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> | |
26fc9cd2 | 36 | #include <linux/magic.h> |
ddbcc7e8 PM |
37 | #include <linux/mm.h> |
38 | #include <linux/mutex.h> | |
39 | #include <linux/mount.h> | |
40 | #include <linux/pagemap.h> | |
a424316c | 41 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
42 | #include <linux/rcupdate.h> |
43 | #include <linux/sched.h> | |
ddbcc7e8 | 44 | #include <linux/slab.h> |
ddbcc7e8 | 45 | #include <linux/spinlock.h> |
96d365e0 | 46 | #include <linux/rwsem.h> |
ddbcc7e8 | 47 | #include <linux/string.h> |
bbcb81d0 | 48 | #include <linux/sort.h> |
81a6a5cd | 49 | #include <linux/kmod.h> |
846c7bb0 BS |
50 | #include <linux/delayacct.h> |
51 | #include <linux/cgroupstats.h> | |
0ac801fe | 52 | #include <linux/hashtable.h> |
096b7fe0 | 53 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 54 | #include <linux/idr.h> |
d1d9fd33 | 55 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 56 | #include <linux/kthread.h> |
776f02fa | 57 | #include <linux/delay.h> |
846c7bb0 | 58 | |
60063497 | 59 | #include <linux/atomic.h> |
ddbcc7e8 | 60 | |
b1a21367 TH |
61 | /* |
62 | * pidlists linger the following amount before being destroyed. The goal | |
63 | * is avoiding frequent destruction in the middle of consecutive read calls | |
64 | * Expiring in the middle is a performance problem not a correctness one. | |
65 | * 1 sec should be enough. | |
66 | */ | |
67 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
68 | ||
8d7e6fb0 TH |
69 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
70 | MAX_CFTYPE_NAME + 2) | |
71 | ||
ace2bee8 TH |
72 | /* |
73 | * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file | |
74 | * creation/removal and hierarchy changing operations including cgroup | |
75 | * creation, removal, css association and controller rebinding. This outer | |
76 | * lock is needed mainly to resolve the circular dependency between kernfs | |
77 | * active ref and cgroup_mutex. cgroup_tree_mutex nests above both. | |
78 | */ | |
79 | static DEFINE_MUTEX(cgroup_tree_mutex); | |
80 | ||
e25e2cbb TH |
81 | /* |
82 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
83 | * hierarchy must be performed while holding it. | |
84 | * | |
0e1d768f TH |
85 | * css_set_rwsem protects task->cgroups pointer, the list of css_set |
86 | * objects, and the chain of tasks off each css_set. | |
e25e2cbb | 87 | * |
0e1d768f TH |
88 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
89 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 90 | */ |
2219449a TH |
91 | #ifdef CONFIG_PROVE_RCU |
92 | DEFINE_MUTEX(cgroup_mutex); | |
0e1d768f TH |
93 | DECLARE_RWSEM(css_set_rwsem); |
94 | EXPORT_SYMBOL_GPL(cgroup_mutex); | |
95 | EXPORT_SYMBOL_GPL(css_set_rwsem); | |
2219449a | 96 | #else |
81a6a5cd | 97 | static DEFINE_MUTEX(cgroup_mutex); |
0e1d768f | 98 | static DECLARE_RWSEM(css_set_rwsem); |
2219449a TH |
99 | #endif |
100 | ||
69e943b7 TH |
101 | /* |
102 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
103 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
104 | */ | |
105 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 106 | |
ace2bee8 | 107 | #define cgroup_assert_mutexes_or_rcu_locked() \ |
87fb54f1 | 108 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
ace2bee8 | 109 | lockdep_is_held(&cgroup_tree_mutex) || \ |
87fb54f1 | 110 | lockdep_is_held(&cgroup_mutex), \ |
ace2bee8 | 111 | "cgroup_[tree_]mutex or RCU read lock required"); |
780cd8b3 | 112 | |
e5fca243 TH |
113 | /* |
114 | * cgroup destruction makes heavy use of work items and there can be a lot | |
115 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
116 | * destruction work items don't end up filling up max_active of system_wq | |
117 | * which may lead to deadlock. | |
118 | */ | |
119 | static struct workqueue_struct *cgroup_destroy_wq; | |
120 | ||
b1a21367 TH |
121 | /* |
122 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
123 | * separate workqueue as flush domain. | |
124 | */ | |
125 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
126 | ||
3ed80a62 | 127 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 128 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 129 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
130 | #include <linux/cgroup_subsys.h> |
131 | }; | |
073219e9 TH |
132 | #undef SUBSYS |
133 | ||
134 | /* array of cgroup subsystem names */ | |
135 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
136 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
137 | #include <linux/cgroup_subsys.h> |
138 | }; | |
073219e9 | 139 | #undef SUBSYS |
ddbcc7e8 | 140 | |
ddbcc7e8 | 141 | /* |
3dd06ffa | 142 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
143 | * unattached - it never has more than a single cgroup, and all tasks are |
144 | * part of that cgroup. | |
ddbcc7e8 | 145 | */ |
a2dd4247 | 146 | struct cgroup_root cgrp_dfl_root; |
9871bf95 | 147 | |
a2dd4247 TH |
148 | /* |
149 | * The default hierarchy always exists but is hidden until mounted for the | |
150 | * first time. This is for backward compatibility. | |
151 | */ | |
152 | static bool cgrp_dfl_root_visible; | |
ddbcc7e8 PM |
153 | |
154 | /* The list of hierarchy roots */ | |
155 | ||
9871bf95 TH |
156 | static LIST_HEAD(cgroup_roots); |
157 | static int cgroup_root_count; | |
ddbcc7e8 | 158 | |
3417ae1f | 159 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 160 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 161 | |
794611a1 LZ |
162 | /* |
163 | * Assign a monotonically increasing serial number to cgroups. It | |
164 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
165 | * numbers. Also, as cgroups are always appended to the parent's | |
166 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
167 | * the ascending serial number order on the list. Protected by |
168 | * cgroup_mutex. | |
794611a1 | 169 | */ |
00356bd5 | 170 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 171 | |
ddbcc7e8 | 172 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
173 | * check for fork/exit handlers to call. This avoids us having to do |
174 | * extra work in the fork/exit path if none of the subsystems need to | |
175 | * be called. | |
ddbcc7e8 | 176 | */ |
8947f9d5 | 177 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 178 | |
628f7cd4 TH |
179 | static struct cftype cgroup_base_files[]; |
180 | ||
59f5296b | 181 | static void cgroup_put(struct cgroup *cgrp); |
3dd06ffa | 182 | static int rebind_subsystems(struct cgroup_root *dst_root, |
5df36032 | 183 | unsigned long ss_mask); |
f20104de | 184 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 185 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
2bb566cb TH |
186 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
187 | bool is_add); | |
b1a21367 | 188 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 189 | |
95109b62 TH |
190 | /** |
191 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
192 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 193 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 194 | * |
ca8bdcaf TH |
195 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
196 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
197 | * the caller is responsible for pinning the returned css if it wants to | |
198 | * keep accessing it outside the said locks. This function may return | |
199 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
200 | */ |
201 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 202 | struct cgroup_subsys *ss) |
95109b62 | 203 | { |
ca8bdcaf | 204 | if (ss) |
aec25020 | 205 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 TH |
206 | lockdep_is_held(&cgroup_tree_mutex) || |
207 | lockdep_is_held(&cgroup_mutex)); | |
ca8bdcaf TH |
208 | else |
209 | return &cgrp->dummy_css; | |
95109b62 | 210 | } |
42809dd4 | 211 | |
ddbcc7e8 | 212 | /* convenient tests for these bits */ |
54766d4a | 213 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 214 | { |
54766d4a | 215 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
216 | } |
217 | ||
59f5296b TH |
218 | struct cgroup_subsys_state *seq_css(struct seq_file *seq) |
219 | { | |
2bd59d48 TH |
220 | struct kernfs_open_file *of = seq->private; |
221 | struct cgroup *cgrp = of->kn->parent->priv; | |
222 | struct cftype *cft = seq_cft(seq); | |
223 | ||
224 | /* | |
225 | * This is open and unprotected implementation of cgroup_css(). | |
226 | * seq_css() is only called from a kernfs file operation which has | |
227 | * an active reference on the file. Because all the subsystem | |
228 | * files are drained before a css is disassociated with a cgroup, | |
229 | * the matching css from the cgroup's subsys table is guaranteed to | |
230 | * be and stay valid until the enclosing operation is complete. | |
231 | */ | |
232 | if (cft->ss) | |
233 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
234 | else | |
235 | return &cgrp->dummy_css; | |
59f5296b TH |
236 | } |
237 | EXPORT_SYMBOL_GPL(seq_css); | |
238 | ||
78574cf9 LZ |
239 | /** |
240 | * cgroup_is_descendant - test ancestry | |
241 | * @cgrp: the cgroup to be tested | |
242 | * @ancestor: possible ancestor of @cgrp | |
243 | * | |
244 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
245 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
246 | * and @ancestor are accessible. | |
247 | */ | |
248 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
249 | { | |
250 | while (cgrp) { | |
251 | if (cgrp == ancestor) | |
252 | return true; | |
253 | cgrp = cgrp->parent; | |
254 | } | |
255 | return false; | |
256 | } | |
ddbcc7e8 | 257 | |
e9685a03 | 258 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
259 | { |
260 | const int bits = | |
bd89aabc PM |
261 | (1 << CGRP_RELEASABLE) | |
262 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
263 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
264 | } |
265 | ||
e9685a03 | 266 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 267 | { |
bd89aabc | 268 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
269 | } |
270 | ||
1c6727af TH |
271 | /** |
272 | * for_each_css - iterate all css's of a cgroup | |
273 | * @css: the iteration cursor | |
274 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
275 | * @cgrp: the target cgroup to iterate css's of | |
276 | * | |
277 | * Should be called under cgroup_mutex. | |
278 | */ | |
279 | #define for_each_css(css, ssid, cgrp) \ | |
280 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
281 | if (!((css) = rcu_dereference_check( \ | |
282 | (cgrp)->subsys[(ssid)], \ | |
ace2bee8 | 283 | lockdep_is_held(&cgroup_tree_mutex) || \ |
1c6727af TH |
284 | lockdep_is_held(&cgroup_mutex)))) { } \ |
285 | else | |
286 | ||
30159ec7 | 287 | /** |
3ed80a62 | 288 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 289 | * @ss: the iteration cursor |
780cd8b3 | 290 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 291 | */ |
780cd8b3 | 292 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
293 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
294 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 295 | |
985ed670 TH |
296 | /* iterate across the hierarchies */ |
297 | #define for_each_root(root) \ | |
5549c497 | 298 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 299 | |
7ae1bad9 TH |
300 | /** |
301 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
302 | * @cgrp: the cgroup to be checked for liveness | |
303 | * | |
47cfcd09 TH |
304 | * On success, returns true; the mutex should be later unlocked. On |
305 | * failure returns false with no lock held. | |
7ae1bad9 | 306 | */ |
b9777cf8 | 307 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
308 | { |
309 | mutex_lock(&cgroup_mutex); | |
54766d4a | 310 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
311 | mutex_unlock(&cgroup_mutex); |
312 | return false; | |
313 | } | |
314 | return true; | |
315 | } | |
7ae1bad9 | 316 | |
81a6a5cd PM |
317 | /* the list of cgroups eligible for automatic release. Protected by |
318 | * release_list_lock */ | |
319 | static LIST_HEAD(release_list); | |
cdcc136f | 320 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
321 | static void cgroup_release_agent(struct work_struct *work); |
322 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 323 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 324 | |
69d0206c TH |
325 | /* |
326 | * A cgroup can be associated with multiple css_sets as different tasks may | |
327 | * belong to different cgroups on different hierarchies. In the other | |
328 | * direction, a css_set is naturally associated with multiple cgroups. | |
329 | * This M:N relationship is represented by the following link structure | |
330 | * which exists for each association and allows traversing the associations | |
331 | * from both sides. | |
332 | */ | |
333 | struct cgrp_cset_link { | |
334 | /* the cgroup and css_set this link associates */ | |
335 | struct cgroup *cgrp; | |
336 | struct css_set *cset; | |
337 | ||
338 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
339 | struct list_head cset_link; | |
340 | ||
341 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
342 | struct list_head cgrp_link; | |
817929ec PM |
343 | }; |
344 | ||
172a2c06 TH |
345 | /* |
346 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
347 | * hierarchies being mounted. It contains a pointer to the root state |
348 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
349 | * reference-counted, to improve performance when child cgroups | |
350 | * haven't been created. | |
351 | */ | |
5024ae29 | 352 | struct css_set init_css_set = { |
172a2c06 TH |
353 | .refcount = ATOMIC_INIT(1), |
354 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
355 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
356 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
357 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
358 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
359 | }; | |
817929ec | 360 | |
172a2c06 | 361 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 362 | |
7717f7ba PM |
363 | /* |
364 | * hash table for cgroup groups. This improves the performance to find | |
365 | * an existing css_set. This hash doesn't (currently) take into | |
366 | * account cgroups in empty hierarchies. | |
367 | */ | |
472b1053 | 368 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 369 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 370 | |
0ac801fe | 371 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 372 | { |
0ac801fe | 373 | unsigned long key = 0UL; |
30159ec7 TH |
374 | struct cgroup_subsys *ss; |
375 | int i; | |
472b1053 | 376 | |
30159ec7 | 377 | for_each_subsys(ss, i) |
0ac801fe LZ |
378 | key += (unsigned long)css[i]; |
379 | key = (key >> 16) ^ key; | |
472b1053 | 380 | |
0ac801fe | 381 | return key; |
472b1053 LZ |
382 | } |
383 | ||
89c5509b | 384 | static void put_css_set_locked(struct css_set *cset, bool taskexit) |
b4f48b63 | 385 | { |
69d0206c | 386 | struct cgrp_cset_link *link, *tmp_link; |
5abb8855 | 387 | |
89c5509b TH |
388 | lockdep_assert_held(&css_set_rwsem); |
389 | ||
390 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 391 | return; |
81a6a5cd | 392 | |
2c6ab6d2 | 393 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
5abb8855 | 394 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
395 | css_set_count--; |
396 | ||
69d0206c | 397 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 398 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 399 | |
69d0206c TH |
400 | list_del(&link->cset_link); |
401 | list_del(&link->cgrp_link); | |
71b5707e | 402 | |
96d365e0 | 403 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
6f3d828f | 404 | if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { |
81a6a5cd | 405 | if (taskexit) |
bd89aabc PM |
406 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
407 | check_for_release(cgrp); | |
81a6a5cd | 408 | } |
2c6ab6d2 PM |
409 | |
410 | kfree(link); | |
81a6a5cd | 411 | } |
2c6ab6d2 | 412 | |
5abb8855 | 413 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
414 | } |
415 | ||
89c5509b TH |
416 | static void put_css_set(struct css_set *cset, bool taskexit) |
417 | { | |
418 | /* | |
419 | * Ensure that the refcount doesn't hit zero while any readers | |
420 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
421 | * rwlock | |
422 | */ | |
423 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
424 | return; | |
425 | ||
426 | down_write(&css_set_rwsem); | |
427 | put_css_set_locked(cset, taskexit); | |
428 | up_write(&css_set_rwsem); | |
429 | } | |
430 | ||
817929ec PM |
431 | /* |
432 | * refcounted get/put for css_set objects | |
433 | */ | |
5abb8855 | 434 | static inline void get_css_set(struct css_set *cset) |
817929ec | 435 | { |
5abb8855 | 436 | atomic_inc(&cset->refcount); |
817929ec PM |
437 | } |
438 | ||
b326f9d0 | 439 | /** |
7717f7ba | 440 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
441 | * @cset: candidate css_set being tested |
442 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
443 | * @new_cgrp: cgroup that's being entered by the task |
444 | * @template: desired set of css pointers in css_set (pre-calculated) | |
445 | * | |
6f4b7e63 | 446 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
447 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
448 | */ | |
5abb8855 TH |
449 | static bool compare_css_sets(struct css_set *cset, |
450 | struct css_set *old_cset, | |
7717f7ba PM |
451 | struct cgroup *new_cgrp, |
452 | struct cgroup_subsys_state *template[]) | |
453 | { | |
454 | struct list_head *l1, *l2; | |
455 | ||
5abb8855 | 456 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) { |
7717f7ba PM |
457 | /* Not all subsystems matched */ |
458 | return false; | |
459 | } | |
460 | ||
461 | /* | |
462 | * Compare cgroup pointers in order to distinguish between | |
463 | * different cgroups in heirarchies with no subsystems. We | |
464 | * could get by with just this check alone (and skip the | |
465 | * memcmp above) but on most setups the memcmp check will | |
466 | * avoid the need for this more expensive check on almost all | |
467 | * candidates. | |
468 | */ | |
469 | ||
69d0206c TH |
470 | l1 = &cset->cgrp_links; |
471 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 472 | while (1) { |
69d0206c | 473 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 474 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
475 | |
476 | l1 = l1->next; | |
477 | l2 = l2->next; | |
478 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
479 | if (l1 == &cset->cgrp_links) { |
480 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
481 | break; |
482 | } else { | |
69d0206c | 483 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
484 | } |
485 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
486 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
487 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
488 | cgrp1 = link1->cgrp; | |
489 | cgrp2 = link2->cgrp; | |
7717f7ba | 490 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 491 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
492 | |
493 | /* | |
494 | * If this hierarchy is the hierarchy of the cgroup | |
495 | * that's changing, then we need to check that this | |
496 | * css_set points to the new cgroup; if it's any other | |
497 | * hierarchy, then this css_set should point to the | |
498 | * same cgroup as the old css_set. | |
499 | */ | |
5abb8855 TH |
500 | if (cgrp1->root == new_cgrp->root) { |
501 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
502 | return false; |
503 | } else { | |
5abb8855 | 504 | if (cgrp1 != cgrp2) |
7717f7ba PM |
505 | return false; |
506 | } | |
507 | } | |
508 | return true; | |
509 | } | |
510 | ||
b326f9d0 TH |
511 | /** |
512 | * find_existing_css_set - init css array and find the matching css_set | |
513 | * @old_cset: the css_set that we're using before the cgroup transition | |
514 | * @cgrp: the cgroup that we're moving into | |
515 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 516 | */ |
5abb8855 TH |
517 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
518 | struct cgroup *cgrp, | |
519 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 520 | { |
3dd06ffa | 521 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 522 | struct cgroup_subsys *ss; |
5abb8855 | 523 | struct css_set *cset; |
0ac801fe | 524 | unsigned long key; |
b326f9d0 | 525 | int i; |
817929ec | 526 | |
aae8aab4 BB |
527 | /* |
528 | * Build the set of subsystem state objects that we want to see in the | |
529 | * new css_set. while subsystems can change globally, the entries here | |
530 | * won't change, so no need for locking. | |
531 | */ | |
30159ec7 | 532 | for_each_subsys(ss, i) { |
3dd06ffa | 533 | if (root->cgrp.subsys_mask & (1UL << i)) { |
817929ec PM |
534 | /* Subsystem is in this hierarchy. So we want |
535 | * the subsystem state from the new | |
536 | * cgroup */ | |
ca8bdcaf | 537 | template[i] = cgroup_css(cgrp, ss); |
817929ec PM |
538 | } else { |
539 | /* Subsystem is not in this hierarchy, so we | |
540 | * don't want to change the subsystem state */ | |
5abb8855 | 541 | template[i] = old_cset->subsys[i]; |
817929ec PM |
542 | } |
543 | } | |
544 | ||
0ac801fe | 545 | key = css_set_hash(template); |
5abb8855 TH |
546 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
547 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
548 | continue; |
549 | ||
550 | /* This css_set matches what we need */ | |
5abb8855 | 551 | return cset; |
472b1053 | 552 | } |
817929ec PM |
553 | |
554 | /* No existing cgroup group matched */ | |
555 | return NULL; | |
556 | } | |
557 | ||
69d0206c | 558 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 559 | { |
69d0206c | 560 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 561 | |
69d0206c TH |
562 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
563 | list_del(&link->cset_link); | |
36553434 LZ |
564 | kfree(link); |
565 | } | |
566 | } | |
567 | ||
69d0206c TH |
568 | /** |
569 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
570 | * @count: the number of links to allocate | |
571 | * @tmp_links: list_head the allocated links are put on | |
572 | * | |
573 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
574 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 575 | */ |
69d0206c | 576 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 577 | { |
69d0206c | 578 | struct cgrp_cset_link *link; |
817929ec | 579 | int i; |
69d0206c TH |
580 | |
581 | INIT_LIST_HEAD(tmp_links); | |
582 | ||
817929ec | 583 | for (i = 0; i < count; i++) { |
f4f4be2b | 584 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 585 | if (!link) { |
69d0206c | 586 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
587 | return -ENOMEM; |
588 | } | |
69d0206c | 589 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
590 | } |
591 | return 0; | |
592 | } | |
593 | ||
c12f65d4 LZ |
594 | /** |
595 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 596 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 597 | * @cset: the css_set to be linked |
c12f65d4 LZ |
598 | * @cgrp: the destination cgroup |
599 | */ | |
69d0206c TH |
600 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
601 | struct cgroup *cgrp) | |
c12f65d4 | 602 | { |
69d0206c | 603 | struct cgrp_cset_link *link; |
c12f65d4 | 604 | |
69d0206c TH |
605 | BUG_ON(list_empty(tmp_links)); |
606 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); | |
607 | link->cset = cset; | |
7717f7ba | 608 | link->cgrp = cgrp; |
69d0206c | 609 | list_move(&link->cset_link, &cgrp->cset_links); |
7717f7ba PM |
610 | /* |
611 | * Always add links to the tail of the list so that the list | |
612 | * is sorted by order of hierarchy creation | |
613 | */ | |
69d0206c | 614 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
615 | } |
616 | ||
b326f9d0 TH |
617 | /** |
618 | * find_css_set - return a new css_set with one cgroup updated | |
619 | * @old_cset: the baseline css_set | |
620 | * @cgrp: the cgroup to be updated | |
621 | * | |
622 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
623 | * substituted into the appropriate hierarchy. | |
817929ec | 624 | */ |
5abb8855 TH |
625 | static struct css_set *find_css_set(struct css_set *old_cset, |
626 | struct cgroup *cgrp) | |
817929ec | 627 | { |
b326f9d0 | 628 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 629 | struct css_set *cset; |
69d0206c TH |
630 | struct list_head tmp_links; |
631 | struct cgrp_cset_link *link; | |
0ac801fe | 632 | unsigned long key; |
472b1053 | 633 | |
b326f9d0 TH |
634 | lockdep_assert_held(&cgroup_mutex); |
635 | ||
817929ec PM |
636 | /* First see if we already have a cgroup group that matches |
637 | * the desired set */ | |
96d365e0 | 638 | down_read(&css_set_rwsem); |
5abb8855 TH |
639 | cset = find_existing_css_set(old_cset, cgrp, template); |
640 | if (cset) | |
641 | get_css_set(cset); | |
96d365e0 | 642 | up_read(&css_set_rwsem); |
817929ec | 643 | |
5abb8855 TH |
644 | if (cset) |
645 | return cset; | |
817929ec | 646 | |
f4f4be2b | 647 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 648 | if (!cset) |
817929ec PM |
649 | return NULL; |
650 | ||
69d0206c | 651 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 652 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 653 | kfree(cset); |
817929ec PM |
654 | return NULL; |
655 | } | |
656 | ||
5abb8855 | 657 | atomic_set(&cset->refcount, 1); |
69d0206c | 658 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 659 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 660 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 661 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 662 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 663 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
664 | |
665 | /* Copy the set of subsystem state objects generated in | |
666 | * find_existing_css_set() */ | |
5abb8855 | 667 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 668 | |
96d365e0 | 669 | down_write(&css_set_rwsem); |
817929ec | 670 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 671 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 672 | struct cgroup *c = link->cgrp; |
69d0206c | 673 | |
7717f7ba PM |
674 | if (c->root == cgrp->root) |
675 | c = cgrp; | |
69d0206c | 676 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 677 | } |
817929ec | 678 | |
69d0206c | 679 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 680 | |
817929ec | 681 | css_set_count++; |
472b1053 LZ |
682 | |
683 | /* Add this cgroup group to the hash table */ | |
5abb8855 TH |
684 | key = css_set_hash(cset->subsys); |
685 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 686 | |
96d365e0 | 687 | up_write(&css_set_rwsem); |
817929ec | 688 | |
5abb8855 | 689 | return cset; |
b4f48b63 PM |
690 | } |
691 | ||
3dd06ffa | 692 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 693 | { |
3dd06ffa | 694 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 695 | |
3dd06ffa | 696 | return root_cgrp->root; |
2bd59d48 TH |
697 | } |
698 | ||
3dd06ffa | 699 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
700 | { |
701 | int id; | |
702 | ||
703 | lockdep_assert_held(&cgroup_mutex); | |
704 | ||
985ed670 | 705 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
706 | if (id < 0) |
707 | return id; | |
708 | ||
709 | root->hierarchy_id = id; | |
710 | return 0; | |
711 | } | |
712 | ||
3dd06ffa | 713 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
714 | { |
715 | lockdep_assert_held(&cgroup_mutex); | |
716 | ||
717 | if (root->hierarchy_id) { | |
718 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
719 | root->hierarchy_id = 0; | |
720 | } | |
721 | } | |
722 | ||
3dd06ffa | 723 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
724 | { |
725 | if (root) { | |
726 | /* hierarhcy ID shoulid already have been released */ | |
727 | WARN_ON_ONCE(root->hierarchy_id); | |
728 | ||
729 | idr_destroy(&root->cgroup_idr); | |
730 | kfree(root); | |
731 | } | |
732 | } | |
733 | ||
3dd06ffa | 734 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 735 | { |
3dd06ffa | 736 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 737 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 738 | |
2bd59d48 | 739 | mutex_lock(&cgroup_tree_mutex); |
2bd59d48 | 740 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 741 | |
776f02fa | 742 | BUG_ON(atomic_read(&root->nr_cgrps)); |
f2e85d57 TH |
743 | BUG_ON(!list_empty(&cgrp->children)); |
744 | ||
f2e85d57 | 745 | /* Rebind all subsystems back to the default hierarchy */ |
3dd06ffa | 746 | rebind_subsystems(&cgrp_dfl_root, cgrp->subsys_mask); |
7717f7ba | 747 | |
7717f7ba | 748 | /* |
f2e85d57 TH |
749 | * Release all the links from cset_links to this hierarchy's |
750 | * root cgroup | |
7717f7ba | 751 | */ |
96d365e0 | 752 | down_write(&css_set_rwsem); |
f2e85d57 TH |
753 | |
754 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
755 | list_del(&link->cset_link); | |
756 | list_del(&link->cgrp_link); | |
757 | kfree(link); | |
758 | } | |
96d365e0 | 759 | up_write(&css_set_rwsem); |
f2e85d57 TH |
760 | |
761 | if (!list_empty(&root->root_list)) { | |
762 | list_del(&root->root_list); | |
763 | cgroup_root_count--; | |
764 | } | |
765 | ||
766 | cgroup_exit_root_id(root); | |
767 | ||
768 | mutex_unlock(&cgroup_mutex); | |
769 | mutex_unlock(&cgroup_tree_mutex); | |
f2e85d57 | 770 | |
2bd59d48 | 771 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
772 | cgroup_free_root(root); |
773 | } | |
774 | ||
ceb6a081 TH |
775 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
776 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 777 | struct cgroup_root *root) |
7717f7ba | 778 | { |
7717f7ba PM |
779 | struct cgroup *res = NULL; |
780 | ||
96d365e0 TH |
781 | lockdep_assert_held(&cgroup_mutex); |
782 | lockdep_assert_held(&css_set_rwsem); | |
783 | ||
5abb8855 | 784 | if (cset == &init_css_set) { |
3dd06ffa | 785 | res = &root->cgrp; |
7717f7ba | 786 | } else { |
69d0206c TH |
787 | struct cgrp_cset_link *link; |
788 | ||
789 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 790 | struct cgroup *c = link->cgrp; |
69d0206c | 791 | |
7717f7ba PM |
792 | if (c->root == root) { |
793 | res = c; | |
794 | break; | |
795 | } | |
796 | } | |
797 | } | |
96d365e0 | 798 | |
7717f7ba PM |
799 | BUG_ON(!res); |
800 | return res; | |
801 | } | |
802 | ||
ddbcc7e8 | 803 | /* |
ceb6a081 TH |
804 | * Return the cgroup for "task" from the given hierarchy. Must be |
805 | * called with cgroup_mutex and css_set_rwsem held. | |
806 | */ | |
807 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 808 | struct cgroup_root *root) |
ceb6a081 TH |
809 | { |
810 | /* | |
811 | * No need to lock the task - since we hold cgroup_mutex the | |
812 | * task can't change groups, so the only thing that can happen | |
813 | * is that it exits and its css is set back to init_css_set. | |
814 | */ | |
815 | return cset_cgroup_from_root(task_css_set(task), root); | |
816 | } | |
817 | ||
ddbcc7e8 | 818 | /* |
ddbcc7e8 PM |
819 | * A task must hold cgroup_mutex to modify cgroups. |
820 | * | |
821 | * Any task can increment and decrement the count field without lock. | |
822 | * So in general, code holding cgroup_mutex can't rely on the count | |
823 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 824 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
825 | * means that no tasks are currently attached, therefore there is no |
826 | * way a task attached to that cgroup can fork (the other way to | |
827 | * increment the count). So code holding cgroup_mutex can safely | |
828 | * assume that if the count is zero, it will stay zero. Similarly, if | |
829 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
830 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
831 | * needs that mutex. | |
832 | * | |
ddbcc7e8 PM |
833 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
834 | * (usually) take cgroup_mutex. These are the two most performance | |
835 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
836 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
837 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
838 | * to the release agent with the name of the cgroup (path relative to |
839 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
840 | * |
841 | * A cgroup can only be deleted if both its 'count' of using tasks | |
842 | * is zero, and its list of 'children' cgroups is empty. Since all | |
843 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 844 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 845 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 846 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
847 | * |
848 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 849 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
850 | */ |
851 | ||
628f7cd4 | 852 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
2bd59d48 | 853 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 854 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 855 | |
8d7e6fb0 TH |
856 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
857 | char *buf) | |
ddbcc7e8 | 858 | { |
8d7e6fb0 TH |
859 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
860 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
861 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
862 | cft->ss->name, cft->name); | |
863 | else | |
864 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
865 | return buf; | |
ddbcc7e8 PM |
866 | } |
867 | ||
f2e85d57 TH |
868 | /** |
869 | * cgroup_file_mode - deduce file mode of a control file | |
870 | * @cft: the control file in question | |
871 | * | |
872 | * returns cft->mode if ->mode is not 0 | |
873 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
874 | * returns S_IRUGO if it has only a read handler | |
875 | * returns S_IWUSR if it has only a write hander | |
876 | */ | |
877 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 878 | { |
f2e85d57 | 879 | umode_t mode = 0; |
65dff759 | 880 | |
f2e85d57 TH |
881 | if (cft->mode) |
882 | return cft->mode; | |
883 | ||
884 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
885 | mode |= S_IRUGO; | |
886 | ||
887 | if (cft->write_u64 || cft->write_s64 || cft->write_string || | |
888 | cft->trigger) | |
889 | mode |= S_IWUSR; | |
890 | ||
891 | return mode; | |
65dff759 LZ |
892 | } |
893 | ||
be445626 LZ |
894 | static void cgroup_free_fn(struct work_struct *work) |
895 | { | |
ea15f8cc | 896 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 | 897 | |
3c9c825b | 898 | atomic_dec(&cgrp->root->nr_cgrps); |
b1a21367 | 899 | cgroup_pidlist_destroy_all(cgrp); |
be445626 | 900 | |
776f02fa TH |
901 | if (cgrp->parent) { |
902 | /* | |
903 | * We get a ref to the parent, and put the ref when this | |
904 | * cgroup is being freed, so it's guaranteed that the | |
905 | * parent won't be destroyed before its children. | |
906 | */ | |
907 | cgroup_put(cgrp->parent); | |
908 | kernfs_put(cgrp->kn); | |
909 | kfree(cgrp); | |
910 | } else { | |
911 | /* | |
3dd06ffa | 912 | * This is root cgroup's refcnt reaching zero, which |
776f02fa TH |
913 | * indicates that the root should be released. |
914 | */ | |
915 | cgroup_destroy_root(cgrp->root); | |
916 | } | |
be445626 LZ |
917 | } |
918 | ||
919 | static void cgroup_free_rcu(struct rcu_head *head) | |
920 | { | |
921 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
922 | ||
ea15f8cc | 923 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 924 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
925 | } |
926 | ||
59f5296b | 927 | static void cgroup_get(struct cgroup *cgrp) |
ddbcc7e8 | 928 | { |
2bd59d48 TH |
929 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
930 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
931 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
932 | } |
933 | ||
59f5296b | 934 | static void cgroup_put(struct cgroup *cgrp) |
05ef1d7c | 935 | { |
2bd59d48 TH |
936 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
937 | return; | |
776f02fa | 938 | if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp))) |
2bd59d48 | 939 | return; |
05ef1d7c | 940 | |
2739d3cc | 941 | /* |
2bd59d48 TH |
942 | * XXX: cgrp->id is only used to look up css's. As cgroup and |
943 | * css's lifetimes will be decoupled, it should be made | |
944 | * per-subsystem and moved to css->id so that lookups are | |
945 | * successful until the target css is released. | |
2739d3cc | 946 | */ |
2bd59d48 TH |
947 | mutex_lock(&cgroup_mutex); |
948 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
949 | mutex_unlock(&cgroup_mutex); | |
950 | cgrp->id = -1; | |
05ef1d7c | 951 | |
2bd59d48 | 952 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 | 953 | } |
05ef1d7c | 954 | |
2739d3cc | 955 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 956 | { |
2bd59d48 | 957 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 958 | |
ace2bee8 | 959 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 960 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
961 | } |
962 | ||
13af07df | 963 | /** |
628f7cd4 | 964 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 965 | * @cgrp: target cgroup |
13af07df AR |
966 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
967 | */ | |
628f7cd4 | 968 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 969 | { |
13af07df | 970 | struct cgroup_subsys *ss; |
b420ba7d | 971 | int i; |
05ef1d7c | 972 | |
b420ba7d | 973 | for_each_subsys(ss, i) { |
0adb0704 | 974 | struct cftype *cfts; |
b420ba7d TH |
975 | |
976 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 977 | continue; |
0adb0704 TH |
978 | list_for_each_entry(cfts, &ss->cfts, node) |
979 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 980 | } |
ddbcc7e8 PM |
981 | } |
982 | ||
3dd06ffa | 983 | static int rebind_subsystems(struct cgroup_root *dst_root, |
5df36032 | 984 | unsigned long ss_mask) |
ddbcc7e8 | 985 | { |
30159ec7 | 986 | struct cgroup_subsys *ss; |
5df36032 | 987 | int ssid, ret; |
ddbcc7e8 | 988 | |
ace2bee8 TH |
989 | lockdep_assert_held(&cgroup_tree_mutex); |
990 | lockdep_assert_held(&cgroup_mutex); | |
ddbcc7e8 | 991 | |
5df36032 TH |
992 | for_each_subsys(ss, ssid) { |
993 | if (!(ss_mask & (1 << ssid))) | |
994 | continue; | |
aae8aab4 | 995 | |
5df36032 | 996 | /* if @ss is on the dummy_root, we can always move it */ |
3dd06ffa | 997 | if (ss->root == &cgrp_dfl_root) |
ddbcc7e8 | 998 | continue; |
30159ec7 | 999 | |
5df36032 | 1000 | /* if @ss has non-root cgroups attached to it, can't move */ |
3dd06ffa | 1001 | if (!list_empty(&ss->root->cgrp.children)) |
3ed80a62 | 1002 | return -EBUSY; |
1d5be6b2 | 1003 | |
5df36032 | 1004 | /* can't move between two non-dummy roots either */ |
3dd06ffa | 1005 | if (dst_root != &cgrp_dfl_root) |
5df36032 | 1006 | return -EBUSY; |
ddbcc7e8 PM |
1007 | } |
1008 | ||
a2dd4247 TH |
1009 | ret = cgroup_populate_dir(&dst_root->cgrp, ss_mask); |
1010 | if (ret) { | |
1011 | if (dst_root != &cgrp_dfl_root) | |
5df36032 | 1012 | return ret; |
ddbcc7e8 | 1013 | |
a2dd4247 TH |
1014 | /* |
1015 | * Rebinding back to the default root is not allowed to | |
1016 | * fail. Using both default and non-default roots should | |
1017 | * be rare. Moving subsystems back and forth even more so. | |
1018 | * Just warn about it and continue. | |
1019 | */ | |
1020 | if (cgrp_dfl_root_visible) { | |
1021 | pr_warning("cgroup: failed to create files (%d) while rebinding 0x%lx to default root\n", | |
1022 | ret, ss_mask); | |
1023 | pr_warning("cgroup: you may retry by moving them to a different hierarchy and unbinding\n"); | |
1024 | } | |
5df36032 | 1025 | } |
3126121f TH |
1026 | |
1027 | /* | |
1028 | * Nothing can fail from this point on. Remove files for the | |
1029 | * removed subsystems and rebind each subsystem. | |
1030 | */ | |
4ac06017 | 1031 | mutex_unlock(&cgroup_mutex); |
5df36032 | 1032 | for_each_subsys(ss, ssid) |
a2dd4247 | 1033 | if (ss_mask & (1 << ssid)) |
3dd06ffa | 1034 | cgroup_clear_dir(&ss->root->cgrp, 1 << ssid); |
4ac06017 | 1035 | mutex_lock(&cgroup_mutex); |
a8a648c4 | 1036 | |
5df36032 | 1037 | for_each_subsys(ss, ssid) { |
3dd06ffa | 1038 | struct cgroup_root *src_root; |
5df36032 | 1039 | struct cgroup_subsys_state *css; |
a8a648c4 | 1040 | |
5df36032 TH |
1041 | if (!(ss_mask & (1 << ssid))) |
1042 | continue; | |
a8a648c4 | 1043 | |
5df36032 | 1044 | src_root = ss->root; |
3dd06ffa | 1045 | css = cgroup_css(&src_root->cgrp, ss); |
a8a648c4 | 1046 | |
3dd06ffa | 1047 | WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss)); |
73e80ed8 | 1048 | |
3dd06ffa TH |
1049 | RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL); |
1050 | rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css); | |
5df36032 | 1051 | ss->root = dst_root; |
3dd06ffa | 1052 | css->cgroup = &dst_root->cgrp; |
73e80ed8 | 1053 | |
3dd06ffa TH |
1054 | src_root->cgrp.subsys_mask &= ~(1 << ssid); |
1055 | dst_root->cgrp.subsys_mask |= 1 << ssid; | |
a8a648c4 | 1056 | |
5df36032 TH |
1057 | if (ss->bind) |
1058 | ss->bind(css); | |
ddbcc7e8 | 1059 | } |
ddbcc7e8 | 1060 | |
a2dd4247 | 1061 | kernfs_activate(dst_root->cgrp.kn); |
ddbcc7e8 PM |
1062 | return 0; |
1063 | } | |
1064 | ||
2bd59d48 TH |
1065 | static int cgroup_show_options(struct seq_file *seq, |
1066 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1067 | { |
3dd06ffa | 1068 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1069 | struct cgroup_subsys *ss; |
b85d2040 | 1070 | int ssid; |
ddbcc7e8 | 1071 | |
b85d2040 | 1072 | for_each_subsys(ss, ssid) |
3dd06ffa | 1073 | if (root->cgrp.subsys_mask & (1 << ssid)) |
b85d2040 | 1074 | seq_printf(seq, ",%s", ss->name); |
873fe09e TH |
1075 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1076 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1077 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1078 | seq_puts(seq, ",noprefix"); |
93438629 | 1079 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1080 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1081 | |
1082 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1083 | if (strlen(root->release_agent_path)) |
1084 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1085 | spin_unlock(&release_agent_path_lock); |
1086 | ||
3dd06ffa | 1087 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1088 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1089 | if (strlen(root->name)) |
1090 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1091 | return 0; |
1092 | } | |
1093 | ||
1094 | struct cgroup_sb_opts { | |
a1a71b45 | 1095 | unsigned long subsys_mask; |
ddbcc7e8 | 1096 | unsigned long flags; |
81a6a5cd | 1097 | char *release_agent; |
2260e7fc | 1098 | bool cpuset_clone_children; |
c6d57f33 | 1099 | char *name; |
2c6ab6d2 PM |
1100 | /* User explicitly requested empty subsystem */ |
1101 | bool none; | |
ddbcc7e8 PM |
1102 | }; |
1103 | ||
aae8aab4 | 1104 | /* |
9871bf95 TH |
1105 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1106 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1107 | * array. This function takes refcounts on subsystems to be used, unless it | |
1108 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1109 | */ |
cf5d5941 | 1110 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1111 | { |
32a8cf23 DL |
1112 | char *token, *o = data; |
1113 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1114 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1115 | struct cgroup_subsys *ss; |
1116 | int i; | |
f9ab5b5b | 1117 | |
aae8aab4 BB |
1118 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1119 | ||
f9ab5b5b | 1120 | #ifdef CONFIG_CPUSETS |
073219e9 | 1121 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1122 | #endif |
ddbcc7e8 | 1123 | |
c6d57f33 | 1124 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1125 | |
1126 | while ((token = strsep(&o, ",")) != NULL) { | |
1127 | if (!*token) | |
1128 | return -EINVAL; | |
32a8cf23 | 1129 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1130 | /* Explicitly have no subsystems */ |
1131 | opts->none = true; | |
32a8cf23 DL |
1132 | continue; |
1133 | } | |
1134 | if (!strcmp(token, "all")) { | |
1135 | /* Mutually exclusive option 'all' + subsystem name */ | |
1136 | if (one_ss) | |
1137 | return -EINVAL; | |
1138 | all_ss = true; | |
1139 | continue; | |
1140 | } | |
873fe09e TH |
1141 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1142 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1143 | continue; | |
1144 | } | |
32a8cf23 | 1145 | if (!strcmp(token, "noprefix")) { |
93438629 | 1146 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1147 | continue; |
1148 | } | |
1149 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1150 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1151 | continue; |
1152 | } | |
03b1cde6 | 1153 | if (!strcmp(token, "xattr")) { |
93438629 | 1154 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1155 | continue; |
1156 | } | |
32a8cf23 | 1157 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1158 | /* Specifying two release agents is forbidden */ |
1159 | if (opts->release_agent) | |
1160 | return -EINVAL; | |
c6d57f33 | 1161 | opts->release_agent = |
e400c285 | 1162 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1163 | if (!opts->release_agent) |
1164 | return -ENOMEM; | |
32a8cf23 DL |
1165 | continue; |
1166 | } | |
1167 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1168 | const char *name = token + 5; |
1169 | /* Can't specify an empty name */ | |
1170 | if (!strlen(name)) | |
1171 | return -EINVAL; | |
1172 | /* Must match [\w.-]+ */ | |
1173 | for (i = 0; i < strlen(name); i++) { | |
1174 | char c = name[i]; | |
1175 | if (isalnum(c)) | |
1176 | continue; | |
1177 | if ((c == '.') || (c == '-') || (c == '_')) | |
1178 | continue; | |
1179 | return -EINVAL; | |
1180 | } | |
1181 | /* Specifying two names is forbidden */ | |
1182 | if (opts->name) | |
1183 | return -EINVAL; | |
1184 | opts->name = kstrndup(name, | |
e400c285 | 1185 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1186 | GFP_KERNEL); |
1187 | if (!opts->name) | |
1188 | return -ENOMEM; | |
32a8cf23 DL |
1189 | |
1190 | continue; | |
1191 | } | |
1192 | ||
30159ec7 | 1193 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1194 | if (strcmp(token, ss->name)) |
1195 | continue; | |
1196 | if (ss->disabled) | |
1197 | continue; | |
1198 | ||
1199 | /* Mutually exclusive option 'all' + subsystem name */ | |
1200 | if (all_ss) | |
1201 | return -EINVAL; | |
a1a71b45 | 1202 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1203 | one_ss = true; |
1204 | ||
1205 | break; | |
1206 | } | |
1207 | if (i == CGROUP_SUBSYS_COUNT) | |
1208 | return -ENOENT; | |
1209 | } | |
1210 | ||
2c6ab6d2 PM |
1211 | /* Consistency checks */ |
1212 | ||
873fe09e TH |
1213 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1214 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1215 | ||
d3ba07c3 TH |
1216 | if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) || |
1217 | opts->cpuset_clone_children || opts->release_agent || | |
1218 | opts->name) { | |
1219 | pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n"); | |
873fe09e TH |
1220 | return -EINVAL; |
1221 | } | |
a2dd4247 TH |
1222 | } else { |
1223 | /* | |
1224 | * If the 'all' option was specified select all the | |
1225 | * subsystems, otherwise if 'none', 'name=' and a subsystem | |
1226 | * name options were not specified, let's default to 'all' | |
1227 | */ | |
1228 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1229 | for_each_subsys(ss, i) | |
1230 | if (!ss->disabled) | |
1231 | set_bit(i, &opts->subsys_mask); | |
873fe09e | 1232 | |
a2dd4247 TH |
1233 | /* |
1234 | * We either have to specify by name or by subsystems. (So | |
1235 | * all empty hierarchies must have a name). | |
1236 | */ | |
1237 | if (!opts->subsys_mask && !opts->name) | |
873fe09e | 1238 | return -EINVAL; |
873fe09e TH |
1239 | } |
1240 | ||
f9ab5b5b LZ |
1241 | /* |
1242 | * Option noprefix was introduced just for backward compatibility | |
1243 | * with the old cpuset, so we allow noprefix only if mounting just | |
1244 | * the cpuset subsystem. | |
1245 | */ | |
93438629 | 1246 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1247 | return -EINVAL; |
1248 | ||
2c6ab6d2 PM |
1249 | |
1250 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1251 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1252 | return -EINVAL; |
1253 | ||
ddbcc7e8 PM |
1254 | return 0; |
1255 | } | |
1256 | ||
2bd59d48 | 1257 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1258 | { |
1259 | int ret = 0; | |
3dd06ffa | 1260 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1261 | struct cgroup_sb_opts opts; |
a1a71b45 | 1262 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1263 | |
873fe09e TH |
1264 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1265 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1266 | return -EINVAL; | |
1267 | } | |
1268 | ||
ace2bee8 | 1269 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1270 | mutex_lock(&cgroup_mutex); |
1271 | ||
1272 | /* See what subsystems are wanted */ | |
1273 | ret = parse_cgroupfs_options(data, &opts); | |
1274 | if (ret) | |
1275 | goto out_unlock; | |
1276 | ||
3dd06ffa | 1277 | if (opts.subsys_mask != root->cgrp.subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1278 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1279 | task_tgid_nr(current), current->comm); | |
1280 | ||
3dd06ffa TH |
1281 | added_mask = opts.subsys_mask & ~root->cgrp.subsys_mask; |
1282 | removed_mask = root->cgrp.subsys_mask & ~opts.subsys_mask; | |
13af07df | 1283 | |
cf5d5941 | 1284 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1285 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1286 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1287 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1288 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1289 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1290 | ret = -EINVAL; |
1291 | goto out_unlock; | |
1292 | } | |
1293 | ||
f172e67c | 1294 | /* remounting is not allowed for populated hierarchies */ |
3dd06ffa | 1295 | if (!list_empty(&root->cgrp.children)) { |
f172e67c | 1296 | ret = -EBUSY; |
0670e08b | 1297 | goto out_unlock; |
cf5d5941 | 1298 | } |
ddbcc7e8 | 1299 | |
5df36032 | 1300 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1301 | if (ret) |
0670e08b | 1302 | goto out_unlock; |
ddbcc7e8 | 1303 | |
3dd06ffa | 1304 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1305 | |
69e943b7 TH |
1306 | if (opts.release_agent) { |
1307 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1308 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1309 | spin_unlock(&release_agent_path_lock); |
1310 | } | |
ddbcc7e8 | 1311 | out_unlock: |
66bdc9cf | 1312 | kfree(opts.release_agent); |
c6d57f33 | 1313 | kfree(opts.name); |
ddbcc7e8 | 1314 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1315 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1316 | return ret; |
1317 | } | |
1318 | ||
afeb0f9f TH |
1319 | /* |
1320 | * To reduce the fork() overhead for systems that are not actually using | |
1321 | * their cgroups capability, we don't maintain the lists running through | |
1322 | * each css_set to its tasks until we see the list actually used - in other | |
1323 | * words after the first mount. | |
1324 | */ | |
1325 | static bool use_task_css_set_links __read_mostly; | |
1326 | ||
1327 | static void cgroup_enable_task_cg_lists(void) | |
1328 | { | |
1329 | struct task_struct *p, *g; | |
1330 | ||
96d365e0 | 1331 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1332 | |
1333 | if (use_task_css_set_links) | |
1334 | goto out_unlock; | |
1335 | ||
1336 | use_task_css_set_links = true; | |
1337 | ||
1338 | /* | |
1339 | * We need tasklist_lock because RCU is not safe against | |
1340 | * while_each_thread(). Besides, a forking task that has passed | |
1341 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1342 | * is not guaranteed to have its child immediately visible in the | |
1343 | * tasklist if we walk through it with RCU. | |
1344 | */ | |
1345 | read_lock(&tasklist_lock); | |
1346 | do_each_thread(g, p) { | |
afeb0f9f TH |
1347 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1348 | task_css_set(p) != &init_css_set); | |
1349 | ||
1350 | /* | |
1351 | * We should check if the process is exiting, otherwise | |
1352 | * it will race with cgroup_exit() in that the list | |
1353 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1354 | * Do it while holding siglock so that we don't end up |
1355 | * racing against cgroup_exit(). | |
afeb0f9f | 1356 | */ |
f153ad11 | 1357 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1358 | if (!(p->flags & PF_EXITING)) { |
1359 | struct css_set *cset = task_css_set(p); | |
1360 | ||
1361 | list_add(&p->cg_list, &cset->tasks); | |
1362 | get_css_set(cset); | |
1363 | } | |
f153ad11 | 1364 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1365 | } while_each_thread(g, p); |
1366 | read_unlock(&tasklist_lock); | |
1367 | out_unlock: | |
96d365e0 | 1368 | up_write(&css_set_rwsem); |
afeb0f9f | 1369 | } |
ddbcc7e8 | 1370 | |
cc31edce PM |
1371 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1372 | { | |
2bd59d48 | 1373 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1374 | INIT_LIST_HEAD(&cgrp->sibling); |
1375 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1376 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1377 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1378 | INIT_LIST_HEAD(&cgrp->pidlists); |
1379 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1380 | cgrp->dummy_css.cgroup = cgrp; |
cc31edce | 1381 | } |
c6d57f33 | 1382 | |
3dd06ffa | 1383 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1384 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1385 | { |
3dd06ffa | 1386 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1387 | |
ddbcc7e8 | 1388 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1389 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1390 | cgrp->root = root; |
cc31edce | 1391 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1392 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1393 | |
c6d57f33 PM |
1394 | root->flags = opts->flags; |
1395 | if (opts->release_agent) | |
1396 | strcpy(root->release_agent_path, opts->release_agent); | |
1397 | if (opts->name) | |
1398 | strcpy(root->name, opts->name); | |
2260e7fc | 1399 | if (opts->cpuset_clone_children) |
3dd06ffa | 1400 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1401 | } |
1402 | ||
3dd06ffa | 1403 | static int cgroup_setup_root(struct cgroup_root *root, unsigned long ss_mask) |
2c6ab6d2 | 1404 | { |
d427dfeb | 1405 | LIST_HEAD(tmp_links); |
3dd06ffa | 1406 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1407 | struct css_set *cset; |
d427dfeb | 1408 | int i, ret; |
2c6ab6d2 | 1409 | |
d427dfeb TH |
1410 | lockdep_assert_held(&cgroup_tree_mutex); |
1411 | lockdep_assert_held(&cgroup_mutex); | |
c6d57f33 | 1412 | |
d427dfeb TH |
1413 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); |
1414 | if (ret < 0) | |
2bd59d48 | 1415 | goto out; |
d427dfeb | 1416 | root_cgrp->id = ret; |
c6d57f33 | 1417 | |
d427dfeb | 1418 | /* |
96d365e0 | 1419 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1420 | * but that's OK - it can only be increased by someone holding |
1421 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1422 | * have some link structures left over | |
1423 | */ | |
1424 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1425 | if (ret) | |
2bd59d48 | 1426 | goto out; |
ddbcc7e8 | 1427 | |
985ed670 | 1428 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1429 | if (ret) |
2bd59d48 | 1430 | goto out; |
ddbcc7e8 | 1431 | |
2bd59d48 TH |
1432 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1433 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1434 | root_cgrp); | |
1435 | if (IS_ERR(root->kf_root)) { | |
1436 | ret = PTR_ERR(root->kf_root); | |
1437 | goto exit_root_id; | |
1438 | } | |
1439 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1440 | |
d427dfeb TH |
1441 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); |
1442 | if (ret) | |
2bd59d48 | 1443 | goto destroy_root; |
ddbcc7e8 | 1444 | |
5df36032 | 1445 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1446 | if (ret) |
2bd59d48 | 1447 | goto destroy_root; |
ddbcc7e8 | 1448 | |
d427dfeb TH |
1449 | /* |
1450 | * There must be no failure case after here, since rebinding takes | |
1451 | * care of subsystems' refcounts, which are explicitly dropped in | |
1452 | * the failure exit path. | |
1453 | */ | |
1454 | list_add(&root->root_list, &cgroup_roots); | |
1455 | cgroup_root_count++; | |
0df6a63f | 1456 | |
d427dfeb | 1457 | /* |
3dd06ffa | 1458 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1459 | * objects. |
1460 | */ | |
96d365e0 | 1461 | down_write(&css_set_rwsem); |
d427dfeb TH |
1462 | hash_for_each(css_set_table, i, cset, hlist) |
1463 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1464 | up_write(&css_set_rwsem); |
ddbcc7e8 | 1465 | |
d427dfeb | 1466 | BUG_ON(!list_empty(&root_cgrp->children)); |
3c9c825b | 1467 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1468 | |
2bd59d48 | 1469 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1470 | ret = 0; |
2bd59d48 | 1471 | goto out; |
d427dfeb | 1472 | |
2bd59d48 TH |
1473 | destroy_root: |
1474 | kernfs_destroy_root(root->kf_root); | |
1475 | root->kf_root = NULL; | |
1476 | exit_root_id: | |
d427dfeb | 1477 | cgroup_exit_root_id(root); |
2bd59d48 | 1478 | out: |
d427dfeb TH |
1479 | free_cgrp_cset_links(&tmp_links); |
1480 | return ret; | |
ddbcc7e8 PM |
1481 | } |
1482 | ||
f7e83571 | 1483 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1484 | int flags, const char *unused_dev_name, |
f7e83571 | 1485 | void *data) |
ddbcc7e8 | 1486 | { |
3dd06ffa | 1487 | struct cgroup_root *root; |
ddbcc7e8 | 1488 | struct cgroup_sb_opts opts; |
2bd59d48 | 1489 | struct dentry *dentry; |
8e30e2b8 | 1490 | int ret; |
c6b3d5bc | 1491 | bool new_sb; |
ddbcc7e8 | 1492 | |
56fde9e0 TH |
1493 | /* |
1494 | * The first time anyone tries to mount a cgroup, enable the list | |
1495 | * linking each css_set to its tasks and fix up all existing tasks. | |
1496 | */ | |
1497 | if (!use_task_css_set_links) | |
1498 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 1499 | |
8e30e2b8 | 1500 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1501 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1502 | |
1503 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1504 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1505 | if (ret) |
8e30e2b8 | 1506 | goto out_unlock; |
e37a06f1 | 1507 | retry: |
2bd59d48 | 1508 | /* look for a matching existing root */ |
a2dd4247 TH |
1509 | if (!opts.subsys_mask && !opts.none && !opts.name) { |
1510 | cgrp_dfl_root_visible = true; | |
1511 | root = &cgrp_dfl_root; | |
1512 | cgroup_get(&root->cgrp); | |
1513 | ret = 0; | |
1514 | goto out_unlock; | |
ddbcc7e8 PM |
1515 | } |
1516 | ||
985ed670 | 1517 | for_each_root(root) { |
2bd59d48 | 1518 | bool name_match = false; |
3126121f | 1519 | |
3dd06ffa | 1520 | if (root == &cgrp_dfl_root) |
985ed670 | 1521 | continue; |
3126121f | 1522 | |
cf5d5941 | 1523 | /* |
2bd59d48 TH |
1524 | * If we asked for a name then it must match. Also, if |
1525 | * name matches but sybsys_mask doesn't, we should fail. | |
1526 | * Remember whether name matched. | |
cf5d5941 | 1527 | */ |
2bd59d48 TH |
1528 | if (opts.name) { |
1529 | if (strcmp(opts.name, root->name)) | |
1530 | continue; | |
1531 | name_match = true; | |
1532 | } | |
ddbcc7e8 | 1533 | |
c6d57f33 | 1534 | /* |
2bd59d48 TH |
1535 | * If we asked for subsystems (or explicitly for no |
1536 | * subsystems) then they must match. | |
c6d57f33 | 1537 | */ |
2bd59d48 | 1538 | if ((opts.subsys_mask || opts.none) && |
3dd06ffa | 1539 | (opts.subsys_mask != root->cgrp.subsys_mask)) { |
2bd59d48 TH |
1540 | if (!name_match) |
1541 | continue; | |
1542 | ret = -EBUSY; | |
1543 | goto out_unlock; | |
1544 | } | |
873fe09e | 1545 | |
c7ba8287 | 1546 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1547 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1548 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1549 | ret = -EINVAL; | |
8e30e2b8 | 1550 | goto out_unlock; |
2a0ff3fb JL |
1551 | } else { |
1552 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1553 | } | |
873fe09e | 1554 | } |
ddbcc7e8 | 1555 | |
776f02fa | 1556 | /* |
3dd06ffa | 1557 | * A root's lifetime is governed by its root cgroup. Zero |
776f02fa TH |
1558 | * ref indicate that the root is being destroyed. Wait for |
1559 | * destruction to complete so that the subsystems are free. | |
1560 | * We can use wait_queue for the wait but this path is | |
1561 | * super cold. Let's just sleep for a bit and retry. | |
1562 | */ | |
3dd06ffa | 1563 | if (!atomic_inc_not_zero(&root->cgrp.refcnt)) { |
776f02fa TH |
1564 | mutex_unlock(&cgroup_mutex); |
1565 | mutex_unlock(&cgroup_tree_mutex); | |
1566 | msleep(10); | |
e37a06f1 LZ |
1567 | mutex_lock(&cgroup_tree_mutex); |
1568 | mutex_lock(&cgroup_mutex); | |
776f02fa TH |
1569 | goto retry; |
1570 | } | |
ddbcc7e8 | 1571 | |
776f02fa | 1572 | ret = 0; |
2bd59d48 | 1573 | goto out_unlock; |
ddbcc7e8 | 1574 | } |
ddbcc7e8 | 1575 | |
817929ec | 1576 | /* |
172a2c06 TH |
1577 | * No such thing, create a new one. name= matching without subsys |
1578 | * specification is allowed for already existing hierarchies but we | |
1579 | * can't create new one without subsys specification. | |
817929ec | 1580 | */ |
172a2c06 TH |
1581 | if (!opts.subsys_mask && !opts.none) { |
1582 | ret = -EINVAL; | |
1583 | goto out_unlock; | |
817929ec | 1584 | } |
817929ec | 1585 | |
172a2c06 TH |
1586 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
1587 | if (!root) { | |
1588 | ret = -ENOMEM; | |
2bd59d48 | 1589 | goto out_unlock; |
839ec545 | 1590 | } |
e5f6a860 | 1591 | |
172a2c06 TH |
1592 | init_cgroup_root(root, &opts); |
1593 | ||
35585573 | 1594 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1595 | if (ret) |
1596 | cgroup_free_root(root); | |
fa3ca07e | 1597 | |
8e30e2b8 | 1598 | out_unlock: |
ddbcc7e8 | 1599 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1600 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 | 1601 | |
c6d57f33 PM |
1602 | kfree(opts.release_agent); |
1603 | kfree(opts.name); | |
03b1cde6 | 1604 | |
2bd59d48 | 1605 | if (ret) |
8e30e2b8 | 1606 | return ERR_PTR(ret); |
2bd59d48 | 1607 | |
26fc9cd2 JZ |
1608 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
1609 | CGROUP_SUPER_MAGIC, &new_sb); | |
c6b3d5bc | 1610 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 1611 | cgroup_put(&root->cgrp); |
2bd59d48 TH |
1612 | return dentry; |
1613 | } | |
1614 | ||
1615 | static void cgroup_kill_sb(struct super_block *sb) | |
1616 | { | |
1617 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 1618 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 1619 | |
3dd06ffa | 1620 | cgroup_put(&root->cgrp); |
2bd59d48 | 1621 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1622 | } |
1623 | ||
1624 | static struct file_system_type cgroup_fs_type = { | |
1625 | .name = "cgroup", | |
f7e83571 | 1626 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1627 | .kill_sb = cgroup_kill_sb, |
1628 | }; | |
1629 | ||
676db4af GK |
1630 | static struct kobject *cgroup_kobj; |
1631 | ||
857a2beb | 1632 | /** |
913ffdb5 | 1633 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1634 | * @task: target task |
857a2beb TH |
1635 | * @buf: the buffer to write the path into |
1636 | * @buflen: the length of the buffer | |
1637 | * | |
913ffdb5 TH |
1638 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1639 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1640 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1641 | * cgroup controller callbacks. | |
1642 | * | |
e61734c5 | 1643 | * Return value is the same as kernfs_path(). |
857a2beb | 1644 | */ |
e61734c5 | 1645 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 1646 | { |
3dd06ffa | 1647 | struct cgroup_root *root; |
913ffdb5 | 1648 | struct cgroup *cgrp; |
e61734c5 TH |
1649 | int hierarchy_id = 1; |
1650 | char *path = NULL; | |
857a2beb TH |
1651 | |
1652 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1653 | down_read(&css_set_rwsem); |
857a2beb | 1654 | |
913ffdb5 TH |
1655 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1656 | ||
857a2beb TH |
1657 | if (root) { |
1658 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1659 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1660 | } else { |
1661 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1662 | if (strlcpy(buf, "/", buflen) < buflen) |
1663 | path = buf; | |
857a2beb TH |
1664 | } |
1665 | ||
96d365e0 | 1666 | up_read(&css_set_rwsem); |
857a2beb | 1667 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1668 | return path; |
857a2beb | 1669 | } |
913ffdb5 | 1670 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1671 | |
b3dc094e | 1672 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 1673 | struct cgroup_taskset { |
b3dc094e TH |
1674 | /* the src and dst cset list running through cset->mg_node */ |
1675 | struct list_head src_csets; | |
1676 | struct list_head dst_csets; | |
1677 | ||
1678 | /* | |
1679 | * Fields for cgroup_taskset_*() iteration. | |
1680 | * | |
1681 | * Before migration is committed, the target migration tasks are on | |
1682 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
1683 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
1684 | * or ->dst_csets depending on whether migration is committed. | |
1685 | * | |
1686 | * ->cur_csets and ->cur_task point to the current task position | |
1687 | * during iteration. | |
1688 | */ | |
1689 | struct list_head *csets; | |
1690 | struct css_set *cur_cset; | |
1691 | struct task_struct *cur_task; | |
2f7ee569 TH |
1692 | }; |
1693 | ||
1694 | /** | |
1695 | * cgroup_taskset_first - reset taskset and return the first task | |
1696 | * @tset: taskset of interest | |
1697 | * | |
1698 | * @tset iteration is initialized and the first task is returned. | |
1699 | */ | |
1700 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1701 | { | |
b3dc094e TH |
1702 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
1703 | tset->cur_task = NULL; | |
1704 | ||
1705 | return cgroup_taskset_next(tset); | |
2f7ee569 | 1706 | } |
2f7ee569 TH |
1707 | |
1708 | /** | |
1709 | * cgroup_taskset_next - iterate to the next task in taskset | |
1710 | * @tset: taskset of interest | |
1711 | * | |
1712 | * Return the next task in @tset. Iteration must have been initialized | |
1713 | * with cgroup_taskset_first(). | |
1714 | */ | |
1715 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1716 | { | |
b3dc094e TH |
1717 | struct css_set *cset = tset->cur_cset; |
1718 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 1719 | |
b3dc094e TH |
1720 | while (&cset->mg_node != tset->csets) { |
1721 | if (!task) | |
1722 | task = list_first_entry(&cset->mg_tasks, | |
1723 | struct task_struct, cg_list); | |
1724 | else | |
1725 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 1726 | |
b3dc094e TH |
1727 | if (&task->cg_list != &cset->mg_tasks) { |
1728 | tset->cur_cset = cset; | |
1729 | tset->cur_task = task; | |
1730 | return task; | |
1731 | } | |
2f7ee569 | 1732 | |
b3dc094e TH |
1733 | cset = list_next_entry(cset, mg_node); |
1734 | task = NULL; | |
1735 | } | |
2f7ee569 | 1736 | |
b3dc094e | 1737 | return NULL; |
2f7ee569 | 1738 | } |
2f7ee569 | 1739 | |
cb0f1fe9 | 1740 | /** |
74a1166d | 1741 | * cgroup_task_migrate - move a task from one cgroup to another. |
cb0f1fe9 TH |
1742 | * @old_cgrp; the cgroup @tsk is being migrated from |
1743 | * @tsk: the task being migrated | |
1744 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 1745 | * |
cb0f1fe9 | 1746 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 1747 | */ |
5abb8855 TH |
1748 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1749 | struct task_struct *tsk, | |
1750 | struct css_set *new_cset) | |
74a1166d | 1751 | { |
5abb8855 | 1752 | struct css_set *old_cset; |
74a1166d | 1753 | |
cb0f1fe9 TH |
1754 | lockdep_assert_held(&cgroup_mutex); |
1755 | lockdep_assert_held(&css_set_rwsem); | |
1756 | ||
74a1166d | 1757 | /* |
026085ef MSB |
1758 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1759 | * setting such that we can't race against cgroup_exit() changing the | |
1760 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1761 | */ |
c84cdf75 | 1762 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1763 | old_cset = task_css_set(tsk); |
74a1166d | 1764 | |
b3dc094e | 1765 | get_css_set(new_cset); |
5abb8855 | 1766 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d | 1767 | |
1b9aba49 TH |
1768 | /* |
1769 | * Use move_tail so that cgroup_taskset_first() still returns the | |
1770 | * leader after migration. This works because cgroup_migrate() | |
1771 | * ensures that the dst_cset of the leader is the first on the | |
1772 | * tset's dst_csets list. | |
1773 | */ | |
1774 | list_move_tail(&tsk->cg_list, &new_cset->mg_tasks); | |
74a1166d BB |
1775 | |
1776 | /* | |
5abb8855 TH |
1777 | * We just gained a reference on old_cset by taking it from the |
1778 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1779 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1780 | */ |
5abb8855 | 1781 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
cb0f1fe9 | 1782 | put_css_set_locked(old_cset, false); |
74a1166d BB |
1783 | } |
1784 | ||
a043e3b2 | 1785 | /** |
1958d2d5 TH |
1786 | * cgroup_migrate_finish - cleanup after attach |
1787 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 1788 | * |
1958d2d5 TH |
1789 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
1790 | * those functions for details. | |
74a1166d | 1791 | */ |
1958d2d5 | 1792 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 1793 | { |
1958d2d5 | 1794 | struct css_set *cset, *tmp_cset; |
74a1166d | 1795 | |
1958d2d5 TH |
1796 | lockdep_assert_held(&cgroup_mutex); |
1797 | ||
1798 | down_write(&css_set_rwsem); | |
1799 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
1800 | cset->mg_src_cgrp = NULL; | |
1801 | cset->mg_dst_cset = NULL; | |
1802 | list_del_init(&cset->mg_preload_node); | |
1803 | put_css_set_locked(cset, false); | |
1804 | } | |
1805 | up_write(&css_set_rwsem); | |
1806 | } | |
1807 | ||
1808 | /** | |
1809 | * cgroup_migrate_add_src - add a migration source css_set | |
1810 | * @src_cset: the source css_set to add | |
1811 | * @dst_cgrp: the destination cgroup | |
1812 | * @preloaded_csets: list of preloaded css_sets | |
1813 | * | |
1814 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
1815 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
1816 | * up by cgroup_migrate_finish(). | |
1817 | * | |
1818 | * This function may be called without holding threadgroup_lock even if the | |
1819 | * target is a process. Threads may be created and destroyed but as long | |
1820 | * as cgroup_mutex is not dropped, no new css_set can be put into play and | |
1821 | * the preloaded css_sets are guaranteed to cover all migrations. | |
1822 | */ | |
1823 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
1824 | struct cgroup *dst_cgrp, | |
1825 | struct list_head *preloaded_csets) | |
1826 | { | |
1827 | struct cgroup *src_cgrp; | |
1828 | ||
1829 | lockdep_assert_held(&cgroup_mutex); | |
1830 | lockdep_assert_held(&css_set_rwsem); | |
1831 | ||
1832 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
1833 | ||
1834 | /* nothing to do if this cset already belongs to the cgroup */ | |
1835 | if (src_cgrp == dst_cgrp) | |
1836 | return; | |
1837 | ||
1838 | if (!list_empty(&src_cset->mg_preload_node)) | |
1839 | return; | |
1840 | ||
1841 | WARN_ON(src_cset->mg_src_cgrp); | |
1842 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
1843 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
1844 | ||
1845 | src_cset->mg_src_cgrp = src_cgrp; | |
1846 | get_css_set(src_cset); | |
1847 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
1848 | } | |
1849 | ||
1850 | /** | |
1851 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
1852 | * @dst_cgrp: the destination cgroup | |
1853 | * @preloaded_csets: list of preloaded source css_sets | |
1854 | * | |
1855 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
1856 | * have been preloaded to @preloaded_csets. This function looks up and | |
1857 | * pins all destination css_sets, links each to its source, and put them on | |
1858 | * @preloaded_csets. | |
1859 | * | |
1860 | * This function must be called after cgroup_migrate_add_src() has been | |
1861 | * called on each migration source css_set. After migration is performed | |
1862 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
1863 | * @preloaded_csets. | |
1864 | */ | |
1865 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
1866 | struct list_head *preloaded_csets) | |
1867 | { | |
1868 | LIST_HEAD(csets); | |
1869 | struct css_set *src_cset; | |
1870 | ||
1871 | lockdep_assert_held(&cgroup_mutex); | |
1872 | ||
1873 | /* look up the dst cset for each src cset and link it to src */ | |
1874 | list_for_each_entry(src_cset, preloaded_csets, mg_preload_node) { | |
1875 | struct css_set *dst_cset; | |
1876 | ||
1877 | dst_cset = find_css_set(src_cset, dst_cgrp); | |
1878 | if (!dst_cset) | |
1879 | goto err; | |
1880 | ||
1881 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
1882 | src_cset->mg_dst_cset = dst_cset; | |
1883 | ||
1884 | if (list_empty(&dst_cset->mg_preload_node)) | |
1885 | list_add(&dst_cset->mg_preload_node, &csets); | |
1886 | else | |
1887 | put_css_set(dst_cset, false); | |
1888 | } | |
1889 | ||
1890 | list_splice(&csets, preloaded_csets); | |
1891 | return 0; | |
1892 | err: | |
1893 | cgroup_migrate_finish(&csets); | |
1894 | return -ENOMEM; | |
1895 | } | |
1896 | ||
1897 | /** | |
1898 | * cgroup_migrate - migrate a process or task to a cgroup | |
1899 | * @cgrp: the destination cgroup | |
1900 | * @leader: the leader of the process or the task to migrate | |
1901 | * @threadgroup: whether @leader points to the whole process or a single task | |
1902 | * | |
1903 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
1904 | * process, the caller must be holding threadgroup_lock of @leader. The | |
1905 | * caller is also responsible for invoking cgroup_migrate_add_src() and | |
1906 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
1907 | * function and following up with cgroup_migrate_finish(). | |
1908 | * | |
1909 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
1910 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
1911 | * failure, when migrating multiple targets, the success or failure can be | |
1912 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
1913 | * actually starting migrating. | |
1914 | */ | |
1915 | static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader, | |
1916 | bool threadgroup) | |
74a1166d | 1917 | { |
b3dc094e TH |
1918 | struct cgroup_taskset tset = { |
1919 | .src_csets = LIST_HEAD_INIT(tset.src_csets), | |
1920 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), | |
1921 | .csets = &tset.src_csets, | |
1922 | }; | |
1c6727af | 1923 | struct cgroup_subsys_state *css, *failed_css = NULL; |
b3dc094e TH |
1924 | struct css_set *cset, *tmp_cset; |
1925 | struct task_struct *task, *tmp_task; | |
1926 | int i, ret; | |
74a1166d | 1927 | |
fb5d2b4c MSB |
1928 | /* |
1929 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1930 | * already PF_EXITING could be freed from underneath us unless we | |
1931 | * take an rcu_read_lock. | |
1932 | */ | |
b3dc094e | 1933 | down_write(&css_set_rwsem); |
fb5d2b4c | 1934 | rcu_read_lock(); |
9db8de37 | 1935 | task = leader; |
74a1166d | 1936 | do { |
9db8de37 TH |
1937 | /* @task either already exited or can't exit until the end */ |
1938 | if (task->flags & PF_EXITING) | |
ea84753c | 1939 | goto next; |
134d3373 | 1940 | |
eaf797ab TH |
1941 | /* leave @task alone if post_fork() hasn't linked it yet */ |
1942 | if (list_empty(&task->cg_list)) | |
ea84753c | 1943 | goto next; |
cd3d0952 | 1944 | |
b3dc094e | 1945 | cset = task_css_set(task); |
1958d2d5 | 1946 | if (!cset->mg_src_cgrp) |
ea84753c | 1947 | goto next; |
b3dc094e | 1948 | |
61d1d219 | 1949 | /* |
1b9aba49 TH |
1950 | * cgroup_taskset_first() must always return the leader. |
1951 | * Take care to avoid disturbing the ordering. | |
61d1d219 | 1952 | */ |
1b9aba49 TH |
1953 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
1954 | if (list_empty(&cset->mg_node)) | |
1955 | list_add_tail(&cset->mg_node, &tset.src_csets); | |
1956 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
1957 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
1958 | &tset.dst_csets); | |
ea84753c | 1959 | next: |
081aa458 LZ |
1960 | if (!threadgroup) |
1961 | break; | |
9db8de37 | 1962 | } while_each_thread(leader, task); |
fb5d2b4c | 1963 | rcu_read_unlock(); |
b3dc094e | 1964 | up_write(&css_set_rwsem); |
74a1166d | 1965 | |
134d3373 | 1966 | /* methods shouldn't be called if no task is actually migrating */ |
b3dc094e TH |
1967 | if (list_empty(&tset.src_csets)) |
1968 | return 0; | |
134d3373 | 1969 | |
1958d2d5 | 1970 | /* check that we can legitimately attach to the cgroup */ |
1c6727af TH |
1971 | for_each_css(css, i, cgrp) { |
1972 | if (css->ss->can_attach) { | |
9db8de37 TH |
1973 | ret = css->ss->can_attach(css, &tset); |
1974 | if (ret) { | |
1c6727af | 1975 | failed_css = css; |
74a1166d BB |
1976 | goto out_cancel_attach; |
1977 | } | |
1978 | } | |
74a1166d BB |
1979 | } |
1980 | ||
1981 | /* | |
1958d2d5 TH |
1982 | * Now that we're guaranteed success, proceed to move all tasks to |
1983 | * the new cgroup. There are no failure cases after here, so this | |
1984 | * is the commit point. | |
74a1166d | 1985 | */ |
cb0f1fe9 | 1986 | down_write(&css_set_rwsem); |
b3dc094e TH |
1987 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
1988 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
1989 | cgroup_task_migrate(cset->mg_src_cgrp, task, | |
1990 | cset->mg_dst_cset); | |
74a1166d | 1991 | } |
cb0f1fe9 | 1992 | up_write(&css_set_rwsem); |
74a1166d BB |
1993 | |
1994 | /* | |
1958d2d5 TH |
1995 | * Migration is committed, all target tasks are now on dst_csets. |
1996 | * Nothing is sensitive to fork() after this point. Notify | |
1997 | * controllers that migration is complete. | |
74a1166d | 1998 | */ |
1958d2d5 | 1999 | tset.csets = &tset.dst_csets; |
74a1166d | 2000 | |
1c6727af TH |
2001 | for_each_css(css, i, cgrp) |
2002 | if (css->ss->attach) | |
2003 | css->ss->attach(css, &tset); | |
74a1166d | 2004 | |
9db8de37 | 2005 | ret = 0; |
b3dc094e TH |
2006 | goto out_release_tset; |
2007 | ||
74a1166d | 2008 | out_cancel_attach: |
b3dc094e TH |
2009 | for_each_css(css, i, cgrp) { |
2010 | if (css == failed_css) | |
2011 | break; | |
2012 | if (css->ss->cancel_attach) | |
2013 | css->ss->cancel_attach(css, &tset); | |
74a1166d | 2014 | } |
b3dc094e TH |
2015 | out_release_tset: |
2016 | down_write(&css_set_rwsem); | |
2017 | list_splice_init(&tset.dst_csets, &tset.src_csets); | |
2018 | list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { | |
1b9aba49 | 2019 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); |
b3dc094e | 2020 | list_del_init(&cset->mg_node); |
b3dc094e TH |
2021 | } |
2022 | up_write(&css_set_rwsem); | |
9db8de37 | 2023 | return ret; |
74a1166d BB |
2024 | } |
2025 | ||
1958d2d5 TH |
2026 | /** |
2027 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2028 | * @dst_cgrp: the cgroup to attach to | |
2029 | * @leader: the task or the leader of the threadgroup to be attached | |
2030 | * @threadgroup: attach the whole threadgroup? | |
2031 | * | |
0e1d768f | 2032 | * Call holding cgroup_mutex and threadgroup_lock of @leader. |
1958d2d5 TH |
2033 | */ |
2034 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2035 | struct task_struct *leader, bool threadgroup) | |
2036 | { | |
2037 | LIST_HEAD(preloaded_csets); | |
2038 | struct task_struct *task; | |
2039 | int ret; | |
2040 | ||
2041 | /* look up all src csets */ | |
2042 | down_read(&css_set_rwsem); | |
2043 | rcu_read_lock(); | |
2044 | task = leader; | |
2045 | do { | |
2046 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2047 | &preloaded_csets); | |
2048 | if (!threadgroup) | |
2049 | break; | |
2050 | } while_each_thread(leader, task); | |
2051 | rcu_read_unlock(); | |
2052 | up_read(&css_set_rwsem); | |
2053 | ||
2054 | /* prepare dst csets and commit */ | |
2055 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2056 | if (!ret) | |
2057 | ret = cgroup_migrate(dst_cgrp, leader, threadgroup); | |
2058 | ||
2059 | cgroup_migrate_finish(&preloaded_csets); | |
2060 | return ret; | |
74a1166d BB |
2061 | } |
2062 | ||
2063 | /* | |
2064 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2065 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2066 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2067 | */ |
74a1166d | 2068 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2069 | { |
bbcb81d0 | 2070 | struct task_struct *tsk; |
c69e8d9c | 2071 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2072 | int ret; |
2073 | ||
74a1166d BB |
2074 | if (!cgroup_lock_live_group(cgrp)) |
2075 | return -ENODEV; | |
2076 | ||
b78949eb MSB |
2077 | retry_find_task: |
2078 | rcu_read_lock(); | |
bbcb81d0 | 2079 | if (pid) { |
73507f33 | 2080 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2081 | if (!tsk) { |
2082 | rcu_read_unlock(); | |
dd4b0a46 | 2083 | ret = -ESRCH; |
b78949eb | 2084 | goto out_unlock_cgroup; |
bbcb81d0 | 2085 | } |
74a1166d BB |
2086 | /* |
2087 | * even if we're attaching all tasks in the thread group, we | |
2088 | * only need to check permissions on one of them. | |
2089 | */ | |
c69e8d9c | 2090 | tcred = __task_cred(tsk); |
14a590c3 EB |
2091 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2092 | !uid_eq(cred->euid, tcred->uid) && | |
2093 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2094 | rcu_read_unlock(); |
b78949eb MSB |
2095 | ret = -EACCES; |
2096 | goto out_unlock_cgroup; | |
bbcb81d0 | 2097 | } |
b78949eb MSB |
2098 | } else |
2099 | tsk = current; | |
cd3d0952 TH |
2100 | |
2101 | if (threadgroup) | |
b78949eb | 2102 | tsk = tsk->group_leader; |
c4c27fbd MG |
2103 | |
2104 | /* | |
14a40ffc | 2105 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2106 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2107 | * with no rt_runtime allocated. Just say no. | |
2108 | */ | |
14a40ffc | 2109 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2110 | ret = -EINVAL; |
2111 | rcu_read_unlock(); | |
2112 | goto out_unlock_cgroup; | |
2113 | } | |
2114 | ||
b78949eb MSB |
2115 | get_task_struct(tsk); |
2116 | rcu_read_unlock(); | |
2117 | ||
2118 | threadgroup_lock(tsk); | |
2119 | if (threadgroup) { | |
2120 | if (!thread_group_leader(tsk)) { | |
2121 | /* | |
2122 | * a race with de_thread from another thread's exec() | |
2123 | * may strip us of our leadership, if this happens, | |
2124 | * there is no choice but to throw this task away and | |
2125 | * try again; this is | |
2126 | * "double-double-toil-and-trouble-check locking". | |
2127 | */ | |
2128 | threadgroup_unlock(tsk); | |
2129 | put_task_struct(tsk); | |
2130 | goto retry_find_task; | |
2131 | } | |
081aa458 LZ |
2132 | } |
2133 | ||
2134 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2135 | ||
cd3d0952 TH |
2136 | threadgroup_unlock(tsk); |
2137 | ||
bbcb81d0 | 2138 | put_task_struct(tsk); |
b78949eb | 2139 | out_unlock_cgroup: |
47cfcd09 | 2140 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
2141 | return ret; |
2142 | } | |
2143 | ||
7ae1bad9 TH |
2144 | /** |
2145 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2146 | * @from: attach to all cgroups of a given task | |
2147 | * @tsk: the task to be attached | |
2148 | */ | |
2149 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2150 | { | |
3dd06ffa | 2151 | struct cgroup_root *root; |
7ae1bad9 TH |
2152 | int retval = 0; |
2153 | ||
47cfcd09 | 2154 | mutex_lock(&cgroup_mutex); |
985ed670 | 2155 | for_each_root(root) { |
96d365e0 TH |
2156 | struct cgroup *from_cgrp; |
2157 | ||
3dd06ffa | 2158 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2159 | continue; |
2160 | ||
96d365e0 TH |
2161 | down_read(&css_set_rwsem); |
2162 | from_cgrp = task_cgroup_from_root(from, root); | |
2163 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2164 | |
6f4b7e63 | 2165 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2166 | if (retval) |
2167 | break; | |
2168 | } | |
47cfcd09 | 2169 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2170 | |
2171 | return retval; | |
2172 | } | |
2173 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2174 | ||
182446d0 TH |
2175 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2176 | struct cftype *cft, u64 pid) | |
74a1166d | 2177 | { |
182446d0 | 2178 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2179 | } |
2180 | ||
182446d0 TH |
2181 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2182 | struct cftype *cft, u64 tgid) | |
af351026 | 2183 | { |
182446d0 | 2184 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2185 | } |
2186 | ||
182446d0 | 2187 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
4d3bb511 | 2188 | struct cftype *cft, char *buffer) |
e788e066 | 2189 | { |
3dd06ffa | 2190 | struct cgroup_root *root = css->cgroup->root; |
5f469907 TH |
2191 | |
2192 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
182446d0 | 2193 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2194 | return -ENODEV; |
69e943b7 | 2195 | spin_lock(&release_agent_path_lock); |
5f469907 TH |
2196 | strlcpy(root->release_agent_path, buffer, |
2197 | sizeof(root->release_agent_path)); | |
69e943b7 | 2198 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2199 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2200 | return 0; |
2201 | } | |
2202 | ||
2da8ca82 | 2203 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2204 | { |
2da8ca82 | 2205 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2206 | |
e788e066 PM |
2207 | if (!cgroup_lock_live_group(cgrp)) |
2208 | return -ENODEV; | |
2209 | seq_puts(seq, cgrp->root->release_agent_path); | |
2210 | seq_putc(seq, '\n'); | |
47cfcd09 | 2211 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2212 | return 0; |
2213 | } | |
2214 | ||
2da8ca82 | 2215 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2216 | { |
2da8ca82 TH |
2217 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2218 | ||
2219 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2220 | return 0; |
2221 | } | |
2222 | ||
2bd59d48 TH |
2223 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2224 | size_t nbytes, loff_t off) | |
355e0c48 | 2225 | { |
2bd59d48 TH |
2226 | struct cgroup *cgrp = of->kn->parent->priv; |
2227 | struct cftype *cft = of->kn->priv; | |
2228 | struct cgroup_subsys_state *css; | |
a742c59d | 2229 | int ret; |
355e0c48 | 2230 | |
2bd59d48 TH |
2231 | /* |
2232 | * kernfs guarantees that a file isn't deleted with operations in | |
2233 | * flight, which means that the matching css is and stays alive and | |
2234 | * doesn't need to be pinned. The RCU locking is not necessary | |
2235 | * either. It's just for the convenience of using cgroup_css(). | |
2236 | */ | |
2237 | rcu_read_lock(); | |
2238 | css = cgroup_css(cgrp, cft->ss); | |
2239 | rcu_read_unlock(); | |
a742c59d TH |
2240 | |
2241 | if (cft->write_string) { | |
2242 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2243 | } else if (cft->write_u64) { | |
2244 | unsigned long long v; | |
2245 | ret = kstrtoull(buf, 0, &v); | |
2246 | if (!ret) | |
2247 | ret = cft->write_u64(css, cft, v); | |
2248 | } else if (cft->write_s64) { | |
2249 | long long v; | |
2250 | ret = kstrtoll(buf, 0, &v); | |
2251 | if (!ret) | |
2252 | ret = cft->write_s64(css, cft, v); | |
2253 | } else if (cft->trigger) { | |
2254 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2255 | } else { |
a742c59d | 2256 | ret = -EINVAL; |
e73d2c61 | 2257 | } |
2bd59d48 | 2258 | |
a742c59d | 2259 | return ret ?: nbytes; |
355e0c48 PM |
2260 | } |
2261 | ||
6612f05b | 2262 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2263 | { |
2bd59d48 | 2264 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2265 | } |
2266 | ||
6612f05b | 2267 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2268 | { |
2bd59d48 | 2269 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2270 | } |
2271 | ||
6612f05b | 2272 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2273 | { |
2bd59d48 | 2274 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2275 | } |
2276 | ||
91796569 | 2277 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2278 | { |
7da11279 TH |
2279 | struct cftype *cft = seq_cft(m); |
2280 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2281 | |
2da8ca82 TH |
2282 | if (cft->seq_show) |
2283 | return cft->seq_show(m, arg); | |
e73d2c61 | 2284 | |
f4c753b7 | 2285 | if (cft->read_u64) |
896f5199 TH |
2286 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2287 | else if (cft->read_s64) | |
2288 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2289 | else | |
2290 | return -EINVAL; | |
2291 | return 0; | |
91796569 PM |
2292 | } |
2293 | ||
2bd59d48 TH |
2294 | static struct kernfs_ops cgroup_kf_single_ops = { |
2295 | .atomic_write_len = PAGE_SIZE, | |
2296 | .write = cgroup_file_write, | |
2297 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2298 | }; |
2299 | ||
2bd59d48 TH |
2300 | static struct kernfs_ops cgroup_kf_ops = { |
2301 | .atomic_write_len = PAGE_SIZE, | |
2302 | .write = cgroup_file_write, | |
2303 | .seq_start = cgroup_seqfile_start, | |
2304 | .seq_next = cgroup_seqfile_next, | |
2305 | .seq_stop = cgroup_seqfile_stop, | |
2306 | .seq_show = cgroup_seqfile_show, | |
2307 | }; | |
ddbcc7e8 PM |
2308 | |
2309 | /* | |
2310 | * cgroup_rename - Only allow simple rename of directories in place. | |
2311 | */ | |
2bd59d48 TH |
2312 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2313 | const char *new_name_str) | |
ddbcc7e8 | 2314 | { |
2bd59d48 | 2315 | struct cgroup *cgrp = kn->priv; |
65dff759 | 2316 | int ret; |
65dff759 | 2317 | |
2bd59d48 | 2318 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2319 | return -ENOTDIR; |
2bd59d48 | 2320 | if (kn->parent != new_parent) |
ddbcc7e8 | 2321 | return -EIO; |
65dff759 | 2322 | |
6db8e85c TH |
2323 | /* |
2324 | * This isn't a proper migration and its usefulness is very | |
2325 | * limited. Disallow if sane_behavior. | |
2326 | */ | |
2327 | if (cgroup_sane_behavior(cgrp)) | |
2328 | return -EPERM; | |
099fca32 | 2329 | |
e1b2dc17 TH |
2330 | /* |
2331 | * We're gonna grab cgroup_tree_mutex which nests outside kernfs | |
2332 | * active_ref. kernfs_rename() doesn't require active_ref | |
2333 | * protection. Break them before grabbing cgroup_tree_mutex. | |
2334 | */ | |
2335 | kernfs_break_active_protection(new_parent); | |
2336 | kernfs_break_active_protection(kn); | |
099fca32 | 2337 | |
2bd59d48 TH |
2338 | mutex_lock(&cgroup_tree_mutex); |
2339 | mutex_lock(&cgroup_mutex); | |
099fca32 | 2340 | |
2bd59d48 | 2341 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 2342 | |
2bd59d48 TH |
2343 | mutex_unlock(&cgroup_mutex); |
2344 | mutex_unlock(&cgroup_tree_mutex); | |
e1b2dc17 TH |
2345 | |
2346 | kernfs_unbreak_active_protection(kn); | |
2347 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 2348 | return ret; |
099fca32 LZ |
2349 | } |
2350 | ||
49957f8e TH |
2351 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
2352 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
2353 | { | |
2354 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
2355 | .ia_uid = current_fsuid(), | |
2356 | .ia_gid = current_fsgid(), }; | |
2357 | ||
2358 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
2359 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
2360 | return 0; | |
2361 | ||
2362 | return kernfs_setattr(kn, &iattr); | |
2363 | } | |
2364 | ||
2bb566cb | 2365 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2366 | { |
8d7e6fb0 | 2367 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2368 | struct kernfs_node *kn; |
2369 | struct lock_class_key *key = NULL; | |
49957f8e | 2370 | int ret; |
05ef1d7c | 2371 | |
2bd59d48 TH |
2372 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2373 | key = &cft->lockdep_key; | |
2374 | #endif | |
2375 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2376 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2377 | NULL, false, key); | |
49957f8e TH |
2378 | if (IS_ERR(kn)) |
2379 | return PTR_ERR(kn); | |
2380 | ||
2381 | ret = cgroup_kn_set_ugid(kn); | |
2382 | if (ret) | |
2383 | kernfs_remove(kn); | |
2384 | return ret; | |
ddbcc7e8 PM |
2385 | } |
2386 | ||
b1f28d31 TH |
2387 | /** |
2388 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2389 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2390 | * @cfts: array of cftypes to be added |
2391 | * @is_add: whether to add or remove | |
2392 | * | |
2393 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2394 | * For removals, this function never fails. If addition fails, this |
2395 | * function doesn't remove files already added. The caller is responsible | |
2396 | * for cleaning up. | |
b1f28d31 | 2397 | */ |
2bb566cb TH |
2398 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2399 | bool is_add) | |
ddbcc7e8 | 2400 | { |
03b1cde6 | 2401 | struct cftype *cft; |
b1f28d31 TH |
2402 | int ret; |
2403 | ||
ace2bee8 | 2404 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2405 | |
2406 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2407 | /* does cft->flags tell us to skip this file on @cgrp? */ |
8cbbf2c9 TH |
2408 | if ((cft->flags & CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
2409 | continue; | |
873fe09e TH |
2410 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2411 | continue; | |
f33fddc2 G |
2412 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2413 | continue; | |
2414 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2415 | continue; | |
2416 | ||
2739d3cc | 2417 | if (is_add) { |
2bb566cb | 2418 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2419 | if (ret) { |
2739d3cc | 2420 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2421 | cft->name, ret); |
2422 | return ret; | |
2423 | } | |
2739d3cc LZ |
2424 | } else { |
2425 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2426 | } |
ddbcc7e8 | 2427 | } |
b1f28d31 | 2428 | return 0; |
ddbcc7e8 PM |
2429 | } |
2430 | ||
21a2d343 | 2431 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2432 | { |
2433 | LIST_HEAD(pending); | |
2bb566cb | 2434 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 2435 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 2436 | struct cgroup_subsys_state *css; |
9ccece80 | 2437 | int ret = 0; |
8e3f6541 | 2438 | |
21a2d343 | 2439 | lockdep_assert_held(&cgroup_tree_mutex); |
8e3f6541 | 2440 | |
21a2d343 | 2441 | /* don't bother if @ss isn't attached */ |
3dd06ffa | 2442 | if (ss->root == &cgrp_dfl_root) |
9ccece80 | 2443 | return 0; |
e8c82d20 | 2444 | |
e8c82d20 | 2445 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2446 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2447 | struct cgroup *cgrp = css->cgroup; |
2448 | ||
e8c82d20 LZ |
2449 | if (cgroup_is_dead(cgrp)) |
2450 | continue; | |
2451 | ||
21a2d343 | 2452 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2453 | if (ret) |
2454 | break; | |
8e3f6541 | 2455 | } |
21a2d343 TH |
2456 | |
2457 | if (is_add && !ret) | |
2458 | kernfs_activate(root->kn); | |
9ccece80 | 2459 | return ret; |
8e3f6541 TH |
2460 | } |
2461 | ||
2da440a2 | 2462 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 2463 | { |
2bb566cb | 2464 | struct cftype *cft; |
8e3f6541 | 2465 | |
2bd59d48 TH |
2466 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2467 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2468 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2469 | kfree(cft->kf_ops); | |
2470 | cft->kf_ops = NULL; | |
2da440a2 | 2471 | cft->ss = NULL; |
2bd59d48 | 2472 | } |
2da440a2 TH |
2473 | } |
2474 | ||
2bd59d48 | 2475 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2476 | { |
2477 | struct cftype *cft; | |
2478 | ||
2bd59d48 TH |
2479 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2480 | struct kernfs_ops *kf_ops; | |
2481 | ||
0adb0704 TH |
2482 | WARN_ON(cft->ss || cft->kf_ops); |
2483 | ||
2bd59d48 TH |
2484 | if (cft->seq_start) |
2485 | kf_ops = &cgroup_kf_ops; | |
2486 | else | |
2487 | kf_ops = &cgroup_kf_single_ops; | |
2488 | ||
2489 | /* | |
2490 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2491 | * make a copy of kf_ops to set its atomic_write_len. | |
2492 | */ | |
2493 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2494 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2495 | if (!kf_ops) { | |
2496 | cgroup_exit_cftypes(cfts); | |
2497 | return -ENOMEM; | |
2498 | } | |
2499 | kf_ops->atomic_write_len = cft->max_write_len; | |
2500 | } | |
8e3f6541 | 2501 | |
2bd59d48 | 2502 | cft->kf_ops = kf_ops; |
2bb566cb | 2503 | cft->ss = ss; |
2bd59d48 | 2504 | } |
2bb566cb | 2505 | |
2bd59d48 | 2506 | return 0; |
2da440a2 TH |
2507 | } |
2508 | ||
21a2d343 TH |
2509 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
2510 | { | |
2511 | lockdep_assert_held(&cgroup_tree_mutex); | |
2512 | ||
2513 | if (!cfts || !cfts[0].ss) | |
2514 | return -ENOENT; | |
2515 | ||
2516 | list_del(&cfts->node); | |
2517 | cgroup_apply_cftypes(cfts, false); | |
2518 | cgroup_exit_cftypes(cfts); | |
2519 | return 0; | |
8e3f6541 | 2520 | } |
8e3f6541 | 2521 | |
79578621 TH |
2522 | /** |
2523 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
2524 | * @cfts: zero-length name terminated array of cftypes |
2525 | * | |
2bb566cb TH |
2526 | * Unregister @cfts. Files described by @cfts are removed from all |
2527 | * existing cgroups and all future cgroups won't have them either. This | |
2528 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
2529 | * |
2530 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 2531 | * registered. |
79578621 | 2532 | */ |
2bb566cb | 2533 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 2534 | { |
21a2d343 | 2535 | int ret; |
79578621 | 2536 | |
21a2d343 TH |
2537 | mutex_lock(&cgroup_tree_mutex); |
2538 | ret = cgroup_rm_cftypes_locked(cfts); | |
2539 | mutex_unlock(&cgroup_tree_mutex); | |
2540 | return ret; | |
80b13586 TH |
2541 | } |
2542 | ||
8e3f6541 TH |
2543 | /** |
2544 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2545 | * @ss: target cgroup subsystem | |
2546 | * @cfts: zero-length name terminated array of cftypes | |
2547 | * | |
2548 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2549 | * existing cgroups to which @ss is attached and all future cgroups will | |
2550 | * have them too. This function can be called anytime whether @ss is | |
2551 | * attached or not. | |
2552 | * | |
2553 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2554 | * function currently returns 0 as long as @cfts registration is successful | |
2555 | * even if some file creation attempts on existing cgroups fail. | |
2556 | */ | |
03b1cde6 | 2557 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 2558 | { |
9ccece80 | 2559 | int ret; |
8e3f6541 | 2560 | |
dc5736ed LZ |
2561 | if (!cfts || cfts[0].name[0] == '\0') |
2562 | return 0; | |
2bb566cb | 2563 | |
2bd59d48 TH |
2564 | ret = cgroup_init_cftypes(ss, cfts); |
2565 | if (ret) | |
2566 | return ret; | |
79578621 | 2567 | |
21a2d343 TH |
2568 | mutex_lock(&cgroup_tree_mutex); |
2569 | ||
0adb0704 | 2570 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 2571 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 2572 | if (ret) |
21a2d343 | 2573 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 2574 | |
21a2d343 | 2575 | mutex_unlock(&cgroup_tree_mutex); |
9ccece80 | 2576 | return ret; |
79578621 TH |
2577 | } |
2578 | ||
a043e3b2 LZ |
2579 | /** |
2580 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2581 | * @cgrp: the cgroup in question | |
2582 | * | |
2583 | * Return the number of tasks in the cgroup. | |
2584 | */ | |
07bc356e | 2585 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2586 | { |
2587 | int count = 0; | |
69d0206c | 2588 | struct cgrp_cset_link *link; |
817929ec | 2589 | |
96d365e0 | 2590 | down_read(&css_set_rwsem); |
69d0206c TH |
2591 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2592 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 2593 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
2594 | return count; |
2595 | } | |
2596 | ||
53fa5261 | 2597 | /** |
492eb21b TH |
2598 | * css_next_child - find the next child of a given css |
2599 | * @pos_css: the current position (%NULL to initiate traversal) | |
2600 | * @parent_css: css whose children to walk | |
53fa5261 | 2601 | * |
492eb21b | 2602 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
2603 | * under either cgroup_mutex or RCU read lock. The only requirement is |
2604 | * that @parent_css and @pos_css are accessible. The next sibling is | |
2605 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 2606 | */ |
492eb21b TH |
2607 | struct cgroup_subsys_state * |
2608 | css_next_child(struct cgroup_subsys_state *pos_css, | |
2609 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 2610 | { |
492eb21b TH |
2611 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
2612 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
2613 | struct cgroup *next; |
2614 | ||
ace2bee8 | 2615 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
2616 | |
2617 | /* | |
2618 | * @pos could already have been removed. Once a cgroup is removed, | |
2619 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
2620 | * changes. As CGRP_DEAD assertion is serialized and happens |
2621 | * before the cgroup is taken off the ->sibling list, if we see it | |
2622 | * unasserted, it's guaranteed that the next sibling hasn't | |
2623 | * finished its grace period even if it's already removed, and thus | |
2624 | * safe to dereference from this RCU critical section. If | |
2625 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
2626 | * to be visible as %true here. | |
3b287a50 TH |
2627 | * |
2628 | * If @pos is dead, its next pointer can't be dereferenced; | |
2629 | * however, as each cgroup is given a monotonically increasing | |
2630 | * unique serial number and always appended to the sibling list, | |
2631 | * the next one can be found by walking the parent's children until | |
2632 | * we see a cgroup with higher serial number than @pos's. While | |
2633 | * this path can be slower, it's taken only when either the current | |
2634 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 2635 | */ |
3b287a50 TH |
2636 | if (!pos) { |
2637 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
2638 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 2639 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
2640 | } else { |
2641 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
2642 | if (next->serial_nr > pos->serial_nr) | |
2643 | break; | |
53fa5261 TH |
2644 | } |
2645 | ||
492eb21b TH |
2646 | if (&next->sibling == &cgrp->children) |
2647 | return NULL; | |
2648 | ||
ca8bdcaf | 2649 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 2650 | } |
53fa5261 | 2651 | |
574bd9f7 | 2652 | /** |
492eb21b | 2653 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 2654 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2655 | * @root: css whose descendants to walk |
574bd9f7 | 2656 | * |
492eb21b | 2657 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
2658 | * to visit for pre-order traversal of @root's descendants. @root is |
2659 | * included in the iteration and the first node to be visited. | |
75501a6d | 2660 | * |
87fb54f1 TH |
2661 | * While this function requires cgroup_mutex or RCU read locking, it |
2662 | * doesn't require the whole traversal to be contained in a single critical | |
2663 | * section. This function will return the correct next descendant as long | |
2664 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 2665 | */ |
492eb21b TH |
2666 | struct cgroup_subsys_state * |
2667 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
2668 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2669 | { |
492eb21b | 2670 | struct cgroup_subsys_state *next; |
574bd9f7 | 2671 | |
ace2bee8 | 2672 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2673 | |
bd8815a6 | 2674 | /* if first iteration, visit @root */ |
7805d000 | 2675 | if (!pos) |
bd8815a6 | 2676 | return root; |
574bd9f7 TH |
2677 | |
2678 | /* visit the first child if exists */ | |
492eb21b | 2679 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
2680 | if (next) |
2681 | return next; | |
2682 | ||
2683 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
2684 | while (pos != root) { |
2685 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 2686 | if (next) |
574bd9f7 | 2687 | return next; |
492eb21b | 2688 | pos = css_parent(pos); |
7805d000 | 2689 | } |
574bd9f7 TH |
2690 | |
2691 | return NULL; | |
2692 | } | |
574bd9f7 | 2693 | |
12a9d2fe | 2694 | /** |
492eb21b TH |
2695 | * css_rightmost_descendant - return the rightmost descendant of a css |
2696 | * @pos: css of interest | |
12a9d2fe | 2697 | * |
492eb21b TH |
2698 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
2699 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 2700 | * subtree of @pos. |
75501a6d | 2701 | * |
87fb54f1 TH |
2702 | * While this function requires cgroup_mutex or RCU read locking, it |
2703 | * doesn't require the whole traversal to be contained in a single critical | |
2704 | * section. This function will return the correct rightmost descendant as | |
2705 | * long as @pos is accessible. | |
12a9d2fe | 2706 | */ |
492eb21b TH |
2707 | struct cgroup_subsys_state * |
2708 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 2709 | { |
492eb21b | 2710 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 2711 | |
ace2bee8 | 2712 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
2713 | |
2714 | do { | |
2715 | last = pos; | |
2716 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
2717 | pos = NULL; | |
492eb21b | 2718 | css_for_each_child(tmp, last) |
12a9d2fe TH |
2719 | pos = tmp; |
2720 | } while (pos); | |
2721 | ||
2722 | return last; | |
2723 | } | |
12a9d2fe | 2724 | |
492eb21b TH |
2725 | static struct cgroup_subsys_state * |
2726 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 2727 | { |
492eb21b | 2728 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
2729 | |
2730 | do { | |
2731 | last = pos; | |
492eb21b | 2732 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
2733 | } while (pos); |
2734 | ||
2735 | return last; | |
2736 | } | |
2737 | ||
2738 | /** | |
492eb21b | 2739 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 2740 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2741 | * @root: css whose descendants to walk |
574bd9f7 | 2742 | * |
492eb21b | 2743 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
2744 | * to visit for post-order traversal of @root's descendants. @root is |
2745 | * included in the iteration and the last node to be visited. | |
75501a6d | 2746 | * |
87fb54f1 TH |
2747 | * While this function requires cgroup_mutex or RCU read locking, it |
2748 | * doesn't require the whole traversal to be contained in a single critical | |
2749 | * section. This function will return the correct next descendant as long | |
2750 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
2751 | * @cgroup. | |
574bd9f7 | 2752 | */ |
492eb21b TH |
2753 | struct cgroup_subsys_state * |
2754 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
2755 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2756 | { |
492eb21b | 2757 | struct cgroup_subsys_state *next; |
574bd9f7 | 2758 | |
ace2bee8 | 2759 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2760 | |
58b79a91 TH |
2761 | /* if first iteration, visit leftmost descendant which may be @root */ |
2762 | if (!pos) | |
2763 | return css_leftmost_descendant(root); | |
574bd9f7 | 2764 | |
bd8815a6 TH |
2765 | /* if we visited @root, we're done */ |
2766 | if (pos == root) | |
2767 | return NULL; | |
2768 | ||
574bd9f7 | 2769 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 2770 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 2771 | if (next) |
492eb21b | 2772 | return css_leftmost_descendant(next); |
574bd9f7 TH |
2773 | |
2774 | /* no sibling left, visit parent */ | |
bd8815a6 | 2775 | return css_parent(pos); |
574bd9f7 | 2776 | } |
574bd9f7 | 2777 | |
0942eeee | 2778 | /** |
72ec7029 | 2779 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
2780 | * @it: the iterator to advance |
2781 | * | |
2782 | * Advance @it to the next css_set to walk. | |
d515876e | 2783 | */ |
72ec7029 | 2784 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
2785 | { |
2786 | struct list_head *l = it->cset_link; | |
2787 | struct cgrp_cset_link *link; | |
2788 | struct css_set *cset; | |
2789 | ||
2790 | /* Advance to the next non-empty css_set */ | |
2791 | do { | |
2792 | l = l->next; | |
72ec7029 | 2793 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
2794 | it->cset_link = NULL; |
2795 | return; | |
2796 | } | |
2797 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
2798 | cset = link->cset; | |
c7561128 TH |
2799 | } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); |
2800 | ||
d515876e | 2801 | it->cset_link = l; |
c7561128 TH |
2802 | |
2803 | if (!list_empty(&cset->tasks)) | |
2804 | it->task = cset->tasks.next; | |
2805 | else | |
2806 | it->task = cset->mg_tasks.next; | |
d515876e TH |
2807 | } |
2808 | ||
0942eeee | 2809 | /** |
72ec7029 TH |
2810 | * css_task_iter_start - initiate task iteration |
2811 | * @css: the css to walk tasks of | |
0942eeee TH |
2812 | * @it: the task iterator to use |
2813 | * | |
72ec7029 TH |
2814 | * Initiate iteration through the tasks of @css. The caller can call |
2815 | * css_task_iter_next() to walk through the tasks until the function | |
2816 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
2817 | * called. | |
0942eeee TH |
2818 | * |
2819 | * Note that this function acquires a lock which is released when the | |
2820 | * iteration finishes. The caller can't sleep while iteration is in | |
2821 | * progress. | |
2822 | */ | |
72ec7029 TH |
2823 | void css_task_iter_start(struct cgroup_subsys_state *css, |
2824 | struct css_task_iter *it) | |
96d365e0 | 2825 | __acquires(css_set_rwsem) |
817929ec | 2826 | { |
56fde9e0 TH |
2827 | /* no one should try to iterate before mounting cgroups */ |
2828 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 2829 | |
96d365e0 | 2830 | down_read(&css_set_rwsem); |
c59cd3d8 | 2831 | |
72ec7029 TH |
2832 | it->origin_css = css; |
2833 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 2834 | |
72ec7029 | 2835 | css_advance_task_iter(it); |
817929ec PM |
2836 | } |
2837 | ||
0942eeee | 2838 | /** |
72ec7029 | 2839 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
2840 | * @it: the task iterator being iterated |
2841 | * | |
2842 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
2843 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
2844 | * reaches the end. | |
0942eeee | 2845 | */ |
72ec7029 | 2846 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
2847 | { |
2848 | struct task_struct *res; | |
2849 | struct list_head *l = it->task; | |
c7561128 TH |
2850 | struct cgrp_cset_link *link = list_entry(it->cset_link, |
2851 | struct cgrp_cset_link, cset_link); | |
817929ec PM |
2852 | |
2853 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 2854 | if (!it->cset_link) |
817929ec PM |
2855 | return NULL; |
2856 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
2857 | |
2858 | /* | |
2859 | * Advance iterator to find next entry. cset->tasks is consumed | |
2860 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
2861 | * next cset. | |
2862 | */ | |
817929ec | 2863 | l = l->next; |
c7561128 TH |
2864 | |
2865 | if (l == &link->cset->tasks) | |
2866 | l = link->cset->mg_tasks.next; | |
2867 | ||
2868 | if (l == &link->cset->mg_tasks) | |
72ec7029 | 2869 | css_advance_task_iter(it); |
c7561128 | 2870 | else |
817929ec | 2871 | it->task = l; |
c7561128 | 2872 | |
817929ec PM |
2873 | return res; |
2874 | } | |
2875 | ||
0942eeee | 2876 | /** |
72ec7029 | 2877 | * css_task_iter_end - finish task iteration |
0942eeee TH |
2878 | * @it: the task iterator to finish |
2879 | * | |
72ec7029 | 2880 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 2881 | */ |
72ec7029 | 2882 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 2883 | __releases(css_set_rwsem) |
31a7df01 | 2884 | { |
96d365e0 | 2885 | up_read(&css_set_rwsem); |
31a7df01 CW |
2886 | } |
2887 | ||
2888 | /** | |
8cc99345 TH |
2889 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
2890 | * @to: cgroup to which the tasks will be moved | |
2891 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 2892 | * |
eaf797ab TH |
2893 | * Locking rules between cgroup_post_fork() and the migration path |
2894 | * guarantee that, if a task is forking while being migrated, the new child | |
2895 | * is guaranteed to be either visible in the source cgroup after the | |
2896 | * parent's migration is complete or put into the target cgroup. No task | |
2897 | * can slip out of migration through forking. | |
31a7df01 | 2898 | */ |
8cc99345 | 2899 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 2900 | { |
952aaa12 TH |
2901 | LIST_HEAD(preloaded_csets); |
2902 | struct cgrp_cset_link *link; | |
72ec7029 | 2903 | struct css_task_iter it; |
e406d1cf | 2904 | struct task_struct *task; |
952aaa12 | 2905 | int ret; |
31a7df01 | 2906 | |
952aaa12 | 2907 | mutex_lock(&cgroup_mutex); |
31a7df01 | 2908 | |
952aaa12 TH |
2909 | /* all tasks in @from are being moved, all csets are source */ |
2910 | down_read(&css_set_rwsem); | |
2911 | list_for_each_entry(link, &from->cset_links, cset_link) | |
2912 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
2913 | up_read(&css_set_rwsem); | |
31a7df01 | 2914 | |
952aaa12 TH |
2915 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
2916 | if (ret) | |
2917 | goto out_err; | |
8cc99345 | 2918 | |
952aaa12 TH |
2919 | /* |
2920 | * Migrate tasks one-by-one until @form is empty. This fails iff | |
2921 | * ->can_attach() fails. | |
2922 | */ | |
e406d1cf TH |
2923 | do { |
2924 | css_task_iter_start(&from->dummy_css, &it); | |
2925 | task = css_task_iter_next(&it); | |
2926 | if (task) | |
2927 | get_task_struct(task); | |
2928 | css_task_iter_end(&it); | |
2929 | ||
2930 | if (task) { | |
952aaa12 | 2931 | ret = cgroup_migrate(to, task, false); |
e406d1cf TH |
2932 | put_task_struct(task); |
2933 | } | |
2934 | } while (task && !ret); | |
952aaa12 TH |
2935 | out_err: |
2936 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 2937 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 2938 | return ret; |
8cc99345 TH |
2939 | } |
2940 | ||
bbcb81d0 | 2941 | /* |
102a775e | 2942 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
2943 | * |
2944 | * Reading this file can return large amounts of data if a cgroup has | |
2945 | * *lots* of attached tasks. So it may need several calls to read(), | |
2946 | * but we cannot guarantee that the information we produce is correct | |
2947 | * unless we produce it entirely atomically. | |
2948 | * | |
bbcb81d0 | 2949 | */ |
bbcb81d0 | 2950 | |
24528255 LZ |
2951 | /* which pidlist file are we talking about? */ |
2952 | enum cgroup_filetype { | |
2953 | CGROUP_FILE_PROCS, | |
2954 | CGROUP_FILE_TASKS, | |
2955 | }; | |
2956 | ||
2957 | /* | |
2958 | * A pidlist is a list of pids that virtually represents the contents of one | |
2959 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
2960 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
2961 | * to the cgroup. | |
2962 | */ | |
2963 | struct cgroup_pidlist { | |
2964 | /* | |
2965 | * used to find which pidlist is wanted. doesn't change as long as | |
2966 | * this particular list stays in the list. | |
2967 | */ | |
2968 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
2969 | /* array of xids */ | |
2970 | pid_t *list; | |
2971 | /* how many elements the above list has */ | |
2972 | int length; | |
24528255 LZ |
2973 | /* each of these stored in a list by its cgroup */ |
2974 | struct list_head links; | |
2975 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
2976 | struct cgroup *owner; | |
b1a21367 TH |
2977 | /* for delayed destruction */ |
2978 | struct delayed_work destroy_dwork; | |
24528255 LZ |
2979 | }; |
2980 | ||
d1d9fd33 BB |
2981 | /* |
2982 | * The following two functions "fix" the issue where there are more pids | |
2983 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
2984 | * TODO: replace with a kernel-wide solution to this problem | |
2985 | */ | |
2986 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
2987 | static void *pidlist_allocate(int count) | |
2988 | { | |
2989 | if (PIDLIST_TOO_LARGE(count)) | |
2990 | return vmalloc(count * sizeof(pid_t)); | |
2991 | else | |
2992 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
2993 | } | |
b1a21367 | 2994 | |
d1d9fd33 BB |
2995 | static void pidlist_free(void *p) |
2996 | { | |
2997 | if (is_vmalloc_addr(p)) | |
2998 | vfree(p); | |
2999 | else | |
3000 | kfree(p); | |
3001 | } | |
d1d9fd33 | 3002 | |
b1a21367 TH |
3003 | /* |
3004 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3005 | * should be left afterwards. | |
3006 | */ | |
3007 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3008 | { | |
3009 | struct cgroup_pidlist *l, *tmp_l; | |
3010 | ||
3011 | mutex_lock(&cgrp->pidlist_mutex); | |
3012 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3013 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3014 | mutex_unlock(&cgrp->pidlist_mutex); | |
3015 | ||
3016 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3017 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3018 | } | |
3019 | ||
3020 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3021 | { | |
3022 | struct delayed_work *dwork = to_delayed_work(work); | |
3023 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3024 | destroy_dwork); | |
3025 | struct cgroup_pidlist *tofree = NULL; | |
3026 | ||
3027 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3028 | |
3029 | /* | |
04502365 TH |
3030 | * Destroy iff we didn't get queued again. The state won't change |
3031 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3032 | */ |
04502365 | 3033 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3034 | list_del(&l->links); |
3035 | pidlist_free(l->list); | |
3036 | put_pid_ns(l->key.ns); | |
3037 | tofree = l; | |
3038 | } | |
3039 | ||
b1a21367 TH |
3040 | mutex_unlock(&l->owner->pidlist_mutex); |
3041 | kfree(tofree); | |
3042 | } | |
3043 | ||
bbcb81d0 | 3044 | /* |
102a775e | 3045 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3046 | * Returns the number of unique elements. |
bbcb81d0 | 3047 | */ |
6ee211ad | 3048 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3049 | { |
102a775e | 3050 | int src, dest = 1; |
102a775e BB |
3051 | |
3052 | /* | |
3053 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3054 | * edge cases first; no work needs to be done for either | |
3055 | */ | |
3056 | if (length == 0 || length == 1) | |
3057 | return length; | |
3058 | /* src and dest walk down the list; dest counts unique elements */ | |
3059 | for (src = 1; src < length; src++) { | |
3060 | /* find next unique element */ | |
3061 | while (list[src] == list[src-1]) { | |
3062 | src++; | |
3063 | if (src == length) | |
3064 | goto after; | |
3065 | } | |
3066 | /* dest always points to where the next unique element goes */ | |
3067 | list[dest] = list[src]; | |
3068 | dest++; | |
3069 | } | |
3070 | after: | |
102a775e BB |
3071 | return dest; |
3072 | } | |
3073 | ||
afb2bc14 TH |
3074 | /* |
3075 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3076 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3077 | * different per namespace, each namespace needs differently sorted list, | |
3078 | * making it impossible to use, for example, single rbtree of member tasks | |
3079 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3080 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3081 | * pidlists keyed by cgroup and namespace. | |
3082 | * | |
3083 | * All this extra complexity was caused by the original implementation | |
3084 | * committing to an entirely unnecessary property. In the long term, we | |
3085 | * want to do away with it. Explicitly scramble sort order if | |
3086 | * sane_behavior so that no such expectation exists in the new interface. | |
3087 | * | |
3088 | * Scrambling is done by swapping every two consecutive bits, which is | |
3089 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3090 | */ | |
3091 | static pid_t pid_fry(pid_t pid) | |
3092 | { | |
3093 | unsigned a = pid & 0x55555555; | |
3094 | unsigned b = pid & 0xAAAAAAAA; | |
3095 | ||
3096 | return (a << 1) | (b >> 1); | |
3097 | } | |
3098 | ||
3099 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3100 | { | |
3101 | if (cgroup_sane_behavior(cgrp)) | |
3102 | return pid_fry(pid); | |
3103 | else | |
3104 | return pid; | |
3105 | } | |
3106 | ||
102a775e BB |
3107 | static int cmppid(const void *a, const void *b) |
3108 | { | |
3109 | return *(pid_t *)a - *(pid_t *)b; | |
3110 | } | |
3111 | ||
afb2bc14 TH |
3112 | static int fried_cmppid(const void *a, const void *b) |
3113 | { | |
3114 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3115 | } | |
3116 | ||
e6b81710 TH |
3117 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3118 | enum cgroup_filetype type) | |
3119 | { | |
3120 | struct cgroup_pidlist *l; | |
3121 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3122 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3123 | ||
3124 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3125 | ||
3126 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3127 | if (l->key.type == type && l->key.ns == ns) | |
3128 | return l; | |
3129 | return NULL; | |
3130 | } | |
3131 | ||
72a8cb30 BB |
3132 | /* |
3133 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3134 | * returns with the lock on that pidlist already held, and takes care | |
3135 | * of the use count, or returns NULL with no locks held if we're out of | |
3136 | * memory. | |
3137 | */ | |
e6b81710 TH |
3138 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3139 | enum cgroup_filetype type) | |
72a8cb30 BB |
3140 | { |
3141 | struct cgroup_pidlist *l; | |
b70cc5fd | 3142 | |
e6b81710 TH |
3143 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3144 | ||
3145 | l = cgroup_pidlist_find(cgrp, type); | |
3146 | if (l) | |
3147 | return l; | |
3148 | ||
72a8cb30 | 3149 | /* entry not found; create a new one */ |
f4f4be2b | 3150 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3151 | if (!l) |
72a8cb30 | 3152 | return l; |
e6b81710 | 3153 | |
b1a21367 | 3154 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3155 | l->key.type = type; |
e6b81710 TH |
3156 | /* don't need task_nsproxy() if we're looking at ourself */ |
3157 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3158 | l->owner = cgrp; |
3159 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3160 | return l; |
3161 | } | |
3162 | ||
102a775e BB |
3163 | /* |
3164 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3165 | */ | |
72a8cb30 BB |
3166 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3167 | struct cgroup_pidlist **lp) | |
102a775e BB |
3168 | { |
3169 | pid_t *array; | |
3170 | int length; | |
3171 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3172 | struct css_task_iter it; |
817929ec | 3173 | struct task_struct *tsk; |
102a775e BB |
3174 | struct cgroup_pidlist *l; |
3175 | ||
4bac00d1 TH |
3176 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3177 | ||
102a775e BB |
3178 | /* |
3179 | * If cgroup gets more users after we read count, we won't have | |
3180 | * enough space - tough. This race is indistinguishable to the | |
3181 | * caller from the case that the additional cgroup users didn't | |
3182 | * show up until sometime later on. | |
3183 | */ | |
3184 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3185 | array = pidlist_allocate(length); |
102a775e BB |
3186 | if (!array) |
3187 | return -ENOMEM; | |
3188 | /* now, populate the array */ | |
72ec7029 TH |
3189 | css_task_iter_start(&cgrp->dummy_css, &it); |
3190 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3191 | if (unlikely(n == length)) |
817929ec | 3192 | break; |
102a775e | 3193 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3194 | if (type == CGROUP_FILE_PROCS) |
3195 | pid = task_tgid_vnr(tsk); | |
3196 | else | |
3197 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3198 | if (pid > 0) /* make sure to only use valid results */ |
3199 | array[n++] = pid; | |
817929ec | 3200 | } |
72ec7029 | 3201 | css_task_iter_end(&it); |
102a775e BB |
3202 | length = n; |
3203 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3204 | if (cgroup_sane_behavior(cgrp)) |
3205 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3206 | else | |
3207 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3208 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3209 | length = pidlist_uniq(array, length); |
e6b81710 | 3210 | |
e6b81710 | 3211 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3212 | if (!l) { |
e6b81710 | 3213 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3214 | pidlist_free(array); |
72a8cb30 | 3215 | return -ENOMEM; |
102a775e | 3216 | } |
e6b81710 TH |
3217 | |
3218 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3219 | pidlist_free(l->list); |
102a775e BB |
3220 | l->list = array; |
3221 | l->length = length; | |
72a8cb30 | 3222 | *lp = l; |
102a775e | 3223 | return 0; |
bbcb81d0 PM |
3224 | } |
3225 | ||
846c7bb0 | 3226 | /** |
a043e3b2 | 3227 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3228 | * @stats: cgroupstats to fill information into |
3229 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3230 | * been requested. | |
a043e3b2 LZ |
3231 | * |
3232 | * Build and fill cgroupstats so that taskstats can export it to user | |
3233 | * space. | |
846c7bb0 BS |
3234 | */ |
3235 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3236 | { | |
2bd59d48 | 3237 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3238 | struct cgroup *cgrp; |
72ec7029 | 3239 | struct css_task_iter it; |
846c7bb0 | 3240 | struct task_struct *tsk; |
33d283be | 3241 | |
2bd59d48 TH |
3242 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3243 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3244 | kernfs_type(kn) != KERNFS_DIR) | |
3245 | return -EINVAL; | |
3246 | ||
bad34660 LZ |
3247 | mutex_lock(&cgroup_mutex); |
3248 | ||
846c7bb0 | 3249 | /* |
2bd59d48 TH |
3250 | * We aren't being called from kernfs and there's no guarantee on |
3251 | * @kn->priv's validity. For this and css_tryget_from_dir(), | |
3252 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
846c7bb0 | 3253 | */ |
2bd59d48 TH |
3254 | rcu_read_lock(); |
3255 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 3256 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 3257 | rcu_read_unlock(); |
bad34660 | 3258 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
3259 | return -ENOENT; |
3260 | } | |
bad34660 | 3261 | rcu_read_unlock(); |
846c7bb0 | 3262 | |
72ec7029 TH |
3263 | css_task_iter_start(&cgrp->dummy_css, &it); |
3264 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3265 | switch (tsk->state) { |
3266 | case TASK_RUNNING: | |
3267 | stats->nr_running++; | |
3268 | break; | |
3269 | case TASK_INTERRUPTIBLE: | |
3270 | stats->nr_sleeping++; | |
3271 | break; | |
3272 | case TASK_UNINTERRUPTIBLE: | |
3273 | stats->nr_uninterruptible++; | |
3274 | break; | |
3275 | case TASK_STOPPED: | |
3276 | stats->nr_stopped++; | |
3277 | break; | |
3278 | default: | |
3279 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3280 | stats->nr_io_wait++; | |
3281 | break; | |
3282 | } | |
3283 | } | |
72ec7029 | 3284 | css_task_iter_end(&it); |
846c7bb0 | 3285 | |
bad34660 | 3286 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 3287 | return 0; |
846c7bb0 BS |
3288 | } |
3289 | ||
8f3ff208 | 3290 | |
bbcb81d0 | 3291 | /* |
102a775e | 3292 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3293 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3294 | * in the cgroup->l->list array. |
bbcb81d0 | 3295 | */ |
cc31edce | 3296 | |
102a775e | 3297 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3298 | { |
cc31edce PM |
3299 | /* |
3300 | * Initially we receive a position value that corresponds to | |
3301 | * one more than the last pid shown (or 0 on the first call or | |
3302 | * after a seek to the start). Use a binary-search to find the | |
3303 | * next pid to display, if any | |
3304 | */ | |
2bd59d48 | 3305 | struct kernfs_open_file *of = s->private; |
7da11279 | 3306 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3307 | struct cgroup_pidlist *l; |
7da11279 | 3308 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3309 | int index = 0, pid = *pos; |
4bac00d1 TH |
3310 | int *iter, ret; |
3311 | ||
3312 | mutex_lock(&cgrp->pidlist_mutex); | |
3313 | ||
3314 | /* | |
5d22444f | 3315 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3316 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3317 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3318 | * could already have been destroyed. |
3319 | */ | |
5d22444f TH |
3320 | if (of->priv) |
3321 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3322 | |
3323 | /* | |
3324 | * Either this is the first start() after open or the matching | |
3325 | * pidlist has been destroyed inbetween. Create a new one. | |
3326 | */ | |
5d22444f TH |
3327 | if (!of->priv) { |
3328 | ret = pidlist_array_load(cgrp, type, | |
3329 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3330 | if (ret) |
3331 | return ERR_PTR(ret); | |
3332 | } | |
5d22444f | 3333 | l = of->priv; |
cc31edce | 3334 | |
cc31edce | 3335 | if (pid) { |
102a775e | 3336 | int end = l->length; |
20777766 | 3337 | |
cc31edce PM |
3338 | while (index < end) { |
3339 | int mid = (index + end) / 2; | |
afb2bc14 | 3340 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3341 | index = mid; |
3342 | break; | |
afb2bc14 | 3343 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3344 | index = mid + 1; |
3345 | else | |
3346 | end = mid; | |
3347 | } | |
3348 | } | |
3349 | /* If we're off the end of the array, we're done */ | |
102a775e | 3350 | if (index >= l->length) |
cc31edce PM |
3351 | return NULL; |
3352 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3353 | iter = l->list + index; |
afb2bc14 | 3354 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3355 | return iter; |
3356 | } | |
3357 | ||
102a775e | 3358 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3359 | { |
2bd59d48 | 3360 | struct kernfs_open_file *of = s->private; |
5d22444f | 3361 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3362 | |
5d22444f TH |
3363 | if (l) |
3364 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3365 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3366 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3367 | } |
3368 | ||
102a775e | 3369 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3370 | { |
2bd59d48 | 3371 | struct kernfs_open_file *of = s->private; |
5d22444f | 3372 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3373 | pid_t *p = v; |
3374 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3375 | /* |
3376 | * Advance to the next pid in the array. If this goes off the | |
3377 | * end, we're done | |
3378 | */ | |
3379 | p++; | |
3380 | if (p >= end) { | |
3381 | return NULL; | |
3382 | } else { | |
7da11279 | 3383 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3384 | return p; |
3385 | } | |
3386 | } | |
3387 | ||
102a775e | 3388 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3389 | { |
3390 | return seq_printf(s, "%d\n", *(int *)v); | |
3391 | } | |
bbcb81d0 | 3392 | |
102a775e BB |
3393 | /* |
3394 | * seq_operations functions for iterating on pidlists through seq_file - | |
3395 | * independent of whether it's tasks or procs | |
3396 | */ | |
3397 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3398 | .start = cgroup_pidlist_start, | |
3399 | .stop = cgroup_pidlist_stop, | |
3400 | .next = cgroup_pidlist_next, | |
3401 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3402 | }; |
3403 | ||
182446d0 TH |
3404 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3405 | struct cftype *cft) | |
81a6a5cd | 3406 | { |
182446d0 | 3407 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3408 | } |
3409 | ||
182446d0 TH |
3410 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3411 | struct cftype *cft, u64 val) | |
6379c106 | 3412 | { |
182446d0 | 3413 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3414 | if (val) |
182446d0 | 3415 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3416 | else |
182446d0 | 3417 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3418 | return 0; |
3419 | } | |
3420 | ||
182446d0 TH |
3421 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3422 | struct cftype *cft) | |
97978e6d | 3423 | { |
182446d0 | 3424 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3425 | } |
3426 | ||
182446d0 TH |
3427 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3428 | struct cftype *cft, u64 val) | |
97978e6d DL |
3429 | { |
3430 | if (val) | |
182446d0 | 3431 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3432 | else |
182446d0 | 3433 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3434 | return 0; |
3435 | } | |
3436 | ||
d5c56ced | 3437 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3438 | { |
d5c56ced | 3439 | .name = "cgroup.procs", |
6612f05b TH |
3440 | .seq_start = cgroup_pidlist_start, |
3441 | .seq_next = cgroup_pidlist_next, | |
3442 | .seq_stop = cgroup_pidlist_stop, | |
3443 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3444 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3445 | .write_u64 = cgroup_procs_write, |
74a1166d | 3446 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3447 | }, |
97978e6d DL |
3448 | { |
3449 | .name = "cgroup.clone_children", | |
873fe09e | 3450 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3451 | .read_u64 = cgroup_clone_children_read, |
3452 | .write_u64 = cgroup_clone_children_write, | |
3453 | }, | |
873fe09e TH |
3454 | { |
3455 | .name = "cgroup.sane_behavior", | |
3456 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3457 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3458 | }, |
d5c56ced TH |
3459 | |
3460 | /* | |
3461 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3462 | * don't exist if sane_behavior. If you're depending on these, be | |
3463 | * prepared to be burned. | |
3464 | */ | |
3465 | { | |
3466 | .name = "tasks", | |
3467 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3468 | .seq_start = cgroup_pidlist_start, |
3469 | .seq_next = cgroup_pidlist_next, | |
3470 | .seq_stop = cgroup_pidlist_stop, | |
3471 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3472 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3473 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3474 | .mode = S_IRUGO | S_IWUSR, |
3475 | }, | |
3476 | { | |
3477 | .name = "notify_on_release", | |
3478 | .flags = CFTYPE_INSANE, | |
3479 | .read_u64 = cgroup_read_notify_on_release, | |
3480 | .write_u64 = cgroup_write_notify_on_release, | |
3481 | }, | |
6e6ff25b TH |
3482 | { |
3483 | .name = "release_agent", | |
cc5943a7 | 3484 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3485 | .seq_show = cgroup_release_agent_show, |
6e6ff25b | 3486 | .write_string = cgroup_release_agent_write, |
5f469907 | 3487 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 3488 | }, |
db0416b6 | 3489 | { } /* terminate */ |
bbcb81d0 PM |
3490 | }; |
3491 | ||
13af07df | 3492 | /** |
628f7cd4 | 3493 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3494 | * @cgrp: target cgroup |
13af07df | 3495 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3496 | * |
3497 | * On failure, no file is added. | |
13af07df | 3498 | */ |
628f7cd4 | 3499 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3500 | { |
ddbcc7e8 | 3501 | struct cgroup_subsys *ss; |
b420ba7d | 3502 | int i, ret = 0; |
bbcb81d0 | 3503 | |
8e3f6541 | 3504 | /* process cftsets of each subsystem */ |
b420ba7d | 3505 | for_each_subsys(ss, i) { |
0adb0704 | 3506 | struct cftype *cfts; |
b420ba7d TH |
3507 | |
3508 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3509 | continue; |
8e3f6541 | 3510 | |
0adb0704 TH |
3511 | list_for_each_entry(cfts, &ss->cfts, node) { |
3512 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
3513 | if (ret < 0) |
3514 | goto err; | |
3515 | } | |
ddbcc7e8 | 3516 | } |
ddbcc7e8 | 3517 | return 0; |
bee55099 TH |
3518 | err: |
3519 | cgroup_clear_dir(cgrp, subsys_mask); | |
3520 | return ret; | |
ddbcc7e8 PM |
3521 | } |
3522 | ||
0c21ead1 TH |
3523 | /* |
3524 | * css destruction is four-stage process. | |
3525 | * | |
3526 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3527 | * Implemented in kill_css(). | |
3528 | * | |
3529 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3530 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3531 | * by invoking offline_css(). After offlining, the base ref is put. | |
3532 | * Implemented in css_killed_work_fn(). | |
3533 | * | |
3534 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3535 | * accessors are inside RCU read sections. css_release() schedules the | |
3536 | * RCU callback. | |
3537 | * | |
3538 | * 4. After the grace period, the css can be freed. Implemented in | |
3539 | * css_free_work_fn(). | |
3540 | * | |
3541 | * It is actually hairier because both step 2 and 4 require process context | |
3542 | * and thus involve punting to css->destroy_work adding two additional | |
3543 | * steps to the already complex sequence. | |
3544 | */ | |
35ef10da | 3545 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3546 | { |
3547 | struct cgroup_subsys_state *css = | |
35ef10da | 3548 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 3549 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3550 | |
0ae78e0b TH |
3551 | if (css->parent) |
3552 | css_put(css->parent); | |
3553 | ||
0c21ead1 | 3554 | css->ss->css_free(css); |
2bd59d48 | 3555 | cgroup_put(cgrp); |
48ddbe19 TH |
3556 | } |
3557 | ||
0c21ead1 | 3558 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3559 | { |
3560 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3561 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3562 | |
35ef10da | 3563 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 3564 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
3565 | } |
3566 | ||
d3daf28d TH |
3567 | static void css_release(struct percpu_ref *ref) |
3568 | { | |
3569 | struct cgroup_subsys_state *css = | |
3570 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3571 | ||
01a97140 | 3572 | RCU_INIT_POINTER(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 3573 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
3574 | } |
3575 | ||
623f926b TH |
3576 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
3577 | struct cgroup *cgrp) | |
ddbcc7e8 | 3578 | { |
bd89aabc | 3579 | css->cgroup = cgrp; |
72c97e54 | 3580 | css->ss = ss; |
ddbcc7e8 | 3581 | css->flags = 0; |
0ae78e0b TH |
3582 | |
3583 | if (cgrp->parent) | |
ca8bdcaf | 3584 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 3585 | else |
38b53aba | 3586 | css->flags |= CSS_ROOT; |
48ddbe19 | 3587 | |
ca8bdcaf | 3588 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
3589 | } |
3590 | ||
2a4ac633 | 3591 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 3592 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3593 | { |
623f926b | 3594 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
3595 | int ret = 0; |
3596 | ||
ace2bee8 | 3597 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3598 | lockdep_assert_held(&cgroup_mutex); |
3599 | ||
92fb9748 | 3600 | if (ss->css_online) |
eb95419b | 3601 | ret = ss->css_online(css); |
ae7f164a | 3602 | if (!ret) { |
eb95419b | 3603 | css->flags |= CSS_ONLINE; |
f20104de | 3604 | css->cgroup->nr_css++; |
aec25020 | 3605 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 3606 | } |
b1929db4 | 3607 | return ret; |
a31f2d3f TH |
3608 | } |
3609 | ||
2a4ac633 | 3610 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 3611 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3612 | { |
623f926b | 3613 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 3614 | |
ace2bee8 | 3615 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3616 | lockdep_assert_held(&cgroup_mutex); |
3617 | ||
3618 | if (!(css->flags & CSS_ONLINE)) | |
3619 | return; | |
3620 | ||
d7eeac19 | 3621 | if (ss->css_offline) |
eb95419b | 3622 | ss->css_offline(css); |
a31f2d3f | 3623 | |
eb95419b | 3624 | css->flags &= ~CSS_ONLINE; |
09a503ea | 3625 | css->cgroup->nr_css--; |
aec25020 | 3626 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); |
a31f2d3f TH |
3627 | } |
3628 | ||
c81c925a TH |
3629 | /** |
3630 | * create_css - create a cgroup_subsys_state | |
3631 | * @cgrp: the cgroup new css will be associated with | |
3632 | * @ss: the subsys of new css | |
3633 | * | |
3634 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
3635 | * css is online and installed in @cgrp with all interface files created. | |
3636 | * Returns 0 on success, -errno on failure. | |
3637 | */ | |
3638 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
3639 | { | |
3640 | struct cgroup *parent = cgrp->parent; | |
3641 | struct cgroup_subsys_state *css; | |
3642 | int err; | |
3643 | ||
c81c925a TH |
3644 | lockdep_assert_held(&cgroup_mutex); |
3645 | ||
3646 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
3647 | if (IS_ERR(css)) | |
3648 | return PTR_ERR(css); | |
3649 | ||
3650 | err = percpu_ref_init(&css->refcnt, css_release); | |
3651 | if (err) | |
3eb59ec6 | 3652 | goto err_free_css; |
c81c925a TH |
3653 | |
3654 | init_css(css, ss, cgrp); | |
3655 | ||
aec25020 | 3656 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a | 3657 | if (err) |
3eb59ec6 | 3658 | goto err_free_percpu_ref; |
c81c925a TH |
3659 | |
3660 | err = online_css(css); | |
3661 | if (err) | |
3eb59ec6 | 3662 | goto err_clear_dir; |
c81c925a | 3663 | |
59f5296b | 3664 | cgroup_get(cgrp); |
c81c925a TH |
3665 | css_get(css->parent); |
3666 | ||
94419627 TH |
3667 | cgrp->subsys_mask |= 1 << ss->id; |
3668 | ||
c81c925a TH |
3669 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
3670 | parent->parent) { | |
3671 | 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", | |
3672 | current->comm, current->pid, ss->name); | |
3673 | if (!strcmp(ss->name, "memory")) | |
3674 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
3675 | ss->warned_broken_hierarchy = true; | |
3676 | } | |
3677 | ||
3678 | return 0; | |
3679 | ||
3eb59ec6 | 3680 | err_clear_dir: |
32d01dc7 | 3681 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3eb59ec6 | 3682 | err_free_percpu_ref: |
c81c925a | 3683 | percpu_ref_cancel_init(&css->refcnt); |
3eb59ec6 | 3684 | err_free_css: |
c81c925a TH |
3685 | ss->css_free(css); |
3686 | return err; | |
3687 | } | |
3688 | ||
2bd59d48 | 3689 | /** |
a043e3b2 LZ |
3690 | * cgroup_create - create a cgroup |
3691 | * @parent: cgroup that will be parent of the new cgroup | |
e61734c5 | 3692 | * @name: name of the new cgroup |
2bd59d48 | 3693 | * @mode: mode to set on new cgroup |
ddbcc7e8 | 3694 | */ |
e61734c5 | 3695 | static long cgroup_create(struct cgroup *parent, const char *name, |
2bd59d48 | 3696 | umode_t mode) |
ddbcc7e8 | 3697 | { |
bd89aabc | 3698 | struct cgroup *cgrp; |
3dd06ffa | 3699 | struct cgroup_root *root = parent->root; |
b58c8998 | 3700 | int ssid, err; |
ddbcc7e8 | 3701 | struct cgroup_subsys *ss; |
2bd59d48 | 3702 | struct kernfs_node *kn; |
ddbcc7e8 | 3703 | |
a2dd4247 TH |
3704 | /* |
3705 | * XXX: The default hierarchy isn't fully implemented yet. Block | |
3706 | * !root cgroup creation on it for now. | |
3707 | */ | |
3708 | if (root == &cgrp_dfl_root) | |
3709 | return -EINVAL; | |
ddbcc7e8 | 3710 | |
0a950f65 | 3711 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
3712 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
3713 | if (!cgrp) | |
ddbcc7e8 PM |
3714 | return -ENOMEM; |
3715 | ||
ace2bee8 | 3716 | mutex_lock(&cgroup_tree_mutex); |
65dff759 | 3717 | |
976c06bc TH |
3718 | /* |
3719 | * Only live parents can have children. Note that the liveliness | |
3720 | * check isn't strictly necessary because cgroup_mkdir() and | |
3721 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
3722 | * anyway so that locking is contained inside cgroup proper and we | |
3723 | * don't get nasty surprises if we ever grow another caller. | |
3724 | */ | |
3725 | if (!cgroup_lock_live_group(parent)) { | |
3726 | err = -ENODEV; | |
ace2bee8 | 3727 | goto err_unlock_tree; |
0ab02ca8 LZ |
3728 | } |
3729 | ||
3730 | /* | |
3731 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
3732 | * a half-baked cgroup. | |
3733 | */ | |
3734 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
3735 | if (cgrp->id < 0) { | |
3736 | err = -ENOMEM; | |
3737 | goto err_unlock; | |
976c06bc TH |
3738 | } |
3739 | ||
cc31edce | 3740 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3741 | |
bd89aabc | 3742 | cgrp->parent = parent; |
0ae78e0b | 3743 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 3744 | cgrp->root = parent->root; |
ddbcc7e8 | 3745 | |
b6abdb0e LZ |
3746 | if (notify_on_release(parent)) |
3747 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3748 | ||
2260e7fc TH |
3749 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
3750 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 3751 | |
2bd59d48 | 3752 | /* create the directory */ |
e61734c5 | 3753 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 TH |
3754 | if (IS_ERR(kn)) { |
3755 | err = PTR_ERR(kn); | |
0ab02ca8 | 3756 | goto err_free_id; |
2bd59d48 TH |
3757 | } |
3758 | cgrp->kn = kn; | |
ddbcc7e8 | 3759 | |
4e139afc | 3760 | /* |
6f30558f TH |
3761 | * This extra ref will be put in cgroup_free_fn() and guarantees |
3762 | * that @cgrp->kn is always accessible. | |
4e139afc | 3763 | */ |
6f30558f | 3764 | kernfs_get(kn); |
ddbcc7e8 | 3765 | |
00356bd5 | 3766 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 3767 | |
4e139afc | 3768 | /* allocation complete, commit to creation */ |
4e139afc | 3769 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3c9c825b | 3770 | atomic_inc(&root->nr_cgrps); |
59f5296b | 3771 | cgroup_get(parent); |
415cf07a | 3772 | |
0d80255e TH |
3773 | /* |
3774 | * @cgrp is now fully operational. If something fails after this | |
3775 | * point, it'll be released via the normal destruction path. | |
3776 | */ | |
4e96ee8e LZ |
3777 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
3778 | ||
49957f8e TH |
3779 | err = cgroup_kn_set_ugid(kn); |
3780 | if (err) | |
3781 | goto err_destroy; | |
3782 | ||
2bb566cb | 3783 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
3784 | if (err) |
3785 | goto err_destroy; | |
3786 | ||
9d403e99 | 3787 | /* let's create and online css's */ |
b85d2040 | 3788 | for_each_subsys(ss, ssid) { |
3dd06ffa | 3789 | if (root->cgrp.subsys_mask & (1 << ssid)) { |
b85d2040 TH |
3790 | err = create_css(cgrp, ss); |
3791 | if (err) | |
3792 | goto err_destroy; | |
3793 | } | |
a8638030 | 3794 | } |
ddbcc7e8 | 3795 | |
2bd59d48 TH |
3796 | kernfs_activate(kn); |
3797 | ||
ddbcc7e8 | 3798 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3799 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
3800 | |
3801 | return 0; | |
3802 | ||
0a950f65 | 3803 | err_free_id: |
4e96ee8e | 3804 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
3805 | err_unlock: |
3806 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
3807 | err_unlock_tree: |
3808 | mutex_unlock(&cgroup_tree_mutex); | |
bd89aabc | 3809 | kfree(cgrp); |
ddbcc7e8 | 3810 | return err; |
4b8b47eb TH |
3811 | |
3812 | err_destroy: | |
3813 | cgroup_destroy_locked(cgrp); | |
3814 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3815 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 3816 | return err; |
ddbcc7e8 PM |
3817 | } |
3818 | ||
2bd59d48 TH |
3819 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
3820 | umode_t mode) | |
ddbcc7e8 | 3821 | { |
2bd59d48 | 3822 | struct cgroup *parent = parent_kn->priv; |
e1b2dc17 | 3823 | int ret; |
ddbcc7e8 | 3824 | |
e1b2dc17 TH |
3825 | /* |
3826 | * cgroup_create() grabs cgroup_tree_mutex which nests outside | |
3827 | * kernfs active_ref and cgroup_create() already synchronizes | |
3828 | * properly against removal through cgroup_lock_live_group(). | |
3829 | * Break it before calling cgroup_create(). | |
3830 | */ | |
3831 | cgroup_get(parent); | |
3832 | kernfs_break_active_protection(parent_kn); | |
ddbcc7e8 | 3833 | |
e1b2dc17 TH |
3834 | ret = cgroup_create(parent, name, mode); |
3835 | ||
3836 | kernfs_unbreak_active_protection(parent_kn); | |
3837 | cgroup_put(parent); | |
3838 | return ret; | |
ddbcc7e8 PM |
3839 | } |
3840 | ||
223dbc38 TH |
3841 | /* |
3842 | * This is called when the refcnt of a css is confirmed to be killed. | |
3843 | * css_tryget() is now guaranteed to fail. | |
3844 | */ | |
3845 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 3846 | { |
223dbc38 TH |
3847 | struct cgroup_subsys_state *css = |
3848 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
3849 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 3850 | |
ace2bee8 | 3851 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
3852 | mutex_lock(&cgroup_mutex); |
3853 | ||
09a503ea TH |
3854 | /* |
3855 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
3856 | * initate destruction. | |
3857 | */ | |
3858 | offline_css(css); | |
3859 | ||
f20104de TH |
3860 | /* |
3861 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
3862 | * be disabled before proceeding to the second phase of cgroup | |
3863 | * destruction. If we are the last one, kick it off. | |
3864 | */ | |
09a503ea | 3865 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
3866 | cgroup_destroy_css_killed(cgrp); |
3867 | ||
3868 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3869 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
3870 | |
3871 | /* | |
3872 | * Put the css refs from kill_css(). Each css holds an extra | |
3873 | * reference to the cgroup's dentry and cgroup removal proceeds | |
3874 | * regardless of css refs. On the last put of each css, whenever | |
3875 | * that may be, the extra dentry ref is put so that dentry | |
3876 | * destruction happens only after all css's are released. | |
3877 | */ | |
3878 | css_put(css); | |
d3daf28d TH |
3879 | } |
3880 | ||
223dbc38 TH |
3881 | /* css kill confirmation processing requires process context, bounce */ |
3882 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
3883 | { |
3884 | struct cgroup_subsys_state *css = | |
3885 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3886 | ||
223dbc38 | 3887 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 3888 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
3889 | } |
3890 | ||
94419627 | 3891 | static void __kill_css(struct cgroup_subsys_state *css) |
edae0c33 | 3892 | { |
94419627 TH |
3893 | lockdep_assert_held(&cgroup_tree_mutex); |
3894 | ||
2bd59d48 TH |
3895 | /* |
3896 | * This must happen before css is disassociated with its cgroup. | |
3897 | * See seq_css() for details. | |
3898 | */ | |
aec25020 | 3899 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 3900 | |
edae0c33 TH |
3901 | /* |
3902 | * Killing would put the base ref, but we need to keep it alive | |
3903 | * until after ->css_offline(). | |
3904 | */ | |
3905 | css_get(css); | |
3906 | ||
3907 | /* | |
3908 | * cgroup core guarantees that, by the time ->css_offline() is | |
3909 | * invoked, no new css reference will be given out via | |
3910 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
3911 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
3912 | * guarantee that the ref is seen as killed on all CPUs on return. | |
3913 | * | |
3914 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
3915 | * css is confirmed to be seen as killed on all CPUs. | |
3916 | */ | |
3917 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
3918 | } |
3919 | ||
94419627 TH |
3920 | /** |
3921 | * kill_css - destroy a css | |
3922 | * @css: css to destroy | |
3923 | * | |
3924 | * This function initiates destruction of @css by removing cgroup interface | |
3925 | * files and putting its base reference. ->css_offline() will be invoked | |
3926 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
3927 | * reference count reaches zero, @css will be released. | |
3928 | */ | |
3929 | static void kill_css(struct cgroup_subsys_state *css) | |
3930 | { | |
3931 | struct cgroup *cgrp = css->cgroup; | |
3932 | ||
3933 | lockdep_assert_held(&cgroup_tree_mutex); | |
3934 | ||
3935 | /* if already killed, noop */ | |
3936 | if (cgrp->subsys_mask & (1 << css->ss->id)) { | |
3937 | cgrp->subsys_mask &= ~(1 << css->ss->id); | |
3938 | __kill_css(css); | |
3939 | } | |
3940 | } | |
3941 | ||
d3daf28d TH |
3942 | /** |
3943 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
3944 | * @cgrp: cgroup to be destroyed | |
3945 | * | |
3946 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
3947 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
3948 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
3949 | * invoked. To satisfy all the requirements, destruction is implemented in | |
3950 | * the following two steps. | |
3951 | * | |
3952 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
3953 | * userland visible parts and start killing the percpu refcnts of | |
3954 | * css's. Set up so that the next stage will be kicked off once all | |
3955 | * the percpu refcnts are confirmed to be killed. | |
3956 | * | |
3957 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
3958 | * rest of destruction. Once all cgroup references are gone, the | |
3959 | * cgroup is RCU-freed. | |
3960 | * | |
3961 | * This function implements s1. After this step, @cgrp is gone as far as | |
3962 | * the userland is concerned and a new cgroup with the same name may be | |
3963 | * created. As cgroup doesn't care about the names internally, this | |
3964 | * doesn't cause any problem. | |
3965 | */ | |
42809dd4 TH |
3966 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
3967 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 3968 | { |
bb78a92f | 3969 | struct cgroup *child; |
2bd59d48 | 3970 | struct cgroup_subsys_state *css; |
ddd69148 | 3971 | bool empty; |
1c6727af | 3972 | int ssid; |
ddbcc7e8 | 3973 | |
ace2bee8 | 3974 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
3975 | lockdep_assert_held(&cgroup_mutex); |
3976 | ||
ddd69148 | 3977 | /* |
96d365e0 | 3978 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 3979 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 3980 | */ |
96d365e0 | 3981 | down_read(&css_set_rwsem); |
bb78a92f | 3982 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 3983 | up_read(&css_set_rwsem); |
ddd69148 | 3984 | if (!empty) |
ddbcc7e8 | 3985 | return -EBUSY; |
a043e3b2 | 3986 | |
bb78a92f HD |
3987 | /* |
3988 | * Make sure there's no live children. We can't test ->children | |
3989 | * emptiness as dead children linger on it while being destroyed; | |
3990 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
3991 | */ | |
3992 | empty = true; | |
3993 | rcu_read_lock(); | |
3994 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
3995 | empty = cgroup_is_dead(child); | |
3996 | if (!empty) | |
3997 | break; | |
3998 | } | |
3999 | rcu_read_unlock(); | |
4000 | if (!empty) | |
4001 | return -EBUSY; | |
4002 | ||
455050d2 TH |
4003 | /* |
4004 | * Mark @cgrp dead. This prevents further task migration and child | |
4005 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 4006 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 4007 | * resume iteration after dropping RCU read lock. See |
492eb21b | 4008 | * css_next_child() for details. |
455050d2 | 4009 | */ |
54766d4a | 4010 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 4011 | |
88703267 | 4012 | /* |
edae0c33 TH |
4013 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
4014 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
4015 | * percpu refs of all css's are confirmed to be killed. This |
4016 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 4017 | */ |
4ac06017 | 4018 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
4019 | for_each_css(css, ssid, cgrp) |
4020 | kill_css(css); | |
4ac06017 | 4021 | mutex_lock(&cgroup_mutex); |
455050d2 | 4022 | |
455050d2 TH |
4023 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
4024 | raw_spin_lock(&release_list_lock); | |
4025 | if (!list_empty(&cgrp->release_list)) | |
4026 | list_del_init(&cgrp->release_list); | |
4027 | raw_spin_unlock(&release_list_lock); | |
4028 | ||
4029 | /* | |
f20104de TH |
4030 | * If @cgrp has css's attached, the second stage of cgroup |
4031 | * destruction is kicked off from css_killed_work_fn() after the | |
4032 | * refs of all attached css's are killed. If @cgrp doesn't have | |
4033 | * any css, we kick it off here. | |
4034 | */ | |
4035 | if (!cgrp->nr_css) | |
4036 | cgroup_destroy_css_killed(cgrp); | |
4037 | ||
2bd59d48 TH |
4038 | /* remove @cgrp directory along with the base files */ |
4039 | mutex_unlock(&cgroup_mutex); | |
4040 | ||
455050d2 | 4041 | /* |
2bd59d48 TH |
4042 | * There are two control paths which try to determine cgroup from |
4043 | * dentry without going through kernfs - cgroupstats_build() and | |
4044 | * css_tryget_from_dir(). Those are supported by RCU protecting | |
4045 | * clearing of cgrp->kn->priv backpointer, which should happen | |
4046 | * after all files under it have been removed. | |
455050d2 | 4047 | */ |
6f30558f | 4048 | kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ |
2bd59d48 | 4049 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); |
2bd59d48 | 4050 | |
4ac06017 | 4051 | mutex_lock(&cgroup_mutex); |
455050d2 | 4052 | |
ea15f8cc TH |
4053 | return 0; |
4054 | }; | |
4055 | ||
d3daf28d | 4056 | /** |
f20104de | 4057 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
4058 | * @work: cgroup->destroy_free_work |
4059 | * | |
4060 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
4061 | * destroyed after all css's are offlined and performs the rest of |
4062 | * destruction. This is the second step of destruction described in the | |
4063 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 4064 | */ |
f20104de | 4065 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 4066 | { |
ea15f8cc | 4067 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 4068 | |
ace2bee8 | 4069 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 4070 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 4071 | |
999cd8a4 | 4072 | /* delete this cgroup from parent->children */ |
eb6fd504 | 4073 | list_del_rcu(&cgrp->sibling); |
ed957793 | 4074 | |
59f5296b | 4075 | cgroup_put(cgrp); |
ddbcc7e8 | 4076 | |
bd89aabc | 4077 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 4078 | check_for_release(parent); |
ddbcc7e8 PM |
4079 | } |
4080 | ||
2bd59d48 | 4081 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4082 | { |
2bd59d48 TH |
4083 | struct cgroup *cgrp = kn->priv; |
4084 | int ret = 0; | |
42809dd4 | 4085 | |
2bd59d48 TH |
4086 | /* |
4087 | * This is self-destruction but @kn can't be removed while this | |
4088 | * callback is in progress. Let's break active protection. Once | |
4089 | * the protection is broken, @cgrp can be destroyed at any point. | |
4090 | * Pin it so that it stays accessible. | |
4091 | */ | |
4092 | cgroup_get(cgrp); | |
4093 | kernfs_break_active_protection(kn); | |
42809dd4 | 4094 | |
ace2bee8 | 4095 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 4096 | mutex_lock(&cgroup_mutex); |
8e3f6541 TH |
4097 | |
4098 | /* | |
2bd59d48 TH |
4099 | * @cgrp might already have been destroyed while we're trying to |
4100 | * grab the mutexes. | |
8e3f6541 | 4101 | */ |
2bd59d48 TH |
4102 | if (!cgroup_is_dead(cgrp)) |
4103 | ret = cgroup_destroy_locked(cgrp); | |
2bb566cb | 4104 | |
42809dd4 | 4105 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4106 | mutex_unlock(&cgroup_tree_mutex); |
2bb566cb | 4107 | |
2bd59d48 TH |
4108 | kernfs_unbreak_active_protection(kn); |
4109 | cgroup_put(cgrp); | |
42809dd4 | 4110 | return ret; |
8e3f6541 TH |
4111 | } |
4112 | ||
2bd59d48 TH |
4113 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4114 | .remount_fs = cgroup_remount, | |
4115 | .show_options = cgroup_show_options, | |
4116 | .mkdir = cgroup_mkdir, | |
4117 | .rmdir = cgroup_rmdir, | |
4118 | .rename = cgroup_rename, | |
4119 | }; | |
4120 | ||
06a11920 | 4121 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4122 | { |
ddbcc7e8 | 4123 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4124 | |
4125 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4126 | |
ace2bee8 | 4127 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
4128 | mutex_lock(&cgroup_mutex); |
4129 | ||
0adb0704 | 4130 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4131 | |
3dd06ffa TH |
4132 | /* Create the root cgroup state for this subsystem */ |
4133 | ss->root = &cgrp_dfl_root; | |
4134 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
4135 | /* We don't handle early failures gracefully */ |
4136 | BUG_ON(IS_ERR(css)); | |
3dd06ffa | 4137 | init_css(css, ss, &cgrp_dfl_root.cgrp); |
ddbcc7e8 | 4138 | |
e8d55fde | 4139 | /* Update the init_css_set to contain a subsys |
817929ec | 4140 | * pointer to this state - since the subsystem is |
e8d55fde | 4141 | * newly registered, all tasks and hence the |
3dd06ffa | 4142 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 4143 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4144 | |
4145 | need_forkexit_callback |= ss->fork || ss->exit; | |
4146 | ||
e8d55fde LZ |
4147 | /* At system boot, before all subsystems have been |
4148 | * registered, no tasks have been forked, so we don't | |
4149 | * need to invoke fork callbacks here. */ | |
4150 | BUG_ON(!list_empty(&init_task.tasks)); | |
4151 | ||
ae7f164a | 4152 | BUG_ON(online_css(css)); |
a8638030 | 4153 | |
3dd06ffa | 4154 | cgrp_dfl_root.cgrp.subsys_mask |= 1 << ss->id; |
cf5d5941 BB |
4155 | |
4156 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4157 | mutex_unlock(&cgroup_tree_mutex); |
cf5d5941 | 4158 | } |
cf5d5941 | 4159 | |
ddbcc7e8 | 4160 | /** |
a043e3b2 LZ |
4161 | * cgroup_init_early - cgroup initialization at system boot |
4162 | * | |
4163 | * Initialize cgroups at system boot, and initialize any | |
4164 | * subsystems that request early init. | |
ddbcc7e8 PM |
4165 | */ |
4166 | int __init cgroup_init_early(void) | |
4167 | { | |
a2dd4247 TH |
4168 | static struct cgroup_sb_opts __initdata opts = |
4169 | { .flags = CGRP_ROOT_SANE_BEHAVIOR }; | |
30159ec7 | 4170 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4171 | int i; |
30159ec7 | 4172 | |
3dd06ffa | 4173 | init_cgroup_root(&cgrp_dfl_root, &opts); |
a4ea1cc9 | 4174 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4175 | |
3ed80a62 | 4176 | for_each_subsys(ss, i) { |
aec25020 | 4177 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4178 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4179 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4180 | ss->id, ss->name); |
073219e9 TH |
4181 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4182 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4183 | ||
aec25020 | 4184 | ss->id = i; |
073219e9 | 4185 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4186 | |
4187 | if (ss->early_init) | |
4188 | cgroup_init_subsys(ss); | |
4189 | } | |
4190 | return 0; | |
4191 | } | |
4192 | ||
4193 | /** | |
a043e3b2 LZ |
4194 | * cgroup_init - cgroup initialization |
4195 | * | |
4196 | * Register cgroup filesystem and /proc file, and initialize | |
4197 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4198 | */ |
4199 | int __init cgroup_init(void) | |
4200 | { | |
30159ec7 | 4201 | struct cgroup_subsys *ss; |
0ac801fe | 4202 | unsigned long key; |
172a2c06 | 4203 | int ssid, err; |
ddbcc7e8 | 4204 | |
2bd59d48 | 4205 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
ddbcc7e8 | 4206 | |
985ed670 | 4207 | mutex_lock(&cgroup_tree_mutex); |
54e7b4eb | 4208 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4209 | |
82fe9b0d TH |
4210 | /* Add init_css_set to the hash table */ |
4211 | key = css_set_hash(init_css_set.subsys); | |
4212 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4213 | ||
3dd06ffa | 4214 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 4215 | |
54e7b4eb | 4216 | mutex_unlock(&cgroup_mutex); |
985ed670 | 4217 | mutex_unlock(&cgroup_tree_mutex); |
54e7b4eb | 4218 | |
172a2c06 TH |
4219 | for_each_subsys(ss, ssid) { |
4220 | if (!ss->early_init) | |
4221 | cgroup_init_subsys(ss); | |
4222 | ||
4223 | /* | |
4224 | * cftype registration needs kmalloc and can't be done | |
4225 | * during early_init. Register base cftypes separately. | |
4226 | */ | |
4227 | if (ss->base_cftypes) | |
4228 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
676db4af GK |
4229 | } |
4230 | ||
676db4af | 4231 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4232 | if (!cgroup_kobj) |
4233 | return -ENOMEM; | |
676db4af | 4234 | |
ddbcc7e8 | 4235 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
4236 | if (err < 0) { |
4237 | kobject_put(cgroup_kobj); | |
2bd59d48 | 4238 | return err; |
676db4af | 4239 | } |
ddbcc7e8 | 4240 | |
46ae220b | 4241 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 4242 | return 0; |
ddbcc7e8 | 4243 | } |
b4f48b63 | 4244 | |
e5fca243 TH |
4245 | static int __init cgroup_wq_init(void) |
4246 | { | |
4247 | /* | |
4248 | * There isn't much point in executing destruction path in | |
4249 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 4250 | * Use 1 for @max_active. |
e5fca243 TH |
4251 | * |
4252 | * We would prefer to do this in cgroup_init() above, but that | |
4253 | * is called before init_workqueues(): so leave this until after. | |
4254 | */ | |
1a11533f | 4255 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 4256 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4257 | |
4258 | /* | |
4259 | * Used to destroy pidlists and separate to serve as flush domain. | |
4260 | * Cap @max_active to 1 too. | |
4261 | */ | |
4262 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4263 | 0, 1); | |
4264 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4265 | ||
e5fca243 TH |
4266 | return 0; |
4267 | } | |
4268 | core_initcall(cgroup_wq_init); | |
4269 | ||
a424316c PM |
4270 | /* |
4271 | * proc_cgroup_show() | |
4272 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4273 | * - Used for /proc/<pid>/cgroup. | |
a424316c PM |
4274 | */ |
4275 | ||
4276 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4277 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4278 | { |
4279 | struct pid *pid; | |
4280 | struct task_struct *tsk; | |
e61734c5 | 4281 | char *buf, *path; |
a424316c | 4282 | int retval; |
3dd06ffa | 4283 | struct cgroup_root *root; |
a424316c PM |
4284 | |
4285 | retval = -ENOMEM; | |
e61734c5 | 4286 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4287 | if (!buf) |
4288 | goto out; | |
4289 | ||
4290 | retval = -ESRCH; | |
4291 | pid = m->private; | |
4292 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4293 | if (!tsk) | |
4294 | goto out_free; | |
4295 | ||
4296 | retval = 0; | |
4297 | ||
4298 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 4299 | down_read(&css_set_rwsem); |
a424316c | 4300 | |
985ed670 | 4301 | for_each_root(root) { |
a424316c | 4302 | struct cgroup_subsys *ss; |
bd89aabc | 4303 | struct cgroup *cgrp; |
b85d2040 | 4304 | int ssid, count = 0; |
a424316c | 4305 | |
a2dd4247 | 4306 | if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) |
985ed670 TH |
4307 | continue; |
4308 | ||
2c6ab6d2 | 4309 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 | 4310 | for_each_subsys(ss, ssid) |
3dd06ffa | 4311 | if (root->cgrp.subsys_mask & (1 << ssid)) |
b85d2040 | 4312 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); |
c6d57f33 PM |
4313 | if (strlen(root->name)) |
4314 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4315 | root->name); | |
a424316c | 4316 | seq_putc(m, ':'); |
7717f7ba | 4317 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
4318 | path = cgroup_path(cgrp, buf, PATH_MAX); |
4319 | if (!path) { | |
4320 | retval = -ENAMETOOLONG; | |
a424316c | 4321 | goto out_unlock; |
e61734c5 TH |
4322 | } |
4323 | seq_puts(m, path); | |
a424316c PM |
4324 | seq_putc(m, '\n'); |
4325 | } | |
4326 | ||
4327 | out_unlock: | |
96d365e0 | 4328 | up_read(&css_set_rwsem); |
a424316c PM |
4329 | mutex_unlock(&cgroup_mutex); |
4330 | put_task_struct(tsk); | |
4331 | out_free: | |
4332 | kfree(buf); | |
4333 | out: | |
4334 | return retval; | |
4335 | } | |
4336 | ||
a424316c PM |
4337 | /* Display information about each subsystem and each hierarchy */ |
4338 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4339 | { | |
30159ec7 | 4340 | struct cgroup_subsys *ss; |
a424316c | 4341 | int i; |
a424316c | 4342 | |
8bab8dde | 4343 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4344 | /* |
4345 | * ideally we don't want subsystems moving around while we do this. | |
4346 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4347 | * subsys/hierarchy state. | |
4348 | */ | |
a424316c | 4349 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4350 | |
4351 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4352 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4353 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 4354 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 4355 | |
a424316c PM |
4356 | mutex_unlock(&cgroup_mutex); |
4357 | return 0; | |
4358 | } | |
4359 | ||
4360 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4361 | { | |
9dce07f1 | 4362 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4363 | } |
4364 | ||
828c0950 | 4365 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4366 | .open = cgroupstats_open, |
4367 | .read = seq_read, | |
4368 | .llseek = seq_lseek, | |
4369 | .release = single_release, | |
4370 | }; | |
4371 | ||
b4f48b63 | 4372 | /** |
eaf797ab | 4373 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 4374 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 4375 | * |
eaf797ab TH |
4376 | * A task is associated with the init_css_set until cgroup_post_fork() |
4377 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
4378 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
4379 | */ |
4380 | void cgroup_fork(struct task_struct *child) | |
4381 | { | |
eaf797ab | 4382 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 4383 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4384 | } |
4385 | ||
817929ec | 4386 | /** |
a043e3b2 LZ |
4387 | * cgroup_post_fork - called on a new task after adding it to the task list |
4388 | * @child: the task in question | |
4389 | * | |
5edee61e TH |
4390 | * Adds the task to the list running through its css_set if necessary and |
4391 | * call the subsystem fork() callbacks. Has to be after the task is | |
4392 | * visible on the task list in case we race with the first call to | |
0942eeee | 4393 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4394 | * list. |
a043e3b2 | 4395 | */ |
817929ec PM |
4396 | void cgroup_post_fork(struct task_struct *child) |
4397 | { | |
30159ec7 | 4398 | struct cgroup_subsys *ss; |
5edee61e TH |
4399 | int i; |
4400 | ||
3ce3230a | 4401 | /* |
eaf797ab TH |
4402 | * This may race against cgroup_enable_task_cg_links(). As that |
4403 | * function sets use_task_css_set_links before grabbing | |
4404 | * tasklist_lock and we just went through tasklist_lock to add | |
4405 | * @child, it's guaranteed that either we see the set | |
4406 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
4407 | * @child during its iteration. | |
4408 | * | |
4409 | * If we won the race, @child is associated with %current's | |
4410 | * css_set. Grabbing css_set_rwsem guarantees both that the | |
4411 | * association is stable, and, on completion of the parent's | |
4412 | * migration, @child is visible in the source of migration or | |
4413 | * already in the destination cgroup. This guarantee is necessary | |
4414 | * when implementing operations which need to migrate all tasks of | |
4415 | * a cgroup to another. | |
4416 | * | |
4417 | * Note that if we lose to cgroup_enable_task_cg_links(), @child | |
4418 | * will remain in init_css_set. This is safe because all tasks are | |
4419 | * in the init_css_set before cg_links is enabled and there's no | |
4420 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 4421 | */ |
817929ec | 4422 | if (use_task_css_set_links) { |
eaf797ab TH |
4423 | struct css_set *cset; |
4424 | ||
96d365e0 | 4425 | down_write(&css_set_rwsem); |
0e1d768f | 4426 | cset = task_css_set(current); |
eaf797ab TH |
4427 | if (list_empty(&child->cg_list)) { |
4428 | rcu_assign_pointer(child->cgroups, cset); | |
4429 | list_add(&child->cg_list, &cset->tasks); | |
4430 | get_css_set(cset); | |
4431 | } | |
96d365e0 | 4432 | up_write(&css_set_rwsem); |
817929ec | 4433 | } |
5edee61e TH |
4434 | |
4435 | /* | |
4436 | * Call ss->fork(). This must happen after @child is linked on | |
4437 | * css_set; otherwise, @child might change state between ->fork() | |
4438 | * and addition to css_set. | |
4439 | */ | |
4440 | if (need_forkexit_callback) { | |
3ed80a62 | 4441 | for_each_subsys(ss, i) |
5edee61e TH |
4442 | if (ss->fork) |
4443 | ss->fork(child); | |
5edee61e | 4444 | } |
817929ec | 4445 | } |
5edee61e | 4446 | |
b4f48b63 PM |
4447 | /** |
4448 | * cgroup_exit - detach cgroup from exiting task | |
4449 | * @tsk: pointer to task_struct of exiting process | |
4450 | * | |
4451 | * Description: Detach cgroup from @tsk and release it. | |
4452 | * | |
4453 | * Note that cgroups marked notify_on_release force every task in | |
4454 | * them to take the global cgroup_mutex mutex when exiting. | |
4455 | * This could impact scaling on very large systems. Be reluctant to | |
4456 | * use notify_on_release cgroups where very high task exit scaling | |
4457 | * is required on large systems. | |
4458 | * | |
0e1d768f TH |
4459 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
4460 | * call cgroup_exit() while the task is still competent to handle | |
4461 | * notify_on_release(), then leave the task attached to the root cgroup in | |
4462 | * each hierarchy for the remainder of its exit. No need to bother with | |
4463 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 4464 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 4465 | */ |
1ec41830 | 4466 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 4467 | { |
30159ec7 | 4468 | struct cgroup_subsys *ss; |
5abb8855 | 4469 | struct css_set *cset; |
eaf797ab | 4470 | bool put_cset = false; |
d41d5a01 | 4471 | int i; |
817929ec PM |
4472 | |
4473 | /* | |
0e1d768f TH |
4474 | * Unlink from @tsk from its css_set. As migration path can't race |
4475 | * with us, we can check cg_list without grabbing css_set_rwsem. | |
817929ec PM |
4476 | */ |
4477 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 4478 | down_write(&css_set_rwsem); |
0e1d768f | 4479 | list_del_init(&tsk->cg_list); |
96d365e0 | 4480 | up_write(&css_set_rwsem); |
0e1d768f | 4481 | put_cset = true; |
817929ec PM |
4482 | } |
4483 | ||
b4f48b63 | 4484 | /* Reassign the task to the init_css_set. */ |
a8ad805c TH |
4485 | cset = task_css_set(tsk); |
4486 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 | 4487 | |
1ec41830 | 4488 | if (need_forkexit_callback) { |
3ed80a62 TH |
4489 | /* see cgroup_post_fork() for details */ |
4490 | for_each_subsys(ss, i) { | |
d41d5a01 | 4491 | if (ss->exit) { |
eb95419b TH |
4492 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4493 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4494 | |
eb95419b | 4495 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4496 | } |
4497 | } | |
4498 | } | |
d41d5a01 | 4499 | |
eaf797ab TH |
4500 | if (put_cset) |
4501 | put_css_set(cset, true); | |
b4f48b63 | 4502 | } |
697f4161 | 4503 | |
bd89aabc | 4504 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4505 | { |
f50daa70 | 4506 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4507 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4508 | /* |
4509 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4510 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4511 | * it now |
4512 | */ | |
81a6a5cd | 4513 | int need_schedule_work = 0; |
f50daa70 | 4514 | |
cdcc136f | 4515 | raw_spin_lock(&release_list_lock); |
54766d4a | 4516 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4517 | list_empty(&cgrp->release_list)) { |
4518 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4519 | need_schedule_work = 1; |
4520 | } | |
cdcc136f | 4521 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4522 | if (need_schedule_work) |
4523 | schedule_work(&release_agent_work); | |
4524 | } | |
4525 | } | |
4526 | ||
81a6a5cd PM |
4527 | /* |
4528 | * Notify userspace when a cgroup is released, by running the | |
4529 | * configured release agent with the name of the cgroup (path | |
4530 | * relative to the root of cgroup file system) as the argument. | |
4531 | * | |
4532 | * Most likely, this user command will try to rmdir this cgroup. | |
4533 | * | |
4534 | * This races with the possibility that some other task will be | |
4535 | * attached to this cgroup before it is removed, or that some other | |
4536 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4537 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4538 | * unused, and this cgroup will be reprieved from its death sentence, | |
4539 | * to continue to serve a useful existence. Next time it's released, | |
4540 | * we will get notified again, if it still has 'notify_on_release' set. | |
4541 | * | |
4542 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4543 | * means only wait until the task is successfully execve()'d. The | |
4544 | * separate release agent task is forked by call_usermodehelper(), | |
4545 | * then control in this thread returns here, without waiting for the | |
4546 | * release agent task. We don't bother to wait because the caller of | |
4547 | * this routine has no use for the exit status of the release agent | |
4548 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4549 | */ |
81a6a5cd PM |
4550 | static void cgroup_release_agent(struct work_struct *work) |
4551 | { | |
4552 | BUG_ON(work != &release_agent_work); | |
4553 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4554 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4555 | while (!list_empty(&release_list)) { |
4556 | char *argv[3], *envp[3]; | |
4557 | int i; | |
e61734c5 | 4558 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 4559 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4560 | struct cgroup, |
4561 | release_list); | |
bd89aabc | 4562 | list_del_init(&cgrp->release_list); |
cdcc136f | 4563 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 4564 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
4565 | if (!pathbuf) |
4566 | goto continue_free; | |
e61734c5 TH |
4567 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
4568 | if (!path) | |
e788e066 PM |
4569 | goto continue_free; |
4570 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4571 | if (!agentbuf) | |
4572 | goto continue_free; | |
81a6a5cd PM |
4573 | |
4574 | i = 0; | |
e788e066 | 4575 | argv[i++] = agentbuf; |
e61734c5 | 4576 | argv[i++] = path; |
81a6a5cd PM |
4577 | argv[i] = NULL; |
4578 | ||
4579 | i = 0; | |
4580 | /* minimal command environment */ | |
4581 | envp[i++] = "HOME=/"; | |
4582 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4583 | envp[i] = NULL; | |
4584 | ||
4585 | /* Drop the lock while we invoke the usermode helper, | |
4586 | * since the exec could involve hitting disk and hence | |
4587 | * be a slow process */ | |
4588 | mutex_unlock(&cgroup_mutex); | |
4589 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4590 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4591 | continue_free: |
4592 | kfree(pathbuf); | |
4593 | kfree(agentbuf); | |
cdcc136f | 4594 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4595 | } |
cdcc136f | 4596 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4597 | mutex_unlock(&cgroup_mutex); |
4598 | } | |
8bab8dde PM |
4599 | |
4600 | static int __init cgroup_disable(char *str) | |
4601 | { | |
30159ec7 | 4602 | struct cgroup_subsys *ss; |
8bab8dde | 4603 | char *token; |
30159ec7 | 4604 | int i; |
8bab8dde PM |
4605 | |
4606 | while ((token = strsep(&str, ",")) != NULL) { | |
4607 | if (!*token) | |
4608 | continue; | |
be45c900 | 4609 | |
3ed80a62 | 4610 | for_each_subsys(ss, i) { |
8bab8dde PM |
4611 | if (!strcmp(token, ss->name)) { |
4612 | ss->disabled = 1; | |
4613 | printk(KERN_INFO "Disabling %s control group" | |
4614 | " subsystem\n", ss->name); | |
4615 | break; | |
4616 | } | |
4617 | } | |
4618 | } | |
4619 | return 1; | |
4620 | } | |
4621 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 4622 | |
b77d7b60 | 4623 | /** |
5a17f543 | 4624 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
4625 | * @dentry: directory dentry of interest |
4626 | * @ss: subsystem of interest | |
b77d7b60 | 4627 | * |
5a17f543 TH |
4628 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
4629 | * to get the corresponding css and return it. If such css doesn't exist | |
4630 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 4631 | */ |
5a17f543 TH |
4632 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
4633 | struct cgroup_subsys *ss) | |
e5d1367f | 4634 | { |
2bd59d48 TH |
4635 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
4636 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 4637 | struct cgroup *cgrp; |
e5d1367f | 4638 | |
35cf0836 | 4639 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
4640 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
4641 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
4642 | return ERR_PTR(-EBADF); |
4643 | ||
5a17f543 TH |
4644 | rcu_read_lock(); |
4645 | ||
2bd59d48 TH |
4646 | /* |
4647 | * This path doesn't originate from kernfs and @kn could already | |
4648 | * have been or be removed at any point. @kn->priv is RCU | |
4649 | * protected for this access. See destroy_locked() for details. | |
4650 | */ | |
4651 | cgrp = rcu_dereference(kn->priv); | |
4652 | if (cgrp) | |
4653 | css = cgroup_css(cgrp, ss); | |
5a17f543 TH |
4654 | |
4655 | if (!css || !css_tryget(css)) | |
4656 | css = ERR_PTR(-ENOENT); | |
4657 | ||
4658 | rcu_read_unlock(); | |
4659 | return css; | |
e5d1367f | 4660 | } |
e5d1367f | 4661 | |
1cb650b9 LZ |
4662 | /** |
4663 | * css_from_id - lookup css by id | |
4664 | * @id: the cgroup id | |
4665 | * @ss: cgroup subsys to be looked into | |
4666 | * | |
4667 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
4668 | * Should be called under rcu_read_lock(). | |
4669 | */ | |
4670 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
4671 | { | |
4672 | struct cgroup *cgrp; | |
4673 | ||
ace2bee8 | 4674 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
4675 | |
4676 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
4677 | if (cgrp) | |
d1625964 | 4678 | return cgroup_css(cgrp, ss); |
1cb650b9 | 4679 | return NULL; |
e5d1367f SE |
4680 | } |
4681 | ||
fe693435 | 4682 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
4683 | static struct cgroup_subsys_state * |
4684 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
4685 | { |
4686 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
4687 | ||
4688 | if (!css) | |
4689 | return ERR_PTR(-ENOMEM); | |
4690 | ||
4691 | return css; | |
4692 | } | |
4693 | ||
eb95419b | 4694 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 4695 | { |
eb95419b | 4696 | kfree(css); |
fe693435 PM |
4697 | } |
4698 | ||
182446d0 TH |
4699 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
4700 | struct cftype *cft) | |
fe693435 | 4701 | { |
182446d0 | 4702 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
4703 | } |
4704 | ||
182446d0 TH |
4705 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
4706 | struct cftype *cft) | |
fe693435 PM |
4707 | { |
4708 | return (u64)(unsigned long)current->cgroups; | |
4709 | } | |
4710 | ||
182446d0 | 4711 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 4712 | struct cftype *cft) |
fe693435 PM |
4713 | { |
4714 | u64 count; | |
4715 | ||
4716 | rcu_read_lock(); | |
a8ad805c | 4717 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
4718 | rcu_read_unlock(); |
4719 | return count; | |
4720 | } | |
4721 | ||
2da8ca82 | 4722 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4723 | { |
69d0206c | 4724 | struct cgrp_cset_link *link; |
5abb8855 | 4725 | struct css_set *cset; |
e61734c5 TH |
4726 | char *name_buf; |
4727 | ||
4728 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
4729 | if (!name_buf) | |
4730 | return -ENOMEM; | |
7717f7ba | 4731 | |
96d365e0 | 4732 | down_read(&css_set_rwsem); |
7717f7ba | 4733 | rcu_read_lock(); |
5abb8855 | 4734 | cset = rcu_dereference(current->cgroups); |
69d0206c | 4735 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 4736 | struct cgroup *c = link->cgrp; |
7717f7ba | 4737 | |
a2dd4247 | 4738 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 4739 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 4740 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
4741 | } |
4742 | rcu_read_unlock(); | |
96d365e0 | 4743 | up_read(&css_set_rwsem); |
e61734c5 | 4744 | kfree(name_buf); |
7717f7ba PM |
4745 | return 0; |
4746 | } | |
4747 | ||
4748 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 4749 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4750 | { |
2da8ca82 | 4751 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 4752 | struct cgrp_cset_link *link; |
7717f7ba | 4753 | |
96d365e0 | 4754 | down_read(&css_set_rwsem); |
182446d0 | 4755 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 4756 | struct css_set *cset = link->cset; |
7717f7ba PM |
4757 | struct task_struct *task; |
4758 | int count = 0; | |
c7561128 | 4759 | |
5abb8855 | 4760 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 4761 | |
5abb8855 | 4762 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
4763 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
4764 | goto overflow; | |
4765 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
4766 | } | |
4767 | ||
4768 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
4769 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
4770 | goto overflow; | |
4771 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 4772 | } |
c7561128 TH |
4773 | continue; |
4774 | overflow: | |
4775 | seq_puts(seq, " ...\n"); | |
7717f7ba | 4776 | } |
96d365e0 | 4777 | up_read(&css_set_rwsem); |
7717f7ba PM |
4778 | return 0; |
4779 | } | |
4780 | ||
182446d0 | 4781 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 4782 | { |
182446d0 | 4783 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
4784 | } |
4785 | ||
4786 | static struct cftype debug_files[] = { | |
fe693435 PM |
4787 | { |
4788 | .name = "taskcount", | |
4789 | .read_u64 = debug_taskcount_read, | |
4790 | }, | |
4791 | ||
4792 | { | |
4793 | .name = "current_css_set", | |
4794 | .read_u64 = current_css_set_read, | |
4795 | }, | |
4796 | ||
4797 | { | |
4798 | .name = "current_css_set_refcount", | |
4799 | .read_u64 = current_css_set_refcount_read, | |
4800 | }, | |
4801 | ||
7717f7ba PM |
4802 | { |
4803 | .name = "current_css_set_cg_links", | |
2da8ca82 | 4804 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
4805 | }, |
4806 | ||
4807 | { | |
4808 | .name = "cgroup_css_links", | |
2da8ca82 | 4809 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
4810 | }, |
4811 | ||
fe693435 PM |
4812 | { |
4813 | .name = "releasable", | |
4814 | .read_u64 = releasable_read, | |
4815 | }, | |
fe693435 | 4816 | |
4baf6e33 TH |
4817 | { } /* terminate */ |
4818 | }; | |
fe693435 | 4819 | |
073219e9 | 4820 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
4821 | .css_alloc = debug_css_alloc, |
4822 | .css_free = debug_css_free, | |
4baf6e33 | 4823 | .base_cftypes = debug_files, |
fe693435 PM |
4824 | }; |
4825 | #endif /* CONFIG_CGROUP_DEBUG */ |