]>
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 | 44 | #include <linux/spinlock.h> |
96d365e0 | 45 | #include <linux/rwsem.h> |
ddbcc7e8 | 46 | #include <linux/string.h> |
bbcb81d0 | 47 | #include <linux/sort.h> |
81a6a5cd | 48 | #include <linux/kmod.h> |
846c7bb0 BS |
49 | #include <linux/delayacct.h> |
50 | #include <linux/cgroupstats.h> | |
0ac801fe | 51 | #include <linux/hashtable.h> |
096b7fe0 | 52 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 53 | #include <linux/idr.h> |
d1d9fd33 | 54 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 55 | #include <linux/kthread.h> |
776f02fa | 56 | #include <linux/delay.h> |
846c7bb0 | 57 | |
60063497 | 58 | #include <linux/atomic.h> |
ddbcc7e8 | 59 | |
b1a21367 TH |
60 | /* |
61 | * pidlists linger the following amount before being destroyed. The goal | |
62 | * is avoiding frequent destruction in the middle of consecutive read calls | |
63 | * Expiring in the middle is a performance problem not a correctness one. | |
64 | * 1 sec should be enough. | |
65 | */ | |
66 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
67 | ||
8d7e6fb0 TH |
68 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
69 | MAX_CFTYPE_NAME + 2) | |
70 | ||
ace2bee8 TH |
71 | /* |
72 | * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file | |
73 | * creation/removal and hierarchy changing operations including cgroup | |
74 | * creation, removal, css association and controller rebinding. This outer | |
75 | * lock is needed mainly to resolve the circular dependency between kernfs | |
76 | * active ref and cgroup_mutex. cgroup_tree_mutex nests above both. | |
77 | */ | |
78 | static DEFINE_MUTEX(cgroup_tree_mutex); | |
79 | ||
e25e2cbb TH |
80 | /* |
81 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
82 | * hierarchy must be performed while holding it. | |
83 | * | |
0e1d768f TH |
84 | * css_set_rwsem protects task->cgroups pointer, the list of css_set |
85 | * objects, and the chain of tasks off each css_set. | |
e25e2cbb | 86 | * |
0e1d768f TH |
87 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
88 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 89 | */ |
2219449a TH |
90 | #ifdef CONFIG_PROVE_RCU |
91 | DEFINE_MUTEX(cgroup_mutex); | |
0e1d768f TH |
92 | DECLARE_RWSEM(css_set_rwsem); |
93 | EXPORT_SYMBOL_GPL(cgroup_mutex); | |
94 | EXPORT_SYMBOL_GPL(css_set_rwsem); | |
2219449a | 95 | #else |
81a6a5cd | 96 | static DEFINE_MUTEX(cgroup_mutex); |
0e1d768f | 97 | static DECLARE_RWSEM(css_set_rwsem); |
2219449a TH |
98 | #endif |
99 | ||
69e943b7 TH |
100 | /* |
101 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
102 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
103 | */ | |
104 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 105 | |
ace2bee8 | 106 | #define cgroup_assert_mutexes_or_rcu_locked() \ |
87fb54f1 | 107 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
ace2bee8 | 108 | lockdep_is_held(&cgroup_tree_mutex) || \ |
87fb54f1 | 109 | lockdep_is_held(&cgroup_mutex), \ |
ace2bee8 | 110 | "cgroup_[tree_]mutex or RCU read lock required"); |
780cd8b3 | 111 | |
e5fca243 TH |
112 | /* |
113 | * cgroup destruction makes heavy use of work items and there can be a lot | |
114 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
115 | * destruction work items don't end up filling up max_active of system_wq | |
116 | * which may lead to deadlock. | |
117 | */ | |
118 | static struct workqueue_struct *cgroup_destroy_wq; | |
119 | ||
b1a21367 TH |
120 | /* |
121 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
122 | * separate workqueue as flush domain. | |
123 | */ | |
124 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
125 | ||
3ed80a62 | 126 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 127 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 128 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
129 | #include <linux/cgroup_subsys.h> |
130 | }; | |
073219e9 TH |
131 | #undef SUBSYS |
132 | ||
133 | /* array of cgroup subsystem names */ | |
134 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
135 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
136 | #include <linux/cgroup_subsys.h> |
137 | }; | |
073219e9 | 138 | #undef SUBSYS |
ddbcc7e8 | 139 | |
ddbcc7e8 | 140 | /* |
3dd06ffa | 141 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
142 | * unattached - it never has more than a single cgroup, and all tasks are |
143 | * part of that cgroup. | |
ddbcc7e8 | 144 | */ |
a2dd4247 | 145 | struct cgroup_root cgrp_dfl_root; |
9871bf95 | 146 | |
a2dd4247 TH |
147 | /* |
148 | * The default hierarchy always exists but is hidden until mounted for the | |
149 | * first time. This is for backward compatibility. | |
150 | */ | |
151 | static bool cgrp_dfl_root_visible; | |
ddbcc7e8 PM |
152 | |
153 | /* The list of hierarchy roots */ | |
154 | ||
9871bf95 TH |
155 | static LIST_HEAD(cgroup_roots); |
156 | static int cgroup_root_count; | |
ddbcc7e8 | 157 | |
3417ae1f | 158 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 159 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 160 | |
794611a1 LZ |
161 | /* |
162 | * Assign a monotonically increasing serial number to cgroups. It | |
163 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
164 | * numbers. Also, as cgroups are always appended to the parent's | |
165 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
166 | * the ascending serial number order on the list. Protected by |
167 | * cgroup_mutex. | |
794611a1 | 168 | */ |
00356bd5 | 169 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 170 | |
ddbcc7e8 | 171 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
172 | * check for fork/exit handlers to call. This avoids us having to do |
173 | * extra work in the fork/exit path if none of the subsystems need to | |
174 | * be called. | |
ddbcc7e8 | 175 | */ |
8947f9d5 | 176 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 177 | |
628f7cd4 TH |
178 | static struct cftype cgroup_base_files[]; |
179 | ||
59f5296b | 180 | static void cgroup_put(struct cgroup *cgrp); |
3dd06ffa | 181 | static int rebind_subsystems(struct cgroup_root *dst_root, |
5df36032 | 182 | unsigned long ss_mask); |
f20104de | 183 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 184 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
2bb566cb TH |
185 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
186 | bool is_add); | |
b1a21367 | 187 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 188 | |
95109b62 TH |
189 | /** |
190 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
191 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 192 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 193 | * |
ca8bdcaf TH |
194 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
195 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
196 | * the caller is responsible for pinning the returned css if it wants to | |
197 | * keep accessing it outside the said locks. This function may return | |
198 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
199 | */ |
200 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 201 | struct cgroup_subsys *ss) |
95109b62 | 202 | { |
ca8bdcaf | 203 | if (ss) |
aec25020 | 204 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 TH |
205 | lockdep_is_held(&cgroup_tree_mutex) || |
206 | lockdep_is_held(&cgroup_mutex)); | |
ca8bdcaf TH |
207 | else |
208 | return &cgrp->dummy_css; | |
95109b62 | 209 | } |
42809dd4 | 210 | |
aec3dfcb TH |
211 | /** |
212 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
213 | * @cgrp: the cgroup of interest | |
214 | * @ss: the subsystem of interest (%NULL returns the dummy_css) | |
215 | * | |
216 | * Similar to cgroup_css() but returns the effctive css, which is defined | |
217 | * as the matching css of the nearest ancestor including self which has @ss | |
218 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
219 | * function is guaranteed to return non-NULL css. | |
220 | */ | |
221 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
222 | struct cgroup_subsys *ss) | |
223 | { | |
224 | lockdep_assert_held(&cgroup_mutex); | |
225 | ||
226 | if (!ss) | |
227 | return &cgrp->dummy_css; | |
228 | ||
229 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
230 | return NULL; | |
231 | ||
232 | while (cgrp->parent && | |
233 | !(cgrp->parent->child_subsys_mask & (1 << ss->id))) | |
234 | cgrp = cgrp->parent; | |
235 | ||
236 | return cgroup_css(cgrp, ss); | |
237 | } | |
238 | ||
ddbcc7e8 | 239 | /* convenient tests for these bits */ |
54766d4a | 240 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 241 | { |
54766d4a | 242 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
243 | } |
244 | ||
59f5296b TH |
245 | struct cgroup_subsys_state *seq_css(struct seq_file *seq) |
246 | { | |
2bd59d48 TH |
247 | struct kernfs_open_file *of = seq->private; |
248 | struct cgroup *cgrp = of->kn->parent->priv; | |
249 | struct cftype *cft = seq_cft(seq); | |
250 | ||
251 | /* | |
252 | * This is open and unprotected implementation of cgroup_css(). | |
253 | * seq_css() is only called from a kernfs file operation which has | |
254 | * an active reference on the file. Because all the subsystem | |
255 | * files are drained before a css is disassociated with a cgroup, | |
256 | * the matching css from the cgroup's subsys table is guaranteed to | |
257 | * be and stay valid until the enclosing operation is complete. | |
258 | */ | |
259 | if (cft->ss) | |
260 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
261 | else | |
262 | return &cgrp->dummy_css; | |
59f5296b TH |
263 | } |
264 | EXPORT_SYMBOL_GPL(seq_css); | |
265 | ||
78574cf9 LZ |
266 | /** |
267 | * cgroup_is_descendant - test ancestry | |
268 | * @cgrp: the cgroup to be tested | |
269 | * @ancestor: possible ancestor of @cgrp | |
270 | * | |
271 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
272 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
273 | * and @ancestor are accessible. | |
274 | */ | |
275 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
276 | { | |
277 | while (cgrp) { | |
278 | if (cgrp == ancestor) | |
279 | return true; | |
280 | cgrp = cgrp->parent; | |
281 | } | |
282 | return false; | |
283 | } | |
ddbcc7e8 | 284 | |
e9685a03 | 285 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
286 | { |
287 | const int bits = | |
bd89aabc PM |
288 | (1 << CGRP_RELEASABLE) | |
289 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
290 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
291 | } |
292 | ||
e9685a03 | 293 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 294 | { |
bd89aabc | 295 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
296 | } |
297 | ||
1c6727af TH |
298 | /** |
299 | * for_each_css - iterate all css's of a cgroup | |
300 | * @css: the iteration cursor | |
301 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
302 | * @cgrp: the target cgroup to iterate css's of | |
303 | * | |
aec3dfcb | 304 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
305 | */ |
306 | #define for_each_css(css, ssid, cgrp) \ | |
307 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
308 | if (!((css) = rcu_dereference_check( \ | |
309 | (cgrp)->subsys[(ssid)], \ | |
ace2bee8 | 310 | lockdep_is_held(&cgroup_tree_mutex) || \ |
1c6727af TH |
311 | lockdep_is_held(&cgroup_mutex)))) { } \ |
312 | else | |
313 | ||
aec3dfcb TH |
314 | /** |
315 | * for_each_e_css - iterate all effective css's of a cgroup | |
316 | * @css: the iteration cursor | |
317 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
318 | * @cgrp: the target cgroup to iterate css's of | |
319 | * | |
320 | * Should be called under cgroup_[tree_]mutex. | |
321 | */ | |
322 | #define for_each_e_css(css, ssid, cgrp) \ | |
323 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
324 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
325 | ; \ | |
326 | else | |
327 | ||
30159ec7 | 328 | /** |
3ed80a62 | 329 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 330 | * @ss: the iteration cursor |
780cd8b3 | 331 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 332 | */ |
780cd8b3 | 333 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
334 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
335 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 336 | |
985ed670 TH |
337 | /* iterate across the hierarchies */ |
338 | #define for_each_root(root) \ | |
5549c497 | 339 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 340 | |
7ae1bad9 TH |
341 | /** |
342 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
343 | * @cgrp: the cgroup to be checked for liveness | |
344 | * | |
47cfcd09 TH |
345 | * On success, returns true; the mutex should be later unlocked. On |
346 | * failure returns false with no lock held. | |
7ae1bad9 | 347 | */ |
b9777cf8 | 348 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
349 | { |
350 | mutex_lock(&cgroup_mutex); | |
54766d4a | 351 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
352 | mutex_unlock(&cgroup_mutex); |
353 | return false; | |
354 | } | |
355 | return true; | |
356 | } | |
7ae1bad9 | 357 | |
81a6a5cd PM |
358 | /* the list of cgroups eligible for automatic release. Protected by |
359 | * release_list_lock */ | |
360 | static LIST_HEAD(release_list); | |
cdcc136f | 361 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
362 | static void cgroup_release_agent(struct work_struct *work); |
363 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 364 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 365 | |
69d0206c TH |
366 | /* |
367 | * A cgroup can be associated with multiple css_sets as different tasks may | |
368 | * belong to different cgroups on different hierarchies. In the other | |
369 | * direction, a css_set is naturally associated with multiple cgroups. | |
370 | * This M:N relationship is represented by the following link structure | |
371 | * which exists for each association and allows traversing the associations | |
372 | * from both sides. | |
373 | */ | |
374 | struct cgrp_cset_link { | |
375 | /* the cgroup and css_set this link associates */ | |
376 | struct cgroup *cgrp; | |
377 | struct css_set *cset; | |
378 | ||
379 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
380 | struct list_head cset_link; | |
381 | ||
382 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
383 | struct list_head cgrp_link; | |
817929ec PM |
384 | }; |
385 | ||
172a2c06 TH |
386 | /* |
387 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
388 | * hierarchies being mounted. It contains a pointer to the root state |
389 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
390 | * reference-counted, to improve performance when child cgroups | |
391 | * haven't been created. | |
392 | */ | |
172a2c06 TH |
393 | static struct css_set init_css_set = { |
394 | .refcount = ATOMIC_INIT(1), | |
395 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
396 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
397 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
398 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
399 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
400 | }; | |
817929ec | 401 | |
172a2c06 | 402 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 403 | |
7717f7ba PM |
404 | /* |
405 | * hash table for cgroup groups. This improves the performance to find | |
406 | * an existing css_set. This hash doesn't (currently) take into | |
407 | * account cgroups in empty hierarchies. | |
408 | */ | |
472b1053 | 409 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 410 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 411 | |
0ac801fe | 412 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 413 | { |
0ac801fe | 414 | unsigned long key = 0UL; |
30159ec7 TH |
415 | struct cgroup_subsys *ss; |
416 | int i; | |
472b1053 | 417 | |
30159ec7 | 418 | for_each_subsys(ss, i) |
0ac801fe LZ |
419 | key += (unsigned long)css[i]; |
420 | key = (key >> 16) ^ key; | |
472b1053 | 421 | |
0ac801fe | 422 | return key; |
472b1053 LZ |
423 | } |
424 | ||
89c5509b | 425 | static void put_css_set_locked(struct css_set *cset, bool taskexit) |
b4f48b63 | 426 | { |
69d0206c | 427 | struct cgrp_cset_link *link, *tmp_link; |
5abb8855 | 428 | |
89c5509b TH |
429 | lockdep_assert_held(&css_set_rwsem); |
430 | ||
431 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 432 | return; |
81a6a5cd | 433 | |
2c6ab6d2 | 434 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
5abb8855 | 435 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
436 | css_set_count--; |
437 | ||
69d0206c | 438 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 439 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 440 | |
69d0206c TH |
441 | list_del(&link->cset_link); |
442 | list_del(&link->cgrp_link); | |
71b5707e | 443 | |
96d365e0 | 444 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
6f3d828f | 445 | if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { |
81a6a5cd | 446 | if (taskexit) |
bd89aabc PM |
447 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
448 | check_for_release(cgrp); | |
81a6a5cd | 449 | } |
2c6ab6d2 PM |
450 | |
451 | kfree(link); | |
81a6a5cd | 452 | } |
2c6ab6d2 | 453 | |
5abb8855 | 454 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
455 | } |
456 | ||
89c5509b TH |
457 | static void put_css_set(struct css_set *cset, bool taskexit) |
458 | { | |
459 | /* | |
460 | * Ensure that the refcount doesn't hit zero while any readers | |
461 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
462 | * rwlock | |
463 | */ | |
464 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
465 | return; | |
466 | ||
467 | down_write(&css_set_rwsem); | |
468 | put_css_set_locked(cset, taskexit); | |
469 | up_write(&css_set_rwsem); | |
470 | } | |
471 | ||
817929ec PM |
472 | /* |
473 | * refcounted get/put for css_set objects | |
474 | */ | |
5abb8855 | 475 | static inline void get_css_set(struct css_set *cset) |
817929ec | 476 | { |
5abb8855 | 477 | atomic_inc(&cset->refcount); |
817929ec PM |
478 | } |
479 | ||
b326f9d0 | 480 | /** |
7717f7ba | 481 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
482 | * @cset: candidate css_set being tested |
483 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
484 | * @new_cgrp: cgroup that's being entered by the task |
485 | * @template: desired set of css pointers in css_set (pre-calculated) | |
486 | * | |
6f4b7e63 | 487 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
488 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
489 | */ | |
5abb8855 TH |
490 | static bool compare_css_sets(struct css_set *cset, |
491 | struct css_set *old_cset, | |
7717f7ba PM |
492 | struct cgroup *new_cgrp, |
493 | struct cgroup_subsys_state *template[]) | |
494 | { | |
495 | struct list_head *l1, *l2; | |
496 | ||
aec3dfcb TH |
497 | /* |
498 | * On the default hierarchy, there can be csets which are | |
499 | * associated with the same set of cgroups but different csses. | |
500 | * Let's first ensure that csses match. | |
501 | */ | |
502 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 503 | return false; |
7717f7ba PM |
504 | |
505 | /* | |
506 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
507 | * different cgroups in hierarchies. As different cgroups may |
508 | * share the same effective css, this comparison is always | |
509 | * necessary. | |
7717f7ba | 510 | */ |
69d0206c TH |
511 | l1 = &cset->cgrp_links; |
512 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 513 | while (1) { |
69d0206c | 514 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 515 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
516 | |
517 | l1 = l1->next; | |
518 | l2 = l2->next; | |
519 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
520 | if (l1 == &cset->cgrp_links) { |
521 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
522 | break; |
523 | } else { | |
69d0206c | 524 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
525 | } |
526 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
527 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
528 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
529 | cgrp1 = link1->cgrp; | |
530 | cgrp2 = link2->cgrp; | |
7717f7ba | 531 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 532 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
533 | |
534 | /* | |
535 | * If this hierarchy is the hierarchy of the cgroup | |
536 | * that's changing, then we need to check that this | |
537 | * css_set points to the new cgroup; if it's any other | |
538 | * hierarchy, then this css_set should point to the | |
539 | * same cgroup as the old css_set. | |
540 | */ | |
5abb8855 TH |
541 | if (cgrp1->root == new_cgrp->root) { |
542 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
543 | return false; |
544 | } else { | |
5abb8855 | 545 | if (cgrp1 != cgrp2) |
7717f7ba PM |
546 | return false; |
547 | } | |
548 | } | |
549 | return true; | |
550 | } | |
551 | ||
b326f9d0 TH |
552 | /** |
553 | * find_existing_css_set - init css array and find the matching css_set | |
554 | * @old_cset: the css_set that we're using before the cgroup transition | |
555 | * @cgrp: the cgroup that we're moving into | |
556 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 557 | */ |
5abb8855 TH |
558 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
559 | struct cgroup *cgrp, | |
560 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 561 | { |
3dd06ffa | 562 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 563 | struct cgroup_subsys *ss; |
5abb8855 | 564 | struct css_set *cset; |
0ac801fe | 565 | unsigned long key; |
b326f9d0 | 566 | int i; |
817929ec | 567 | |
aae8aab4 BB |
568 | /* |
569 | * Build the set of subsystem state objects that we want to see in the | |
570 | * new css_set. while subsystems can change globally, the entries here | |
571 | * won't change, so no need for locking. | |
572 | */ | |
30159ec7 | 573 | for_each_subsys(ss, i) { |
f392e51c | 574 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
575 | /* |
576 | * @ss is in this hierarchy, so we want the | |
577 | * effective css from @cgrp. | |
578 | */ | |
579 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 580 | } else { |
aec3dfcb TH |
581 | /* |
582 | * @ss is not in this hierarchy, so we don't want | |
583 | * to change the css. | |
584 | */ | |
5abb8855 | 585 | template[i] = old_cset->subsys[i]; |
817929ec PM |
586 | } |
587 | } | |
588 | ||
0ac801fe | 589 | key = css_set_hash(template); |
5abb8855 TH |
590 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
591 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
592 | continue; |
593 | ||
594 | /* This css_set matches what we need */ | |
5abb8855 | 595 | return cset; |
472b1053 | 596 | } |
817929ec PM |
597 | |
598 | /* No existing cgroup group matched */ | |
599 | return NULL; | |
600 | } | |
601 | ||
69d0206c | 602 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 603 | { |
69d0206c | 604 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 605 | |
69d0206c TH |
606 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
607 | list_del(&link->cset_link); | |
36553434 LZ |
608 | kfree(link); |
609 | } | |
610 | } | |
611 | ||
69d0206c TH |
612 | /** |
613 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
614 | * @count: the number of links to allocate | |
615 | * @tmp_links: list_head the allocated links are put on | |
616 | * | |
617 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
618 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 619 | */ |
69d0206c | 620 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 621 | { |
69d0206c | 622 | struct cgrp_cset_link *link; |
817929ec | 623 | int i; |
69d0206c TH |
624 | |
625 | INIT_LIST_HEAD(tmp_links); | |
626 | ||
817929ec | 627 | for (i = 0; i < count; i++) { |
f4f4be2b | 628 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 629 | if (!link) { |
69d0206c | 630 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
631 | return -ENOMEM; |
632 | } | |
69d0206c | 633 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
634 | } |
635 | return 0; | |
636 | } | |
637 | ||
c12f65d4 LZ |
638 | /** |
639 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 640 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 641 | * @cset: the css_set to be linked |
c12f65d4 LZ |
642 | * @cgrp: the destination cgroup |
643 | */ | |
69d0206c TH |
644 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
645 | struct cgroup *cgrp) | |
c12f65d4 | 646 | { |
69d0206c | 647 | struct cgrp_cset_link *link; |
c12f65d4 | 648 | |
69d0206c TH |
649 | BUG_ON(list_empty(tmp_links)); |
650 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); | |
651 | link->cset = cset; | |
7717f7ba | 652 | link->cgrp = cgrp; |
69d0206c | 653 | list_move(&link->cset_link, &cgrp->cset_links); |
7717f7ba PM |
654 | /* |
655 | * Always add links to the tail of the list so that the list | |
656 | * is sorted by order of hierarchy creation | |
657 | */ | |
69d0206c | 658 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
659 | } |
660 | ||
b326f9d0 TH |
661 | /** |
662 | * find_css_set - return a new css_set with one cgroup updated | |
663 | * @old_cset: the baseline css_set | |
664 | * @cgrp: the cgroup to be updated | |
665 | * | |
666 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
667 | * substituted into the appropriate hierarchy. | |
817929ec | 668 | */ |
5abb8855 TH |
669 | static struct css_set *find_css_set(struct css_set *old_cset, |
670 | struct cgroup *cgrp) | |
817929ec | 671 | { |
b326f9d0 | 672 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 673 | struct css_set *cset; |
69d0206c TH |
674 | struct list_head tmp_links; |
675 | struct cgrp_cset_link *link; | |
0ac801fe | 676 | unsigned long key; |
472b1053 | 677 | |
b326f9d0 TH |
678 | lockdep_assert_held(&cgroup_mutex); |
679 | ||
817929ec PM |
680 | /* First see if we already have a cgroup group that matches |
681 | * the desired set */ | |
96d365e0 | 682 | down_read(&css_set_rwsem); |
5abb8855 TH |
683 | cset = find_existing_css_set(old_cset, cgrp, template); |
684 | if (cset) | |
685 | get_css_set(cset); | |
96d365e0 | 686 | up_read(&css_set_rwsem); |
817929ec | 687 | |
5abb8855 TH |
688 | if (cset) |
689 | return cset; | |
817929ec | 690 | |
f4f4be2b | 691 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 692 | if (!cset) |
817929ec PM |
693 | return NULL; |
694 | ||
69d0206c | 695 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 696 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 697 | kfree(cset); |
817929ec PM |
698 | return NULL; |
699 | } | |
700 | ||
5abb8855 | 701 | atomic_set(&cset->refcount, 1); |
69d0206c | 702 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 703 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 704 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 705 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 706 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 707 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
708 | |
709 | /* Copy the set of subsystem state objects generated in | |
710 | * find_existing_css_set() */ | |
5abb8855 | 711 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 712 | |
96d365e0 | 713 | down_write(&css_set_rwsem); |
817929ec | 714 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 715 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 716 | struct cgroup *c = link->cgrp; |
69d0206c | 717 | |
7717f7ba PM |
718 | if (c->root == cgrp->root) |
719 | c = cgrp; | |
69d0206c | 720 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 721 | } |
817929ec | 722 | |
69d0206c | 723 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 724 | |
817929ec | 725 | css_set_count++; |
472b1053 LZ |
726 | |
727 | /* Add this cgroup group to the hash table */ | |
5abb8855 TH |
728 | key = css_set_hash(cset->subsys); |
729 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 730 | |
96d365e0 | 731 | up_write(&css_set_rwsem); |
817929ec | 732 | |
5abb8855 | 733 | return cset; |
b4f48b63 PM |
734 | } |
735 | ||
3dd06ffa | 736 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 737 | { |
3dd06ffa | 738 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 739 | |
3dd06ffa | 740 | return root_cgrp->root; |
2bd59d48 TH |
741 | } |
742 | ||
3dd06ffa | 743 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
744 | { |
745 | int id; | |
746 | ||
747 | lockdep_assert_held(&cgroup_mutex); | |
748 | ||
985ed670 | 749 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
750 | if (id < 0) |
751 | return id; | |
752 | ||
753 | root->hierarchy_id = id; | |
754 | return 0; | |
755 | } | |
756 | ||
3dd06ffa | 757 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
758 | { |
759 | lockdep_assert_held(&cgroup_mutex); | |
760 | ||
761 | if (root->hierarchy_id) { | |
762 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
763 | root->hierarchy_id = 0; | |
764 | } | |
765 | } | |
766 | ||
3dd06ffa | 767 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
768 | { |
769 | if (root) { | |
770 | /* hierarhcy ID shoulid already have been released */ | |
771 | WARN_ON_ONCE(root->hierarchy_id); | |
772 | ||
773 | idr_destroy(&root->cgroup_idr); | |
774 | kfree(root); | |
775 | } | |
776 | } | |
777 | ||
3dd06ffa | 778 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 779 | { |
3dd06ffa | 780 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 781 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 782 | |
2bd59d48 | 783 | mutex_lock(&cgroup_tree_mutex); |
2bd59d48 | 784 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 785 | |
776f02fa | 786 | BUG_ON(atomic_read(&root->nr_cgrps)); |
f2e85d57 TH |
787 | BUG_ON(!list_empty(&cgrp->children)); |
788 | ||
f2e85d57 | 789 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 790 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 791 | |
7717f7ba | 792 | /* |
f2e85d57 TH |
793 | * Release all the links from cset_links to this hierarchy's |
794 | * root cgroup | |
7717f7ba | 795 | */ |
96d365e0 | 796 | down_write(&css_set_rwsem); |
f2e85d57 TH |
797 | |
798 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
799 | list_del(&link->cset_link); | |
800 | list_del(&link->cgrp_link); | |
801 | kfree(link); | |
802 | } | |
96d365e0 | 803 | up_write(&css_set_rwsem); |
f2e85d57 TH |
804 | |
805 | if (!list_empty(&root->root_list)) { | |
806 | list_del(&root->root_list); | |
807 | cgroup_root_count--; | |
808 | } | |
809 | ||
810 | cgroup_exit_root_id(root); | |
811 | ||
812 | mutex_unlock(&cgroup_mutex); | |
813 | mutex_unlock(&cgroup_tree_mutex); | |
f2e85d57 | 814 | |
2bd59d48 | 815 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
816 | cgroup_free_root(root); |
817 | } | |
818 | ||
ceb6a081 TH |
819 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
820 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 821 | struct cgroup_root *root) |
7717f7ba | 822 | { |
7717f7ba PM |
823 | struct cgroup *res = NULL; |
824 | ||
96d365e0 TH |
825 | lockdep_assert_held(&cgroup_mutex); |
826 | lockdep_assert_held(&css_set_rwsem); | |
827 | ||
5abb8855 | 828 | if (cset == &init_css_set) { |
3dd06ffa | 829 | res = &root->cgrp; |
7717f7ba | 830 | } else { |
69d0206c TH |
831 | struct cgrp_cset_link *link; |
832 | ||
833 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 834 | struct cgroup *c = link->cgrp; |
69d0206c | 835 | |
7717f7ba PM |
836 | if (c->root == root) { |
837 | res = c; | |
838 | break; | |
839 | } | |
840 | } | |
841 | } | |
96d365e0 | 842 | |
7717f7ba PM |
843 | BUG_ON(!res); |
844 | return res; | |
845 | } | |
846 | ||
ddbcc7e8 | 847 | /* |
ceb6a081 TH |
848 | * Return the cgroup for "task" from the given hierarchy. Must be |
849 | * called with cgroup_mutex and css_set_rwsem held. | |
850 | */ | |
851 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 852 | struct cgroup_root *root) |
ceb6a081 TH |
853 | { |
854 | /* | |
855 | * No need to lock the task - since we hold cgroup_mutex the | |
856 | * task can't change groups, so the only thing that can happen | |
857 | * is that it exits and its css is set back to init_css_set. | |
858 | */ | |
859 | return cset_cgroup_from_root(task_css_set(task), root); | |
860 | } | |
861 | ||
ddbcc7e8 | 862 | /* |
ddbcc7e8 PM |
863 | * A task must hold cgroup_mutex to modify cgroups. |
864 | * | |
865 | * Any task can increment and decrement the count field without lock. | |
866 | * So in general, code holding cgroup_mutex can't rely on the count | |
867 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 868 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
869 | * means that no tasks are currently attached, therefore there is no |
870 | * way a task attached to that cgroup can fork (the other way to | |
871 | * increment the count). So code holding cgroup_mutex can safely | |
872 | * assume that if the count is zero, it will stay zero. Similarly, if | |
873 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
874 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
875 | * needs that mutex. | |
876 | * | |
ddbcc7e8 PM |
877 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
878 | * (usually) take cgroup_mutex. These are the two most performance | |
879 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
880 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
881 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
882 | * to the release agent with the name of the cgroup (path relative to |
883 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
884 | * |
885 | * A cgroup can only be deleted if both its 'count' of using tasks | |
886 | * is zero, and its list of 'children' cgroups is empty. Since all | |
887 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 888 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 889 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 890 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
891 | * |
892 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 893 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
894 | */ |
895 | ||
628f7cd4 | 896 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
2bd59d48 | 897 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 898 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 899 | |
8d7e6fb0 TH |
900 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
901 | char *buf) | |
ddbcc7e8 | 902 | { |
8d7e6fb0 TH |
903 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
904 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
905 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
906 | cft->ss->name, cft->name); | |
907 | else | |
908 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
909 | return buf; | |
ddbcc7e8 PM |
910 | } |
911 | ||
f2e85d57 TH |
912 | /** |
913 | * cgroup_file_mode - deduce file mode of a control file | |
914 | * @cft: the control file in question | |
915 | * | |
916 | * returns cft->mode if ->mode is not 0 | |
917 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
918 | * returns S_IRUGO if it has only a read handler | |
919 | * returns S_IWUSR if it has only a write hander | |
920 | */ | |
921 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 922 | { |
f2e85d57 | 923 | umode_t mode = 0; |
65dff759 | 924 | |
f2e85d57 TH |
925 | if (cft->mode) |
926 | return cft->mode; | |
927 | ||
928 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
929 | mode |= S_IRUGO; | |
930 | ||
931 | if (cft->write_u64 || cft->write_s64 || cft->write_string || | |
932 | cft->trigger) | |
933 | mode |= S_IWUSR; | |
934 | ||
935 | return mode; | |
65dff759 LZ |
936 | } |
937 | ||
be445626 LZ |
938 | static void cgroup_free_fn(struct work_struct *work) |
939 | { | |
ea15f8cc | 940 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 | 941 | |
3c9c825b | 942 | atomic_dec(&cgrp->root->nr_cgrps); |
b1a21367 | 943 | cgroup_pidlist_destroy_all(cgrp); |
be445626 | 944 | |
776f02fa TH |
945 | if (cgrp->parent) { |
946 | /* | |
947 | * We get a ref to the parent, and put the ref when this | |
948 | * cgroup is being freed, so it's guaranteed that the | |
949 | * parent won't be destroyed before its children. | |
950 | */ | |
951 | cgroup_put(cgrp->parent); | |
952 | kernfs_put(cgrp->kn); | |
953 | kfree(cgrp); | |
954 | } else { | |
955 | /* | |
3dd06ffa | 956 | * This is root cgroup's refcnt reaching zero, which |
776f02fa TH |
957 | * indicates that the root should be released. |
958 | */ | |
959 | cgroup_destroy_root(cgrp->root); | |
960 | } | |
be445626 LZ |
961 | } |
962 | ||
963 | static void cgroup_free_rcu(struct rcu_head *head) | |
964 | { | |
965 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
966 | ||
ea15f8cc | 967 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 968 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
969 | } |
970 | ||
59f5296b | 971 | static void cgroup_get(struct cgroup *cgrp) |
ddbcc7e8 | 972 | { |
2bd59d48 TH |
973 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
974 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
975 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
976 | } |
977 | ||
59f5296b | 978 | static void cgroup_put(struct cgroup *cgrp) |
05ef1d7c | 979 | { |
2bd59d48 TH |
980 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
981 | return; | |
776f02fa | 982 | if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp))) |
2bd59d48 | 983 | return; |
05ef1d7c | 984 | |
2739d3cc | 985 | /* |
2bd59d48 TH |
986 | * XXX: cgrp->id is only used to look up css's. As cgroup and |
987 | * css's lifetimes will be decoupled, it should be made | |
988 | * per-subsystem and moved to css->id so that lookups are | |
989 | * successful until the target css is released. | |
2739d3cc | 990 | */ |
2bd59d48 TH |
991 | mutex_lock(&cgroup_mutex); |
992 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
993 | mutex_unlock(&cgroup_mutex); | |
994 | cgrp->id = -1; | |
05ef1d7c | 995 | |
2bd59d48 | 996 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 | 997 | } |
05ef1d7c | 998 | |
2739d3cc | 999 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1000 | { |
2bd59d48 | 1001 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1002 | |
ace2bee8 | 1003 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 1004 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1005 | } |
1006 | ||
13af07df | 1007 | /** |
628f7cd4 | 1008 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 1009 | * @cgrp: target cgroup |
13af07df AR |
1010 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
1011 | */ | |
628f7cd4 | 1012 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 1013 | { |
13af07df | 1014 | struct cgroup_subsys *ss; |
b420ba7d | 1015 | int i; |
05ef1d7c | 1016 | |
b420ba7d | 1017 | for_each_subsys(ss, i) { |
0adb0704 | 1018 | struct cftype *cfts; |
b420ba7d TH |
1019 | |
1020 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 1021 | continue; |
0adb0704 TH |
1022 | list_for_each_entry(cfts, &ss->cfts, node) |
1023 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 1024 | } |
ddbcc7e8 PM |
1025 | } |
1026 | ||
3dd06ffa | 1027 | static int rebind_subsystems(struct cgroup_root *dst_root, |
5df36032 | 1028 | unsigned long ss_mask) |
ddbcc7e8 | 1029 | { |
30159ec7 | 1030 | struct cgroup_subsys *ss; |
5df36032 | 1031 | int ssid, ret; |
ddbcc7e8 | 1032 | |
ace2bee8 TH |
1033 | lockdep_assert_held(&cgroup_tree_mutex); |
1034 | lockdep_assert_held(&cgroup_mutex); | |
ddbcc7e8 | 1035 | |
5df36032 TH |
1036 | for_each_subsys(ss, ssid) { |
1037 | if (!(ss_mask & (1 << ssid))) | |
1038 | continue; | |
aae8aab4 | 1039 | |
5df36032 | 1040 | /* if @ss is on the dummy_root, we can always move it */ |
3dd06ffa | 1041 | if (ss->root == &cgrp_dfl_root) |
ddbcc7e8 | 1042 | continue; |
30159ec7 | 1043 | |
5df36032 | 1044 | /* if @ss has non-root cgroups attached to it, can't move */ |
3dd06ffa | 1045 | if (!list_empty(&ss->root->cgrp.children)) |
3ed80a62 | 1046 | return -EBUSY; |
1d5be6b2 | 1047 | |
5df36032 | 1048 | /* can't move between two non-dummy roots either */ |
3dd06ffa | 1049 | if (dst_root != &cgrp_dfl_root) |
5df36032 | 1050 | return -EBUSY; |
ddbcc7e8 PM |
1051 | } |
1052 | ||
a2dd4247 TH |
1053 | ret = cgroup_populate_dir(&dst_root->cgrp, ss_mask); |
1054 | if (ret) { | |
1055 | if (dst_root != &cgrp_dfl_root) | |
5df36032 | 1056 | return ret; |
ddbcc7e8 | 1057 | |
a2dd4247 TH |
1058 | /* |
1059 | * Rebinding back to the default root is not allowed to | |
1060 | * fail. Using both default and non-default roots should | |
1061 | * be rare. Moving subsystems back and forth even more so. | |
1062 | * Just warn about it and continue. | |
1063 | */ | |
1064 | if (cgrp_dfl_root_visible) { | |
1065 | pr_warning("cgroup: failed to create files (%d) while rebinding 0x%lx to default root\n", | |
1066 | ret, ss_mask); | |
1067 | pr_warning("cgroup: you may retry by moving them to a different hierarchy and unbinding\n"); | |
1068 | } | |
5df36032 | 1069 | } |
3126121f TH |
1070 | |
1071 | /* | |
1072 | * Nothing can fail from this point on. Remove files for the | |
1073 | * removed subsystems and rebind each subsystem. | |
1074 | */ | |
4ac06017 | 1075 | mutex_unlock(&cgroup_mutex); |
5df36032 | 1076 | for_each_subsys(ss, ssid) |
a2dd4247 | 1077 | if (ss_mask & (1 << ssid)) |
3dd06ffa | 1078 | cgroup_clear_dir(&ss->root->cgrp, 1 << ssid); |
4ac06017 | 1079 | mutex_lock(&cgroup_mutex); |
a8a648c4 | 1080 | |
5df36032 | 1081 | for_each_subsys(ss, ssid) { |
3dd06ffa | 1082 | struct cgroup_root *src_root; |
5df36032 | 1083 | struct cgroup_subsys_state *css; |
a8a648c4 | 1084 | |
5df36032 TH |
1085 | if (!(ss_mask & (1 << ssid))) |
1086 | continue; | |
a8a648c4 | 1087 | |
5df36032 | 1088 | src_root = ss->root; |
3dd06ffa | 1089 | css = cgroup_css(&src_root->cgrp, ss); |
a8a648c4 | 1090 | |
3dd06ffa | 1091 | WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss)); |
73e80ed8 | 1092 | |
3dd06ffa TH |
1093 | RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL); |
1094 | rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css); | |
5df36032 | 1095 | ss->root = dst_root; |
3dd06ffa | 1096 | css->cgroup = &dst_root->cgrp; |
73e80ed8 | 1097 | |
f392e51c TH |
1098 | src_root->subsys_mask &= ~(1 << ssid); |
1099 | src_root->cgrp.child_subsys_mask &= ~(1 << ssid); | |
1100 | ||
1101 | dst_root->subsys_mask |= 1 << ssid; | |
1102 | dst_root->cgrp.child_subsys_mask |= 1 << ssid; | |
a8a648c4 | 1103 | |
5df36032 TH |
1104 | if (ss->bind) |
1105 | ss->bind(css); | |
ddbcc7e8 | 1106 | } |
ddbcc7e8 | 1107 | |
a2dd4247 | 1108 | kernfs_activate(dst_root->cgrp.kn); |
ddbcc7e8 PM |
1109 | return 0; |
1110 | } | |
1111 | ||
2bd59d48 TH |
1112 | static int cgroup_show_options(struct seq_file *seq, |
1113 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1114 | { |
3dd06ffa | 1115 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1116 | struct cgroup_subsys *ss; |
b85d2040 | 1117 | int ssid; |
ddbcc7e8 | 1118 | |
b85d2040 | 1119 | for_each_subsys(ss, ssid) |
f392e51c | 1120 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 1121 | seq_printf(seq, ",%s", ss->name); |
873fe09e TH |
1122 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1123 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1124 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1125 | seq_puts(seq, ",noprefix"); |
93438629 | 1126 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1127 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1128 | |
1129 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1130 | if (strlen(root->release_agent_path)) |
1131 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1132 | spin_unlock(&release_agent_path_lock); |
1133 | ||
3dd06ffa | 1134 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1135 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1136 | if (strlen(root->name)) |
1137 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1138 | return 0; |
1139 | } | |
1140 | ||
1141 | struct cgroup_sb_opts { | |
a1a71b45 | 1142 | unsigned long subsys_mask; |
ddbcc7e8 | 1143 | unsigned long flags; |
81a6a5cd | 1144 | char *release_agent; |
2260e7fc | 1145 | bool cpuset_clone_children; |
c6d57f33 | 1146 | char *name; |
2c6ab6d2 PM |
1147 | /* User explicitly requested empty subsystem */ |
1148 | bool none; | |
ddbcc7e8 PM |
1149 | }; |
1150 | ||
aae8aab4 | 1151 | /* |
9871bf95 TH |
1152 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1153 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1154 | * array. This function takes refcounts on subsystems to be used, unless it | |
1155 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1156 | */ |
cf5d5941 | 1157 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1158 | { |
32a8cf23 DL |
1159 | char *token, *o = data; |
1160 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1161 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1162 | struct cgroup_subsys *ss; |
1163 | int i; | |
f9ab5b5b | 1164 | |
aae8aab4 BB |
1165 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1166 | ||
f9ab5b5b | 1167 | #ifdef CONFIG_CPUSETS |
073219e9 | 1168 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1169 | #endif |
ddbcc7e8 | 1170 | |
c6d57f33 | 1171 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1172 | |
1173 | while ((token = strsep(&o, ",")) != NULL) { | |
1174 | if (!*token) | |
1175 | return -EINVAL; | |
32a8cf23 | 1176 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1177 | /* Explicitly have no subsystems */ |
1178 | opts->none = true; | |
32a8cf23 DL |
1179 | continue; |
1180 | } | |
1181 | if (!strcmp(token, "all")) { | |
1182 | /* Mutually exclusive option 'all' + subsystem name */ | |
1183 | if (one_ss) | |
1184 | return -EINVAL; | |
1185 | all_ss = true; | |
1186 | continue; | |
1187 | } | |
873fe09e TH |
1188 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1189 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1190 | continue; | |
1191 | } | |
32a8cf23 | 1192 | if (!strcmp(token, "noprefix")) { |
93438629 | 1193 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1194 | continue; |
1195 | } | |
1196 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1197 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1198 | continue; |
1199 | } | |
03b1cde6 | 1200 | if (!strcmp(token, "xattr")) { |
93438629 | 1201 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1202 | continue; |
1203 | } | |
32a8cf23 | 1204 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1205 | /* Specifying two release agents is forbidden */ |
1206 | if (opts->release_agent) | |
1207 | return -EINVAL; | |
c6d57f33 | 1208 | opts->release_agent = |
e400c285 | 1209 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1210 | if (!opts->release_agent) |
1211 | return -ENOMEM; | |
32a8cf23 DL |
1212 | continue; |
1213 | } | |
1214 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1215 | const char *name = token + 5; |
1216 | /* Can't specify an empty name */ | |
1217 | if (!strlen(name)) | |
1218 | return -EINVAL; | |
1219 | /* Must match [\w.-]+ */ | |
1220 | for (i = 0; i < strlen(name); i++) { | |
1221 | char c = name[i]; | |
1222 | if (isalnum(c)) | |
1223 | continue; | |
1224 | if ((c == '.') || (c == '-') || (c == '_')) | |
1225 | continue; | |
1226 | return -EINVAL; | |
1227 | } | |
1228 | /* Specifying two names is forbidden */ | |
1229 | if (opts->name) | |
1230 | return -EINVAL; | |
1231 | opts->name = kstrndup(name, | |
e400c285 | 1232 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1233 | GFP_KERNEL); |
1234 | if (!opts->name) | |
1235 | return -ENOMEM; | |
32a8cf23 DL |
1236 | |
1237 | continue; | |
1238 | } | |
1239 | ||
30159ec7 | 1240 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1241 | if (strcmp(token, ss->name)) |
1242 | continue; | |
1243 | if (ss->disabled) | |
1244 | continue; | |
1245 | ||
1246 | /* Mutually exclusive option 'all' + subsystem name */ | |
1247 | if (all_ss) | |
1248 | return -EINVAL; | |
a1a71b45 | 1249 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1250 | one_ss = true; |
1251 | ||
1252 | break; | |
1253 | } | |
1254 | if (i == CGROUP_SUBSYS_COUNT) | |
1255 | return -ENOENT; | |
1256 | } | |
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 | ||
d3ba07c3 TH |
1263 | if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) || |
1264 | opts->cpuset_clone_children || opts->release_agent || | |
1265 | opts->name) { | |
1266 | pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n"); | |
873fe09e TH |
1267 | return -EINVAL; |
1268 | } | |
a2dd4247 TH |
1269 | } else { |
1270 | /* | |
1271 | * If the 'all' option was specified select all the | |
1272 | * subsystems, otherwise if 'none', 'name=' and a subsystem | |
1273 | * name options were not specified, let's default to 'all' | |
1274 | */ | |
1275 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1276 | for_each_subsys(ss, i) | |
1277 | if (!ss->disabled) | |
1278 | set_bit(i, &opts->subsys_mask); | |
873fe09e | 1279 | |
a2dd4247 TH |
1280 | /* |
1281 | * We either have to specify by name or by subsystems. (So | |
1282 | * all empty hierarchies must have a name). | |
1283 | */ | |
1284 | if (!opts->subsys_mask && !opts->name) | |
873fe09e | 1285 | return -EINVAL; |
873fe09e TH |
1286 | } |
1287 | ||
f9ab5b5b LZ |
1288 | /* |
1289 | * Option noprefix was introduced just for backward compatibility | |
1290 | * with the old cpuset, so we allow noprefix only if mounting just | |
1291 | * the cpuset subsystem. | |
1292 | */ | |
93438629 | 1293 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1294 | return -EINVAL; |
1295 | ||
2c6ab6d2 PM |
1296 | |
1297 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1298 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1299 | return -EINVAL; |
1300 | ||
ddbcc7e8 PM |
1301 | return 0; |
1302 | } | |
1303 | ||
2bd59d48 | 1304 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1305 | { |
1306 | int ret = 0; | |
3dd06ffa | 1307 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1308 | struct cgroup_sb_opts opts; |
a1a71b45 | 1309 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1310 | |
873fe09e TH |
1311 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1312 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1313 | return -EINVAL; | |
1314 | } | |
1315 | ||
ace2bee8 | 1316 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1317 | mutex_lock(&cgroup_mutex); |
1318 | ||
1319 | /* See what subsystems are wanted */ | |
1320 | ret = parse_cgroupfs_options(data, &opts); | |
1321 | if (ret) | |
1322 | goto out_unlock; | |
1323 | ||
f392e51c | 1324 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1325 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1326 | task_tgid_nr(current), current->comm); | |
1327 | ||
f392e51c TH |
1328 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1329 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1330 | |
cf5d5941 | 1331 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1332 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1333 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1334 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1335 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1336 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1337 | ret = -EINVAL; |
1338 | goto out_unlock; | |
1339 | } | |
1340 | ||
f172e67c | 1341 | /* remounting is not allowed for populated hierarchies */ |
3dd06ffa | 1342 | if (!list_empty(&root->cgrp.children)) { |
f172e67c | 1343 | ret = -EBUSY; |
0670e08b | 1344 | goto out_unlock; |
cf5d5941 | 1345 | } |
ddbcc7e8 | 1346 | |
5df36032 | 1347 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1348 | if (ret) |
0670e08b | 1349 | goto out_unlock; |
ddbcc7e8 | 1350 | |
3dd06ffa | 1351 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1352 | |
69e943b7 TH |
1353 | if (opts.release_agent) { |
1354 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1355 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1356 | spin_unlock(&release_agent_path_lock); |
1357 | } | |
ddbcc7e8 | 1358 | out_unlock: |
66bdc9cf | 1359 | kfree(opts.release_agent); |
c6d57f33 | 1360 | kfree(opts.name); |
ddbcc7e8 | 1361 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1362 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1363 | return ret; |
1364 | } | |
1365 | ||
afeb0f9f TH |
1366 | /* |
1367 | * To reduce the fork() overhead for systems that are not actually using | |
1368 | * their cgroups capability, we don't maintain the lists running through | |
1369 | * each css_set to its tasks until we see the list actually used - in other | |
1370 | * words after the first mount. | |
1371 | */ | |
1372 | static bool use_task_css_set_links __read_mostly; | |
1373 | ||
1374 | static void cgroup_enable_task_cg_lists(void) | |
1375 | { | |
1376 | struct task_struct *p, *g; | |
1377 | ||
96d365e0 | 1378 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1379 | |
1380 | if (use_task_css_set_links) | |
1381 | goto out_unlock; | |
1382 | ||
1383 | use_task_css_set_links = true; | |
1384 | ||
1385 | /* | |
1386 | * We need tasklist_lock because RCU is not safe against | |
1387 | * while_each_thread(). Besides, a forking task that has passed | |
1388 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1389 | * is not guaranteed to have its child immediately visible in the | |
1390 | * tasklist if we walk through it with RCU. | |
1391 | */ | |
1392 | read_lock(&tasklist_lock); | |
1393 | do_each_thread(g, p) { | |
afeb0f9f TH |
1394 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1395 | task_css_set(p) != &init_css_set); | |
1396 | ||
1397 | /* | |
1398 | * We should check if the process is exiting, otherwise | |
1399 | * it will race with cgroup_exit() in that the list | |
1400 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1401 | * Do it while holding siglock so that we don't end up |
1402 | * racing against cgroup_exit(). | |
afeb0f9f | 1403 | */ |
f153ad11 | 1404 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1405 | if (!(p->flags & PF_EXITING)) { |
1406 | struct css_set *cset = task_css_set(p); | |
1407 | ||
1408 | list_add(&p->cg_list, &cset->tasks); | |
1409 | get_css_set(cset); | |
1410 | } | |
f153ad11 | 1411 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1412 | } while_each_thread(g, p); |
1413 | read_unlock(&tasklist_lock); | |
1414 | out_unlock: | |
96d365e0 | 1415 | up_write(&css_set_rwsem); |
afeb0f9f | 1416 | } |
ddbcc7e8 | 1417 | |
cc31edce PM |
1418 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1419 | { | |
2bd59d48 | 1420 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1421 | INIT_LIST_HEAD(&cgrp->sibling); |
1422 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1423 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1424 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1425 | INIT_LIST_HEAD(&cgrp->pidlists); |
1426 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1427 | cgrp->dummy_css.cgroup = cgrp; |
cc31edce | 1428 | } |
c6d57f33 | 1429 | |
3dd06ffa | 1430 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1431 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1432 | { |
3dd06ffa | 1433 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1434 | |
ddbcc7e8 | 1435 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1436 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1437 | cgrp->root = root; |
cc31edce | 1438 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1439 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1440 | |
c6d57f33 PM |
1441 | root->flags = opts->flags; |
1442 | if (opts->release_agent) | |
1443 | strcpy(root->release_agent_path, opts->release_agent); | |
1444 | if (opts->name) | |
1445 | strcpy(root->name, opts->name); | |
2260e7fc | 1446 | if (opts->cpuset_clone_children) |
3dd06ffa | 1447 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1448 | } |
1449 | ||
3dd06ffa | 1450 | static int cgroup_setup_root(struct cgroup_root *root, unsigned long ss_mask) |
2c6ab6d2 | 1451 | { |
d427dfeb | 1452 | LIST_HEAD(tmp_links); |
3dd06ffa | 1453 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1454 | struct css_set *cset; |
d427dfeb | 1455 | int i, ret; |
2c6ab6d2 | 1456 | |
d427dfeb TH |
1457 | lockdep_assert_held(&cgroup_tree_mutex); |
1458 | lockdep_assert_held(&cgroup_mutex); | |
c6d57f33 | 1459 | |
d427dfeb TH |
1460 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); |
1461 | if (ret < 0) | |
2bd59d48 | 1462 | goto out; |
d427dfeb | 1463 | root_cgrp->id = ret; |
c6d57f33 | 1464 | |
d427dfeb | 1465 | /* |
96d365e0 | 1466 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1467 | * but that's OK - it can only be increased by someone holding |
1468 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1469 | * have some link structures left over | |
1470 | */ | |
1471 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1472 | if (ret) | |
2bd59d48 | 1473 | goto out; |
ddbcc7e8 | 1474 | |
985ed670 | 1475 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1476 | if (ret) |
2bd59d48 | 1477 | goto out; |
ddbcc7e8 | 1478 | |
2bd59d48 TH |
1479 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1480 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1481 | root_cgrp); | |
1482 | if (IS_ERR(root->kf_root)) { | |
1483 | ret = PTR_ERR(root->kf_root); | |
1484 | goto exit_root_id; | |
1485 | } | |
1486 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1487 | |
d427dfeb TH |
1488 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); |
1489 | if (ret) | |
2bd59d48 | 1490 | goto destroy_root; |
ddbcc7e8 | 1491 | |
5df36032 | 1492 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1493 | if (ret) |
2bd59d48 | 1494 | goto destroy_root; |
ddbcc7e8 | 1495 | |
d427dfeb TH |
1496 | /* |
1497 | * There must be no failure case after here, since rebinding takes | |
1498 | * care of subsystems' refcounts, which are explicitly dropped in | |
1499 | * the failure exit path. | |
1500 | */ | |
1501 | list_add(&root->root_list, &cgroup_roots); | |
1502 | cgroup_root_count++; | |
0df6a63f | 1503 | |
d427dfeb | 1504 | /* |
3dd06ffa | 1505 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1506 | * objects. |
1507 | */ | |
96d365e0 | 1508 | down_write(&css_set_rwsem); |
d427dfeb TH |
1509 | hash_for_each(css_set_table, i, cset, hlist) |
1510 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1511 | up_write(&css_set_rwsem); |
ddbcc7e8 | 1512 | |
d427dfeb | 1513 | BUG_ON(!list_empty(&root_cgrp->children)); |
3c9c825b | 1514 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1515 | |
2bd59d48 | 1516 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1517 | ret = 0; |
2bd59d48 | 1518 | goto out; |
d427dfeb | 1519 | |
2bd59d48 TH |
1520 | destroy_root: |
1521 | kernfs_destroy_root(root->kf_root); | |
1522 | root->kf_root = NULL; | |
1523 | exit_root_id: | |
d427dfeb | 1524 | cgroup_exit_root_id(root); |
2bd59d48 | 1525 | out: |
d427dfeb TH |
1526 | free_cgrp_cset_links(&tmp_links); |
1527 | return ret; | |
ddbcc7e8 PM |
1528 | } |
1529 | ||
f7e83571 | 1530 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1531 | int flags, const char *unused_dev_name, |
f7e83571 | 1532 | void *data) |
ddbcc7e8 | 1533 | { |
3dd06ffa | 1534 | struct cgroup_root *root; |
ddbcc7e8 | 1535 | struct cgroup_sb_opts opts; |
2bd59d48 | 1536 | struct dentry *dentry; |
8e30e2b8 | 1537 | int ret; |
c6b3d5bc | 1538 | bool new_sb; |
ddbcc7e8 | 1539 | |
56fde9e0 TH |
1540 | /* |
1541 | * The first time anyone tries to mount a cgroup, enable the list | |
1542 | * linking each css_set to its tasks and fix up all existing tasks. | |
1543 | */ | |
1544 | if (!use_task_css_set_links) | |
1545 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 1546 | |
8e30e2b8 | 1547 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1548 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1549 | |
1550 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1551 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1552 | if (ret) |
8e30e2b8 | 1553 | goto out_unlock; |
e37a06f1 | 1554 | retry: |
2bd59d48 | 1555 | /* look for a matching existing root */ |
a2dd4247 TH |
1556 | if (!opts.subsys_mask && !opts.none && !opts.name) { |
1557 | cgrp_dfl_root_visible = true; | |
1558 | root = &cgrp_dfl_root; | |
1559 | cgroup_get(&root->cgrp); | |
1560 | ret = 0; | |
1561 | goto out_unlock; | |
ddbcc7e8 PM |
1562 | } |
1563 | ||
985ed670 | 1564 | for_each_root(root) { |
2bd59d48 | 1565 | bool name_match = false; |
3126121f | 1566 | |
3dd06ffa | 1567 | if (root == &cgrp_dfl_root) |
985ed670 | 1568 | continue; |
3126121f | 1569 | |
cf5d5941 | 1570 | /* |
2bd59d48 TH |
1571 | * If we asked for a name then it must match. Also, if |
1572 | * name matches but sybsys_mask doesn't, we should fail. | |
1573 | * Remember whether name matched. | |
cf5d5941 | 1574 | */ |
2bd59d48 TH |
1575 | if (opts.name) { |
1576 | if (strcmp(opts.name, root->name)) | |
1577 | continue; | |
1578 | name_match = true; | |
1579 | } | |
ddbcc7e8 | 1580 | |
c6d57f33 | 1581 | /* |
2bd59d48 TH |
1582 | * If we asked for subsystems (or explicitly for no |
1583 | * subsystems) then they must match. | |
c6d57f33 | 1584 | */ |
2bd59d48 | 1585 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 1586 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
1587 | if (!name_match) |
1588 | continue; | |
1589 | ret = -EBUSY; | |
1590 | goto out_unlock; | |
1591 | } | |
873fe09e | 1592 | |
c7ba8287 | 1593 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1594 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1595 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1596 | ret = -EINVAL; | |
8e30e2b8 | 1597 | goto out_unlock; |
2a0ff3fb JL |
1598 | } else { |
1599 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1600 | } | |
873fe09e | 1601 | } |
ddbcc7e8 | 1602 | |
776f02fa | 1603 | /* |
3dd06ffa | 1604 | * A root's lifetime is governed by its root cgroup. Zero |
776f02fa TH |
1605 | * ref indicate that the root is being destroyed. Wait for |
1606 | * destruction to complete so that the subsystems are free. | |
1607 | * We can use wait_queue for the wait but this path is | |
1608 | * super cold. Let's just sleep for a bit and retry. | |
1609 | */ | |
3dd06ffa | 1610 | if (!atomic_inc_not_zero(&root->cgrp.refcnt)) { |
776f02fa TH |
1611 | mutex_unlock(&cgroup_mutex); |
1612 | mutex_unlock(&cgroup_tree_mutex); | |
1613 | msleep(10); | |
e37a06f1 LZ |
1614 | mutex_lock(&cgroup_tree_mutex); |
1615 | mutex_lock(&cgroup_mutex); | |
776f02fa TH |
1616 | goto retry; |
1617 | } | |
ddbcc7e8 | 1618 | |
776f02fa | 1619 | ret = 0; |
2bd59d48 | 1620 | goto out_unlock; |
ddbcc7e8 | 1621 | } |
ddbcc7e8 | 1622 | |
817929ec | 1623 | /* |
172a2c06 TH |
1624 | * No such thing, create a new one. name= matching without subsys |
1625 | * specification is allowed for already existing hierarchies but we | |
1626 | * can't create new one without subsys specification. | |
817929ec | 1627 | */ |
172a2c06 TH |
1628 | if (!opts.subsys_mask && !opts.none) { |
1629 | ret = -EINVAL; | |
1630 | goto out_unlock; | |
817929ec | 1631 | } |
817929ec | 1632 | |
172a2c06 TH |
1633 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
1634 | if (!root) { | |
1635 | ret = -ENOMEM; | |
2bd59d48 | 1636 | goto out_unlock; |
839ec545 | 1637 | } |
e5f6a860 | 1638 | |
172a2c06 TH |
1639 | init_cgroup_root(root, &opts); |
1640 | ||
35585573 | 1641 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1642 | if (ret) |
1643 | cgroup_free_root(root); | |
fa3ca07e | 1644 | |
8e30e2b8 | 1645 | out_unlock: |
ddbcc7e8 | 1646 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1647 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 | 1648 | |
c6d57f33 PM |
1649 | kfree(opts.release_agent); |
1650 | kfree(opts.name); | |
03b1cde6 | 1651 | |
2bd59d48 | 1652 | if (ret) |
8e30e2b8 | 1653 | return ERR_PTR(ret); |
2bd59d48 | 1654 | |
c6b3d5bc LZ |
1655 | dentry = kernfs_mount(fs_type, flags, root->kf_root, &new_sb); |
1656 | if (IS_ERR(dentry) || !new_sb) | |
3dd06ffa | 1657 | cgroup_put(&root->cgrp); |
2bd59d48 TH |
1658 | return dentry; |
1659 | } | |
1660 | ||
1661 | static void cgroup_kill_sb(struct super_block *sb) | |
1662 | { | |
1663 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 1664 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 1665 | |
3dd06ffa | 1666 | cgroup_put(&root->cgrp); |
2bd59d48 | 1667 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1668 | } |
1669 | ||
1670 | static struct file_system_type cgroup_fs_type = { | |
1671 | .name = "cgroup", | |
f7e83571 | 1672 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1673 | .kill_sb = cgroup_kill_sb, |
1674 | }; | |
1675 | ||
676db4af GKH |
1676 | static struct kobject *cgroup_kobj; |
1677 | ||
857a2beb | 1678 | /** |
913ffdb5 | 1679 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1680 | * @task: target task |
857a2beb TH |
1681 | * @buf: the buffer to write the path into |
1682 | * @buflen: the length of the buffer | |
1683 | * | |
913ffdb5 TH |
1684 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1685 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1686 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1687 | * cgroup controller callbacks. | |
1688 | * | |
e61734c5 | 1689 | * Return value is the same as kernfs_path(). |
857a2beb | 1690 | */ |
e61734c5 | 1691 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 1692 | { |
3dd06ffa | 1693 | struct cgroup_root *root; |
913ffdb5 | 1694 | struct cgroup *cgrp; |
e61734c5 TH |
1695 | int hierarchy_id = 1; |
1696 | char *path = NULL; | |
857a2beb TH |
1697 | |
1698 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1699 | down_read(&css_set_rwsem); |
857a2beb | 1700 | |
913ffdb5 TH |
1701 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1702 | ||
857a2beb TH |
1703 | if (root) { |
1704 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1705 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1706 | } else { |
1707 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1708 | if (strlcpy(buf, "/", buflen) < buflen) |
1709 | path = buf; | |
857a2beb TH |
1710 | } |
1711 | ||
96d365e0 | 1712 | up_read(&css_set_rwsem); |
857a2beb | 1713 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1714 | return path; |
857a2beb | 1715 | } |
913ffdb5 | 1716 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1717 | |
b3dc094e | 1718 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 1719 | struct cgroup_taskset { |
b3dc094e TH |
1720 | /* the src and dst cset list running through cset->mg_node */ |
1721 | struct list_head src_csets; | |
1722 | struct list_head dst_csets; | |
1723 | ||
1724 | /* | |
1725 | * Fields for cgroup_taskset_*() iteration. | |
1726 | * | |
1727 | * Before migration is committed, the target migration tasks are on | |
1728 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
1729 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
1730 | * or ->dst_csets depending on whether migration is committed. | |
1731 | * | |
1732 | * ->cur_csets and ->cur_task point to the current task position | |
1733 | * during iteration. | |
1734 | */ | |
1735 | struct list_head *csets; | |
1736 | struct css_set *cur_cset; | |
1737 | struct task_struct *cur_task; | |
2f7ee569 TH |
1738 | }; |
1739 | ||
1740 | /** | |
1741 | * cgroup_taskset_first - reset taskset and return the first task | |
1742 | * @tset: taskset of interest | |
1743 | * | |
1744 | * @tset iteration is initialized and the first task is returned. | |
1745 | */ | |
1746 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1747 | { | |
b3dc094e TH |
1748 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
1749 | tset->cur_task = NULL; | |
1750 | ||
1751 | return cgroup_taskset_next(tset); | |
2f7ee569 | 1752 | } |
2f7ee569 TH |
1753 | |
1754 | /** | |
1755 | * cgroup_taskset_next - iterate to the next task in taskset | |
1756 | * @tset: taskset of interest | |
1757 | * | |
1758 | * Return the next task in @tset. Iteration must have been initialized | |
1759 | * with cgroup_taskset_first(). | |
1760 | */ | |
1761 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1762 | { | |
b3dc094e TH |
1763 | struct css_set *cset = tset->cur_cset; |
1764 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 1765 | |
b3dc094e TH |
1766 | while (&cset->mg_node != tset->csets) { |
1767 | if (!task) | |
1768 | task = list_first_entry(&cset->mg_tasks, | |
1769 | struct task_struct, cg_list); | |
1770 | else | |
1771 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 1772 | |
b3dc094e TH |
1773 | if (&task->cg_list != &cset->mg_tasks) { |
1774 | tset->cur_cset = cset; | |
1775 | tset->cur_task = task; | |
1776 | return task; | |
1777 | } | |
2f7ee569 | 1778 | |
b3dc094e TH |
1779 | cset = list_next_entry(cset, mg_node); |
1780 | task = NULL; | |
1781 | } | |
2f7ee569 | 1782 | |
b3dc094e | 1783 | return NULL; |
2f7ee569 | 1784 | } |
2f7ee569 | 1785 | |
cb0f1fe9 | 1786 | /** |
74a1166d | 1787 | * cgroup_task_migrate - move a task from one cgroup to another. |
cb0f1fe9 TH |
1788 | * @old_cgrp; the cgroup @tsk is being migrated from |
1789 | * @tsk: the task being migrated | |
1790 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 1791 | * |
cb0f1fe9 | 1792 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 1793 | */ |
5abb8855 TH |
1794 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1795 | struct task_struct *tsk, | |
1796 | struct css_set *new_cset) | |
74a1166d | 1797 | { |
5abb8855 | 1798 | struct css_set *old_cset; |
74a1166d | 1799 | |
cb0f1fe9 TH |
1800 | lockdep_assert_held(&cgroup_mutex); |
1801 | lockdep_assert_held(&css_set_rwsem); | |
1802 | ||
74a1166d | 1803 | /* |
026085ef MSB |
1804 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1805 | * setting such that we can't race against cgroup_exit() changing the | |
1806 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1807 | */ |
c84cdf75 | 1808 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1809 | old_cset = task_css_set(tsk); |
74a1166d | 1810 | |
b3dc094e | 1811 | get_css_set(new_cset); |
5abb8855 | 1812 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d | 1813 | |
1b9aba49 TH |
1814 | /* |
1815 | * Use move_tail so that cgroup_taskset_first() still returns the | |
1816 | * leader after migration. This works because cgroup_migrate() | |
1817 | * ensures that the dst_cset of the leader is the first on the | |
1818 | * tset's dst_csets list. | |
1819 | */ | |
1820 | list_move_tail(&tsk->cg_list, &new_cset->mg_tasks); | |
74a1166d BB |
1821 | |
1822 | /* | |
5abb8855 TH |
1823 | * We just gained a reference on old_cset by taking it from the |
1824 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1825 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1826 | */ |
5abb8855 | 1827 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
cb0f1fe9 | 1828 | put_css_set_locked(old_cset, false); |
74a1166d BB |
1829 | } |
1830 | ||
a043e3b2 | 1831 | /** |
1958d2d5 TH |
1832 | * cgroup_migrate_finish - cleanup after attach |
1833 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 1834 | * |
1958d2d5 TH |
1835 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
1836 | * those functions for details. | |
74a1166d | 1837 | */ |
1958d2d5 | 1838 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 1839 | { |
1958d2d5 | 1840 | struct css_set *cset, *tmp_cset; |
74a1166d | 1841 | |
1958d2d5 TH |
1842 | lockdep_assert_held(&cgroup_mutex); |
1843 | ||
1844 | down_write(&css_set_rwsem); | |
1845 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
1846 | cset->mg_src_cgrp = NULL; | |
1847 | cset->mg_dst_cset = NULL; | |
1848 | list_del_init(&cset->mg_preload_node); | |
1849 | put_css_set_locked(cset, false); | |
1850 | } | |
1851 | up_write(&css_set_rwsem); | |
1852 | } | |
1853 | ||
1854 | /** | |
1855 | * cgroup_migrate_add_src - add a migration source css_set | |
1856 | * @src_cset: the source css_set to add | |
1857 | * @dst_cgrp: the destination cgroup | |
1858 | * @preloaded_csets: list of preloaded css_sets | |
1859 | * | |
1860 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
1861 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
1862 | * up by cgroup_migrate_finish(). | |
1863 | * | |
1864 | * This function may be called without holding threadgroup_lock even if the | |
1865 | * target is a process. Threads may be created and destroyed but as long | |
1866 | * as cgroup_mutex is not dropped, no new css_set can be put into play and | |
1867 | * the preloaded css_sets are guaranteed to cover all migrations. | |
1868 | */ | |
1869 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
1870 | struct cgroup *dst_cgrp, | |
1871 | struct list_head *preloaded_csets) | |
1872 | { | |
1873 | struct cgroup *src_cgrp; | |
1874 | ||
1875 | lockdep_assert_held(&cgroup_mutex); | |
1876 | lockdep_assert_held(&css_set_rwsem); | |
1877 | ||
1878 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
1879 | ||
1880 | /* nothing to do if this cset already belongs to the cgroup */ | |
1881 | if (src_cgrp == dst_cgrp) | |
1882 | return; | |
1883 | ||
1884 | if (!list_empty(&src_cset->mg_preload_node)) | |
1885 | return; | |
1886 | ||
1887 | WARN_ON(src_cset->mg_src_cgrp); | |
1888 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
1889 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
1890 | ||
1891 | src_cset->mg_src_cgrp = src_cgrp; | |
1892 | get_css_set(src_cset); | |
1893 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
1894 | } | |
1895 | ||
1896 | /** | |
1897 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
1898 | * @dst_cgrp: the destination cgroup | |
1899 | * @preloaded_csets: list of preloaded source css_sets | |
1900 | * | |
1901 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
1902 | * have been preloaded to @preloaded_csets. This function looks up and | |
1903 | * pins all destination css_sets, links each to its source, and put them on | |
1904 | * @preloaded_csets. | |
1905 | * | |
1906 | * This function must be called after cgroup_migrate_add_src() has been | |
1907 | * called on each migration source css_set. After migration is performed | |
1908 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
1909 | * @preloaded_csets. | |
1910 | */ | |
1911 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
1912 | struct list_head *preloaded_csets) | |
1913 | { | |
1914 | LIST_HEAD(csets); | |
1915 | struct css_set *src_cset; | |
1916 | ||
1917 | lockdep_assert_held(&cgroup_mutex); | |
1918 | ||
1919 | /* look up the dst cset for each src cset and link it to src */ | |
1920 | list_for_each_entry(src_cset, preloaded_csets, mg_preload_node) { | |
1921 | struct css_set *dst_cset; | |
1922 | ||
1923 | dst_cset = find_css_set(src_cset, dst_cgrp); | |
1924 | if (!dst_cset) | |
1925 | goto err; | |
1926 | ||
1927 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
1928 | src_cset->mg_dst_cset = dst_cset; | |
1929 | ||
1930 | if (list_empty(&dst_cset->mg_preload_node)) | |
1931 | list_add(&dst_cset->mg_preload_node, &csets); | |
1932 | else | |
1933 | put_css_set(dst_cset, false); | |
1934 | } | |
1935 | ||
1936 | list_splice(&csets, preloaded_csets); | |
1937 | return 0; | |
1938 | err: | |
1939 | cgroup_migrate_finish(&csets); | |
1940 | return -ENOMEM; | |
1941 | } | |
1942 | ||
1943 | /** | |
1944 | * cgroup_migrate - migrate a process or task to a cgroup | |
1945 | * @cgrp: the destination cgroup | |
1946 | * @leader: the leader of the process or the task to migrate | |
1947 | * @threadgroup: whether @leader points to the whole process or a single task | |
1948 | * | |
1949 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
1950 | * process, the caller must be holding threadgroup_lock of @leader. The | |
1951 | * caller is also responsible for invoking cgroup_migrate_add_src() and | |
1952 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
1953 | * function and following up with cgroup_migrate_finish(). | |
1954 | * | |
1955 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
1956 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
1957 | * failure, when migrating multiple targets, the success or failure can be | |
1958 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
1959 | * actually starting migrating. | |
1960 | */ | |
1961 | static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader, | |
1962 | bool threadgroup) | |
74a1166d | 1963 | { |
b3dc094e TH |
1964 | struct cgroup_taskset tset = { |
1965 | .src_csets = LIST_HEAD_INIT(tset.src_csets), | |
1966 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), | |
1967 | .csets = &tset.src_csets, | |
1968 | }; | |
1c6727af | 1969 | struct cgroup_subsys_state *css, *failed_css = NULL; |
b3dc094e TH |
1970 | struct css_set *cset, *tmp_cset; |
1971 | struct task_struct *task, *tmp_task; | |
1972 | int i, ret; | |
74a1166d | 1973 | |
fb5d2b4c MSB |
1974 | /* |
1975 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1976 | * already PF_EXITING could be freed from underneath us unless we | |
1977 | * take an rcu_read_lock. | |
1978 | */ | |
b3dc094e | 1979 | down_write(&css_set_rwsem); |
fb5d2b4c | 1980 | rcu_read_lock(); |
9db8de37 | 1981 | task = leader; |
74a1166d | 1982 | do { |
9db8de37 TH |
1983 | /* @task either already exited or can't exit until the end */ |
1984 | if (task->flags & PF_EXITING) | |
ea84753c | 1985 | goto next; |
134d3373 | 1986 | |
eaf797ab TH |
1987 | /* leave @task alone if post_fork() hasn't linked it yet */ |
1988 | if (list_empty(&task->cg_list)) | |
ea84753c | 1989 | goto next; |
cd3d0952 | 1990 | |
b3dc094e | 1991 | cset = task_css_set(task); |
1958d2d5 | 1992 | if (!cset->mg_src_cgrp) |
ea84753c | 1993 | goto next; |
b3dc094e | 1994 | |
61d1d219 | 1995 | /* |
1b9aba49 TH |
1996 | * cgroup_taskset_first() must always return the leader. |
1997 | * Take care to avoid disturbing the ordering. | |
61d1d219 | 1998 | */ |
1b9aba49 TH |
1999 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2000 | if (list_empty(&cset->mg_node)) | |
2001 | list_add_tail(&cset->mg_node, &tset.src_csets); | |
2002 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2003 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2004 | &tset.dst_csets); | |
ea84753c | 2005 | next: |
081aa458 LZ |
2006 | if (!threadgroup) |
2007 | break; | |
9db8de37 | 2008 | } while_each_thread(leader, task); |
fb5d2b4c | 2009 | rcu_read_unlock(); |
b3dc094e | 2010 | up_write(&css_set_rwsem); |
74a1166d | 2011 | |
134d3373 | 2012 | /* methods shouldn't be called if no task is actually migrating */ |
b3dc094e TH |
2013 | if (list_empty(&tset.src_csets)) |
2014 | return 0; | |
134d3373 | 2015 | |
1958d2d5 | 2016 | /* check that we can legitimately attach to the cgroup */ |
aec3dfcb | 2017 | for_each_e_css(css, i, cgrp) { |
1c6727af | 2018 | if (css->ss->can_attach) { |
9db8de37 TH |
2019 | ret = css->ss->can_attach(css, &tset); |
2020 | if (ret) { | |
1c6727af | 2021 | failed_css = css; |
74a1166d BB |
2022 | goto out_cancel_attach; |
2023 | } | |
2024 | } | |
74a1166d BB |
2025 | } |
2026 | ||
2027 | /* | |
1958d2d5 TH |
2028 | * Now that we're guaranteed success, proceed to move all tasks to |
2029 | * the new cgroup. There are no failure cases after here, so this | |
2030 | * is the commit point. | |
74a1166d | 2031 | */ |
cb0f1fe9 | 2032 | down_write(&css_set_rwsem); |
b3dc094e TH |
2033 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
2034 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
2035 | cgroup_task_migrate(cset->mg_src_cgrp, task, | |
2036 | cset->mg_dst_cset); | |
74a1166d | 2037 | } |
cb0f1fe9 | 2038 | up_write(&css_set_rwsem); |
74a1166d BB |
2039 | |
2040 | /* | |
1958d2d5 TH |
2041 | * Migration is committed, all target tasks are now on dst_csets. |
2042 | * Nothing is sensitive to fork() after this point. Notify | |
2043 | * controllers that migration is complete. | |
74a1166d | 2044 | */ |
1958d2d5 | 2045 | tset.csets = &tset.dst_csets; |
74a1166d | 2046 | |
aec3dfcb | 2047 | for_each_e_css(css, i, cgrp) |
1c6727af TH |
2048 | if (css->ss->attach) |
2049 | css->ss->attach(css, &tset); | |
74a1166d | 2050 | |
9db8de37 | 2051 | ret = 0; |
b3dc094e TH |
2052 | goto out_release_tset; |
2053 | ||
74a1166d | 2054 | out_cancel_attach: |
aec3dfcb | 2055 | for_each_e_css(css, i, cgrp) { |
b3dc094e TH |
2056 | if (css == failed_css) |
2057 | break; | |
2058 | if (css->ss->cancel_attach) | |
2059 | css->ss->cancel_attach(css, &tset); | |
74a1166d | 2060 | } |
b3dc094e TH |
2061 | out_release_tset: |
2062 | down_write(&css_set_rwsem); | |
2063 | list_splice_init(&tset.dst_csets, &tset.src_csets); | |
2064 | list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { | |
1b9aba49 | 2065 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); |
b3dc094e | 2066 | list_del_init(&cset->mg_node); |
b3dc094e TH |
2067 | } |
2068 | up_write(&css_set_rwsem); | |
9db8de37 | 2069 | return ret; |
74a1166d BB |
2070 | } |
2071 | ||
1958d2d5 TH |
2072 | /** |
2073 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2074 | * @dst_cgrp: the cgroup to attach to | |
2075 | * @leader: the task or the leader of the threadgroup to be attached | |
2076 | * @threadgroup: attach the whole threadgroup? | |
2077 | * | |
0e1d768f | 2078 | * Call holding cgroup_mutex and threadgroup_lock of @leader. |
1958d2d5 TH |
2079 | */ |
2080 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2081 | struct task_struct *leader, bool threadgroup) | |
2082 | { | |
2083 | LIST_HEAD(preloaded_csets); | |
2084 | struct task_struct *task; | |
2085 | int ret; | |
2086 | ||
2087 | /* look up all src csets */ | |
2088 | down_read(&css_set_rwsem); | |
2089 | rcu_read_lock(); | |
2090 | task = leader; | |
2091 | do { | |
2092 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2093 | &preloaded_csets); | |
2094 | if (!threadgroup) | |
2095 | break; | |
2096 | } while_each_thread(leader, task); | |
2097 | rcu_read_unlock(); | |
2098 | up_read(&css_set_rwsem); | |
2099 | ||
2100 | /* prepare dst csets and commit */ | |
2101 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2102 | if (!ret) | |
2103 | ret = cgroup_migrate(dst_cgrp, leader, threadgroup); | |
2104 | ||
2105 | cgroup_migrate_finish(&preloaded_csets); | |
2106 | return ret; | |
74a1166d BB |
2107 | } |
2108 | ||
2109 | /* | |
2110 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2111 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2112 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2113 | */ |
74a1166d | 2114 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2115 | { |
bbcb81d0 | 2116 | struct task_struct *tsk; |
c69e8d9c | 2117 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2118 | int ret; |
2119 | ||
74a1166d BB |
2120 | if (!cgroup_lock_live_group(cgrp)) |
2121 | return -ENODEV; | |
2122 | ||
b78949eb MSB |
2123 | retry_find_task: |
2124 | rcu_read_lock(); | |
bbcb81d0 | 2125 | if (pid) { |
73507f33 | 2126 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2127 | if (!tsk) { |
2128 | rcu_read_unlock(); | |
dd4b0a46 | 2129 | ret = -ESRCH; |
b78949eb | 2130 | goto out_unlock_cgroup; |
bbcb81d0 | 2131 | } |
74a1166d BB |
2132 | /* |
2133 | * even if we're attaching all tasks in the thread group, we | |
2134 | * only need to check permissions on one of them. | |
2135 | */ | |
c69e8d9c | 2136 | tcred = __task_cred(tsk); |
14a590c3 EB |
2137 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2138 | !uid_eq(cred->euid, tcred->uid) && | |
2139 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2140 | rcu_read_unlock(); |
b78949eb MSB |
2141 | ret = -EACCES; |
2142 | goto out_unlock_cgroup; | |
bbcb81d0 | 2143 | } |
b78949eb MSB |
2144 | } else |
2145 | tsk = current; | |
cd3d0952 TH |
2146 | |
2147 | if (threadgroup) | |
b78949eb | 2148 | tsk = tsk->group_leader; |
c4c27fbd MG |
2149 | |
2150 | /* | |
14a40ffc | 2151 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2152 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2153 | * with no rt_runtime allocated. Just say no. | |
2154 | */ | |
14a40ffc | 2155 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2156 | ret = -EINVAL; |
2157 | rcu_read_unlock(); | |
2158 | goto out_unlock_cgroup; | |
2159 | } | |
2160 | ||
b78949eb MSB |
2161 | get_task_struct(tsk); |
2162 | rcu_read_unlock(); | |
2163 | ||
2164 | threadgroup_lock(tsk); | |
2165 | if (threadgroup) { | |
2166 | if (!thread_group_leader(tsk)) { | |
2167 | /* | |
2168 | * a race with de_thread from another thread's exec() | |
2169 | * may strip us of our leadership, if this happens, | |
2170 | * there is no choice but to throw this task away and | |
2171 | * try again; this is | |
2172 | * "double-double-toil-and-trouble-check locking". | |
2173 | */ | |
2174 | threadgroup_unlock(tsk); | |
2175 | put_task_struct(tsk); | |
2176 | goto retry_find_task; | |
2177 | } | |
081aa458 LZ |
2178 | } |
2179 | ||
2180 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2181 | ||
cd3d0952 TH |
2182 | threadgroup_unlock(tsk); |
2183 | ||
bbcb81d0 | 2184 | put_task_struct(tsk); |
b78949eb | 2185 | out_unlock_cgroup: |
47cfcd09 | 2186 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
2187 | return ret; |
2188 | } | |
2189 | ||
7ae1bad9 TH |
2190 | /** |
2191 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2192 | * @from: attach to all cgroups of a given task | |
2193 | * @tsk: the task to be attached | |
2194 | */ | |
2195 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2196 | { | |
3dd06ffa | 2197 | struct cgroup_root *root; |
7ae1bad9 TH |
2198 | int retval = 0; |
2199 | ||
47cfcd09 | 2200 | mutex_lock(&cgroup_mutex); |
985ed670 | 2201 | for_each_root(root) { |
96d365e0 TH |
2202 | struct cgroup *from_cgrp; |
2203 | ||
3dd06ffa | 2204 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2205 | continue; |
2206 | ||
96d365e0 TH |
2207 | down_read(&css_set_rwsem); |
2208 | from_cgrp = task_cgroup_from_root(from, root); | |
2209 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2210 | |
6f4b7e63 | 2211 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2212 | if (retval) |
2213 | break; | |
2214 | } | |
47cfcd09 | 2215 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2216 | |
2217 | return retval; | |
2218 | } | |
2219 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2220 | ||
182446d0 TH |
2221 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2222 | struct cftype *cft, u64 pid) | |
74a1166d | 2223 | { |
182446d0 | 2224 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2225 | } |
2226 | ||
182446d0 TH |
2227 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2228 | struct cftype *cft, u64 tgid) | |
af351026 | 2229 | { |
182446d0 | 2230 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2231 | } |
2232 | ||
182446d0 | 2233 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
4d3bb511 | 2234 | struct cftype *cft, char *buffer) |
e788e066 | 2235 | { |
3dd06ffa | 2236 | struct cgroup_root *root = css->cgroup->root; |
5f469907 TH |
2237 | |
2238 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
182446d0 | 2239 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2240 | return -ENODEV; |
69e943b7 | 2241 | spin_lock(&release_agent_path_lock); |
5f469907 TH |
2242 | strlcpy(root->release_agent_path, buffer, |
2243 | sizeof(root->release_agent_path)); | |
69e943b7 | 2244 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2245 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2246 | return 0; |
2247 | } | |
2248 | ||
2da8ca82 | 2249 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2250 | { |
2da8ca82 | 2251 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2252 | |
e788e066 PM |
2253 | if (!cgroup_lock_live_group(cgrp)) |
2254 | return -ENODEV; | |
2255 | seq_puts(seq, cgrp->root->release_agent_path); | |
2256 | seq_putc(seq, '\n'); | |
47cfcd09 | 2257 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2258 | return 0; |
2259 | } | |
2260 | ||
2da8ca82 | 2261 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2262 | { |
2da8ca82 TH |
2263 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2264 | ||
2265 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2266 | return 0; |
2267 | } | |
2268 | ||
2bd59d48 TH |
2269 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2270 | size_t nbytes, loff_t off) | |
355e0c48 | 2271 | { |
2bd59d48 TH |
2272 | struct cgroup *cgrp = of->kn->parent->priv; |
2273 | struct cftype *cft = of->kn->priv; | |
2274 | struct cgroup_subsys_state *css; | |
a742c59d | 2275 | int ret; |
355e0c48 | 2276 | |
2bd59d48 TH |
2277 | /* |
2278 | * kernfs guarantees that a file isn't deleted with operations in | |
2279 | * flight, which means that the matching css is and stays alive and | |
2280 | * doesn't need to be pinned. The RCU locking is not necessary | |
2281 | * either. It's just for the convenience of using cgroup_css(). | |
2282 | */ | |
2283 | rcu_read_lock(); | |
2284 | css = cgroup_css(cgrp, cft->ss); | |
2285 | rcu_read_unlock(); | |
a742c59d TH |
2286 | |
2287 | if (cft->write_string) { | |
2288 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2289 | } else if (cft->write_u64) { | |
2290 | unsigned long long v; | |
2291 | ret = kstrtoull(buf, 0, &v); | |
2292 | if (!ret) | |
2293 | ret = cft->write_u64(css, cft, v); | |
2294 | } else if (cft->write_s64) { | |
2295 | long long v; | |
2296 | ret = kstrtoll(buf, 0, &v); | |
2297 | if (!ret) | |
2298 | ret = cft->write_s64(css, cft, v); | |
2299 | } else if (cft->trigger) { | |
2300 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2301 | } else { |
a742c59d | 2302 | ret = -EINVAL; |
e73d2c61 | 2303 | } |
2bd59d48 | 2304 | |
a742c59d | 2305 | return ret ?: nbytes; |
355e0c48 PM |
2306 | } |
2307 | ||
6612f05b | 2308 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2309 | { |
2bd59d48 | 2310 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2311 | } |
2312 | ||
6612f05b | 2313 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2314 | { |
2bd59d48 | 2315 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2316 | } |
2317 | ||
6612f05b | 2318 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2319 | { |
2bd59d48 | 2320 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2321 | } |
2322 | ||
91796569 | 2323 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2324 | { |
7da11279 TH |
2325 | struct cftype *cft = seq_cft(m); |
2326 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2327 | |
2da8ca82 TH |
2328 | if (cft->seq_show) |
2329 | return cft->seq_show(m, arg); | |
e73d2c61 | 2330 | |
f4c753b7 | 2331 | if (cft->read_u64) |
896f5199 TH |
2332 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2333 | else if (cft->read_s64) | |
2334 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2335 | else | |
2336 | return -EINVAL; | |
2337 | return 0; | |
91796569 PM |
2338 | } |
2339 | ||
2bd59d48 TH |
2340 | static struct kernfs_ops cgroup_kf_single_ops = { |
2341 | .atomic_write_len = PAGE_SIZE, | |
2342 | .write = cgroup_file_write, | |
2343 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2344 | }; |
2345 | ||
2bd59d48 TH |
2346 | static struct kernfs_ops cgroup_kf_ops = { |
2347 | .atomic_write_len = PAGE_SIZE, | |
2348 | .write = cgroup_file_write, | |
2349 | .seq_start = cgroup_seqfile_start, | |
2350 | .seq_next = cgroup_seqfile_next, | |
2351 | .seq_stop = cgroup_seqfile_stop, | |
2352 | .seq_show = cgroup_seqfile_show, | |
2353 | }; | |
ddbcc7e8 PM |
2354 | |
2355 | /* | |
2356 | * cgroup_rename - Only allow simple rename of directories in place. | |
2357 | */ | |
2bd59d48 TH |
2358 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2359 | const char *new_name_str) | |
ddbcc7e8 | 2360 | { |
2bd59d48 | 2361 | struct cgroup *cgrp = kn->priv; |
65dff759 | 2362 | int ret; |
65dff759 | 2363 | |
2bd59d48 | 2364 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2365 | return -ENOTDIR; |
2bd59d48 | 2366 | if (kn->parent != new_parent) |
ddbcc7e8 | 2367 | return -EIO; |
65dff759 | 2368 | |
6db8e85c TH |
2369 | /* |
2370 | * This isn't a proper migration and its usefulness is very | |
2371 | * limited. Disallow if sane_behavior. | |
2372 | */ | |
2373 | if (cgroup_sane_behavior(cgrp)) | |
2374 | return -EPERM; | |
099fca32 | 2375 | |
e1b2dc17 TH |
2376 | /* |
2377 | * We're gonna grab cgroup_tree_mutex which nests outside kernfs | |
2378 | * active_ref. kernfs_rename() doesn't require active_ref | |
2379 | * protection. Break them before grabbing cgroup_tree_mutex. | |
2380 | */ | |
2381 | kernfs_break_active_protection(new_parent); | |
2382 | kernfs_break_active_protection(kn); | |
099fca32 | 2383 | |
2bd59d48 TH |
2384 | mutex_lock(&cgroup_tree_mutex); |
2385 | mutex_lock(&cgroup_mutex); | |
099fca32 | 2386 | |
2bd59d48 | 2387 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 2388 | |
2bd59d48 TH |
2389 | mutex_unlock(&cgroup_mutex); |
2390 | mutex_unlock(&cgroup_tree_mutex); | |
e1b2dc17 TH |
2391 | |
2392 | kernfs_unbreak_active_protection(kn); | |
2393 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 2394 | return ret; |
099fca32 LZ |
2395 | } |
2396 | ||
49957f8e TH |
2397 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
2398 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
2399 | { | |
2400 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
2401 | .ia_uid = current_fsuid(), | |
2402 | .ia_gid = current_fsgid(), }; | |
2403 | ||
2404 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
2405 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
2406 | return 0; | |
2407 | ||
2408 | return kernfs_setattr(kn, &iattr); | |
2409 | } | |
2410 | ||
2bb566cb | 2411 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2412 | { |
8d7e6fb0 | 2413 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2414 | struct kernfs_node *kn; |
2415 | struct lock_class_key *key = NULL; | |
49957f8e | 2416 | int ret; |
05ef1d7c | 2417 | |
2bd59d48 TH |
2418 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2419 | key = &cft->lockdep_key; | |
2420 | #endif | |
2421 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2422 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2423 | NULL, false, key); | |
49957f8e TH |
2424 | if (IS_ERR(kn)) |
2425 | return PTR_ERR(kn); | |
2426 | ||
2427 | ret = cgroup_kn_set_ugid(kn); | |
2428 | if (ret) | |
2429 | kernfs_remove(kn); | |
2430 | return ret; | |
ddbcc7e8 PM |
2431 | } |
2432 | ||
b1f28d31 TH |
2433 | /** |
2434 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2435 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2436 | * @cfts: array of cftypes to be added |
2437 | * @is_add: whether to add or remove | |
2438 | * | |
2439 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2440 | * For removals, this function never fails. If addition fails, this |
2441 | * function doesn't remove files already added. The caller is responsible | |
2442 | * for cleaning up. | |
b1f28d31 | 2443 | */ |
2bb566cb TH |
2444 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2445 | bool is_add) | |
ddbcc7e8 | 2446 | { |
03b1cde6 | 2447 | struct cftype *cft; |
b1f28d31 TH |
2448 | int ret; |
2449 | ||
ace2bee8 | 2450 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2451 | |
2452 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2453 | /* does cft->flags tell us to skip this file on @cgrp? */ |
8cbbf2c9 TH |
2454 | if ((cft->flags & CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
2455 | continue; | |
873fe09e TH |
2456 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2457 | continue; | |
f33fddc2 G |
2458 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2459 | continue; | |
2460 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2461 | continue; | |
2462 | ||
2739d3cc | 2463 | if (is_add) { |
2bb566cb | 2464 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2465 | if (ret) { |
2739d3cc | 2466 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2467 | cft->name, ret); |
2468 | return ret; | |
2469 | } | |
2739d3cc LZ |
2470 | } else { |
2471 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2472 | } |
ddbcc7e8 | 2473 | } |
b1f28d31 | 2474 | return 0; |
ddbcc7e8 PM |
2475 | } |
2476 | ||
21a2d343 | 2477 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2478 | { |
2479 | LIST_HEAD(pending); | |
2bb566cb | 2480 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 2481 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 2482 | struct cgroup_subsys_state *css; |
9ccece80 | 2483 | int ret = 0; |
8e3f6541 | 2484 | |
21a2d343 | 2485 | lockdep_assert_held(&cgroup_tree_mutex); |
8e3f6541 | 2486 | |
e8c82d20 | 2487 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2488 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2489 | struct cgroup *cgrp = css->cgroup; |
2490 | ||
e8c82d20 LZ |
2491 | if (cgroup_is_dead(cgrp)) |
2492 | continue; | |
2493 | ||
21a2d343 | 2494 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2495 | if (ret) |
2496 | break; | |
8e3f6541 | 2497 | } |
21a2d343 TH |
2498 | |
2499 | if (is_add && !ret) | |
2500 | kernfs_activate(root->kn); | |
9ccece80 | 2501 | return ret; |
8e3f6541 TH |
2502 | } |
2503 | ||
2da440a2 | 2504 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 2505 | { |
2bb566cb | 2506 | struct cftype *cft; |
8e3f6541 | 2507 | |
2bd59d48 TH |
2508 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2509 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2510 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2511 | kfree(cft->kf_ops); | |
2512 | cft->kf_ops = NULL; | |
2da440a2 | 2513 | cft->ss = NULL; |
2bd59d48 | 2514 | } |
2da440a2 TH |
2515 | } |
2516 | ||
2bd59d48 | 2517 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2518 | { |
2519 | struct cftype *cft; | |
2520 | ||
2bd59d48 TH |
2521 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2522 | struct kernfs_ops *kf_ops; | |
2523 | ||
0adb0704 TH |
2524 | WARN_ON(cft->ss || cft->kf_ops); |
2525 | ||
2bd59d48 TH |
2526 | if (cft->seq_start) |
2527 | kf_ops = &cgroup_kf_ops; | |
2528 | else | |
2529 | kf_ops = &cgroup_kf_single_ops; | |
2530 | ||
2531 | /* | |
2532 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2533 | * make a copy of kf_ops to set its atomic_write_len. | |
2534 | */ | |
2535 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2536 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2537 | if (!kf_ops) { | |
2538 | cgroup_exit_cftypes(cfts); | |
2539 | return -ENOMEM; | |
2540 | } | |
2541 | kf_ops->atomic_write_len = cft->max_write_len; | |
2542 | } | |
8e3f6541 | 2543 | |
2bd59d48 | 2544 | cft->kf_ops = kf_ops; |
2bb566cb | 2545 | cft->ss = ss; |
2bd59d48 | 2546 | } |
2bb566cb | 2547 | |
2bd59d48 | 2548 | return 0; |
2da440a2 TH |
2549 | } |
2550 | ||
21a2d343 TH |
2551 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
2552 | { | |
2553 | lockdep_assert_held(&cgroup_tree_mutex); | |
2554 | ||
2555 | if (!cfts || !cfts[0].ss) | |
2556 | return -ENOENT; | |
2557 | ||
2558 | list_del(&cfts->node); | |
2559 | cgroup_apply_cftypes(cfts, false); | |
2560 | cgroup_exit_cftypes(cfts); | |
2561 | return 0; | |
8e3f6541 | 2562 | } |
8e3f6541 | 2563 | |
79578621 TH |
2564 | /** |
2565 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
2566 | * @cfts: zero-length name terminated array of cftypes |
2567 | * | |
2bb566cb TH |
2568 | * Unregister @cfts. Files described by @cfts are removed from all |
2569 | * existing cgroups and all future cgroups won't have them either. This | |
2570 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
2571 | * |
2572 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 2573 | * registered. |
79578621 | 2574 | */ |
2bb566cb | 2575 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 2576 | { |
21a2d343 | 2577 | int ret; |
79578621 | 2578 | |
21a2d343 TH |
2579 | mutex_lock(&cgroup_tree_mutex); |
2580 | ret = cgroup_rm_cftypes_locked(cfts); | |
2581 | mutex_unlock(&cgroup_tree_mutex); | |
2582 | return ret; | |
80b13586 TH |
2583 | } |
2584 | ||
8e3f6541 TH |
2585 | /** |
2586 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2587 | * @ss: target cgroup subsystem | |
2588 | * @cfts: zero-length name terminated array of cftypes | |
2589 | * | |
2590 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2591 | * existing cgroups to which @ss is attached and all future cgroups will | |
2592 | * have them too. This function can be called anytime whether @ss is | |
2593 | * attached or not. | |
2594 | * | |
2595 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2596 | * function currently returns 0 as long as @cfts registration is successful | |
2597 | * even if some file creation attempts on existing cgroups fail. | |
2598 | */ | |
03b1cde6 | 2599 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 2600 | { |
9ccece80 | 2601 | int ret; |
8e3f6541 | 2602 | |
dc5736ed LZ |
2603 | if (!cfts || cfts[0].name[0] == '\0') |
2604 | return 0; | |
2bb566cb | 2605 | |
2bd59d48 TH |
2606 | ret = cgroup_init_cftypes(ss, cfts); |
2607 | if (ret) | |
2608 | return ret; | |
79578621 | 2609 | |
21a2d343 TH |
2610 | mutex_lock(&cgroup_tree_mutex); |
2611 | ||
0adb0704 | 2612 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 2613 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 2614 | if (ret) |
21a2d343 | 2615 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 2616 | |
21a2d343 | 2617 | mutex_unlock(&cgroup_tree_mutex); |
9ccece80 | 2618 | return ret; |
79578621 TH |
2619 | } |
2620 | ||
a043e3b2 LZ |
2621 | /** |
2622 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2623 | * @cgrp: the cgroup in question | |
2624 | * | |
2625 | * Return the number of tasks in the cgroup. | |
2626 | */ | |
07bc356e | 2627 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2628 | { |
2629 | int count = 0; | |
69d0206c | 2630 | struct cgrp_cset_link *link; |
817929ec | 2631 | |
96d365e0 | 2632 | down_read(&css_set_rwsem); |
69d0206c TH |
2633 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2634 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 2635 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
2636 | return count; |
2637 | } | |
2638 | ||
53fa5261 | 2639 | /** |
492eb21b TH |
2640 | * css_next_child - find the next child of a given css |
2641 | * @pos_css: the current position (%NULL to initiate traversal) | |
2642 | * @parent_css: css whose children to walk | |
53fa5261 | 2643 | * |
492eb21b | 2644 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
2645 | * under either cgroup_mutex or RCU read lock. The only requirement is |
2646 | * that @parent_css and @pos_css are accessible. The next sibling is | |
2647 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 2648 | */ |
492eb21b TH |
2649 | struct cgroup_subsys_state * |
2650 | css_next_child(struct cgroup_subsys_state *pos_css, | |
2651 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 2652 | { |
492eb21b TH |
2653 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
2654 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
2655 | struct cgroup *next; |
2656 | ||
ace2bee8 | 2657 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
2658 | |
2659 | /* | |
2660 | * @pos could already have been removed. Once a cgroup is removed, | |
2661 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
2662 | * changes. As CGRP_DEAD assertion is serialized and happens |
2663 | * before the cgroup is taken off the ->sibling list, if we see it | |
2664 | * unasserted, it's guaranteed that the next sibling hasn't | |
2665 | * finished its grace period even if it's already removed, and thus | |
2666 | * safe to dereference from this RCU critical section. If | |
2667 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
2668 | * to be visible as %true here. | |
3b287a50 TH |
2669 | * |
2670 | * If @pos is dead, its next pointer can't be dereferenced; | |
2671 | * however, as each cgroup is given a monotonically increasing | |
2672 | * unique serial number and always appended to the sibling list, | |
2673 | * the next one can be found by walking the parent's children until | |
2674 | * we see a cgroup with higher serial number than @pos's. While | |
2675 | * this path can be slower, it's taken only when either the current | |
2676 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 2677 | */ |
3b287a50 TH |
2678 | if (!pos) { |
2679 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
2680 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 2681 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
2682 | } else { |
2683 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
2684 | if (next->serial_nr > pos->serial_nr) | |
2685 | break; | |
53fa5261 TH |
2686 | } |
2687 | ||
492eb21b TH |
2688 | if (&next->sibling == &cgrp->children) |
2689 | return NULL; | |
2690 | ||
ca8bdcaf | 2691 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 2692 | } |
53fa5261 | 2693 | |
574bd9f7 | 2694 | /** |
492eb21b | 2695 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 2696 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2697 | * @root: css whose descendants to walk |
574bd9f7 | 2698 | * |
492eb21b | 2699 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
2700 | * to visit for pre-order traversal of @root's descendants. @root is |
2701 | * included in the iteration and the first node to be visited. | |
75501a6d | 2702 | * |
87fb54f1 TH |
2703 | * While this function requires cgroup_mutex or RCU read locking, it |
2704 | * doesn't require the whole traversal to be contained in a single critical | |
2705 | * section. This function will return the correct next descendant as long | |
2706 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 2707 | */ |
492eb21b TH |
2708 | struct cgroup_subsys_state * |
2709 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
2710 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2711 | { |
492eb21b | 2712 | struct cgroup_subsys_state *next; |
574bd9f7 | 2713 | |
ace2bee8 | 2714 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2715 | |
bd8815a6 | 2716 | /* if first iteration, visit @root */ |
7805d000 | 2717 | if (!pos) |
bd8815a6 | 2718 | return root; |
574bd9f7 TH |
2719 | |
2720 | /* visit the first child if exists */ | |
492eb21b | 2721 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
2722 | if (next) |
2723 | return next; | |
2724 | ||
2725 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
2726 | while (pos != root) { |
2727 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 2728 | if (next) |
574bd9f7 | 2729 | return next; |
492eb21b | 2730 | pos = css_parent(pos); |
7805d000 | 2731 | } |
574bd9f7 TH |
2732 | |
2733 | return NULL; | |
2734 | } | |
574bd9f7 | 2735 | |
12a9d2fe | 2736 | /** |
492eb21b TH |
2737 | * css_rightmost_descendant - return the rightmost descendant of a css |
2738 | * @pos: css of interest | |
12a9d2fe | 2739 | * |
492eb21b TH |
2740 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
2741 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 2742 | * subtree of @pos. |
75501a6d | 2743 | * |
87fb54f1 TH |
2744 | * While this function requires cgroup_mutex or RCU read locking, it |
2745 | * doesn't require the whole traversal to be contained in a single critical | |
2746 | * section. This function will return the correct rightmost descendant as | |
2747 | * long as @pos is accessible. | |
12a9d2fe | 2748 | */ |
492eb21b TH |
2749 | struct cgroup_subsys_state * |
2750 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 2751 | { |
492eb21b | 2752 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 2753 | |
ace2bee8 | 2754 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
2755 | |
2756 | do { | |
2757 | last = pos; | |
2758 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
2759 | pos = NULL; | |
492eb21b | 2760 | css_for_each_child(tmp, last) |
12a9d2fe TH |
2761 | pos = tmp; |
2762 | } while (pos); | |
2763 | ||
2764 | return last; | |
2765 | } | |
12a9d2fe | 2766 | |
492eb21b TH |
2767 | static struct cgroup_subsys_state * |
2768 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 2769 | { |
492eb21b | 2770 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
2771 | |
2772 | do { | |
2773 | last = pos; | |
492eb21b | 2774 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
2775 | } while (pos); |
2776 | ||
2777 | return last; | |
2778 | } | |
2779 | ||
2780 | /** | |
492eb21b | 2781 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 2782 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2783 | * @root: css whose descendants to walk |
574bd9f7 | 2784 | * |
492eb21b | 2785 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
2786 | * to visit for post-order traversal of @root's descendants. @root is |
2787 | * included in the iteration and the last node to be visited. | |
75501a6d | 2788 | * |
87fb54f1 TH |
2789 | * While this function requires cgroup_mutex or RCU read locking, it |
2790 | * doesn't require the whole traversal to be contained in a single critical | |
2791 | * section. This function will return the correct next descendant as long | |
2792 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
2793 | * @cgroup. | |
574bd9f7 | 2794 | */ |
492eb21b TH |
2795 | struct cgroup_subsys_state * |
2796 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
2797 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2798 | { |
492eb21b | 2799 | struct cgroup_subsys_state *next; |
574bd9f7 | 2800 | |
ace2bee8 | 2801 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2802 | |
58b79a91 TH |
2803 | /* if first iteration, visit leftmost descendant which may be @root */ |
2804 | if (!pos) | |
2805 | return css_leftmost_descendant(root); | |
574bd9f7 | 2806 | |
bd8815a6 TH |
2807 | /* if we visited @root, we're done */ |
2808 | if (pos == root) | |
2809 | return NULL; | |
2810 | ||
574bd9f7 | 2811 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 2812 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 2813 | if (next) |
492eb21b | 2814 | return css_leftmost_descendant(next); |
574bd9f7 TH |
2815 | |
2816 | /* no sibling left, visit parent */ | |
bd8815a6 | 2817 | return css_parent(pos); |
574bd9f7 | 2818 | } |
574bd9f7 | 2819 | |
0942eeee | 2820 | /** |
72ec7029 | 2821 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
2822 | * @it: the iterator to advance |
2823 | * | |
2824 | * Advance @it to the next css_set to walk. | |
d515876e | 2825 | */ |
72ec7029 | 2826 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
2827 | { |
2828 | struct list_head *l = it->cset_link; | |
2829 | struct cgrp_cset_link *link; | |
2830 | struct css_set *cset; | |
2831 | ||
2832 | /* Advance to the next non-empty css_set */ | |
2833 | do { | |
2834 | l = l->next; | |
72ec7029 | 2835 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
2836 | it->cset_link = NULL; |
2837 | return; | |
2838 | } | |
2839 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
2840 | cset = link->cset; | |
c7561128 TH |
2841 | } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); |
2842 | ||
d515876e | 2843 | it->cset_link = l; |
c7561128 TH |
2844 | |
2845 | if (!list_empty(&cset->tasks)) | |
2846 | it->task = cset->tasks.next; | |
2847 | else | |
2848 | it->task = cset->mg_tasks.next; | |
d515876e TH |
2849 | } |
2850 | ||
0942eeee | 2851 | /** |
72ec7029 TH |
2852 | * css_task_iter_start - initiate task iteration |
2853 | * @css: the css to walk tasks of | |
0942eeee TH |
2854 | * @it: the task iterator to use |
2855 | * | |
72ec7029 TH |
2856 | * Initiate iteration through the tasks of @css. The caller can call |
2857 | * css_task_iter_next() to walk through the tasks until the function | |
2858 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
2859 | * called. | |
0942eeee TH |
2860 | * |
2861 | * Note that this function acquires a lock which is released when the | |
2862 | * iteration finishes. The caller can't sleep while iteration is in | |
2863 | * progress. | |
2864 | */ | |
72ec7029 TH |
2865 | void css_task_iter_start(struct cgroup_subsys_state *css, |
2866 | struct css_task_iter *it) | |
96d365e0 | 2867 | __acquires(css_set_rwsem) |
817929ec | 2868 | { |
56fde9e0 TH |
2869 | /* no one should try to iterate before mounting cgroups */ |
2870 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 2871 | |
96d365e0 | 2872 | down_read(&css_set_rwsem); |
c59cd3d8 | 2873 | |
72ec7029 TH |
2874 | it->origin_css = css; |
2875 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 2876 | |
72ec7029 | 2877 | css_advance_task_iter(it); |
817929ec PM |
2878 | } |
2879 | ||
0942eeee | 2880 | /** |
72ec7029 | 2881 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
2882 | * @it: the task iterator being iterated |
2883 | * | |
2884 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
2885 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
2886 | * reaches the end. | |
0942eeee | 2887 | */ |
72ec7029 | 2888 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
2889 | { |
2890 | struct task_struct *res; | |
2891 | struct list_head *l = it->task; | |
c7561128 TH |
2892 | struct cgrp_cset_link *link = list_entry(it->cset_link, |
2893 | struct cgrp_cset_link, cset_link); | |
817929ec PM |
2894 | |
2895 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 2896 | if (!it->cset_link) |
817929ec PM |
2897 | return NULL; |
2898 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
2899 | |
2900 | /* | |
2901 | * Advance iterator to find next entry. cset->tasks is consumed | |
2902 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
2903 | * next cset. | |
2904 | */ | |
817929ec | 2905 | l = l->next; |
c7561128 TH |
2906 | |
2907 | if (l == &link->cset->tasks) | |
2908 | l = link->cset->mg_tasks.next; | |
2909 | ||
2910 | if (l == &link->cset->mg_tasks) | |
72ec7029 | 2911 | css_advance_task_iter(it); |
c7561128 | 2912 | else |
817929ec | 2913 | it->task = l; |
c7561128 | 2914 | |
817929ec PM |
2915 | return res; |
2916 | } | |
2917 | ||
0942eeee | 2918 | /** |
72ec7029 | 2919 | * css_task_iter_end - finish task iteration |
0942eeee TH |
2920 | * @it: the task iterator to finish |
2921 | * | |
72ec7029 | 2922 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 2923 | */ |
72ec7029 | 2924 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 2925 | __releases(css_set_rwsem) |
31a7df01 | 2926 | { |
96d365e0 | 2927 | up_read(&css_set_rwsem); |
31a7df01 CW |
2928 | } |
2929 | ||
2930 | /** | |
8cc99345 TH |
2931 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
2932 | * @to: cgroup to which the tasks will be moved | |
2933 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 2934 | * |
eaf797ab TH |
2935 | * Locking rules between cgroup_post_fork() and the migration path |
2936 | * guarantee that, if a task is forking while being migrated, the new child | |
2937 | * is guaranteed to be either visible in the source cgroup after the | |
2938 | * parent's migration is complete or put into the target cgroup. No task | |
2939 | * can slip out of migration through forking. | |
31a7df01 | 2940 | */ |
8cc99345 | 2941 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 2942 | { |
952aaa12 TH |
2943 | LIST_HEAD(preloaded_csets); |
2944 | struct cgrp_cset_link *link; | |
72ec7029 | 2945 | struct css_task_iter it; |
e406d1cf | 2946 | struct task_struct *task; |
952aaa12 | 2947 | int ret; |
31a7df01 | 2948 | |
952aaa12 | 2949 | mutex_lock(&cgroup_mutex); |
31a7df01 | 2950 | |
952aaa12 TH |
2951 | /* all tasks in @from are being moved, all csets are source */ |
2952 | down_read(&css_set_rwsem); | |
2953 | list_for_each_entry(link, &from->cset_links, cset_link) | |
2954 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
2955 | up_read(&css_set_rwsem); | |
31a7df01 | 2956 | |
952aaa12 TH |
2957 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
2958 | if (ret) | |
2959 | goto out_err; | |
8cc99345 | 2960 | |
952aaa12 TH |
2961 | /* |
2962 | * Migrate tasks one-by-one until @form is empty. This fails iff | |
2963 | * ->can_attach() fails. | |
2964 | */ | |
e406d1cf TH |
2965 | do { |
2966 | css_task_iter_start(&from->dummy_css, &it); | |
2967 | task = css_task_iter_next(&it); | |
2968 | if (task) | |
2969 | get_task_struct(task); | |
2970 | css_task_iter_end(&it); | |
2971 | ||
2972 | if (task) { | |
952aaa12 | 2973 | ret = cgroup_migrate(to, task, false); |
e406d1cf TH |
2974 | put_task_struct(task); |
2975 | } | |
2976 | } while (task && !ret); | |
952aaa12 TH |
2977 | out_err: |
2978 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 2979 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 2980 | return ret; |
8cc99345 TH |
2981 | } |
2982 | ||
bbcb81d0 | 2983 | /* |
102a775e | 2984 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
2985 | * |
2986 | * Reading this file can return large amounts of data if a cgroup has | |
2987 | * *lots* of attached tasks. So it may need several calls to read(), | |
2988 | * but we cannot guarantee that the information we produce is correct | |
2989 | * unless we produce it entirely atomically. | |
2990 | * | |
bbcb81d0 | 2991 | */ |
bbcb81d0 | 2992 | |
24528255 LZ |
2993 | /* which pidlist file are we talking about? */ |
2994 | enum cgroup_filetype { | |
2995 | CGROUP_FILE_PROCS, | |
2996 | CGROUP_FILE_TASKS, | |
2997 | }; | |
2998 | ||
2999 | /* | |
3000 | * A pidlist is a list of pids that virtually represents the contents of one | |
3001 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3002 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3003 | * to the cgroup. | |
3004 | */ | |
3005 | struct cgroup_pidlist { | |
3006 | /* | |
3007 | * used to find which pidlist is wanted. doesn't change as long as | |
3008 | * this particular list stays in the list. | |
3009 | */ | |
3010 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3011 | /* array of xids */ | |
3012 | pid_t *list; | |
3013 | /* how many elements the above list has */ | |
3014 | int length; | |
24528255 LZ |
3015 | /* each of these stored in a list by its cgroup */ |
3016 | struct list_head links; | |
3017 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3018 | struct cgroup *owner; | |
b1a21367 TH |
3019 | /* for delayed destruction */ |
3020 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3021 | }; |
3022 | ||
d1d9fd33 BB |
3023 | /* |
3024 | * The following two functions "fix" the issue where there are more pids | |
3025 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3026 | * TODO: replace with a kernel-wide solution to this problem | |
3027 | */ | |
3028 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3029 | static void *pidlist_allocate(int count) | |
3030 | { | |
3031 | if (PIDLIST_TOO_LARGE(count)) | |
3032 | return vmalloc(count * sizeof(pid_t)); | |
3033 | else | |
3034 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3035 | } | |
b1a21367 | 3036 | |
d1d9fd33 BB |
3037 | static void pidlist_free(void *p) |
3038 | { | |
3039 | if (is_vmalloc_addr(p)) | |
3040 | vfree(p); | |
3041 | else | |
3042 | kfree(p); | |
3043 | } | |
d1d9fd33 | 3044 | |
b1a21367 TH |
3045 | /* |
3046 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3047 | * should be left afterwards. | |
3048 | */ | |
3049 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3050 | { | |
3051 | struct cgroup_pidlist *l, *tmp_l; | |
3052 | ||
3053 | mutex_lock(&cgrp->pidlist_mutex); | |
3054 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3055 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3056 | mutex_unlock(&cgrp->pidlist_mutex); | |
3057 | ||
3058 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3059 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3060 | } | |
3061 | ||
3062 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3063 | { | |
3064 | struct delayed_work *dwork = to_delayed_work(work); | |
3065 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3066 | destroy_dwork); | |
3067 | struct cgroup_pidlist *tofree = NULL; | |
3068 | ||
3069 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3070 | |
3071 | /* | |
04502365 TH |
3072 | * Destroy iff we didn't get queued again. The state won't change |
3073 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3074 | */ |
04502365 | 3075 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3076 | list_del(&l->links); |
3077 | pidlist_free(l->list); | |
3078 | put_pid_ns(l->key.ns); | |
3079 | tofree = l; | |
3080 | } | |
3081 | ||
b1a21367 TH |
3082 | mutex_unlock(&l->owner->pidlist_mutex); |
3083 | kfree(tofree); | |
3084 | } | |
3085 | ||
bbcb81d0 | 3086 | /* |
102a775e | 3087 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3088 | * Returns the number of unique elements. |
bbcb81d0 | 3089 | */ |
6ee211ad | 3090 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3091 | { |
102a775e | 3092 | int src, dest = 1; |
102a775e BB |
3093 | |
3094 | /* | |
3095 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3096 | * edge cases first; no work needs to be done for either | |
3097 | */ | |
3098 | if (length == 0 || length == 1) | |
3099 | return length; | |
3100 | /* src and dest walk down the list; dest counts unique elements */ | |
3101 | for (src = 1; src < length; src++) { | |
3102 | /* find next unique element */ | |
3103 | while (list[src] == list[src-1]) { | |
3104 | src++; | |
3105 | if (src == length) | |
3106 | goto after; | |
3107 | } | |
3108 | /* dest always points to where the next unique element goes */ | |
3109 | list[dest] = list[src]; | |
3110 | dest++; | |
3111 | } | |
3112 | after: | |
102a775e BB |
3113 | return dest; |
3114 | } | |
3115 | ||
afb2bc14 TH |
3116 | /* |
3117 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3118 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3119 | * different per namespace, each namespace needs differently sorted list, | |
3120 | * making it impossible to use, for example, single rbtree of member tasks | |
3121 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3122 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3123 | * pidlists keyed by cgroup and namespace. | |
3124 | * | |
3125 | * All this extra complexity was caused by the original implementation | |
3126 | * committing to an entirely unnecessary property. In the long term, we | |
3127 | * want to do away with it. Explicitly scramble sort order if | |
3128 | * sane_behavior so that no such expectation exists in the new interface. | |
3129 | * | |
3130 | * Scrambling is done by swapping every two consecutive bits, which is | |
3131 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3132 | */ | |
3133 | static pid_t pid_fry(pid_t pid) | |
3134 | { | |
3135 | unsigned a = pid & 0x55555555; | |
3136 | unsigned b = pid & 0xAAAAAAAA; | |
3137 | ||
3138 | return (a << 1) | (b >> 1); | |
3139 | } | |
3140 | ||
3141 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3142 | { | |
3143 | if (cgroup_sane_behavior(cgrp)) | |
3144 | return pid_fry(pid); | |
3145 | else | |
3146 | return pid; | |
3147 | } | |
3148 | ||
102a775e BB |
3149 | static int cmppid(const void *a, const void *b) |
3150 | { | |
3151 | return *(pid_t *)a - *(pid_t *)b; | |
3152 | } | |
3153 | ||
afb2bc14 TH |
3154 | static int fried_cmppid(const void *a, const void *b) |
3155 | { | |
3156 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3157 | } | |
3158 | ||
e6b81710 TH |
3159 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3160 | enum cgroup_filetype type) | |
3161 | { | |
3162 | struct cgroup_pidlist *l; | |
3163 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3164 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3165 | ||
3166 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3167 | ||
3168 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3169 | if (l->key.type == type && l->key.ns == ns) | |
3170 | return l; | |
3171 | return NULL; | |
3172 | } | |
3173 | ||
72a8cb30 BB |
3174 | /* |
3175 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3176 | * returns with the lock on that pidlist already held, and takes care | |
3177 | * of the use count, or returns NULL with no locks held if we're out of | |
3178 | * memory. | |
3179 | */ | |
e6b81710 TH |
3180 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3181 | enum cgroup_filetype type) | |
72a8cb30 BB |
3182 | { |
3183 | struct cgroup_pidlist *l; | |
b70cc5fd | 3184 | |
e6b81710 TH |
3185 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3186 | ||
3187 | l = cgroup_pidlist_find(cgrp, type); | |
3188 | if (l) | |
3189 | return l; | |
3190 | ||
72a8cb30 | 3191 | /* entry not found; create a new one */ |
f4f4be2b | 3192 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3193 | if (!l) |
72a8cb30 | 3194 | return l; |
e6b81710 | 3195 | |
b1a21367 | 3196 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3197 | l->key.type = type; |
e6b81710 TH |
3198 | /* don't need task_nsproxy() if we're looking at ourself */ |
3199 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3200 | l->owner = cgrp; |
3201 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3202 | return l; |
3203 | } | |
3204 | ||
102a775e BB |
3205 | /* |
3206 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3207 | */ | |
72a8cb30 BB |
3208 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3209 | struct cgroup_pidlist **lp) | |
102a775e BB |
3210 | { |
3211 | pid_t *array; | |
3212 | int length; | |
3213 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3214 | struct css_task_iter it; |
817929ec | 3215 | struct task_struct *tsk; |
102a775e BB |
3216 | struct cgroup_pidlist *l; |
3217 | ||
4bac00d1 TH |
3218 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3219 | ||
102a775e BB |
3220 | /* |
3221 | * If cgroup gets more users after we read count, we won't have | |
3222 | * enough space - tough. This race is indistinguishable to the | |
3223 | * caller from the case that the additional cgroup users didn't | |
3224 | * show up until sometime later on. | |
3225 | */ | |
3226 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3227 | array = pidlist_allocate(length); |
102a775e BB |
3228 | if (!array) |
3229 | return -ENOMEM; | |
3230 | /* now, populate the array */ | |
72ec7029 TH |
3231 | css_task_iter_start(&cgrp->dummy_css, &it); |
3232 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3233 | if (unlikely(n == length)) |
817929ec | 3234 | break; |
102a775e | 3235 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3236 | if (type == CGROUP_FILE_PROCS) |
3237 | pid = task_tgid_vnr(tsk); | |
3238 | else | |
3239 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3240 | if (pid > 0) /* make sure to only use valid results */ |
3241 | array[n++] = pid; | |
817929ec | 3242 | } |
72ec7029 | 3243 | css_task_iter_end(&it); |
102a775e BB |
3244 | length = n; |
3245 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3246 | if (cgroup_sane_behavior(cgrp)) |
3247 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3248 | else | |
3249 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3250 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3251 | length = pidlist_uniq(array, length); |
e6b81710 | 3252 | |
e6b81710 | 3253 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3254 | if (!l) { |
e6b81710 | 3255 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3256 | pidlist_free(array); |
72a8cb30 | 3257 | return -ENOMEM; |
102a775e | 3258 | } |
e6b81710 TH |
3259 | |
3260 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3261 | pidlist_free(l->list); |
102a775e BB |
3262 | l->list = array; |
3263 | l->length = length; | |
72a8cb30 | 3264 | *lp = l; |
102a775e | 3265 | return 0; |
bbcb81d0 PM |
3266 | } |
3267 | ||
846c7bb0 | 3268 | /** |
a043e3b2 | 3269 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3270 | * @stats: cgroupstats to fill information into |
3271 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3272 | * been requested. | |
a043e3b2 LZ |
3273 | * |
3274 | * Build and fill cgroupstats so that taskstats can export it to user | |
3275 | * space. | |
846c7bb0 BS |
3276 | */ |
3277 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3278 | { | |
2bd59d48 | 3279 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3280 | struct cgroup *cgrp; |
72ec7029 | 3281 | struct css_task_iter it; |
846c7bb0 | 3282 | struct task_struct *tsk; |
33d283be | 3283 | |
2bd59d48 TH |
3284 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3285 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3286 | kernfs_type(kn) != KERNFS_DIR) | |
3287 | return -EINVAL; | |
3288 | ||
bad34660 LZ |
3289 | mutex_lock(&cgroup_mutex); |
3290 | ||
846c7bb0 | 3291 | /* |
2bd59d48 TH |
3292 | * We aren't being called from kernfs and there's no guarantee on |
3293 | * @kn->priv's validity. For this and css_tryget_from_dir(), | |
3294 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
846c7bb0 | 3295 | */ |
2bd59d48 TH |
3296 | rcu_read_lock(); |
3297 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 3298 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 3299 | rcu_read_unlock(); |
bad34660 | 3300 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
3301 | return -ENOENT; |
3302 | } | |
bad34660 | 3303 | rcu_read_unlock(); |
846c7bb0 | 3304 | |
72ec7029 TH |
3305 | css_task_iter_start(&cgrp->dummy_css, &it); |
3306 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3307 | switch (tsk->state) { |
3308 | case TASK_RUNNING: | |
3309 | stats->nr_running++; | |
3310 | break; | |
3311 | case TASK_INTERRUPTIBLE: | |
3312 | stats->nr_sleeping++; | |
3313 | break; | |
3314 | case TASK_UNINTERRUPTIBLE: | |
3315 | stats->nr_uninterruptible++; | |
3316 | break; | |
3317 | case TASK_STOPPED: | |
3318 | stats->nr_stopped++; | |
3319 | break; | |
3320 | default: | |
3321 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3322 | stats->nr_io_wait++; | |
3323 | break; | |
3324 | } | |
3325 | } | |
72ec7029 | 3326 | css_task_iter_end(&it); |
846c7bb0 | 3327 | |
bad34660 | 3328 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 3329 | return 0; |
846c7bb0 BS |
3330 | } |
3331 | ||
8f3ff208 | 3332 | |
bbcb81d0 | 3333 | /* |
102a775e | 3334 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3335 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3336 | * in the cgroup->l->list array. |
bbcb81d0 | 3337 | */ |
cc31edce | 3338 | |
102a775e | 3339 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3340 | { |
cc31edce PM |
3341 | /* |
3342 | * Initially we receive a position value that corresponds to | |
3343 | * one more than the last pid shown (or 0 on the first call or | |
3344 | * after a seek to the start). Use a binary-search to find the | |
3345 | * next pid to display, if any | |
3346 | */ | |
2bd59d48 | 3347 | struct kernfs_open_file *of = s->private; |
7da11279 | 3348 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3349 | struct cgroup_pidlist *l; |
7da11279 | 3350 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3351 | int index = 0, pid = *pos; |
4bac00d1 TH |
3352 | int *iter, ret; |
3353 | ||
3354 | mutex_lock(&cgrp->pidlist_mutex); | |
3355 | ||
3356 | /* | |
5d22444f | 3357 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3358 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3359 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3360 | * could already have been destroyed. |
3361 | */ | |
5d22444f TH |
3362 | if (of->priv) |
3363 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3364 | |
3365 | /* | |
3366 | * Either this is the first start() after open or the matching | |
3367 | * pidlist has been destroyed inbetween. Create a new one. | |
3368 | */ | |
5d22444f TH |
3369 | if (!of->priv) { |
3370 | ret = pidlist_array_load(cgrp, type, | |
3371 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3372 | if (ret) |
3373 | return ERR_PTR(ret); | |
3374 | } | |
5d22444f | 3375 | l = of->priv; |
cc31edce | 3376 | |
cc31edce | 3377 | if (pid) { |
102a775e | 3378 | int end = l->length; |
20777766 | 3379 | |
cc31edce PM |
3380 | while (index < end) { |
3381 | int mid = (index + end) / 2; | |
afb2bc14 | 3382 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3383 | index = mid; |
3384 | break; | |
afb2bc14 | 3385 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3386 | index = mid + 1; |
3387 | else | |
3388 | end = mid; | |
3389 | } | |
3390 | } | |
3391 | /* If we're off the end of the array, we're done */ | |
102a775e | 3392 | if (index >= l->length) |
cc31edce PM |
3393 | return NULL; |
3394 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3395 | iter = l->list + index; |
afb2bc14 | 3396 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3397 | return iter; |
3398 | } | |
3399 | ||
102a775e | 3400 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3401 | { |
2bd59d48 | 3402 | struct kernfs_open_file *of = s->private; |
5d22444f | 3403 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3404 | |
5d22444f TH |
3405 | if (l) |
3406 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3407 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3408 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3409 | } |
3410 | ||
102a775e | 3411 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3412 | { |
2bd59d48 | 3413 | struct kernfs_open_file *of = s->private; |
5d22444f | 3414 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3415 | pid_t *p = v; |
3416 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3417 | /* |
3418 | * Advance to the next pid in the array. If this goes off the | |
3419 | * end, we're done | |
3420 | */ | |
3421 | p++; | |
3422 | if (p >= end) { | |
3423 | return NULL; | |
3424 | } else { | |
7da11279 | 3425 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3426 | return p; |
3427 | } | |
3428 | } | |
3429 | ||
102a775e | 3430 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3431 | { |
3432 | return seq_printf(s, "%d\n", *(int *)v); | |
3433 | } | |
bbcb81d0 | 3434 | |
102a775e BB |
3435 | /* |
3436 | * seq_operations functions for iterating on pidlists through seq_file - | |
3437 | * independent of whether it's tasks or procs | |
3438 | */ | |
3439 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3440 | .start = cgroup_pidlist_start, | |
3441 | .stop = cgroup_pidlist_stop, | |
3442 | .next = cgroup_pidlist_next, | |
3443 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3444 | }; |
3445 | ||
182446d0 TH |
3446 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3447 | struct cftype *cft) | |
81a6a5cd | 3448 | { |
182446d0 | 3449 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3450 | } |
3451 | ||
182446d0 TH |
3452 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3453 | struct cftype *cft, u64 val) | |
6379c106 | 3454 | { |
182446d0 | 3455 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3456 | if (val) |
182446d0 | 3457 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3458 | else |
182446d0 | 3459 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3460 | return 0; |
3461 | } | |
3462 | ||
182446d0 TH |
3463 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3464 | struct cftype *cft) | |
97978e6d | 3465 | { |
182446d0 | 3466 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3467 | } |
3468 | ||
182446d0 TH |
3469 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3470 | struct cftype *cft, u64 val) | |
97978e6d DL |
3471 | { |
3472 | if (val) | |
182446d0 | 3473 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3474 | else |
182446d0 | 3475 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3476 | return 0; |
3477 | } | |
3478 | ||
d5c56ced | 3479 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3480 | { |
d5c56ced | 3481 | .name = "cgroup.procs", |
6612f05b TH |
3482 | .seq_start = cgroup_pidlist_start, |
3483 | .seq_next = cgroup_pidlist_next, | |
3484 | .seq_stop = cgroup_pidlist_stop, | |
3485 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3486 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3487 | .write_u64 = cgroup_procs_write, |
74a1166d | 3488 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3489 | }, |
97978e6d DL |
3490 | { |
3491 | .name = "cgroup.clone_children", | |
873fe09e | 3492 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3493 | .read_u64 = cgroup_clone_children_read, |
3494 | .write_u64 = cgroup_clone_children_write, | |
3495 | }, | |
873fe09e TH |
3496 | { |
3497 | .name = "cgroup.sane_behavior", | |
3498 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3499 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3500 | }, |
d5c56ced TH |
3501 | |
3502 | /* | |
3503 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3504 | * don't exist if sane_behavior. If you're depending on these, be | |
3505 | * prepared to be burned. | |
3506 | */ | |
3507 | { | |
3508 | .name = "tasks", | |
3509 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3510 | .seq_start = cgroup_pidlist_start, |
3511 | .seq_next = cgroup_pidlist_next, | |
3512 | .seq_stop = cgroup_pidlist_stop, | |
3513 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3514 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3515 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3516 | .mode = S_IRUGO | S_IWUSR, |
3517 | }, | |
3518 | { | |
3519 | .name = "notify_on_release", | |
3520 | .flags = CFTYPE_INSANE, | |
3521 | .read_u64 = cgroup_read_notify_on_release, | |
3522 | .write_u64 = cgroup_write_notify_on_release, | |
3523 | }, | |
6e6ff25b TH |
3524 | { |
3525 | .name = "release_agent", | |
cc5943a7 | 3526 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3527 | .seq_show = cgroup_release_agent_show, |
6e6ff25b | 3528 | .write_string = cgroup_release_agent_write, |
5f469907 | 3529 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 3530 | }, |
db0416b6 | 3531 | { } /* terminate */ |
bbcb81d0 PM |
3532 | }; |
3533 | ||
13af07df | 3534 | /** |
628f7cd4 | 3535 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3536 | * @cgrp: target cgroup |
13af07df | 3537 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3538 | * |
3539 | * On failure, no file is added. | |
13af07df | 3540 | */ |
628f7cd4 | 3541 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3542 | { |
ddbcc7e8 | 3543 | struct cgroup_subsys *ss; |
b420ba7d | 3544 | int i, ret = 0; |
bbcb81d0 | 3545 | |
8e3f6541 | 3546 | /* process cftsets of each subsystem */ |
b420ba7d | 3547 | for_each_subsys(ss, i) { |
0adb0704 | 3548 | struct cftype *cfts; |
b420ba7d TH |
3549 | |
3550 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3551 | continue; |
8e3f6541 | 3552 | |
0adb0704 TH |
3553 | list_for_each_entry(cfts, &ss->cfts, node) { |
3554 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
3555 | if (ret < 0) |
3556 | goto err; | |
3557 | } | |
ddbcc7e8 | 3558 | } |
ddbcc7e8 | 3559 | return 0; |
bee55099 TH |
3560 | err: |
3561 | cgroup_clear_dir(cgrp, subsys_mask); | |
3562 | return ret; | |
ddbcc7e8 PM |
3563 | } |
3564 | ||
0c21ead1 TH |
3565 | /* |
3566 | * css destruction is four-stage process. | |
3567 | * | |
3568 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3569 | * Implemented in kill_css(). | |
3570 | * | |
3571 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3572 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3573 | * by invoking offline_css(). After offlining, the base ref is put. | |
3574 | * Implemented in css_killed_work_fn(). | |
3575 | * | |
3576 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3577 | * accessors are inside RCU read sections. css_release() schedules the | |
3578 | * RCU callback. | |
3579 | * | |
3580 | * 4. After the grace period, the css can be freed. Implemented in | |
3581 | * css_free_work_fn(). | |
3582 | * | |
3583 | * It is actually hairier because both step 2 and 4 require process context | |
3584 | * and thus involve punting to css->destroy_work adding two additional | |
3585 | * steps to the already complex sequence. | |
3586 | */ | |
35ef10da | 3587 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3588 | { |
3589 | struct cgroup_subsys_state *css = | |
35ef10da | 3590 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 3591 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3592 | |
0ae78e0b TH |
3593 | if (css->parent) |
3594 | css_put(css->parent); | |
3595 | ||
0c21ead1 | 3596 | css->ss->css_free(css); |
2bd59d48 | 3597 | cgroup_put(cgrp); |
48ddbe19 TH |
3598 | } |
3599 | ||
0c21ead1 | 3600 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3601 | { |
3602 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3603 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3604 | |
35ef10da | 3605 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 3606 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
3607 | } |
3608 | ||
d3daf28d TH |
3609 | static void css_release(struct percpu_ref *ref) |
3610 | { | |
3611 | struct cgroup_subsys_state *css = | |
3612 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3613 | ||
01a97140 | 3614 | RCU_INIT_POINTER(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 3615 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
3616 | } |
3617 | ||
623f926b TH |
3618 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
3619 | struct cgroup *cgrp) | |
ddbcc7e8 | 3620 | { |
bd89aabc | 3621 | css->cgroup = cgrp; |
72c97e54 | 3622 | css->ss = ss; |
ddbcc7e8 | 3623 | css->flags = 0; |
0ae78e0b TH |
3624 | |
3625 | if (cgrp->parent) | |
ca8bdcaf | 3626 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 3627 | else |
38b53aba | 3628 | css->flags |= CSS_ROOT; |
48ddbe19 | 3629 | |
ca8bdcaf | 3630 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
3631 | } |
3632 | ||
2a4ac633 | 3633 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 3634 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3635 | { |
623f926b | 3636 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
3637 | int ret = 0; |
3638 | ||
ace2bee8 | 3639 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3640 | lockdep_assert_held(&cgroup_mutex); |
3641 | ||
92fb9748 | 3642 | if (ss->css_online) |
eb95419b | 3643 | ret = ss->css_online(css); |
ae7f164a | 3644 | if (!ret) { |
eb95419b | 3645 | css->flags |= CSS_ONLINE; |
f20104de | 3646 | css->cgroup->nr_css++; |
aec25020 | 3647 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 3648 | } |
b1929db4 | 3649 | return ret; |
a31f2d3f TH |
3650 | } |
3651 | ||
2a4ac633 | 3652 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 3653 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3654 | { |
623f926b | 3655 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 3656 | |
ace2bee8 | 3657 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3658 | lockdep_assert_held(&cgroup_mutex); |
3659 | ||
3660 | if (!(css->flags & CSS_ONLINE)) | |
3661 | return; | |
3662 | ||
d7eeac19 | 3663 | if (ss->css_offline) |
eb95419b | 3664 | ss->css_offline(css); |
a31f2d3f | 3665 | |
eb95419b | 3666 | css->flags &= ~CSS_ONLINE; |
09a503ea | 3667 | css->cgroup->nr_css--; |
aec25020 | 3668 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); |
a31f2d3f TH |
3669 | } |
3670 | ||
c81c925a TH |
3671 | /** |
3672 | * create_css - create a cgroup_subsys_state | |
3673 | * @cgrp: the cgroup new css will be associated with | |
3674 | * @ss: the subsys of new css | |
3675 | * | |
3676 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
3677 | * css is online and installed in @cgrp with all interface files created. | |
3678 | * Returns 0 on success, -errno on failure. | |
3679 | */ | |
3680 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
3681 | { | |
3682 | struct cgroup *parent = cgrp->parent; | |
3683 | struct cgroup_subsys_state *css; | |
3684 | int err; | |
3685 | ||
c81c925a TH |
3686 | lockdep_assert_held(&cgroup_mutex); |
3687 | ||
3688 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
3689 | if (IS_ERR(css)) | |
3690 | return PTR_ERR(css); | |
3691 | ||
3692 | err = percpu_ref_init(&css->refcnt, css_release); | |
3693 | if (err) | |
3eb59ec6 | 3694 | goto err_free_css; |
c81c925a TH |
3695 | |
3696 | init_css(css, ss, cgrp); | |
3697 | ||
aec25020 | 3698 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a | 3699 | if (err) |
3eb59ec6 | 3700 | goto err_free_percpu_ref; |
c81c925a TH |
3701 | |
3702 | err = online_css(css); | |
3703 | if (err) | |
3eb59ec6 | 3704 | goto err_clear_dir; |
c81c925a | 3705 | |
59f5296b | 3706 | cgroup_get(cgrp); |
c81c925a TH |
3707 | css_get(css->parent); |
3708 | ||
3709 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
3710 | parent->parent) { | |
3711 | 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", | |
3712 | current->comm, current->pid, ss->name); | |
3713 | if (!strcmp(ss->name, "memory")) | |
3714 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
3715 | ss->warned_broken_hierarchy = true; | |
3716 | } | |
3717 | ||
3718 | return 0; | |
3719 | ||
3eb59ec6 | 3720 | err_clear_dir: |
32d01dc7 | 3721 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3eb59ec6 | 3722 | err_free_percpu_ref: |
c81c925a | 3723 | percpu_ref_cancel_init(&css->refcnt); |
3eb59ec6 | 3724 | err_free_css: |
c81c925a TH |
3725 | ss->css_free(css); |
3726 | return err; | |
3727 | } | |
3728 | ||
2bd59d48 | 3729 | /** |
a043e3b2 LZ |
3730 | * cgroup_create - create a cgroup |
3731 | * @parent: cgroup that will be parent of the new cgroup | |
e61734c5 | 3732 | * @name: name of the new cgroup |
2bd59d48 | 3733 | * @mode: mode to set on new cgroup |
ddbcc7e8 | 3734 | */ |
e61734c5 | 3735 | static long cgroup_create(struct cgroup *parent, const char *name, |
2bd59d48 | 3736 | umode_t mode) |
ddbcc7e8 | 3737 | { |
bd89aabc | 3738 | struct cgroup *cgrp; |
3dd06ffa | 3739 | struct cgroup_root *root = parent->root; |
b58c8998 | 3740 | int ssid, err; |
ddbcc7e8 | 3741 | struct cgroup_subsys *ss; |
2bd59d48 | 3742 | struct kernfs_node *kn; |
ddbcc7e8 | 3743 | |
a2dd4247 TH |
3744 | /* |
3745 | * XXX: The default hierarchy isn't fully implemented yet. Block | |
3746 | * !root cgroup creation on it for now. | |
3747 | */ | |
3748 | if (root == &cgrp_dfl_root) | |
3749 | return -EINVAL; | |
ddbcc7e8 | 3750 | |
0a950f65 | 3751 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
3752 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
3753 | if (!cgrp) | |
ddbcc7e8 PM |
3754 | return -ENOMEM; |
3755 | ||
ace2bee8 | 3756 | mutex_lock(&cgroup_tree_mutex); |
65dff759 | 3757 | |
976c06bc TH |
3758 | /* |
3759 | * Only live parents can have children. Note that the liveliness | |
3760 | * check isn't strictly necessary because cgroup_mkdir() and | |
3761 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
3762 | * anyway so that locking is contained inside cgroup proper and we | |
3763 | * don't get nasty surprises if we ever grow another caller. | |
3764 | */ | |
3765 | if (!cgroup_lock_live_group(parent)) { | |
3766 | err = -ENODEV; | |
ace2bee8 | 3767 | goto err_unlock_tree; |
0ab02ca8 LZ |
3768 | } |
3769 | ||
3770 | /* | |
3771 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
3772 | * a half-baked cgroup. | |
3773 | */ | |
3774 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
3775 | if (cgrp->id < 0) { | |
3776 | err = -ENOMEM; | |
3777 | goto err_unlock; | |
976c06bc TH |
3778 | } |
3779 | ||
cc31edce | 3780 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3781 | |
bd89aabc | 3782 | cgrp->parent = parent; |
0ae78e0b | 3783 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 3784 | cgrp->root = parent->root; |
ddbcc7e8 | 3785 | |
b6abdb0e LZ |
3786 | if (notify_on_release(parent)) |
3787 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3788 | ||
2260e7fc TH |
3789 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
3790 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 3791 | |
2bd59d48 | 3792 | /* create the directory */ |
e61734c5 | 3793 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 TH |
3794 | if (IS_ERR(kn)) { |
3795 | err = PTR_ERR(kn); | |
0ab02ca8 | 3796 | goto err_free_id; |
2bd59d48 TH |
3797 | } |
3798 | cgrp->kn = kn; | |
ddbcc7e8 | 3799 | |
4e139afc | 3800 | /* |
6f30558f TH |
3801 | * This extra ref will be put in cgroup_free_fn() and guarantees |
3802 | * that @cgrp->kn is always accessible. | |
4e139afc | 3803 | */ |
6f30558f | 3804 | kernfs_get(kn); |
ddbcc7e8 | 3805 | |
00356bd5 | 3806 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 3807 | |
4e139afc | 3808 | /* allocation complete, commit to creation */ |
4e139afc | 3809 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3c9c825b | 3810 | atomic_inc(&root->nr_cgrps); |
59f5296b | 3811 | cgroup_get(parent); |
415cf07a | 3812 | |
0d80255e TH |
3813 | /* |
3814 | * @cgrp is now fully operational. If something fails after this | |
3815 | * point, it'll be released via the normal destruction path. | |
3816 | */ | |
4e96ee8e LZ |
3817 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
3818 | ||
49957f8e TH |
3819 | err = cgroup_kn_set_ugid(kn); |
3820 | if (err) | |
3821 | goto err_destroy; | |
3822 | ||
2bb566cb | 3823 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
3824 | if (err) |
3825 | goto err_destroy; | |
3826 | ||
9d403e99 | 3827 | /* let's create and online css's */ |
b85d2040 | 3828 | for_each_subsys(ss, ssid) { |
f392e51c | 3829 | if (parent->child_subsys_mask & (1 << ssid)) { |
b85d2040 TH |
3830 | err = create_css(cgrp, ss); |
3831 | if (err) | |
3832 | goto err_destroy; | |
3833 | } | |
a8638030 | 3834 | } |
ddbcc7e8 | 3835 | |
f392e51c TH |
3836 | cgrp->child_subsys_mask = parent->child_subsys_mask; |
3837 | ||
2bd59d48 TH |
3838 | kernfs_activate(kn); |
3839 | ||
ddbcc7e8 | 3840 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3841 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
3842 | |
3843 | return 0; | |
3844 | ||
0a950f65 | 3845 | err_free_id: |
4e96ee8e | 3846 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
3847 | err_unlock: |
3848 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
3849 | err_unlock_tree: |
3850 | mutex_unlock(&cgroup_tree_mutex); | |
bd89aabc | 3851 | kfree(cgrp); |
ddbcc7e8 | 3852 | return err; |
4b8b47eb TH |
3853 | |
3854 | err_destroy: | |
3855 | cgroup_destroy_locked(cgrp); | |
3856 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3857 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 3858 | return err; |
ddbcc7e8 PM |
3859 | } |
3860 | ||
2bd59d48 TH |
3861 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
3862 | umode_t mode) | |
ddbcc7e8 | 3863 | { |
2bd59d48 | 3864 | struct cgroup *parent = parent_kn->priv; |
e1b2dc17 | 3865 | int ret; |
ddbcc7e8 | 3866 | |
e1b2dc17 TH |
3867 | /* |
3868 | * cgroup_create() grabs cgroup_tree_mutex which nests outside | |
3869 | * kernfs active_ref and cgroup_create() already synchronizes | |
3870 | * properly against removal through cgroup_lock_live_group(). | |
3871 | * Break it before calling cgroup_create(). | |
3872 | */ | |
3873 | cgroup_get(parent); | |
3874 | kernfs_break_active_protection(parent_kn); | |
ddbcc7e8 | 3875 | |
e1b2dc17 TH |
3876 | ret = cgroup_create(parent, name, mode); |
3877 | ||
3878 | kernfs_unbreak_active_protection(parent_kn); | |
3879 | cgroup_put(parent); | |
3880 | return ret; | |
ddbcc7e8 PM |
3881 | } |
3882 | ||
223dbc38 TH |
3883 | /* |
3884 | * This is called when the refcnt of a css is confirmed to be killed. | |
3885 | * css_tryget() is now guaranteed to fail. | |
3886 | */ | |
3887 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 3888 | { |
223dbc38 TH |
3889 | struct cgroup_subsys_state *css = |
3890 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
3891 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 3892 | |
ace2bee8 | 3893 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
3894 | mutex_lock(&cgroup_mutex); |
3895 | ||
09a503ea TH |
3896 | /* |
3897 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
3898 | * initate destruction. | |
3899 | */ | |
3900 | offline_css(css); | |
3901 | ||
f20104de TH |
3902 | /* |
3903 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
3904 | * be disabled before proceeding to the second phase of cgroup | |
3905 | * destruction. If we are the last one, kick it off. | |
3906 | */ | |
09a503ea | 3907 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
3908 | cgroup_destroy_css_killed(cgrp); |
3909 | ||
3910 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3911 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
3912 | |
3913 | /* | |
3914 | * Put the css refs from kill_css(). Each css holds an extra | |
3915 | * reference to the cgroup's dentry and cgroup removal proceeds | |
3916 | * regardless of css refs. On the last put of each css, whenever | |
3917 | * that may be, the extra dentry ref is put so that dentry | |
3918 | * destruction happens only after all css's are released. | |
3919 | */ | |
3920 | css_put(css); | |
d3daf28d TH |
3921 | } |
3922 | ||
223dbc38 TH |
3923 | /* css kill confirmation processing requires process context, bounce */ |
3924 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
3925 | { |
3926 | struct cgroup_subsys_state *css = | |
3927 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3928 | ||
223dbc38 | 3929 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 3930 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
3931 | } |
3932 | ||
f392e51c TH |
3933 | /** |
3934 | * kill_css - destroy a css | |
3935 | * @css: css to destroy | |
3936 | * | |
3937 | * This function initiates destruction of @css by removing cgroup interface | |
3938 | * files and putting its base reference. ->css_offline() will be invoked | |
3939 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
3940 | * reference count reaches zero, @css will be released. | |
3941 | */ | |
3942 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 3943 | { |
94419627 TH |
3944 | lockdep_assert_held(&cgroup_tree_mutex); |
3945 | ||
2bd59d48 TH |
3946 | /* |
3947 | * This must happen before css is disassociated with its cgroup. | |
3948 | * See seq_css() for details. | |
3949 | */ | |
aec25020 | 3950 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 3951 | |
edae0c33 TH |
3952 | /* |
3953 | * Killing would put the base ref, but we need to keep it alive | |
3954 | * until after ->css_offline(). | |
3955 | */ | |
3956 | css_get(css); | |
3957 | ||
3958 | /* | |
3959 | * cgroup core guarantees that, by the time ->css_offline() is | |
3960 | * invoked, no new css reference will be given out via | |
3961 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
3962 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
3963 | * guarantee that the ref is seen as killed on all CPUs on return. | |
3964 | * | |
3965 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
3966 | * css is confirmed to be seen as killed on all CPUs. | |
3967 | */ | |
3968 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
3969 | } |
3970 | ||
3971 | /** | |
3972 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
3973 | * @cgrp: cgroup to be destroyed | |
3974 | * | |
3975 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
3976 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
3977 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
3978 | * invoked. To satisfy all the requirements, destruction is implemented in | |
3979 | * the following two steps. | |
3980 | * | |
3981 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
3982 | * userland visible parts and start killing the percpu refcnts of | |
3983 | * css's. Set up so that the next stage will be kicked off once all | |
3984 | * the percpu refcnts are confirmed to be killed. | |
3985 | * | |
3986 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
3987 | * rest of destruction. Once all cgroup references are gone, the | |
3988 | * cgroup is RCU-freed. | |
3989 | * | |
3990 | * This function implements s1. After this step, @cgrp is gone as far as | |
3991 | * the userland is concerned and a new cgroup with the same name may be | |
3992 | * created. As cgroup doesn't care about the names internally, this | |
3993 | * doesn't cause any problem. | |
3994 | */ | |
42809dd4 TH |
3995 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
3996 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 3997 | { |
bb78a92f | 3998 | struct cgroup *child; |
2bd59d48 | 3999 | struct cgroup_subsys_state *css; |
ddd69148 | 4000 | bool empty; |
1c6727af | 4001 | int ssid; |
ddbcc7e8 | 4002 | |
ace2bee8 | 4003 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
4004 | lockdep_assert_held(&cgroup_mutex); |
4005 | ||
ddd69148 | 4006 | /* |
96d365e0 | 4007 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 4008 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 4009 | */ |
96d365e0 | 4010 | down_read(&css_set_rwsem); |
bb78a92f | 4011 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 4012 | up_read(&css_set_rwsem); |
ddd69148 | 4013 | if (!empty) |
ddbcc7e8 | 4014 | return -EBUSY; |
a043e3b2 | 4015 | |
bb78a92f HD |
4016 | /* |
4017 | * Make sure there's no live children. We can't test ->children | |
4018 | * emptiness as dead children linger on it while being destroyed; | |
4019 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
4020 | */ | |
4021 | empty = true; | |
4022 | rcu_read_lock(); | |
4023 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
4024 | empty = cgroup_is_dead(child); | |
4025 | if (!empty) | |
4026 | break; | |
4027 | } | |
4028 | rcu_read_unlock(); | |
4029 | if (!empty) | |
4030 | return -EBUSY; | |
4031 | ||
455050d2 TH |
4032 | /* |
4033 | * Mark @cgrp dead. This prevents further task migration and child | |
4034 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 4035 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 4036 | * resume iteration after dropping RCU read lock. See |
492eb21b | 4037 | * css_next_child() for details. |
455050d2 | 4038 | */ |
54766d4a | 4039 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 4040 | |
88703267 | 4041 | /* |
edae0c33 TH |
4042 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
4043 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
4044 | * percpu refs of all css's are confirmed to be killed. This |
4045 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 4046 | */ |
4ac06017 | 4047 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
4048 | for_each_css(css, ssid, cgrp) |
4049 | kill_css(css); | |
4ac06017 | 4050 | mutex_lock(&cgroup_mutex); |
455050d2 | 4051 | |
455050d2 TH |
4052 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
4053 | raw_spin_lock(&release_list_lock); | |
4054 | if (!list_empty(&cgrp->release_list)) | |
4055 | list_del_init(&cgrp->release_list); | |
4056 | raw_spin_unlock(&release_list_lock); | |
4057 | ||
4058 | /* | |
f20104de TH |
4059 | * If @cgrp has css's attached, the second stage of cgroup |
4060 | * destruction is kicked off from css_killed_work_fn() after the | |
4061 | * refs of all attached css's are killed. If @cgrp doesn't have | |
4062 | * any css, we kick it off here. | |
4063 | */ | |
4064 | if (!cgrp->nr_css) | |
4065 | cgroup_destroy_css_killed(cgrp); | |
4066 | ||
2bd59d48 TH |
4067 | /* remove @cgrp directory along with the base files */ |
4068 | mutex_unlock(&cgroup_mutex); | |
4069 | ||
455050d2 | 4070 | /* |
2bd59d48 TH |
4071 | * There are two control paths which try to determine cgroup from |
4072 | * dentry without going through kernfs - cgroupstats_build() and | |
4073 | * css_tryget_from_dir(). Those are supported by RCU protecting | |
4074 | * clearing of cgrp->kn->priv backpointer, which should happen | |
4075 | * after all files under it have been removed. | |
455050d2 | 4076 | */ |
6f30558f | 4077 | kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ |
2bd59d48 | 4078 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); |
2bd59d48 | 4079 | |
4ac06017 | 4080 | mutex_lock(&cgroup_mutex); |
455050d2 | 4081 | |
ea15f8cc TH |
4082 | return 0; |
4083 | }; | |
4084 | ||
d3daf28d | 4085 | /** |
f20104de | 4086 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
4087 | * @work: cgroup->destroy_free_work |
4088 | * | |
4089 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
4090 | * destroyed after all css's are offlined and performs the rest of |
4091 | * destruction. This is the second step of destruction described in the | |
4092 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 4093 | */ |
f20104de | 4094 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 4095 | { |
ea15f8cc | 4096 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 4097 | |
ace2bee8 | 4098 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 4099 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 4100 | |
999cd8a4 | 4101 | /* delete this cgroup from parent->children */ |
eb6fd504 | 4102 | list_del_rcu(&cgrp->sibling); |
ed957793 | 4103 | |
59f5296b | 4104 | cgroup_put(cgrp); |
ddbcc7e8 | 4105 | |
bd89aabc | 4106 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 4107 | check_for_release(parent); |
ddbcc7e8 PM |
4108 | } |
4109 | ||
2bd59d48 | 4110 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4111 | { |
2bd59d48 TH |
4112 | struct cgroup *cgrp = kn->priv; |
4113 | int ret = 0; | |
42809dd4 | 4114 | |
2bd59d48 TH |
4115 | /* |
4116 | * This is self-destruction but @kn can't be removed while this | |
4117 | * callback is in progress. Let's break active protection. Once | |
4118 | * the protection is broken, @cgrp can be destroyed at any point. | |
4119 | * Pin it so that it stays accessible. | |
4120 | */ | |
4121 | cgroup_get(cgrp); | |
4122 | kernfs_break_active_protection(kn); | |
42809dd4 | 4123 | |
ace2bee8 | 4124 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 4125 | mutex_lock(&cgroup_mutex); |
8e3f6541 TH |
4126 | |
4127 | /* | |
2bd59d48 TH |
4128 | * @cgrp might already have been destroyed while we're trying to |
4129 | * grab the mutexes. | |
8e3f6541 | 4130 | */ |
2bd59d48 TH |
4131 | if (!cgroup_is_dead(cgrp)) |
4132 | ret = cgroup_destroy_locked(cgrp); | |
2bb566cb | 4133 | |
42809dd4 | 4134 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4135 | mutex_unlock(&cgroup_tree_mutex); |
2bb566cb | 4136 | |
2bd59d48 TH |
4137 | kernfs_unbreak_active_protection(kn); |
4138 | cgroup_put(cgrp); | |
42809dd4 | 4139 | return ret; |
8e3f6541 TH |
4140 | } |
4141 | ||
2bd59d48 TH |
4142 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4143 | .remount_fs = cgroup_remount, | |
4144 | .show_options = cgroup_show_options, | |
4145 | .mkdir = cgroup_mkdir, | |
4146 | .rmdir = cgroup_rmdir, | |
4147 | .rename = cgroup_rename, | |
4148 | }; | |
4149 | ||
06a11920 | 4150 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4151 | { |
ddbcc7e8 | 4152 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4153 | |
4154 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4155 | |
ace2bee8 | 4156 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
4157 | mutex_lock(&cgroup_mutex); |
4158 | ||
0adb0704 | 4159 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4160 | |
3dd06ffa TH |
4161 | /* Create the root cgroup state for this subsystem */ |
4162 | ss->root = &cgrp_dfl_root; | |
4163 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
4164 | /* We don't handle early failures gracefully */ |
4165 | BUG_ON(IS_ERR(css)); | |
3dd06ffa | 4166 | init_css(css, ss, &cgrp_dfl_root.cgrp); |
ddbcc7e8 | 4167 | |
e8d55fde | 4168 | /* Update the init_css_set to contain a subsys |
817929ec | 4169 | * pointer to this state - since the subsystem is |
e8d55fde | 4170 | * newly registered, all tasks and hence the |
3dd06ffa | 4171 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 4172 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4173 | |
4174 | need_forkexit_callback |= ss->fork || ss->exit; | |
4175 | ||
e8d55fde LZ |
4176 | /* At system boot, before all subsystems have been |
4177 | * registered, no tasks have been forked, so we don't | |
4178 | * need to invoke fork callbacks here. */ | |
4179 | BUG_ON(!list_empty(&init_task.tasks)); | |
4180 | ||
ae7f164a | 4181 | BUG_ON(online_css(css)); |
a8638030 | 4182 | |
f392e51c | 4183 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
cf5d5941 BB |
4184 | |
4185 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4186 | mutex_unlock(&cgroup_tree_mutex); |
cf5d5941 | 4187 | } |
cf5d5941 | 4188 | |
ddbcc7e8 | 4189 | /** |
a043e3b2 LZ |
4190 | * cgroup_init_early - cgroup initialization at system boot |
4191 | * | |
4192 | * Initialize cgroups at system boot, and initialize any | |
4193 | * subsystems that request early init. | |
ddbcc7e8 PM |
4194 | */ |
4195 | int __init cgroup_init_early(void) | |
4196 | { | |
a2dd4247 TH |
4197 | static struct cgroup_sb_opts __initdata opts = |
4198 | { .flags = CGRP_ROOT_SANE_BEHAVIOR }; | |
30159ec7 | 4199 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4200 | int i; |
30159ec7 | 4201 | |
3dd06ffa | 4202 | init_cgroup_root(&cgrp_dfl_root, &opts); |
a4ea1cc9 | 4203 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4204 | |
3ed80a62 | 4205 | for_each_subsys(ss, i) { |
aec25020 | 4206 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4207 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4208 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4209 | ss->id, ss->name); |
073219e9 TH |
4210 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4211 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4212 | ||
aec25020 | 4213 | ss->id = i; |
073219e9 | 4214 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4215 | |
4216 | if (ss->early_init) | |
4217 | cgroup_init_subsys(ss); | |
4218 | } | |
4219 | return 0; | |
4220 | } | |
4221 | ||
4222 | /** | |
a043e3b2 LZ |
4223 | * cgroup_init - cgroup initialization |
4224 | * | |
4225 | * Register cgroup filesystem and /proc file, and initialize | |
4226 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4227 | */ |
4228 | int __init cgroup_init(void) | |
4229 | { | |
30159ec7 | 4230 | struct cgroup_subsys *ss; |
0ac801fe | 4231 | unsigned long key; |
172a2c06 | 4232 | int ssid, err; |
ddbcc7e8 | 4233 | |
2bd59d48 | 4234 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
ddbcc7e8 | 4235 | |
985ed670 | 4236 | mutex_lock(&cgroup_tree_mutex); |
54e7b4eb | 4237 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4238 | |
82fe9b0d TH |
4239 | /* Add init_css_set to the hash table */ |
4240 | key = css_set_hash(init_css_set.subsys); | |
4241 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4242 | ||
3dd06ffa | 4243 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 4244 | |
54e7b4eb | 4245 | mutex_unlock(&cgroup_mutex); |
985ed670 | 4246 | mutex_unlock(&cgroup_tree_mutex); |
54e7b4eb | 4247 | |
172a2c06 TH |
4248 | for_each_subsys(ss, ssid) { |
4249 | if (!ss->early_init) | |
4250 | cgroup_init_subsys(ss); | |
4251 | ||
4252 | /* | |
4253 | * cftype registration needs kmalloc and can't be done | |
4254 | * during early_init. Register base cftypes separately. | |
4255 | */ | |
4256 | if (ss->base_cftypes) | |
4257 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
676db4af GKH |
4258 | } |
4259 | ||
676db4af | 4260 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4261 | if (!cgroup_kobj) |
4262 | return -ENOMEM; | |
676db4af | 4263 | |
ddbcc7e8 | 4264 | err = register_filesystem(&cgroup_fs_type); |
676db4af GKH |
4265 | if (err < 0) { |
4266 | kobject_put(cgroup_kobj); | |
2bd59d48 | 4267 | return err; |
676db4af | 4268 | } |
ddbcc7e8 | 4269 | |
46ae220b | 4270 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 4271 | return 0; |
ddbcc7e8 | 4272 | } |
b4f48b63 | 4273 | |
e5fca243 TH |
4274 | static int __init cgroup_wq_init(void) |
4275 | { | |
4276 | /* | |
4277 | * There isn't much point in executing destruction path in | |
4278 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 4279 | * Use 1 for @max_active. |
e5fca243 TH |
4280 | * |
4281 | * We would prefer to do this in cgroup_init() above, but that | |
4282 | * is called before init_workqueues(): so leave this until after. | |
4283 | */ | |
1a11533f | 4284 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 4285 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4286 | |
4287 | /* | |
4288 | * Used to destroy pidlists and separate to serve as flush domain. | |
4289 | * Cap @max_active to 1 too. | |
4290 | */ | |
4291 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4292 | 0, 1); | |
4293 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4294 | ||
e5fca243 TH |
4295 | return 0; |
4296 | } | |
4297 | core_initcall(cgroup_wq_init); | |
4298 | ||
a424316c PM |
4299 | /* |
4300 | * proc_cgroup_show() | |
4301 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4302 | * - Used for /proc/<pid>/cgroup. | |
a424316c PM |
4303 | */ |
4304 | ||
4305 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4306 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4307 | { |
4308 | struct pid *pid; | |
4309 | struct task_struct *tsk; | |
e61734c5 | 4310 | char *buf, *path; |
a424316c | 4311 | int retval; |
3dd06ffa | 4312 | struct cgroup_root *root; |
a424316c PM |
4313 | |
4314 | retval = -ENOMEM; | |
e61734c5 | 4315 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4316 | if (!buf) |
4317 | goto out; | |
4318 | ||
4319 | retval = -ESRCH; | |
4320 | pid = m->private; | |
4321 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4322 | if (!tsk) | |
4323 | goto out_free; | |
4324 | ||
4325 | retval = 0; | |
4326 | ||
4327 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 4328 | down_read(&css_set_rwsem); |
a424316c | 4329 | |
985ed670 | 4330 | for_each_root(root) { |
a424316c | 4331 | struct cgroup_subsys *ss; |
bd89aabc | 4332 | struct cgroup *cgrp; |
b85d2040 | 4333 | int ssid, count = 0; |
a424316c | 4334 | |
a2dd4247 | 4335 | if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) |
985ed670 TH |
4336 | continue; |
4337 | ||
2c6ab6d2 | 4338 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 | 4339 | for_each_subsys(ss, ssid) |
f392e51c | 4340 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 4341 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); |
c6d57f33 PM |
4342 | if (strlen(root->name)) |
4343 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4344 | root->name); | |
a424316c | 4345 | seq_putc(m, ':'); |
7717f7ba | 4346 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
4347 | path = cgroup_path(cgrp, buf, PATH_MAX); |
4348 | if (!path) { | |
4349 | retval = -ENAMETOOLONG; | |
a424316c | 4350 | goto out_unlock; |
e61734c5 TH |
4351 | } |
4352 | seq_puts(m, path); | |
a424316c PM |
4353 | seq_putc(m, '\n'); |
4354 | } | |
4355 | ||
4356 | out_unlock: | |
96d365e0 | 4357 | up_read(&css_set_rwsem); |
a424316c PM |
4358 | mutex_unlock(&cgroup_mutex); |
4359 | put_task_struct(tsk); | |
4360 | out_free: | |
4361 | kfree(buf); | |
4362 | out: | |
4363 | return retval; | |
4364 | } | |
4365 | ||
a424316c PM |
4366 | /* Display information about each subsystem and each hierarchy */ |
4367 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4368 | { | |
30159ec7 | 4369 | struct cgroup_subsys *ss; |
a424316c | 4370 | int i; |
a424316c | 4371 | |
8bab8dde | 4372 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4373 | /* |
4374 | * ideally we don't want subsystems moving around while we do this. | |
4375 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4376 | * subsys/hierarchy state. | |
4377 | */ | |
a424316c | 4378 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4379 | |
4380 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4381 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4382 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 4383 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 4384 | |
a424316c PM |
4385 | mutex_unlock(&cgroup_mutex); |
4386 | return 0; | |
4387 | } | |
4388 | ||
4389 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4390 | { | |
9dce07f1 | 4391 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4392 | } |
4393 | ||
828c0950 | 4394 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4395 | .open = cgroupstats_open, |
4396 | .read = seq_read, | |
4397 | .llseek = seq_lseek, | |
4398 | .release = single_release, | |
4399 | }; | |
4400 | ||
b4f48b63 | 4401 | /** |
eaf797ab | 4402 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 4403 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 4404 | * |
eaf797ab TH |
4405 | * A task is associated with the init_css_set until cgroup_post_fork() |
4406 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
4407 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
4408 | */ |
4409 | void cgroup_fork(struct task_struct *child) | |
4410 | { | |
eaf797ab | 4411 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 4412 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4413 | } |
4414 | ||
817929ec | 4415 | /** |
a043e3b2 LZ |
4416 | * cgroup_post_fork - called on a new task after adding it to the task list |
4417 | * @child: the task in question | |
4418 | * | |
5edee61e TH |
4419 | * Adds the task to the list running through its css_set if necessary and |
4420 | * call the subsystem fork() callbacks. Has to be after the task is | |
4421 | * visible on the task list in case we race with the first call to | |
0942eeee | 4422 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4423 | * list. |
a043e3b2 | 4424 | */ |
817929ec PM |
4425 | void cgroup_post_fork(struct task_struct *child) |
4426 | { | |
30159ec7 | 4427 | struct cgroup_subsys *ss; |
5edee61e TH |
4428 | int i; |
4429 | ||
3ce3230a | 4430 | /* |
eaf797ab TH |
4431 | * This may race against cgroup_enable_task_cg_links(). As that |
4432 | * function sets use_task_css_set_links before grabbing | |
4433 | * tasklist_lock and we just went through tasklist_lock to add | |
4434 | * @child, it's guaranteed that either we see the set | |
4435 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
4436 | * @child during its iteration. | |
4437 | * | |
4438 | * If we won the race, @child is associated with %current's | |
4439 | * css_set. Grabbing css_set_rwsem guarantees both that the | |
4440 | * association is stable, and, on completion of the parent's | |
4441 | * migration, @child is visible in the source of migration or | |
4442 | * already in the destination cgroup. This guarantee is necessary | |
4443 | * when implementing operations which need to migrate all tasks of | |
4444 | * a cgroup to another. | |
4445 | * | |
4446 | * Note that if we lose to cgroup_enable_task_cg_links(), @child | |
4447 | * will remain in init_css_set. This is safe because all tasks are | |
4448 | * in the init_css_set before cg_links is enabled and there's no | |
4449 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 4450 | */ |
817929ec | 4451 | if (use_task_css_set_links) { |
eaf797ab TH |
4452 | struct css_set *cset; |
4453 | ||
96d365e0 | 4454 | down_write(&css_set_rwsem); |
0e1d768f | 4455 | cset = task_css_set(current); |
eaf797ab TH |
4456 | if (list_empty(&child->cg_list)) { |
4457 | rcu_assign_pointer(child->cgroups, cset); | |
4458 | list_add(&child->cg_list, &cset->tasks); | |
4459 | get_css_set(cset); | |
4460 | } | |
96d365e0 | 4461 | up_write(&css_set_rwsem); |
817929ec | 4462 | } |
5edee61e TH |
4463 | |
4464 | /* | |
4465 | * Call ss->fork(). This must happen after @child is linked on | |
4466 | * css_set; otherwise, @child might change state between ->fork() | |
4467 | * and addition to css_set. | |
4468 | */ | |
4469 | if (need_forkexit_callback) { | |
3ed80a62 | 4470 | for_each_subsys(ss, i) |
5edee61e TH |
4471 | if (ss->fork) |
4472 | ss->fork(child); | |
5edee61e | 4473 | } |
817929ec | 4474 | } |
5edee61e | 4475 | |
b4f48b63 PM |
4476 | /** |
4477 | * cgroup_exit - detach cgroup from exiting task | |
4478 | * @tsk: pointer to task_struct of exiting process | |
4479 | * | |
4480 | * Description: Detach cgroup from @tsk and release it. | |
4481 | * | |
4482 | * Note that cgroups marked notify_on_release force every task in | |
4483 | * them to take the global cgroup_mutex mutex when exiting. | |
4484 | * This could impact scaling on very large systems. Be reluctant to | |
4485 | * use notify_on_release cgroups where very high task exit scaling | |
4486 | * is required on large systems. | |
4487 | * | |
0e1d768f TH |
4488 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
4489 | * call cgroup_exit() while the task is still competent to handle | |
4490 | * notify_on_release(), then leave the task attached to the root cgroup in | |
4491 | * each hierarchy for the remainder of its exit. No need to bother with | |
4492 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 4493 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 4494 | */ |
1ec41830 | 4495 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 4496 | { |
30159ec7 | 4497 | struct cgroup_subsys *ss; |
5abb8855 | 4498 | struct css_set *cset; |
eaf797ab | 4499 | bool put_cset = false; |
d41d5a01 | 4500 | int i; |
817929ec PM |
4501 | |
4502 | /* | |
0e1d768f TH |
4503 | * Unlink from @tsk from its css_set. As migration path can't race |
4504 | * with us, we can check cg_list without grabbing css_set_rwsem. | |
817929ec PM |
4505 | */ |
4506 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 4507 | down_write(&css_set_rwsem); |
0e1d768f | 4508 | list_del_init(&tsk->cg_list); |
96d365e0 | 4509 | up_write(&css_set_rwsem); |
0e1d768f | 4510 | put_cset = true; |
817929ec PM |
4511 | } |
4512 | ||
b4f48b63 | 4513 | /* Reassign the task to the init_css_set. */ |
a8ad805c TH |
4514 | cset = task_css_set(tsk); |
4515 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 | 4516 | |
1ec41830 | 4517 | if (need_forkexit_callback) { |
3ed80a62 TH |
4518 | /* see cgroup_post_fork() for details */ |
4519 | for_each_subsys(ss, i) { | |
d41d5a01 | 4520 | if (ss->exit) { |
eb95419b TH |
4521 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4522 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4523 | |
eb95419b | 4524 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4525 | } |
4526 | } | |
4527 | } | |
d41d5a01 | 4528 | |
eaf797ab TH |
4529 | if (put_cset) |
4530 | put_css_set(cset, true); | |
b4f48b63 | 4531 | } |
697f4161 | 4532 | |
bd89aabc | 4533 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4534 | { |
f50daa70 | 4535 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4536 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4537 | /* |
4538 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4539 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4540 | * it now |
4541 | */ | |
81a6a5cd | 4542 | int need_schedule_work = 0; |
f50daa70 | 4543 | |
cdcc136f | 4544 | raw_spin_lock(&release_list_lock); |
54766d4a | 4545 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4546 | list_empty(&cgrp->release_list)) { |
4547 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4548 | need_schedule_work = 1; |
4549 | } | |
cdcc136f | 4550 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4551 | if (need_schedule_work) |
4552 | schedule_work(&release_agent_work); | |
4553 | } | |
4554 | } | |
4555 | ||
81a6a5cd PM |
4556 | /* |
4557 | * Notify userspace when a cgroup is released, by running the | |
4558 | * configured release agent with the name of the cgroup (path | |
4559 | * relative to the root of cgroup file system) as the argument. | |
4560 | * | |
4561 | * Most likely, this user command will try to rmdir this cgroup. | |
4562 | * | |
4563 | * This races with the possibility that some other task will be | |
4564 | * attached to this cgroup before it is removed, or that some other | |
4565 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4566 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4567 | * unused, and this cgroup will be reprieved from its death sentence, | |
4568 | * to continue to serve a useful existence. Next time it's released, | |
4569 | * we will get notified again, if it still has 'notify_on_release' set. | |
4570 | * | |
4571 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4572 | * means only wait until the task is successfully execve()'d. The | |
4573 | * separate release agent task is forked by call_usermodehelper(), | |
4574 | * then control in this thread returns here, without waiting for the | |
4575 | * release agent task. We don't bother to wait because the caller of | |
4576 | * this routine has no use for the exit status of the release agent | |
4577 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4578 | */ |
81a6a5cd PM |
4579 | static void cgroup_release_agent(struct work_struct *work) |
4580 | { | |
4581 | BUG_ON(work != &release_agent_work); | |
4582 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4583 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4584 | while (!list_empty(&release_list)) { |
4585 | char *argv[3], *envp[3]; | |
4586 | int i; | |
e61734c5 | 4587 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 4588 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4589 | struct cgroup, |
4590 | release_list); | |
bd89aabc | 4591 | list_del_init(&cgrp->release_list); |
cdcc136f | 4592 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 4593 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
4594 | if (!pathbuf) |
4595 | goto continue_free; | |
e61734c5 TH |
4596 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
4597 | if (!path) | |
e788e066 PM |
4598 | goto continue_free; |
4599 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4600 | if (!agentbuf) | |
4601 | goto continue_free; | |
81a6a5cd PM |
4602 | |
4603 | i = 0; | |
e788e066 | 4604 | argv[i++] = agentbuf; |
e61734c5 | 4605 | argv[i++] = path; |
81a6a5cd PM |
4606 | argv[i] = NULL; |
4607 | ||
4608 | i = 0; | |
4609 | /* minimal command environment */ | |
4610 | envp[i++] = "HOME=/"; | |
4611 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4612 | envp[i] = NULL; | |
4613 | ||
4614 | /* Drop the lock while we invoke the usermode helper, | |
4615 | * since the exec could involve hitting disk and hence | |
4616 | * be a slow process */ | |
4617 | mutex_unlock(&cgroup_mutex); | |
4618 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4619 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4620 | continue_free: |
4621 | kfree(pathbuf); | |
4622 | kfree(agentbuf); | |
cdcc136f | 4623 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4624 | } |
cdcc136f | 4625 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4626 | mutex_unlock(&cgroup_mutex); |
4627 | } | |
8bab8dde PM |
4628 | |
4629 | static int __init cgroup_disable(char *str) | |
4630 | { | |
30159ec7 | 4631 | struct cgroup_subsys *ss; |
8bab8dde | 4632 | char *token; |
30159ec7 | 4633 | int i; |
8bab8dde PM |
4634 | |
4635 | while ((token = strsep(&str, ",")) != NULL) { | |
4636 | if (!*token) | |
4637 | continue; | |
be45c900 | 4638 | |
3ed80a62 | 4639 | for_each_subsys(ss, i) { |
8bab8dde PM |
4640 | if (!strcmp(token, ss->name)) { |
4641 | ss->disabled = 1; | |
4642 | printk(KERN_INFO "Disabling %s control group" | |
4643 | " subsystem\n", ss->name); | |
4644 | break; | |
4645 | } | |
4646 | } | |
4647 | } | |
4648 | return 1; | |
4649 | } | |
4650 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 4651 | |
b77d7b60 | 4652 | /** |
5a17f543 | 4653 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
4654 | * @dentry: directory dentry of interest |
4655 | * @ss: subsystem of interest | |
b77d7b60 | 4656 | * |
5a17f543 TH |
4657 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
4658 | * to get the corresponding css and return it. If such css doesn't exist | |
4659 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 4660 | */ |
5a17f543 TH |
4661 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
4662 | struct cgroup_subsys *ss) | |
e5d1367f | 4663 | { |
2bd59d48 TH |
4664 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
4665 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 4666 | struct cgroup *cgrp; |
e5d1367f | 4667 | |
35cf0836 | 4668 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
4669 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
4670 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
4671 | return ERR_PTR(-EBADF); |
4672 | ||
5a17f543 TH |
4673 | rcu_read_lock(); |
4674 | ||
2bd59d48 TH |
4675 | /* |
4676 | * This path doesn't originate from kernfs and @kn could already | |
4677 | * have been or be removed at any point. @kn->priv is RCU | |
4678 | * protected for this access. See destroy_locked() for details. | |
4679 | */ | |
4680 | cgrp = rcu_dereference(kn->priv); | |
4681 | if (cgrp) | |
4682 | css = cgroup_css(cgrp, ss); | |
5a17f543 TH |
4683 | |
4684 | if (!css || !css_tryget(css)) | |
4685 | css = ERR_PTR(-ENOENT); | |
4686 | ||
4687 | rcu_read_unlock(); | |
4688 | return css; | |
e5d1367f | 4689 | } |
e5d1367f | 4690 | |
1cb650b9 LZ |
4691 | /** |
4692 | * css_from_id - lookup css by id | |
4693 | * @id: the cgroup id | |
4694 | * @ss: cgroup subsys to be looked into | |
4695 | * | |
4696 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
4697 | * Should be called under rcu_read_lock(). | |
4698 | */ | |
4699 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
4700 | { | |
4701 | struct cgroup *cgrp; | |
4702 | ||
ace2bee8 | 4703 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
4704 | |
4705 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
4706 | if (cgrp) | |
d1625964 | 4707 | return cgroup_css(cgrp, ss); |
1cb650b9 | 4708 | return NULL; |
e5d1367f SE |
4709 | } |
4710 | ||
fe693435 | 4711 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
4712 | static struct cgroup_subsys_state * |
4713 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
4714 | { |
4715 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
4716 | ||
4717 | if (!css) | |
4718 | return ERR_PTR(-ENOMEM); | |
4719 | ||
4720 | return css; | |
4721 | } | |
4722 | ||
eb95419b | 4723 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 4724 | { |
eb95419b | 4725 | kfree(css); |
fe693435 PM |
4726 | } |
4727 | ||
182446d0 TH |
4728 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
4729 | struct cftype *cft) | |
fe693435 | 4730 | { |
182446d0 | 4731 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
4732 | } |
4733 | ||
182446d0 TH |
4734 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
4735 | struct cftype *cft) | |
fe693435 PM |
4736 | { |
4737 | return (u64)(unsigned long)current->cgroups; | |
4738 | } | |
4739 | ||
182446d0 | 4740 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 4741 | struct cftype *cft) |
fe693435 PM |
4742 | { |
4743 | u64 count; | |
4744 | ||
4745 | rcu_read_lock(); | |
a8ad805c | 4746 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
4747 | rcu_read_unlock(); |
4748 | return count; | |
4749 | } | |
4750 | ||
2da8ca82 | 4751 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4752 | { |
69d0206c | 4753 | struct cgrp_cset_link *link; |
5abb8855 | 4754 | struct css_set *cset; |
e61734c5 TH |
4755 | char *name_buf; |
4756 | ||
4757 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
4758 | if (!name_buf) | |
4759 | return -ENOMEM; | |
7717f7ba | 4760 | |
96d365e0 | 4761 | down_read(&css_set_rwsem); |
7717f7ba | 4762 | rcu_read_lock(); |
5abb8855 | 4763 | cset = rcu_dereference(current->cgroups); |
69d0206c | 4764 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 4765 | struct cgroup *c = link->cgrp; |
7717f7ba | 4766 | |
a2dd4247 | 4767 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 4768 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 4769 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
4770 | } |
4771 | rcu_read_unlock(); | |
96d365e0 | 4772 | up_read(&css_set_rwsem); |
e61734c5 | 4773 | kfree(name_buf); |
7717f7ba PM |
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 | 4782 | |
96d365e0 | 4783 | down_read(&css_set_rwsem); |
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; | |
c7561128 | 4788 | |
5abb8855 | 4789 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 4790 | |
5abb8855 | 4791 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
4792 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
4793 | goto overflow; | |
4794 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
4795 | } | |
4796 | ||
4797 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
4798 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
4799 | goto overflow; | |
4800 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 4801 | } |
c7561128 TH |
4802 | continue; |
4803 | overflow: | |
4804 | seq_puts(seq, " ...\n"); | |
7717f7ba | 4805 | } |
96d365e0 | 4806 | up_read(&css_set_rwsem); |
7717f7ba PM |
4807 | return 0; |
4808 | } | |
4809 | ||
182446d0 | 4810 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 4811 | { |
182446d0 | 4812 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
4813 | } |
4814 | ||
4815 | static struct cftype debug_files[] = { | |
fe693435 PM |
4816 | { |
4817 | .name = "taskcount", | |
4818 | .read_u64 = debug_taskcount_read, | |
4819 | }, | |
4820 | ||
4821 | { | |
4822 | .name = "current_css_set", | |
4823 | .read_u64 = current_css_set_read, | |
4824 | }, | |
4825 | ||
4826 | { | |
4827 | .name = "current_css_set_refcount", | |
4828 | .read_u64 = current_css_set_refcount_read, | |
4829 | }, | |
4830 | ||
7717f7ba PM |
4831 | { |
4832 | .name = "current_css_set_cg_links", | |
2da8ca82 | 4833 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
4834 | }, |
4835 | ||
4836 | { | |
4837 | .name = "cgroup_css_links", | |
2da8ca82 | 4838 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
4839 | }, |
4840 | ||
fe693435 PM |
4841 | { |
4842 | .name = "releasable", | |
4843 | .read_u64 = releasable_read, | |
4844 | }, | |
fe693435 | 4845 | |
4baf6e33 TH |
4846 | { } /* terminate */ |
4847 | }; | |
fe693435 | 4848 | |
073219e9 | 4849 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
4850 | .css_alloc = debug_css_alloc, |
4851 | .css_free = debug_css_free, | |
4baf6e33 | 4852 | .base_cftypes = debug_files, |
fe693435 PM |
4853 | }; |
4854 | #endif /* CONFIG_CGROUP_DEBUG */ |