |
4800 | return 0; | |
4801 | } | |
4802 | ||
182446d0 | 4803 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 4804 | { |
182446d0 | 4805 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
4806 | } |
4807 | ||
4808 | static struct cftype debug_files[] = { | |
fe693435 PM |
4809 | { |
4810 | .name = "taskcount", | |
4811 | .read_u64 = debug_taskcount_read, | |
4812 | }, | |
4813 | ||
4814 | { | |
4815 | .name = "current_css_set", | |
4816 | .read_u64 = current_css_set_read, | |
4817 | }, | |
4818 | ||
4819 | { | |
4820 | .name = "current_css_set_refcount", | |
4821 | .read_u64 = current_css_set_refcount_read, | |
4822 | }, | |
4823 | ||
7717f7ba PM |
4824 | { |
4825 | .name = "current_css_set_cg_links", | |
2da8ca82 | 4826 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
4827 | }, |
4828 | ||
4829 | { | |
4830 | .name = "cgroup_css_links", | |
2da8ca82 | 4831 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
4832 | }, |
4833 | ||
fe693435 PM |
4834 | { |
4835 | .name = "releasable", | |
4836 | .read_u64 = releasable_read, | |
4837 | }, | |
fe693435 | 4838 | |
4baf6e33 TH |
4839 | { } /* terminate */ |
4840 | }; | |
fe693435 | 4841 | |
073219e9 | 4842 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
4843 | .css_alloc = debug_css_alloc, |
4844 | .css_free = debug_css_free, | |
4baf6e33 | 4845 | .base_cftypes = debug_files, |
fe693435 PM |
4846 | }; |
4847 | #endif /* CONFIG_CGROUP_DEBUG */ |