Commit | Line | Data |
---|---|---|
ddbcc7e8 | 1 | /* |
ddbcc7e8 PM |
2 | * Generic process-grouping system. |
3 | * | |
4 | * Based originally on the cpuset system, extracted by Paul Menage | |
5 | * Copyright (C) 2006 Google, Inc | |
6 | * | |
0dea1168 KS |
7 | * Notifications support |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
ddbcc7e8 PM |
11 | * Copyright notices from the original cpuset code: |
12 | * -------------------------------------------------- | |
13 | * Copyright (C) 2003 BULL SA. | |
14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
15 | * | |
16 | * Portions derived from Patrick Mochel's sysfs code. | |
17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
18 | * | |
19 | * 2003-10-10 Written by Simon Derr. | |
20 | * 2003-10-22 Updates by Stephen Hemminger. | |
21 | * 2004 May-July Rework by Paul Jackson. | |
22 | * --------------------------------------------------- | |
23 | * | |
24 | * This file is subject to the terms and conditions of the GNU General Public | |
25 | * License. See the file COPYING in the main directory of the Linux | |
26 | * distribution for more details. | |
27 | */ | |
28 | ||
29 | #include <linux/cgroup.h> | |
2ce9738b | 30 | #include <linux/cred.h> |
c6d57f33 | 31 | #include <linux/ctype.h> |
ddbcc7e8 | 32 | #include <linux/errno.h> |
2ce9738b | 33 | #include <linux/init_task.h> |
ddbcc7e8 PM |
34 | #include <linux/kernel.h> |
35 | #include <linux/list.h> | |
36 | #include <linux/mm.h> | |
37 | #include <linux/mutex.h> | |
38 | #include <linux/mount.h> | |
39 | #include <linux/pagemap.h> | |
a424316c | 40 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
41 | #include <linux/rcupdate.h> |
42 | #include <linux/sched.h> | |
ddbcc7e8 | 43 | #include <linux/slab.h> |
ddbcc7e8 | 44 | #include <linux/spinlock.h> |
96d365e0 | 45 | #include <linux/rwsem.h> |
ddbcc7e8 | 46 | #include <linux/string.h> |
bbcb81d0 | 47 | #include <linux/sort.h> |
81a6a5cd | 48 | #include <linux/kmod.h> |
846c7bb0 BS |
49 | #include <linux/delayacct.h> |
50 | #include <linux/cgroupstats.h> | |
0ac801fe | 51 | #include <linux/hashtable.h> |
096b7fe0 | 52 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 53 | #include <linux/idr.h> |
d1d9fd33 | 54 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
081aa458 | 55 | #include <linux/flex_array.h> /* used in cgroup_attach_task */ |
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. | |
e25e2cbb | 84 | */ |
2219449a TH |
85 | #ifdef CONFIG_PROVE_RCU |
86 | DEFINE_MUTEX(cgroup_mutex); | |
8af01f56 | 87 | EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for lockdep */ |
2219449a | 88 | #else |
81a6a5cd | 89 | static DEFINE_MUTEX(cgroup_mutex); |
2219449a TH |
90 | #endif |
91 | ||
69e943b7 TH |
92 | /* |
93 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
94 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
95 | */ | |
96 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
97 | ||
ace2bee8 | 98 | #define cgroup_assert_mutexes_or_rcu_locked() \ |
87fb54f1 | 99 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
ace2bee8 | 100 | lockdep_is_held(&cgroup_tree_mutex) || \ |
87fb54f1 | 101 | lockdep_is_held(&cgroup_mutex), \ |
ace2bee8 | 102 | "cgroup_[tree_]mutex or RCU read lock required"); |
87fb54f1 | 103 | |
e5fca243 TH |
104 | /* |
105 | * cgroup destruction makes heavy use of work items and there can be a lot | |
106 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
107 | * destruction work items don't end up filling up max_active of system_wq | |
108 | * which may lead to deadlock. | |
109 | */ | |
110 | static struct workqueue_struct *cgroup_destroy_wq; | |
111 | ||
b1a21367 TH |
112 | /* |
113 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
114 | * separate workqueue as flush domain. | |
115 | */ | |
116 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
117 | ||
3ed80a62 | 118 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 119 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 120 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
121 | #include <linux/cgroup_subsys.h> |
122 | }; | |
073219e9 TH |
123 | #undef SUBSYS |
124 | ||
125 | /* array of cgroup subsystem names */ | |
126 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
127 | static const char *cgroup_subsys_name[] = { | |
128 | #include <linux/cgroup_subsys.h> | |
129 | }; | |
130 | #undef SUBSYS | |
ddbcc7e8 | 131 | |
ddbcc7e8 | 132 | /* |
9871bf95 TH |
133 | * The dummy hierarchy, reserved for the subsystems that are otherwise |
134 | * unattached - it never has more than a single cgroup, and all tasks are | |
135 | * part of that cgroup. | |
ddbcc7e8 | 136 | */ |
9871bf95 TH |
137 | static struct cgroupfs_root cgroup_dummy_root; |
138 | ||
139 | /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */ | |
140 | static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup; | |
ddbcc7e8 PM |
141 | |
142 | /* The list of hierarchy roots */ | |
143 | ||
9871bf95 TH |
144 | static LIST_HEAD(cgroup_roots); |
145 | static int cgroup_root_count; | |
ddbcc7e8 | 146 | |
3417ae1f | 147 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 148 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 149 | |
794611a1 LZ |
150 | /* |
151 | * Assign a monotonically increasing serial number to cgroups. It | |
152 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
153 | * numbers. Also, as cgroups are always appended to the parent's | |
154 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
155 | * the ascending serial number order on the list. Protected by |
156 | * cgroup_mutex. | |
794611a1 | 157 | */ |
00356bd5 | 158 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 159 | |
ddbcc7e8 | 160 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
161 | * check for fork/exit handlers to call. This avoids us having to do |
162 | * extra work in the fork/exit path if none of the subsystems need to | |
163 | * be called. | |
ddbcc7e8 | 164 | */ |
8947f9d5 | 165 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 166 | |
628f7cd4 TH |
167 | static struct cftype cgroup_base_files[]; |
168 | ||
59f5296b | 169 | static void cgroup_put(struct cgroup *cgrp); |
f2e85d57 TH |
170 | static int rebind_subsystems(struct cgroupfs_root *root, |
171 | unsigned long added_mask, unsigned removed_mask); | |
f20104de | 172 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 173 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
2bb566cb TH |
174 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
175 | bool is_add); | |
b1a21367 | 176 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 177 | |
95109b62 TH |
178 | /** |
179 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
180 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 181 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 182 | * |
ca8bdcaf TH |
183 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
184 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
185 | * the caller is responsible for pinning the returned css if it wants to | |
186 | * keep accessing it outside the said locks. This function may return | |
187 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
188 | */ |
189 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 190 | struct cgroup_subsys *ss) |
95109b62 | 191 | { |
ca8bdcaf | 192 | if (ss) |
aec25020 | 193 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 TH |
194 | lockdep_is_held(&cgroup_tree_mutex) || |
195 | lockdep_is_held(&cgroup_mutex)); | |
ca8bdcaf TH |
196 | else |
197 | return &cgrp->dummy_css; | |
95109b62 | 198 | } |
42809dd4 | 199 | |
ddbcc7e8 | 200 | /* convenient tests for these bits */ |
54766d4a | 201 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 202 | { |
54766d4a | 203 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
204 | } |
205 | ||
59f5296b TH |
206 | struct cgroup_subsys_state *seq_css(struct seq_file *seq) |
207 | { | |
2bd59d48 TH |
208 | struct kernfs_open_file *of = seq->private; |
209 | struct cgroup *cgrp = of->kn->parent->priv; | |
210 | struct cftype *cft = seq_cft(seq); | |
211 | ||
212 | /* | |
213 | * This is open and unprotected implementation of cgroup_css(). | |
214 | * seq_css() is only called from a kernfs file operation which has | |
215 | * an active reference on the file. Because all the subsystem | |
216 | * files are drained before a css is disassociated with a cgroup, | |
217 | * the matching css from the cgroup's subsys table is guaranteed to | |
218 | * be and stay valid until the enclosing operation is complete. | |
219 | */ | |
220 | if (cft->ss) | |
221 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
222 | else | |
223 | return &cgrp->dummy_css; | |
59f5296b TH |
224 | } |
225 | EXPORT_SYMBOL_GPL(seq_css); | |
226 | ||
78574cf9 LZ |
227 | /** |
228 | * cgroup_is_descendant - test ancestry | |
229 | * @cgrp: the cgroup to be tested | |
230 | * @ancestor: possible ancestor of @cgrp | |
231 | * | |
232 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
233 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
234 | * and @ancestor are accessible. | |
235 | */ | |
236 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
237 | { | |
238 | while (cgrp) { | |
239 | if (cgrp == ancestor) | |
240 | return true; | |
241 | cgrp = cgrp->parent; | |
242 | } | |
243 | return false; | |
244 | } | |
245 | EXPORT_SYMBOL_GPL(cgroup_is_descendant); | |
ddbcc7e8 | 246 | |
e9685a03 | 247 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
248 | { |
249 | const int bits = | |
bd89aabc PM |
250 | (1 << CGRP_RELEASABLE) | |
251 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
252 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
253 | } |
254 | ||
e9685a03 | 255 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 256 | { |
bd89aabc | 257 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
258 | } |
259 | ||
1c6727af TH |
260 | /** |
261 | * for_each_css - iterate all css's of a cgroup | |
262 | * @css: the iteration cursor | |
263 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
264 | * @cgrp: the target cgroup to iterate css's of | |
265 | * | |
266 | * Should be called under cgroup_mutex. | |
267 | */ | |
268 | #define for_each_css(css, ssid, cgrp) \ | |
269 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
270 | if (!((css) = rcu_dereference_check( \ | |
271 | (cgrp)->subsys[(ssid)], \ | |
ace2bee8 | 272 | lockdep_is_held(&cgroup_tree_mutex) || \ |
1c6727af TH |
273 | lockdep_is_held(&cgroup_mutex)))) { } \ |
274 | else | |
275 | ||
30159ec7 | 276 | /** |
3ed80a62 | 277 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 278 | * @ss: the iteration cursor |
780cd8b3 | 279 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 280 | */ |
780cd8b3 | 281 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
282 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
283 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 284 | |
5549c497 TH |
285 | /* iterate across the active hierarchies */ |
286 | #define for_each_active_root(root) \ | |
287 | list_for_each_entry((root), &cgroup_roots, root_list) | |
ddbcc7e8 | 288 | |
7ae1bad9 TH |
289 | /** |
290 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
291 | * @cgrp: the cgroup to be checked for liveness | |
292 | * | |
47cfcd09 TH |
293 | * On success, returns true; the mutex should be later unlocked. On |
294 | * failure returns false with no lock held. | |
7ae1bad9 | 295 | */ |
b9777cf8 | 296 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
297 | { |
298 | mutex_lock(&cgroup_mutex); | |
54766d4a | 299 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
300 | mutex_unlock(&cgroup_mutex); |
301 | return false; | |
302 | } | |
303 | return true; | |
304 | } | |
7ae1bad9 | 305 | |
81a6a5cd PM |
306 | /* the list of cgroups eligible for automatic release. Protected by |
307 | * release_list_lock */ | |
308 | static LIST_HEAD(release_list); | |
cdcc136f | 309 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
310 | static void cgroup_release_agent(struct work_struct *work); |
311 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 312 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 313 | |
69d0206c TH |
314 | /* |
315 | * A cgroup can be associated with multiple css_sets as different tasks may | |
316 | * belong to different cgroups on different hierarchies. In the other | |
317 | * direction, a css_set is naturally associated with multiple cgroups. | |
318 | * This M:N relationship is represented by the following link structure | |
319 | * which exists for each association and allows traversing the associations | |
320 | * from both sides. | |
321 | */ | |
322 | struct cgrp_cset_link { | |
323 | /* the cgroup and css_set this link associates */ | |
324 | struct cgroup *cgrp; | |
325 | struct css_set *cset; | |
326 | ||
327 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
328 | struct list_head cset_link; | |
329 | ||
330 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
331 | struct list_head cgrp_link; | |
817929ec PM |
332 | }; |
333 | ||
334 | /* The default css_set - used by init and its children prior to any | |
335 | * hierarchies being mounted. It contains a pointer to the root state | |
336 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
337 | * reference-counted, to improve performance when child cgroups | |
338 | * haven't been created. | |
339 | */ | |
340 | ||
341 | static struct css_set init_css_set; | |
69d0206c | 342 | static struct cgrp_cset_link init_cgrp_cset_link; |
817929ec | 343 | |
0942eeee | 344 | /* |
96d365e0 TH |
345 | * css_set_rwsem protects the list of css_set objects, and the chain of |
346 | * tasks off each css_set. | |
0942eeee | 347 | */ |
96d365e0 | 348 | static DECLARE_RWSEM(css_set_rwsem); |
817929ec PM |
349 | static int css_set_count; |
350 | ||
7717f7ba PM |
351 | /* |
352 | * hash table for cgroup groups. This improves the performance to find | |
353 | * an existing css_set. This hash doesn't (currently) take into | |
354 | * account cgroups in empty hierarchies. | |
355 | */ | |
472b1053 | 356 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 357 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 358 | |
0ac801fe | 359 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 360 | { |
0ac801fe | 361 | unsigned long key = 0UL; |
30159ec7 TH |
362 | struct cgroup_subsys *ss; |
363 | int i; | |
472b1053 | 364 | |
30159ec7 | 365 | for_each_subsys(ss, i) |
0ac801fe LZ |
366 | key += (unsigned long)css[i]; |
367 | key = (key >> 16) ^ key; | |
472b1053 | 368 | |
0ac801fe | 369 | return key; |
472b1053 LZ |
370 | } |
371 | ||
5abb8855 | 372 | static void __put_css_set(struct css_set *cset, int taskexit) |
b4f48b63 | 373 | { |
69d0206c | 374 | struct cgrp_cset_link *link, *tmp_link; |
5abb8855 | 375 | |
146aa1bd LJ |
376 | /* |
377 | * Ensure that the refcount doesn't hit zero while any readers | |
378 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
379 | * rwlock | |
380 | */ | |
5abb8855 | 381 | if (atomic_add_unless(&cset->refcount, -1, 1)) |
146aa1bd | 382 | return; |
96d365e0 | 383 | down_write(&css_set_rwsem); |
5abb8855 | 384 | if (!atomic_dec_and_test(&cset->refcount)) { |
96d365e0 | 385 | up_write(&css_set_rwsem); |
146aa1bd LJ |
386 | return; |
387 | } | |
81a6a5cd | 388 | |
2c6ab6d2 | 389 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
5abb8855 | 390 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
391 | css_set_count--; |
392 | ||
69d0206c | 393 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 394 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 395 | |
69d0206c TH |
396 | list_del(&link->cset_link); |
397 | list_del(&link->cgrp_link); | |
71b5707e | 398 | |
96d365e0 | 399 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
6f3d828f | 400 | if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { |
81a6a5cd | 401 | if (taskexit) |
bd89aabc PM |
402 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
403 | check_for_release(cgrp); | |
81a6a5cd | 404 | } |
2c6ab6d2 PM |
405 | |
406 | kfree(link); | |
81a6a5cd | 407 | } |
2c6ab6d2 | 408 | |
96d365e0 | 409 | up_write(&css_set_rwsem); |
5abb8855 | 410 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
411 | } |
412 | ||
817929ec PM |
413 | /* |
414 | * refcounted get/put for css_set objects | |
415 | */ | |
5abb8855 | 416 | static inline void get_css_set(struct css_set *cset) |
817929ec | 417 | { |
5abb8855 | 418 | atomic_inc(&cset->refcount); |
817929ec PM |
419 | } |
420 | ||
5abb8855 | 421 | static inline void put_css_set(struct css_set *cset) |
817929ec | 422 | { |
5abb8855 | 423 | __put_css_set(cset, 0); |
817929ec PM |
424 | } |
425 | ||
5abb8855 | 426 | static inline void put_css_set_taskexit(struct css_set *cset) |
81a6a5cd | 427 | { |
5abb8855 | 428 | __put_css_set(cset, 1); |
81a6a5cd PM |
429 | } |
430 | ||
b326f9d0 | 431 | /** |
7717f7ba | 432 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
433 | * @cset: candidate css_set being tested |
434 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
435 | * @new_cgrp: cgroup that's being entered by the task |
436 | * @template: desired set of css pointers in css_set (pre-calculated) | |
437 | * | |
6f4b7e63 | 438 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
439 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
440 | */ | |
5abb8855 TH |
441 | static bool compare_css_sets(struct css_set *cset, |
442 | struct css_set *old_cset, | |
7717f7ba PM |
443 | struct cgroup *new_cgrp, |
444 | struct cgroup_subsys_state *template[]) | |
445 | { | |
446 | struct list_head *l1, *l2; | |
447 | ||
5abb8855 | 448 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) { |
7717f7ba PM |
449 | /* Not all subsystems matched */ |
450 | return false; | |
451 | } | |
452 | ||
453 | /* | |
454 | * Compare cgroup pointers in order to distinguish between | |
455 | * different cgroups in heirarchies with no subsystems. We | |
456 | * could get by with just this check alone (and skip the | |
457 | * memcmp above) but on most setups the memcmp check will | |
458 | * avoid the need for this more expensive check on almost all | |
459 | * candidates. | |
460 | */ | |
461 | ||
69d0206c TH |
462 | l1 = &cset->cgrp_links; |
463 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 464 | while (1) { |
69d0206c | 465 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 466 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
467 | |
468 | l1 = l1->next; | |
469 | l2 = l2->next; | |
470 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
471 | if (l1 == &cset->cgrp_links) { |
472 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
473 | break; |
474 | } else { | |
69d0206c | 475 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
476 | } |
477 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
478 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
479 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
480 | cgrp1 = link1->cgrp; | |
481 | cgrp2 = link2->cgrp; | |
7717f7ba | 482 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 483 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
484 | |
485 | /* | |
486 | * If this hierarchy is the hierarchy of the cgroup | |
487 | * that's changing, then we need to check that this | |
488 | * css_set points to the new cgroup; if it's any other | |
489 | * hierarchy, then this css_set should point to the | |
490 | * same cgroup as the old css_set. | |
491 | */ | |
5abb8855 TH |
492 | if (cgrp1->root == new_cgrp->root) { |
493 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
494 | return false; |
495 | } else { | |
5abb8855 | 496 | if (cgrp1 != cgrp2) |
7717f7ba PM |
497 | return false; |
498 | } | |
499 | } | |
500 | return true; | |
501 | } | |
502 | ||
b326f9d0 TH |
503 | /** |
504 | * find_existing_css_set - init css array and find the matching css_set | |
505 | * @old_cset: the css_set that we're using before the cgroup transition | |
506 | * @cgrp: the cgroup that we're moving into | |
507 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 508 | */ |
5abb8855 TH |
509 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
510 | struct cgroup *cgrp, | |
511 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 512 | { |
bd89aabc | 513 | struct cgroupfs_root *root = cgrp->root; |
30159ec7 | 514 | struct cgroup_subsys *ss; |
5abb8855 | 515 | struct css_set *cset; |
0ac801fe | 516 | unsigned long key; |
b326f9d0 | 517 | int i; |
817929ec | 518 | |
aae8aab4 BB |
519 | /* |
520 | * Build the set of subsystem state objects that we want to see in the | |
521 | * new css_set. while subsystems can change globally, the entries here | |
522 | * won't change, so no need for locking. | |
523 | */ | |
30159ec7 | 524 | for_each_subsys(ss, i) { |
a1a71b45 | 525 | if (root->subsys_mask & (1UL << i)) { |
817929ec PM |
526 | /* Subsystem is in this hierarchy. So we want |
527 | * the subsystem state from the new | |
528 | * cgroup */ | |
ca8bdcaf | 529 | template[i] = cgroup_css(cgrp, ss); |
817929ec PM |
530 | } else { |
531 | /* Subsystem is not in this hierarchy, so we | |
532 | * don't want to change the subsystem state */ | |
5abb8855 | 533 | template[i] = old_cset->subsys[i]; |
817929ec PM |
534 | } |
535 | } | |
536 | ||
0ac801fe | 537 | key = css_set_hash(template); |
5abb8855 TH |
538 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
539 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
540 | continue; |
541 | ||
542 | /* This css_set matches what we need */ | |
5abb8855 | 543 | return cset; |
472b1053 | 544 | } |
817929ec PM |
545 | |
546 | /* No existing cgroup group matched */ | |
547 | return NULL; | |
548 | } | |
549 | ||
69d0206c | 550 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 551 | { |
69d0206c | 552 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 553 | |
69d0206c TH |
554 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
555 | list_del(&link->cset_link); | |
36553434 LZ |
556 | kfree(link); |
557 | } | |
558 | } | |
559 | ||
69d0206c TH |
560 | /** |
561 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
562 | * @count: the number of links to allocate | |
563 | * @tmp_links: list_head the allocated links are put on | |
564 | * | |
565 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
566 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 567 | */ |
69d0206c | 568 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 569 | { |
69d0206c | 570 | struct cgrp_cset_link *link; |
817929ec | 571 | int i; |
69d0206c TH |
572 | |
573 | INIT_LIST_HEAD(tmp_links); | |
574 | ||
817929ec | 575 | for (i = 0; i < count; i++) { |
f4f4be2b | 576 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 577 | if (!link) { |
69d0206c | 578 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
579 | return -ENOMEM; |
580 | } | |
69d0206c | 581 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
582 | } |
583 | return 0; | |
584 | } | |
585 | ||
c12f65d4 LZ |
586 | /** |
587 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 588 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 589 | * @cset: the css_set to be linked |
c12f65d4 LZ |
590 | * @cgrp: the destination cgroup |
591 | */ | |
69d0206c TH |
592 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
593 | struct cgroup *cgrp) | |
c12f65d4 | 594 | { |
69d0206c | 595 | struct cgrp_cset_link *link; |
c12f65d4 | 596 | |
69d0206c TH |
597 | BUG_ON(list_empty(tmp_links)); |
598 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); | |
599 | link->cset = cset; | |
7717f7ba | 600 | link->cgrp = cgrp; |
69d0206c | 601 | list_move(&link->cset_link, &cgrp->cset_links); |
7717f7ba PM |
602 | /* |
603 | * Always add links to the tail of the list so that the list | |
604 | * is sorted by order of hierarchy creation | |
605 | */ | |
69d0206c | 606 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
607 | } |
608 | ||
b326f9d0 TH |
609 | /** |
610 | * find_css_set - return a new css_set with one cgroup updated | |
611 | * @old_cset: the baseline css_set | |
612 | * @cgrp: the cgroup to be updated | |
613 | * | |
614 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
615 | * substituted into the appropriate hierarchy. | |
817929ec | 616 | */ |
5abb8855 TH |
617 | static struct css_set *find_css_set(struct css_set *old_cset, |
618 | struct cgroup *cgrp) | |
817929ec | 619 | { |
b326f9d0 | 620 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 621 | struct css_set *cset; |
69d0206c TH |
622 | struct list_head tmp_links; |
623 | struct cgrp_cset_link *link; | |
0ac801fe | 624 | unsigned long key; |
472b1053 | 625 | |
b326f9d0 TH |
626 | lockdep_assert_held(&cgroup_mutex); |
627 | ||
817929ec PM |
628 | /* First see if we already have a cgroup group that matches |
629 | * the desired set */ | |
96d365e0 | 630 | down_read(&css_set_rwsem); |
5abb8855 TH |
631 | cset = find_existing_css_set(old_cset, cgrp, template); |
632 | if (cset) | |
633 | get_css_set(cset); | |
96d365e0 | 634 | up_read(&css_set_rwsem); |
817929ec | 635 | |
5abb8855 TH |
636 | if (cset) |
637 | return cset; | |
817929ec | 638 | |
f4f4be2b | 639 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 640 | if (!cset) |
817929ec PM |
641 | return NULL; |
642 | ||
69d0206c | 643 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 644 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 645 | kfree(cset); |
817929ec PM |
646 | return NULL; |
647 | } | |
648 | ||
5abb8855 | 649 | atomic_set(&cset->refcount, 1); |
69d0206c | 650 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 TH |
651 | INIT_LIST_HEAD(&cset->tasks); |
652 | INIT_HLIST_NODE(&cset->hlist); | |
817929ec PM |
653 | |
654 | /* Copy the set of subsystem state objects generated in | |
655 | * find_existing_css_set() */ | |
5abb8855 | 656 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 657 | |
96d365e0 | 658 | down_write(&css_set_rwsem); |
817929ec | 659 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 660 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 661 | struct cgroup *c = link->cgrp; |
69d0206c | 662 | |
7717f7ba PM |
663 | if (c->root == cgrp->root) |
664 | c = cgrp; | |
69d0206c | 665 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 666 | } |
817929ec | 667 | |
69d0206c | 668 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 669 | |
817929ec | 670 | css_set_count++; |
472b1053 LZ |
671 | |
672 | /* Add this cgroup group to the hash table */ | |
5abb8855 TH |
673 | key = css_set_hash(cset->subsys); |
674 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 675 | |
96d365e0 | 676 | up_write(&css_set_rwsem); |
817929ec | 677 | |
5abb8855 | 678 | return cset; |
b4f48b63 PM |
679 | } |
680 | ||
2bd59d48 TH |
681 | static struct cgroupfs_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
682 | { | |
683 | struct cgroup *top_cgrp = kf_root->kn->priv; | |
684 | ||
685 | return top_cgrp->root; | |
686 | } | |
687 | ||
f2e85d57 TH |
688 | static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end) |
689 | { | |
690 | int id; | |
691 | ||
692 | lockdep_assert_held(&cgroup_mutex); | |
693 | ||
694 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end, | |
695 | GFP_KERNEL); | |
696 | if (id < 0) | |
697 | return id; | |
698 | ||
699 | root->hierarchy_id = id; | |
700 | return 0; | |
701 | } | |
702 | ||
703 | static void cgroup_exit_root_id(struct cgroupfs_root *root) | |
704 | { | |
705 | lockdep_assert_held(&cgroup_mutex); | |
706 | ||
707 | if (root->hierarchy_id) { | |
708 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
709 | root->hierarchy_id = 0; | |
710 | } | |
711 | } | |
712 | ||
713 | static void cgroup_free_root(struct cgroupfs_root *root) | |
714 | { | |
715 | if (root) { | |
716 | /* hierarhcy ID shoulid already have been released */ | |
717 | WARN_ON_ONCE(root->hierarchy_id); | |
718 | ||
719 | idr_destroy(&root->cgroup_idr); | |
720 | kfree(root); | |
721 | } | |
722 | } | |
723 | ||
776f02fa | 724 | static void cgroup_destroy_root(struct cgroupfs_root *root) |
59f5296b | 725 | { |
f2e85d57 TH |
726 | struct cgroup *cgrp = &root->top_cgroup; |
727 | struct cgrp_cset_link *link, *tmp_link; | |
f2e85d57 | 728 | |
2bd59d48 | 729 | mutex_lock(&cgroup_tree_mutex); |
2bd59d48 | 730 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 731 | |
776f02fa | 732 | BUG_ON(atomic_read(&root->nr_cgrps)); |
f2e85d57 TH |
733 | BUG_ON(!list_empty(&cgrp->children)); |
734 | ||
f2e85d57 | 735 | /* Rebind all subsystems back to the default hierarchy */ |
35585573 | 736 | WARN_ON(rebind_subsystems(root, 0, root->subsys_mask)); |
f2e85d57 TH |
737 | |
738 | /* | |
739 | * Release all the links from cset_links to this hierarchy's | |
740 | * root cgroup | |
741 | */ | |
96d365e0 | 742 | down_write(&css_set_rwsem); |
f2e85d57 TH |
743 | |
744 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
745 | list_del(&link->cset_link); | |
746 | list_del(&link->cgrp_link); | |
747 | kfree(link); | |
748 | } | |
96d365e0 | 749 | up_write(&css_set_rwsem); |
f2e85d57 TH |
750 | |
751 | if (!list_empty(&root->root_list)) { | |
752 | list_del(&root->root_list); | |
753 | cgroup_root_count--; | |
754 | } | |
755 | ||
756 | cgroup_exit_root_id(root); | |
757 | ||
758 | mutex_unlock(&cgroup_mutex); | |
759 | mutex_unlock(&cgroup_tree_mutex); | |
f2e85d57 | 760 | |
2bd59d48 | 761 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
762 | cgroup_free_root(root); |
763 | } | |
764 | ||
7717f7ba PM |
765 | /* |
766 | * Return the cgroup for "task" from the given hierarchy. Must be | |
96d365e0 | 767 | * called with cgroup_mutex and css_set_rwsem held. |
7717f7ba PM |
768 | */ |
769 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
770 | struct cgroupfs_root *root) | |
771 | { | |
5abb8855 | 772 | struct css_set *cset; |
7717f7ba PM |
773 | struct cgroup *res = NULL; |
774 | ||
96d365e0 TH |
775 | lockdep_assert_held(&cgroup_mutex); |
776 | lockdep_assert_held(&css_set_rwsem); | |
777 | ||
7717f7ba PM |
778 | /* |
779 | * No need to lock the task - since we hold cgroup_mutex the | |
780 | * task can't change groups, so the only thing that can happen | |
781 | * is that it exits and its css is set back to init_css_set. | |
782 | */ | |
a8ad805c | 783 | cset = task_css_set(task); |
5abb8855 | 784 | if (cset == &init_css_set) { |
7717f7ba PM |
785 | res = &root->top_cgroup; |
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 PM |
803 | /* |
804 | * There is one global cgroup mutex. We also require taking | |
805 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
806 | * See "The task_lock() exception", at the end of this comment. | |
807 | * | |
808 | * A task must hold cgroup_mutex to modify cgroups. | |
809 | * | |
810 | * Any task can increment and decrement the count field without lock. | |
811 | * So in general, code holding cgroup_mutex can't rely on the count | |
812 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 813 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
814 | * means that no tasks are currently attached, therefore there is no |
815 | * way a task attached to that cgroup can fork (the other way to | |
816 | * increment the count). So code holding cgroup_mutex can safely | |
817 | * assume that if the count is zero, it will stay zero. Similarly, if | |
818 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
819 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
820 | * needs that mutex. | |
821 | * | |
ddbcc7e8 PM |
822 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
823 | * (usually) take cgroup_mutex. These are the two most performance | |
824 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
825 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
826 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
827 | * to the release agent with the name of the cgroup (path relative to |
828 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
829 | * |
830 | * A cgroup can only be deleted if both its 'count' of using tasks | |
831 | * is zero, and its list of 'children' cgroups is empty. Since all | |
832 | * tasks in the system use _some_ cgroup, and since there is always at | |
833 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
834 | * always has either children cgroups and/or using tasks. So we don't | |
835 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
836 | * | |
837 | * The task_lock() exception | |
838 | * | |
839 | * The need for this exception arises from the action of | |
d0b2fdd2 | 840 | * cgroup_attach_task(), which overwrites one task's cgroup pointer with |
a043e3b2 | 841 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
842 | * several performance critical places that need to reference |
843 | * task->cgroup without the expense of grabbing a system global | |
844 | * mutex. Therefore except as noted below, when dereferencing or, as | |
d0b2fdd2 | 845 | * in cgroup_attach_task(), modifying a task's cgroup pointer we use |
ddbcc7e8 PM |
846 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
847 | * the task_struct routinely used for such matters. | |
848 | * | |
849 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 850 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
851 | */ |
852 | ||
628f7cd4 | 853 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
2bd59d48 | 854 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 855 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 856 | |
8d7e6fb0 TH |
857 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
858 | char *buf) | |
859 | { | |
860 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && | |
861 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
862 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
863 | cft->ss->name, cft->name); | |
864 | else | |
865 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
866 | return buf; | |
867 | } | |
868 | ||
f2e85d57 TH |
869 | /** |
870 | * cgroup_file_mode - deduce file mode of a control file | |
871 | * @cft: the control file in question | |
872 | * | |
873 | * returns cft->mode if ->mode is not 0 | |
874 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
875 | * returns S_IRUGO if it has only a read handler | |
876 | * returns S_IWUSR if it has only a write hander | |
877 | */ | |
878 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
879 | { | |
880 | umode_t mode = 0; | |
881 | ||
882 | if (cft->mode) | |
883 | return cft->mode; | |
884 | ||
885 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
886 | mode |= S_IRUGO; | |
887 | ||
888 | if (cft->write_u64 || cft->write_s64 || cft->write_string || | |
889 | cft->trigger) | |
890 | mode |= S_IWUSR; | |
891 | ||
892 | return mode; | |
893 | } | |
894 | ||
be445626 LZ |
895 | static void cgroup_free_fn(struct work_struct *work) |
896 | { | |
ea15f8cc | 897 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 | 898 | |
3c9c825b | 899 | atomic_dec(&cgrp->root->nr_cgrps); |
b1a21367 | 900 | cgroup_pidlist_destroy_all(cgrp); |
be445626 | 901 | |
776f02fa TH |
902 | if (cgrp->parent) { |
903 | /* | |
904 | * We get a ref to the parent, and put the ref when this | |
905 | * cgroup is being freed, so it's guaranteed that the | |
906 | * parent won't be destroyed before its children. | |
907 | */ | |
908 | cgroup_put(cgrp->parent); | |
909 | kernfs_put(cgrp->kn); | |
910 | kfree(cgrp); | |
911 | } else { | |
912 | /* | |
913 | * This is top cgroup's refcnt reaching zero, which | |
914 | * indicates that the root should be released. | |
915 | */ | |
916 | cgroup_destroy_root(cgrp->root); | |
917 | } | |
be445626 LZ |
918 | } |
919 | ||
920 | static void cgroup_free_rcu(struct rcu_head *head) | |
921 | { | |
922 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
923 | ||
ea15f8cc | 924 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 925 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
926 | } |
927 | ||
59f5296b TH |
928 | static void cgroup_get(struct cgroup *cgrp) |
929 | { | |
2bd59d48 TH |
930 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
931 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
932 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
933 | } |
934 | ||
59f5296b TH |
935 | static void cgroup_put(struct cgroup *cgrp) |
936 | { | |
2bd59d48 TH |
937 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
938 | return; | |
776f02fa | 939 | if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp))) |
2bd59d48 | 940 | return; |
59f5296b | 941 | |
2bd59d48 TH |
942 | /* |
943 | * XXX: cgrp->id is only used to look up css's. As cgroup and | |
944 | * css's lifetimes will be decoupled, it should be made | |
945 | * per-subsystem and moved to css->id so that lookups are | |
946 | * successful until the target css is released. | |
947 | */ | |
948 | mutex_lock(&cgroup_mutex); | |
949 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
950 | mutex_unlock(&cgroup_mutex); | |
951 | cgrp->id = -1; | |
ddbcc7e8 | 952 | |
2bd59d48 | 953 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 PM |
954 | } |
955 | ||
2739d3cc | 956 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 957 | { |
2bd59d48 | 958 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 959 | |
ace2bee8 | 960 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 961 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
962 | } |
963 | ||
13af07df | 964 | /** |
628f7cd4 | 965 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 966 | * @cgrp: target cgroup |
13af07df AR |
967 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
968 | */ | |
628f7cd4 | 969 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 970 | { |
13af07df | 971 | struct cgroup_subsys *ss; |
b420ba7d | 972 | int i; |
05ef1d7c | 973 | |
b420ba7d | 974 | for_each_subsys(ss, i) { |
0adb0704 | 975 | struct cftype *cfts; |
b420ba7d TH |
976 | |
977 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 978 | continue; |
0adb0704 TH |
979 | list_for_each_entry(cfts, &ss->cfts, node) |
980 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 981 | } |
ddbcc7e8 PM |
982 | } |
983 | ||
ddbcc7e8 | 984 | static int rebind_subsystems(struct cgroupfs_root *root, |
a8a648c4 | 985 | unsigned long added_mask, unsigned removed_mask) |
ddbcc7e8 | 986 | { |
bd89aabc | 987 | struct cgroup *cgrp = &root->top_cgroup; |
30159ec7 | 988 | struct cgroup_subsys *ss; |
3126121f | 989 | int i, ret; |
ddbcc7e8 | 990 | |
ace2bee8 TH |
991 | lockdep_assert_held(&cgroup_tree_mutex); |
992 | lockdep_assert_held(&cgroup_mutex); | |
aae8aab4 | 993 | |
ddbcc7e8 | 994 | /* Check that any added subsystems are currently free */ |
3ed80a62 TH |
995 | for_each_subsys(ss, i) |
996 | if ((added_mask & (1 << i)) && ss->root != &cgroup_dummy_root) | |
997 | return -EBUSY; | |
ddbcc7e8 | 998 | |
3126121f TH |
999 | ret = cgroup_populate_dir(cgrp, added_mask); |
1000 | if (ret) | |
3ed80a62 | 1001 | return ret; |
3126121f TH |
1002 | |
1003 | /* | |
1004 | * Nothing can fail from this point on. Remove files for the | |
1005 | * removed subsystems and rebind each subsystem. | |
1006 | */ | |
4ac06017 | 1007 | mutex_unlock(&cgroup_mutex); |
3126121f | 1008 | cgroup_clear_dir(cgrp, removed_mask); |
4ac06017 | 1009 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1010 | |
30159ec7 | 1011 | for_each_subsys(ss, i) { |
ddbcc7e8 | 1012 | unsigned long bit = 1UL << i; |
30159ec7 | 1013 | |
a1a71b45 | 1014 | if (bit & added_mask) { |
ddbcc7e8 | 1015 | /* We're binding this subsystem to this hierarchy */ |
ca8bdcaf TH |
1016 | BUG_ON(cgroup_css(cgrp, ss)); |
1017 | BUG_ON(!cgroup_css(cgroup_dummy_top, ss)); | |
1018 | BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top); | |
a8a648c4 | 1019 | |
73e80ed8 | 1020 | rcu_assign_pointer(cgrp->subsys[i], |
ca8bdcaf TH |
1021 | cgroup_css(cgroup_dummy_top, ss)); |
1022 | cgroup_css(cgrp, ss)->cgroup = cgrp; | |
a8a648c4 | 1023 | |
b2aa30f7 | 1024 | ss->root = root; |
ddbcc7e8 | 1025 | if (ss->bind) |
ca8bdcaf | 1026 | ss->bind(cgroup_css(cgrp, ss)); |
a8a648c4 | 1027 | |
cf5d5941 | 1028 | /* refcount was already taken, and we're keeping it */ |
a8a648c4 | 1029 | root->subsys_mask |= bit; |
a1a71b45 | 1030 | } else if (bit & removed_mask) { |
ddbcc7e8 | 1031 | /* We're removing this subsystem */ |
ca8bdcaf TH |
1032 | BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss)); |
1033 | BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp); | |
a8a648c4 | 1034 | |
ddbcc7e8 | 1035 | if (ss->bind) |
ca8bdcaf | 1036 | ss->bind(cgroup_css(cgroup_dummy_top, ss)); |
73e80ed8 | 1037 | |
ca8bdcaf | 1038 | cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top; |
73e80ed8 TH |
1039 | RCU_INIT_POINTER(cgrp->subsys[i], NULL); |
1040 | ||
9871bf95 | 1041 | cgroup_subsys[i]->root = &cgroup_dummy_root; |
a8a648c4 | 1042 | root->subsys_mask &= ~bit; |
ddbcc7e8 PM |
1043 | } |
1044 | } | |
ddbcc7e8 | 1045 | |
2bd59d48 | 1046 | kernfs_activate(cgrp->kn); |
ddbcc7e8 PM |
1047 | return 0; |
1048 | } | |
1049 | ||
2bd59d48 TH |
1050 | static int cgroup_show_options(struct seq_file *seq, |
1051 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1052 | { |
2bd59d48 | 1053 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1054 | struct cgroup_subsys *ss; |
b85d2040 | 1055 | int ssid; |
ddbcc7e8 | 1056 | |
b85d2040 TH |
1057 | for_each_subsys(ss, ssid) |
1058 | if (root->subsys_mask & (1 << ssid)) | |
1059 | seq_printf(seq, ",%s", ss->name); | |
873fe09e TH |
1060 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1061 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1062 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1063 | seq_puts(seq, ",noprefix"); |
93438629 | 1064 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1065 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1066 | |
1067 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1068 | if (strlen(root->release_agent_path)) |
1069 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1070 | spin_unlock(&release_agent_path_lock); |
1071 | ||
2260e7fc | 1072 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags)) |
97978e6d | 1073 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1074 | if (strlen(root->name)) |
1075 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1076 | return 0; |
1077 | } | |
1078 | ||
1079 | struct cgroup_sb_opts { | |
a1a71b45 | 1080 | unsigned long subsys_mask; |
ddbcc7e8 | 1081 | unsigned long flags; |
81a6a5cd | 1082 | char *release_agent; |
2260e7fc | 1083 | bool cpuset_clone_children; |
c6d57f33 | 1084 | char *name; |
2c6ab6d2 PM |
1085 | /* User explicitly requested empty subsystem */ |
1086 | bool none; | |
ddbcc7e8 PM |
1087 | }; |
1088 | ||
aae8aab4 | 1089 | /* |
9871bf95 TH |
1090 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1091 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1092 | * array. This function takes refcounts on subsystems to be used, unless it | |
1093 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1094 | */ |
cf5d5941 | 1095 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1096 | { |
32a8cf23 DL |
1097 | char *token, *o = data; |
1098 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1099 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1100 | struct cgroup_subsys *ss; |
1101 | int i; | |
f9ab5b5b | 1102 | |
aae8aab4 BB |
1103 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1104 | ||
f9ab5b5b | 1105 | #ifdef CONFIG_CPUSETS |
073219e9 | 1106 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1107 | #endif |
ddbcc7e8 | 1108 | |
c6d57f33 | 1109 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1110 | |
1111 | while ((token = strsep(&o, ",")) != NULL) { | |
1112 | if (!*token) | |
1113 | return -EINVAL; | |
32a8cf23 | 1114 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1115 | /* Explicitly have no subsystems */ |
1116 | opts->none = true; | |
32a8cf23 DL |
1117 | continue; |
1118 | } | |
1119 | if (!strcmp(token, "all")) { | |
1120 | /* Mutually exclusive option 'all' + subsystem name */ | |
1121 | if (one_ss) | |
1122 | return -EINVAL; | |
1123 | all_ss = true; | |
1124 | continue; | |
1125 | } | |
873fe09e TH |
1126 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1127 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1128 | continue; | |
1129 | } | |
32a8cf23 | 1130 | if (!strcmp(token, "noprefix")) { |
93438629 | 1131 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1132 | continue; |
1133 | } | |
1134 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1135 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1136 | continue; |
1137 | } | |
03b1cde6 | 1138 | if (!strcmp(token, "xattr")) { |
93438629 | 1139 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1140 | continue; |
1141 | } | |
32a8cf23 | 1142 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1143 | /* Specifying two release agents is forbidden */ |
1144 | if (opts->release_agent) | |
1145 | return -EINVAL; | |
c6d57f33 | 1146 | opts->release_agent = |
e400c285 | 1147 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1148 | if (!opts->release_agent) |
1149 | return -ENOMEM; | |
32a8cf23 DL |
1150 | continue; |
1151 | } | |
1152 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1153 | const char *name = token + 5; |
1154 | /* Can't specify an empty name */ | |
1155 | if (!strlen(name)) | |
1156 | return -EINVAL; | |
1157 | /* Must match [\w.-]+ */ | |
1158 | for (i = 0; i < strlen(name); i++) { | |
1159 | char c = name[i]; | |
1160 | if (isalnum(c)) | |
1161 | continue; | |
1162 | if ((c == '.') || (c == '-') || (c == '_')) | |
1163 | continue; | |
1164 | return -EINVAL; | |
1165 | } | |
1166 | /* Specifying two names is forbidden */ | |
1167 | if (opts->name) | |
1168 | return -EINVAL; | |
1169 | opts->name = kstrndup(name, | |
e400c285 | 1170 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1171 | GFP_KERNEL); |
1172 | if (!opts->name) | |
1173 | return -ENOMEM; | |
32a8cf23 DL |
1174 | |
1175 | continue; | |
1176 | } | |
1177 | ||
30159ec7 | 1178 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1179 | if (strcmp(token, ss->name)) |
1180 | continue; | |
1181 | if (ss->disabled) | |
1182 | continue; | |
1183 | ||
1184 | /* Mutually exclusive option 'all' + subsystem name */ | |
1185 | if (all_ss) | |
1186 | return -EINVAL; | |
a1a71b45 | 1187 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1188 | one_ss = true; |
1189 | ||
1190 | break; | |
1191 | } | |
1192 | if (i == CGROUP_SUBSYS_COUNT) | |
1193 | return -ENOENT; | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1198 | * otherwise if 'none', 'name=' and a subsystem name options |
1199 | * were not specified, let's default to 'all' | |
32a8cf23 | 1200 | */ |
30159ec7 TH |
1201 | if (all_ss || (!one_ss && !opts->none && !opts->name)) |
1202 | for_each_subsys(ss, i) | |
1203 | if (!ss->disabled) | |
1204 | set_bit(i, &opts->subsys_mask); | |
ddbcc7e8 | 1205 | |
2c6ab6d2 PM |
1206 | /* Consistency checks */ |
1207 | ||
873fe09e TH |
1208 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1209 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1210 | ||
d3ba07c3 TH |
1211 | if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) || |
1212 | opts->cpuset_clone_children || opts->release_agent || | |
1213 | opts->name) { | |
1214 | pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n"); | |
873fe09e TH |
1215 | return -EINVAL; |
1216 | } | |
873fe09e TH |
1217 | } |
1218 | ||
f9ab5b5b LZ |
1219 | /* |
1220 | * Option noprefix was introduced just for backward compatibility | |
1221 | * with the old cpuset, so we allow noprefix only if mounting just | |
1222 | * the cpuset subsystem. | |
1223 | */ | |
93438629 | 1224 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1225 | return -EINVAL; |
1226 | ||
2c6ab6d2 PM |
1227 | |
1228 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1229 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1230 | return -EINVAL; |
1231 | ||
1232 | /* | |
1233 | * We either have to specify by name or by subsystems. (So all | |
1234 | * empty hierarchies must have a name). | |
1235 | */ | |
a1a71b45 | 1236 | if (!opts->subsys_mask && !opts->name) |
ddbcc7e8 PM |
1237 | return -EINVAL; |
1238 | ||
1239 | return 0; | |
1240 | } | |
1241 | ||
2bd59d48 | 1242 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1243 | { |
1244 | int ret = 0; | |
2bd59d48 | 1245 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1246 | struct cgroup_sb_opts opts; |
a1a71b45 | 1247 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1248 | |
873fe09e TH |
1249 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1250 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1251 | return -EINVAL; | |
1252 | } | |
1253 | ||
ace2bee8 | 1254 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1255 | mutex_lock(&cgroup_mutex); |
1256 | ||
1257 | /* See what subsystems are wanted */ | |
1258 | ret = parse_cgroupfs_options(data, &opts); | |
1259 | if (ret) | |
1260 | goto out_unlock; | |
1261 | ||
a8a648c4 | 1262 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1263 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1264 | task_tgid_nr(current), current->comm); | |
1265 | ||
a1a71b45 AR |
1266 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1267 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1268 | |
cf5d5941 | 1269 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1270 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1271 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1272 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1273 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1274 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1275 | ret = -EINVAL; |
1276 | goto out_unlock; | |
1277 | } | |
1278 | ||
f172e67c | 1279 | /* remounting is not allowed for populated hierarchies */ |
3c9c825b | 1280 | if (!list_empty(&root->top_cgroup.children)) { |
f172e67c | 1281 | ret = -EBUSY; |
0670e08b | 1282 | goto out_unlock; |
cf5d5941 | 1283 | } |
ddbcc7e8 | 1284 | |
a8a648c4 | 1285 | ret = rebind_subsystems(root, added_mask, removed_mask); |
3126121f | 1286 | if (ret) |
0670e08b | 1287 | goto out_unlock; |
ddbcc7e8 | 1288 | |
69e943b7 TH |
1289 | if (opts.release_agent) { |
1290 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1291 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1292 | spin_unlock(&release_agent_path_lock); |
1293 | } | |
ddbcc7e8 | 1294 | out_unlock: |
66bdc9cf | 1295 | kfree(opts.release_agent); |
c6d57f33 | 1296 | kfree(opts.name); |
ddbcc7e8 | 1297 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1298 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1299 | return ret; |
1300 | } | |
1301 | ||
afeb0f9f TH |
1302 | /* |
1303 | * To reduce the fork() overhead for systems that are not actually using | |
1304 | * their cgroups capability, we don't maintain the lists running through | |
1305 | * each css_set to its tasks until we see the list actually used - in other | |
1306 | * words after the first mount. | |
1307 | */ | |
1308 | static bool use_task_css_set_links __read_mostly; | |
1309 | ||
1310 | static void cgroup_enable_task_cg_lists(void) | |
1311 | { | |
1312 | struct task_struct *p, *g; | |
1313 | ||
96d365e0 | 1314 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1315 | |
1316 | if (use_task_css_set_links) | |
1317 | goto out_unlock; | |
1318 | ||
1319 | use_task_css_set_links = true; | |
1320 | ||
1321 | /* | |
1322 | * We need tasklist_lock because RCU is not safe against | |
1323 | * while_each_thread(). Besides, a forking task that has passed | |
1324 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1325 | * is not guaranteed to have its child immediately visible in the | |
1326 | * tasklist if we walk through it with RCU. | |
1327 | */ | |
1328 | read_lock(&tasklist_lock); | |
1329 | do_each_thread(g, p) { | |
1330 | task_lock(p); | |
1331 | ||
1332 | WARN_ON_ONCE(!list_empty(&p->cg_list) || | |
1333 | task_css_set(p) != &init_css_set); | |
1334 | ||
1335 | /* | |
1336 | * We should check if the process is exiting, otherwise | |
1337 | * it will race with cgroup_exit() in that the list | |
1338 | * entry won't be deleted though the process has exited. | |
1339 | */ | |
1340 | if (!(p->flags & PF_EXITING)) | |
1341 | list_add(&p->cg_list, &task_css_set(p)->tasks); | |
1342 | ||
1343 | task_unlock(p); | |
1344 | } while_each_thread(g, p); | |
1345 | read_unlock(&tasklist_lock); | |
1346 | out_unlock: | |
96d365e0 | 1347 | up_write(&css_set_rwsem); |
afeb0f9f TH |
1348 | } |
1349 | ||
cc31edce PM |
1350 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1351 | { | |
2bd59d48 | 1352 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1353 | INIT_LIST_HEAD(&cgrp->sibling); |
1354 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1355 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1356 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1357 | INIT_LIST_HEAD(&cgrp->pidlists); |
1358 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1359 | cgrp->dummy_css.cgroup = cgrp; |
cc31edce | 1360 | } |
c6d57f33 | 1361 | |
ddbcc7e8 PM |
1362 | static void init_cgroup_root(struct cgroupfs_root *root) |
1363 | { | |
bd89aabc | 1364 | struct cgroup *cgrp = &root->top_cgroup; |
b0ca5a84 | 1365 | |
ddbcc7e8 | 1366 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1367 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1368 | cgrp->root = root; |
cc31edce | 1369 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1370 | idr_init(&root->cgroup_idr); |
ddbcc7e8 PM |
1371 | } |
1372 | ||
c6d57f33 PM |
1373 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1374 | { | |
1375 | struct cgroupfs_root *root; | |
1376 | ||
a1a71b45 | 1377 | if (!opts->subsys_mask && !opts->none) |
2bd59d48 | 1378 | return ERR_PTR(-EINVAL); |
c6d57f33 PM |
1379 | |
1380 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1381 | if (!root) | |
1382 | return ERR_PTR(-ENOMEM); | |
1383 | ||
1384 | init_cgroup_root(root); | |
2c6ab6d2 | 1385 | |
c6d57f33 PM |
1386 | root->flags = opts->flags; |
1387 | if (opts->release_agent) | |
1388 | strcpy(root->release_agent_path, opts->release_agent); | |
1389 | if (opts->name) | |
1390 | strcpy(root->name, opts->name); | |
2260e7fc TH |
1391 | if (opts->cpuset_clone_children) |
1392 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1393 | return root; |
1394 | } | |
1395 | ||
35585573 | 1396 | static int cgroup_setup_root(struct cgroupfs_root *root, unsigned long ss_mask) |
d427dfeb TH |
1397 | { |
1398 | LIST_HEAD(tmp_links); | |
d427dfeb | 1399 | struct cgroup *root_cgrp = &root->top_cgroup; |
d427dfeb | 1400 | struct css_set *cset; |
d427dfeb TH |
1401 | int i, ret; |
1402 | ||
1403 | lockdep_assert_held(&cgroup_tree_mutex); | |
1404 | lockdep_assert_held(&cgroup_mutex); | |
d427dfeb TH |
1405 | |
1406 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); | |
1407 | if (ret < 0) | |
2bd59d48 | 1408 | goto out; |
d427dfeb TH |
1409 | root_cgrp->id = ret; |
1410 | ||
d427dfeb | 1411 | /* |
96d365e0 | 1412 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1413 | * but that's OK - it can only be increased by someone holding |
1414 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1415 | * have some link structures left over | |
1416 | */ | |
1417 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1418 | if (ret) | |
2bd59d48 | 1419 | goto out; |
d427dfeb TH |
1420 | |
1421 | /* ID 0 is reserved for dummy root, 1 for unified hierarchy */ | |
1422 | ret = cgroup_init_root_id(root, 2, 0); | |
1423 | if (ret) | |
2bd59d48 | 1424 | goto out; |
d427dfeb | 1425 | |
2bd59d48 TH |
1426 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1427 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1428 | root_cgrp); | |
1429 | if (IS_ERR(root->kf_root)) { | |
1430 | ret = PTR_ERR(root->kf_root); | |
1431 | goto exit_root_id; | |
1432 | } | |
1433 | root_cgrp->kn = root->kf_root->kn; | |
d427dfeb TH |
1434 | |
1435 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); | |
1436 | if (ret) | |
2bd59d48 | 1437 | goto destroy_root; |
d427dfeb | 1438 | |
35585573 | 1439 | ret = rebind_subsystems(root, ss_mask, 0); |
d427dfeb | 1440 | if (ret) |
2bd59d48 | 1441 | goto destroy_root; |
d427dfeb TH |
1442 | |
1443 | /* | |
1444 | * There must be no failure case after here, since rebinding takes | |
1445 | * care of subsystems' refcounts, which are explicitly dropped in | |
1446 | * the failure exit path. | |
1447 | */ | |
1448 | list_add(&root->root_list, &cgroup_roots); | |
1449 | cgroup_root_count++; | |
1450 | ||
1451 | /* | |
1452 | * Link the top cgroup in this hierarchy into all the css_set | |
1453 | * objects. | |
1454 | */ | |
96d365e0 | 1455 | down_write(&css_set_rwsem); |
d427dfeb TH |
1456 | hash_for_each(css_set_table, i, cset, hlist) |
1457 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1458 | up_write(&css_set_rwsem); |
d427dfeb TH |
1459 | |
1460 | BUG_ON(!list_empty(&root_cgrp->children)); | |
3c9c825b | 1461 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
d427dfeb | 1462 | |
2bd59d48 | 1463 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1464 | ret = 0; |
2bd59d48 | 1465 | goto out; |
d427dfeb | 1466 | |
2bd59d48 TH |
1467 | destroy_root: |
1468 | kernfs_destroy_root(root->kf_root); | |
1469 | root->kf_root = NULL; | |
1470 | exit_root_id: | |
d427dfeb | 1471 | cgroup_exit_root_id(root); |
2bd59d48 | 1472 | out: |
d427dfeb TH |
1473 | free_cgrp_cset_links(&tmp_links); |
1474 | return ret; | |
1475 | } | |
1476 | ||
f7e83571 | 1477 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1478 | int flags, const char *unused_dev_name, |
f7e83571 | 1479 | void *data) |
ddbcc7e8 | 1480 | { |
2bd59d48 | 1481 | struct cgroupfs_root *root; |
ddbcc7e8 | 1482 | struct cgroup_sb_opts opts; |
2bd59d48 | 1483 | struct dentry *dentry; |
8e30e2b8 | 1484 | int ret; |
56fde9e0 TH |
1485 | |
1486 | /* | |
1487 | * The first time anyone tries to mount a cgroup, enable the list | |
1488 | * linking each css_set to its tasks and fix up all existing tasks. | |
1489 | */ | |
1490 | if (!use_task_css_set_links) | |
1491 | cgroup_enable_task_cg_lists(); | |
776f02fa | 1492 | retry: |
8e30e2b8 | 1493 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1494 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1495 | |
1496 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1497 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1498 | if (ret) |
8e30e2b8 | 1499 | goto out_unlock; |
ddbcc7e8 | 1500 | |
2bd59d48 TH |
1501 | /* look for a matching existing root */ |
1502 | for_each_active_root(root) { | |
1503 | bool name_match = false; | |
ddbcc7e8 | 1504 | |
2bd59d48 TH |
1505 | /* |
1506 | * If we asked for a name then it must match. Also, if | |
1507 | * name matches but sybsys_mask doesn't, we should fail. | |
1508 | * Remember whether name matched. | |
1509 | */ | |
1510 | if (opts.name) { | |
1511 | if (strcmp(opts.name, root->name)) | |
1512 | continue; | |
1513 | name_match = true; | |
1514 | } | |
ddbcc7e8 | 1515 | |
c6d57f33 | 1516 | /* |
2bd59d48 TH |
1517 | * If we asked for subsystems (or explicitly for no |
1518 | * subsystems) then they must match. | |
c6d57f33 | 1519 | */ |
2bd59d48 TH |
1520 | if ((opts.subsys_mask || opts.none) && |
1521 | (opts.subsys_mask != root->subsys_mask)) { | |
1522 | if (!name_match) | |
1523 | continue; | |
1524 | ret = -EBUSY; | |
1525 | goto out_unlock; | |
1526 | } | |
873fe09e | 1527 | |
c7ba8287 | 1528 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1529 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1530 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1531 | ret = -EINVAL; | |
8e30e2b8 | 1532 | goto out_unlock; |
2a0ff3fb JL |
1533 | } else { |
1534 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1535 | } | |
873fe09e | 1536 | } |
2bd59d48 | 1537 | |
776f02fa TH |
1538 | /* |
1539 | * A root's lifetime is governed by its top cgroup. Zero | |
1540 | * ref indicate that the root is being destroyed. Wait for | |
1541 | * destruction to complete so that the subsystems are free. | |
1542 | * We can use wait_queue for the wait but this path is | |
1543 | * super cold. Let's just sleep for a bit and retry. | |
1544 | */ | |
1545 | if (!atomic_inc_not_zero(&root->top_cgroup.refcnt)) { | |
1546 | mutex_unlock(&cgroup_mutex); | |
1547 | mutex_unlock(&cgroup_tree_mutex); | |
1548 | msleep(10); | |
1549 | goto retry; | |
1550 | } | |
1551 | ||
1552 | ret = 0; | |
2bd59d48 | 1553 | goto out_unlock; |
ddbcc7e8 PM |
1554 | } |
1555 | ||
2bd59d48 TH |
1556 | /* no such thing, create a new one */ |
1557 | root = cgroup_root_from_opts(&opts); | |
1558 | if (IS_ERR(root)) { | |
1559 | ret = PTR_ERR(root); | |
1560 | goto out_unlock; | |
1561 | } | |
1562 | ||
35585573 | 1563 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1564 | if (ret) |
1565 | cgroup_free_root(root); | |
1566 | ||
8e30e2b8 | 1567 | out_unlock: |
e25e2cbb | 1568 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1569 | mutex_unlock(&cgroup_tree_mutex); |
8e30e2b8 | 1570 | |
c6d57f33 PM |
1571 | kfree(opts.release_agent); |
1572 | kfree(opts.name); | |
8e30e2b8 | 1573 | |
2bd59d48 | 1574 | if (ret) |
8e30e2b8 | 1575 | return ERR_PTR(ret); |
2bd59d48 TH |
1576 | |
1577 | dentry = kernfs_mount(fs_type, flags, root->kf_root); | |
1578 | if (IS_ERR(dentry)) | |
776f02fa | 1579 | cgroup_put(&root->top_cgroup); |
2bd59d48 TH |
1580 | return dentry; |
1581 | } | |
1582 | ||
1583 | static void cgroup_kill_sb(struct super_block *sb) | |
1584 | { | |
1585 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
1586 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); | |
1587 | ||
776f02fa | 1588 | cgroup_put(&root->top_cgroup); |
2bd59d48 | 1589 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1590 | } |
1591 | ||
ddbcc7e8 PM |
1592 | static struct file_system_type cgroup_fs_type = { |
1593 | .name = "cgroup", | |
f7e83571 | 1594 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1595 | .kill_sb = cgroup_kill_sb, |
1596 | }; | |
1597 | ||
676db4af GK |
1598 | static struct kobject *cgroup_kobj; |
1599 | ||
857a2beb | 1600 | /** |
913ffdb5 | 1601 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1602 | * @task: target task |
857a2beb TH |
1603 | * @buf: the buffer to write the path into |
1604 | * @buflen: the length of the buffer | |
1605 | * | |
913ffdb5 TH |
1606 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1607 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1608 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1609 | * cgroup controller callbacks. | |
1610 | * | |
e61734c5 | 1611 | * Return value is the same as kernfs_path(). |
857a2beb | 1612 | */ |
e61734c5 | 1613 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb TH |
1614 | { |
1615 | struct cgroupfs_root *root; | |
913ffdb5 | 1616 | struct cgroup *cgrp; |
e61734c5 TH |
1617 | int hierarchy_id = 1; |
1618 | char *path = NULL; | |
857a2beb TH |
1619 | |
1620 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1621 | down_read(&css_set_rwsem); |
857a2beb | 1622 | |
913ffdb5 TH |
1623 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1624 | ||
857a2beb TH |
1625 | if (root) { |
1626 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1627 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1628 | } else { |
1629 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1630 | if (strlcpy(buf, "/", buflen) < buflen) |
1631 | path = buf; | |
857a2beb TH |
1632 | } |
1633 | ||
96d365e0 | 1634 | up_read(&css_set_rwsem); |
857a2beb | 1635 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1636 | return path; |
857a2beb | 1637 | } |
913ffdb5 | 1638 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1639 | |
2f7ee569 TH |
1640 | /* |
1641 | * Control Group taskset | |
1642 | */ | |
134d3373 TH |
1643 | struct task_and_cgroup { |
1644 | struct task_struct *task; | |
1645 | struct cgroup *cgrp; | |
6f4b7e63 | 1646 | struct css_set *cset; |
134d3373 TH |
1647 | }; |
1648 | ||
2f7ee569 TH |
1649 | struct cgroup_taskset { |
1650 | struct task_and_cgroup single; | |
1651 | struct flex_array *tc_array; | |
1652 | int tc_array_len; | |
1653 | int idx; | |
1654 | struct cgroup *cur_cgrp; | |
1655 | }; | |
1656 | ||
1657 | /** | |
1658 | * cgroup_taskset_first - reset taskset and return the first task | |
1659 | * @tset: taskset of interest | |
1660 | * | |
1661 | * @tset iteration is initialized and the first task is returned. | |
1662 | */ | |
1663 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1664 | { | |
1665 | if (tset->tc_array) { | |
1666 | tset->idx = 0; | |
1667 | return cgroup_taskset_next(tset); | |
1668 | } else { | |
1669 | tset->cur_cgrp = tset->single.cgrp; | |
1670 | return tset->single.task; | |
1671 | } | |
1672 | } | |
1673 | EXPORT_SYMBOL_GPL(cgroup_taskset_first); | |
1674 | ||
1675 | /** | |
1676 | * cgroup_taskset_next - iterate to the next task in taskset | |
1677 | * @tset: taskset of interest | |
1678 | * | |
1679 | * Return the next task in @tset. Iteration must have been initialized | |
1680 | * with cgroup_taskset_first(). | |
1681 | */ | |
1682 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1683 | { | |
1684 | struct task_and_cgroup *tc; | |
1685 | ||
1686 | if (!tset->tc_array || tset->idx >= tset->tc_array_len) | |
1687 | return NULL; | |
1688 | ||
1689 | tc = flex_array_get(tset->tc_array, tset->idx++); | |
1690 | tset->cur_cgrp = tc->cgrp; | |
1691 | return tc->task; | |
1692 | } | |
1693 | EXPORT_SYMBOL_GPL(cgroup_taskset_next); | |
1694 | ||
1695 | /** | |
d99c8727 | 1696 | * cgroup_taskset_cur_css - return the matching css for the current task |
2f7ee569 | 1697 | * @tset: taskset of interest |
d99c8727 | 1698 | * @subsys_id: the ID of the target subsystem |
2f7ee569 | 1699 | * |
d99c8727 TH |
1700 | * Return the css for the current (last returned) task of @tset for |
1701 | * subsystem specified by @subsys_id. This function must be preceded by | |
1702 | * either cgroup_taskset_first() or cgroup_taskset_next(). | |
2f7ee569 | 1703 | */ |
d99c8727 TH |
1704 | struct cgroup_subsys_state *cgroup_taskset_cur_css(struct cgroup_taskset *tset, |
1705 | int subsys_id) | |
2f7ee569 | 1706 | { |
ca8bdcaf | 1707 | return cgroup_css(tset->cur_cgrp, cgroup_subsys[subsys_id]); |
2f7ee569 | 1708 | } |
d99c8727 | 1709 | EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css); |
2f7ee569 TH |
1710 | |
1711 | /** | |
1712 | * cgroup_taskset_size - return the number of tasks in taskset | |
1713 | * @tset: taskset of interest | |
1714 | */ | |
1715 | int cgroup_taskset_size(struct cgroup_taskset *tset) | |
1716 | { | |
1717 | return tset->tc_array ? tset->tc_array_len : 1; | |
1718 | } | |
1719 | EXPORT_SYMBOL_GPL(cgroup_taskset_size); | |
1720 | ||
1721 | ||
74a1166d BB |
1722 | /* |
1723 | * cgroup_task_migrate - move a task from one cgroup to another. | |
1724 | * | |
d0b2fdd2 | 1725 | * Must be called with cgroup_mutex and threadgroup locked. |
74a1166d | 1726 | */ |
5abb8855 TH |
1727 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1728 | struct task_struct *tsk, | |
1729 | struct css_set *new_cset) | |
74a1166d | 1730 | { |
5abb8855 | 1731 | struct css_set *old_cset; |
74a1166d BB |
1732 | |
1733 | /* | |
026085ef MSB |
1734 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1735 | * setting such that we can't race against cgroup_exit() changing the | |
1736 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1737 | */ |
c84cdf75 | 1738 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1739 | old_cset = task_css_set(tsk); |
74a1166d | 1740 | |
74a1166d | 1741 | task_lock(tsk); |
5abb8855 | 1742 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d BB |
1743 | task_unlock(tsk); |
1744 | ||
96d365e0 | 1745 | down_write(&css_set_rwsem); |
56fde9e0 | 1746 | list_move(&tsk->cg_list, &new_cset->tasks); |
96d365e0 | 1747 | up_write(&css_set_rwsem); |
74a1166d BB |
1748 | |
1749 | /* | |
5abb8855 TH |
1750 | * We just gained a reference on old_cset by taking it from the |
1751 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1752 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1753 | */ |
5abb8855 TH |
1754 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
1755 | put_css_set(old_cset); | |
74a1166d BB |
1756 | } |
1757 | ||
a043e3b2 | 1758 | /** |
081aa458 | 1759 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup |
74a1166d | 1760 | * @cgrp: the cgroup to attach to |
081aa458 LZ |
1761 | * @tsk: the task or the leader of the threadgroup to be attached |
1762 | * @threadgroup: attach the whole threadgroup? | |
74a1166d | 1763 | * |
257058ae | 1764 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
081aa458 | 1765 | * task_lock of @tsk or each thread in the threadgroup individually in turn. |
74a1166d | 1766 | */ |
47cfcd09 TH |
1767 | static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk, |
1768 | bool threadgroup) | |
74a1166d BB |
1769 | { |
1770 | int retval, i, group_size; | |
74a1166d | 1771 | struct cgroupfs_root *root = cgrp->root; |
1c6727af | 1772 | struct cgroup_subsys_state *css, *failed_css = NULL; |
74a1166d | 1773 | /* threadgroup list cursor and array */ |
081aa458 | 1774 | struct task_struct *leader = tsk; |
134d3373 | 1775 | struct task_and_cgroup *tc; |
d846687d | 1776 | struct flex_array *group; |
2f7ee569 | 1777 | struct cgroup_taskset tset = { }; |
74a1166d BB |
1778 | |
1779 | /* | |
1780 | * step 0: in order to do expensive, possibly blocking operations for | |
1781 | * every thread, we cannot iterate the thread group list, since it needs | |
1782 | * rcu or tasklist locked. instead, build an array of all threads in the | |
257058ae TH |
1783 | * group - group_rwsem prevents new threads from appearing, and if |
1784 | * threads exit, this will just be an over-estimate. | |
74a1166d | 1785 | */ |
081aa458 LZ |
1786 | if (threadgroup) |
1787 | group_size = get_nr_threads(tsk); | |
1788 | else | |
1789 | group_size = 1; | |
d846687d | 1790 | /* flex_array supports very large thread-groups better than kmalloc. */ |
134d3373 | 1791 | group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL); |
74a1166d BB |
1792 | if (!group) |
1793 | return -ENOMEM; | |
d846687d | 1794 | /* pre-allocate to guarantee space while iterating in rcu read-side. */ |
3ac1707a | 1795 | retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL); |
d846687d BB |
1796 | if (retval) |
1797 | goto out_free_group_list; | |
74a1166d | 1798 | |
74a1166d | 1799 | i = 0; |
fb5d2b4c MSB |
1800 | /* |
1801 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1802 | * already PF_EXITING could be freed from underneath us unless we | |
1803 | * take an rcu_read_lock. | |
1804 | */ | |
96d365e0 | 1805 | down_read(&css_set_rwsem); |
fb5d2b4c | 1806 | rcu_read_lock(); |
74a1166d | 1807 | do { |
134d3373 TH |
1808 | struct task_and_cgroup ent; |
1809 | ||
cd3d0952 TH |
1810 | /* @tsk either already exited or can't exit until the end */ |
1811 | if (tsk->flags & PF_EXITING) | |
ea84753c | 1812 | goto next; |
cd3d0952 | 1813 | |
74a1166d BB |
1814 | /* as per above, nr_threads may decrease, but not increase. */ |
1815 | BUG_ON(i >= group_size); | |
134d3373 TH |
1816 | ent.task = tsk; |
1817 | ent.cgrp = task_cgroup_from_root(tsk, root); | |
892a2b90 MSB |
1818 | /* nothing to do if this task is already in the cgroup */ |
1819 | if (ent.cgrp == cgrp) | |
ea84753c | 1820 | goto next; |
61d1d219 MSB |
1821 | /* |
1822 | * saying GFP_ATOMIC has no effect here because we did prealloc | |
1823 | * earlier, but it's good form to communicate our expectations. | |
1824 | */ | |
134d3373 | 1825 | retval = flex_array_put(group, i, &ent, GFP_ATOMIC); |
d846687d | 1826 | BUG_ON(retval != 0); |
74a1166d | 1827 | i++; |
ea84753c | 1828 | next: |
081aa458 LZ |
1829 | if (!threadgroup) |
1830 | break; | |
74a1166d | 1831 | } while_each_thread(leader, tsk); |
fb5d2b4c | 1832 | rcu_read_unlock(); |
96d365e0 | 1833 | up_read(&css_set_rwsem); |
74a1166d BB |
1834 | /* remember the number of threads in the array for later. */ |
1835 | group_size = i; | |
2f7ee569 TH |
1836 | tset.tc_array = group; |
1837 | tset.tc_array_len = group_size; | |
74a1166d | 1838 | |
134d3373 TH |
1839 | /* methods shouldn't be called if no task is actually migrating */ |
1840 | retval = 0; | |
892a2b90 | 1841 | if (!group_size) |
b07ef774 | 1842 | goto out_free_group_list; |
134d3373 | 1843 | |
74a1166d BB |
1844 | /* |
1845 | * step 1: check that we can legitimately attach to the cgroup. | |
1846 | */ | |
1c6727af TH |
1847 | for_each_css(css, i, cgrp) { |
1848 | if (css->ss->can_attach) { | |
1849 | retval = css->ss->can_attach(css, &tset); | |
74a1166d | 1850 | if (retval) { |
1c6727af | 1851 | failed_css = css; |
74a1166d BB |
1852 | goto out_cancel_attach; |
1853 | } | |
1854 | } | |
74a1166d BB |
1855 | } |
1856 | ||
1857 | /* | |
1858 | * step 2: make sure css_sets exist for all threads to be migrated. | |
1859 | * we use find_css_set, which allocates a new one if necessary. | |
1860 | */ | |
74a1166d | 1861 | for (i = 0; i < group_size; i++) { |
a8ad805c TH |
1862 | struct css_set *old_cset; |
1863 | ||
134d3373 | 1864 | tc = flex_array_get(group, i); |
a8ad805c | 1865 | old_cset = task_css_set(tc->task); |
6f4b7e63 LZ |
1866 | tc->cset = find_css_set(old_cset, cgrp); |
1867 | if (!tc->cset) { | |
61d1d219 MSB |
1868 | retval = -ENOMEM; |
1869 | goto out_put_css_set_refs; | |
74a1166d BB |
1870 | } |
1871 | } | |
1872 | ||
1873 | /* | |
494c167c TH |
1874 | * step 3: now that we're guaranteed success wrt the css_sets, |
1875 | * proceed to move all tasks to the new cgroup. There are no | |
1876 | * failure cases after here, so this is the commit point. | |
74a1166d | 1877 | */ |
74a1166d | 1878 | for (i = 0; i < group_size; i++) { |
134d3373 | 1879 | tc = flex_array_get(group, i); |
6f4b7e63 | 1880 | cgroup_task_migrate(tc->cgrp, tc->task, tc->cset); |
74a1166d BB |
1881 | } |
1882 | /* nothing is sensitive to fork() after this point. */ | |
1883 | ||
1884 | /* | |
494c167c | 1885 | * step 4: do subsystem attach callbacks. |
74a1166d | 1886 | */ |
1c6727af TH |
1887 | for_each_css(css, i, cgrp) |
1888 | if (css->ss->attach) | |
1889 | css->ss->attach(css, &tset); | |
74a1166d BB |
1890 | |
1891 | /* | |
1892 | * step 5: success! and cleanup | |
1893 | */ | |
74a1166d | 1894 | retval = 0; |
61d1d219 MSB |
1895 | out_put_css_set_refs: |
1896 | if (retval) { | |
1897 | for (i = 0; i < group_size; i++) { | |
1898 | tc = flex_array_get(group, i); | |
6f4b7e63 | 1899 | if (!tc->cset) |
61d1d219 | 1900 | break; |
6f4b7e63 | 1901 | put_css_set(tc->cset); |
61d1d219 | 1902 | } |
74a1166d BB |
1903 | } |
1904 | out_cancel_attach: | |
74a1166d | 1905 | if (retval) { |
1c6727af TH |
1906 | for_each_css(css, i, cgrp) { |
1907 | if (css == failed_css) | |
74a1166d | 1908 | break; |
1c6727af TH |
1909 | if (css->ss->cancel_attach) |
1910 | css->ss->cancel_attach(css, &tset); | |
74a1166d BB |
1911 | } |
1912 | } | |
74a1166d | 1913 | out_free_group_list: |
d846687d | 1914 | flex_array_free(group); |
74a1166d BB |
1915 | return retval; |
1916 | } | |
1917 | ||
1918 | /* | |
1919 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
1920 | * function to attach either it or all tasks in its threadgroup. Will lock |
1921 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 1922 | */ |
74a1166d | 1923 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 1924 | { |
bbcb81d0 | 1925 | struct task_struct *tsk; |
c69e8d9c | 1926 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
1927 | int ret; |
1928 | ||
74a1166d BB |
1929 | if (!cgroup_lock_live_group(cgrp)) |
1930 | return -ENODEV; | |
1931 | ||
b78949eb MSB |
1932 | retry_find_task: |
1933 | rcu_read_lock(); | |
bbcb81d0 | 1934 | if (pid) { |
73507f33 | 1935 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
1936 | if (!tsk) { |
1937 | rcu_read_unlock(); | |
dd4b0a46 | 1938 | ret = -ESRCH; |
b78949eb | 1939 | goto out_unlock_cgroup; |
bbcb81d0 | 1940 | } |
74a1166d BB |
1941 | /* |
1942 | * even if we're attaching all tasks in the thread group, we | |
1943 | * only need to check permissions on one of them. | |
1944 | */ | |
c69e8d9c | 1945 | tcred = __task_cred(tsk); |
14a590c3 EB |
1946 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
1947 | !uid_eq(cred->euid, tcred->uid) && | |
1948 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 1949 | rcu_read_unlock(); |
b78949eb MSB |
1950 | ret = -EACCES; |
1951 | goto out_unlock_cgroup; | |
bbcb81d0 | 1952 | } |
b78949eb MSB |
1953 | } else |
1954 | tsk = current; | |
cd3d0952 TH |
1955 | |
1956 | if (threadgroup) | |
b78949eb | 1957 | tsk = tsk->group_leader; |
c4c27fbd MG |
1958 | |
1959 | /* | |
14a40ffc | 1960 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
1961 | * trapped in a cpuset, or RT worker may be born in a cgroup |
1962 | * with no rt_runtime allocated. Just say no. | |
1963 | */ | |
14a40ffc | 1964 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
1965 | ret = -EINVAL; |
1966 | rcu_read_unlock(); | |
1967 | goto out_unlock_cgroup; | |
1968 | } | |
1969 | ||
b78949eb MSB |
1970 | get_task_struct(tsk); |
1971 | rcu_read_unlock(); | |
1972 | ||
1973 | threadgroup_lock(tsk); | |
1974 | if (threadgroup) { | |
1975 | if (!thread_group_leader(tsk)) { | |
1976 | /* | |
1977 | * a race with de_thread from another thread's exec() | |
1978 | * may strip us of our leadership, if this happens, | |
1979 | * there is no choice but to throw this task away and | |
1980 | * try again; this is | |
1981 | * "double-double-toil-and-trouble-check locking". | |
1982 | */ | |
1983 | threadgroup_unlock(tsk); | |
1984 | put_task_struct(tsk); | |
1985 | goto retry_find_task; | |
1986 | } | |
081aa458 LZ |
1987 | } |
1988 | ||
1989 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
1990 | ||
cd3d0952 TH |
1991 | threadgroup_unlock(tsk); |
1992 | ||
bbcb81d0 | 1993 | put_task_struct(tsk); |
b78949eb | 1994 | out_unlock_cgroup: |
47cfcd09 | 1995 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
1996 | return ret; |
1997 | } | |
1998 | ||
7ae1bad9 TH |
1999 | /** |
2000 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2001 | * @from: attach to all cgroups of a given task | |
2002 | * @tsk: the task to be attached | |
2003 | */ | |
2004 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2005 | { | |
2006 | struct cgroupfs_root *root; | |
2007 | int retval = 0; | |
2008 | ||
47cfcd09 | 2009 | mutex_lock(&cgroup_mutex); |
7ae1bad9 | 2010 | for_each_active_root(root) { |
96d365e0 TH |
2011 | struct cgroup *from_cgrp; |
2012 | ||
2013 | down_read(&css_set_rwsem); | |
2014 | from_cgrp = task_cgroup_from_root(from, root); | |
2015 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2016 | |
6f4b7e63 | 2017 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2018 | if (retval) |
2019 | break; | |
2020 | } | |
47cfcd09 | 2021 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2022 | |
2023 | return retval; | |
2024 | } | |
2025 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2026 | ||
182446d0 TH |
2027 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2028 | struct cftype *cft, u64 pid) | |
74a1166d | 2029 | { |
182446d0 | 2030 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2031 | } |
2032 | ||
182446d0 TH |
2033 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2034 | struct cftype *cft, u64 tgid) | |
af351026 | 2035 | { |
182446d0 | 2036 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2037 | } |
2038 | ||
182446d0 TH |
2039 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
2040 | struct cftype *cft, const char *buffer) | |
e788e066 | 2041 | { |
5f469907 TH |
2042 | struct cgroupfs_root *root = css->cgroup->root; |
2043 | ||
2044 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
182446d0 | 2045 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2046 | return -ENODEV; |
69e943b7 | 2047 | spin_lock(&release_agent_path_lock); |
5f469907 TH |
2048 | strlcpy(root->release_agent_path, buffer, |
2049 | sizeof(root->release_agent_path)); | |
69e943b7 | 2050 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2051 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2052 | return 0; |
2053 | } | |
2054 | ||
2da8ca82 | 2055 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2056 | { |
2da8ca82 | 2057 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2058 | |
e788e066 PM |
2059 | if (!cgroup_lock_live_group(cgrp)) |
2060 | return -ENODEV; | |
2061 | seq_puts(seq, cgrp->root->release_agent_path); | |
2062 | seq_putc(seq, '\n'); | |
47cfcd09 | 2063 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2064 | return 0; |
2065 | } | |
2066 | ||
2da8ca82 | 2067 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2068 | { |
2da8ca82 TH |
2069 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2070 | ||
2071 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2072 | return 0; |
2073 | } | |
2074 | ||
2bd59d48 TH |
2075 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2076 | size_t nbytes, loff_t off) | |
355e0c48 | 2077 | { |
2bd59d48 TH |
2078 | struct cgroup *cgrp = of->kn->parent->priv; |
2079 | struct cftype *cft = of->kn->priv; | |
2080 | struct cgroup_subsys_state *css; | |
a742c59d | 2081 | int ret; |
355e0c48 | 2082 | |
2bd59d48 TH |
2083 | /* |
2084 | * kernfs guarantees that a file isn't deleted with operations in | |
2085 | * flight, which means that the matching css is and stays alive and | |
2086 | * doesn't need to be pinned. The RCU locking is not necessary | |
2087 | * either. It's just for the convenience of using cgroup_css(). | |
2088 | */ | |
2089 | rcu_read_lock(); | |
2090 | css = cgroup_css(cgrp, cft->ss); | |
2091 | rcu_read_unlock(); | |
a742c59d TH |
2092 | |
2093 | if (cft->write_string) { | |
2094 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2095 | } else if (cft->write_u64) { | |
2096 | unsigned long long v; | |
2097 | ret = kstrtoull(buf, 0, &v); | |
2098 | if (!ret) | |
2099 | ret = cft->write_u64(css, cft, v); | |
2100 | } else if (cft->write_s64) { | |
2101 | long long v; | |
2102 | ret = kstrtoll(buf, 0, &v); | |
2103 | if (!ret) | |
2104 | ret = cft->write_s64(css, cft, v); | |
2105 | } else if (cft->trigger) { | |
2106 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2107 | } else { |
a742c59d | 2108 | ret = -EINVAL; |
e73d2c61 | 2109 | } |
2bd59d48 | 2110 | |
a742c59d | 2111 | return ret ?: nbytes; |
355e0c48 PM |
2112 | } |
2113 | ||
6612f05b | 2114 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2115 | { |
2bd59d48 | 2116 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2117 | } |
2118 | ||
6612f05b | 2119 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2120 | { |
2bd59d48 | 2121 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2122 | } |
2123 | ||
6612f05b | 2124 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2125 | { |
2bd59d48 | 2126 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2127 | } |
2128 | ||
91796569 | 2129 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2130 | { |
7da11279 TH |
2131 | struct cftype *cft = seq_cft(m); |
2132 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2133 | |
2da8ca82 TH |
2134 | if (cft->seq_show) |
2135 | return cft->seq_show(m, arg); | |
e73d2c61 | 2136 | |
f4c753b7 | 2137 | if (cft->read_u64) |
896f5199 TH |
2138 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2139 | else if (cft->read_s64) | |
2140 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2141 | else | |
2142 | return -EINVAL; | |
2143 | return 0; | |
91796569 PM |
2144 | } |
2145 | ||
2bd59d48 TH |
2146 | static struct kernfs_ops cgroup_kf_single_ops = { |
2147 | .atomic_write_len = PAGE_SIZE, | |
2148 | .write = cgroup_file_write, | |
2149 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2150 | }; |
2151 | ||
2bd59d48 TH |
2152 | static struct kernfs_ops cgroup_kf_ops = { |
2153 | .atomic_write_len = PAGE_SIZE, | |
2154 | .write = cgroup_file_write, | |
2155 | .seq_start = cgroup_seqfile_start, | |
2156 | .seq_next = cgroup_seqfile_next, | |
2157 | .seq_stop = cgroup_seqfile_stop, | |
2158 | .seq_show = cgroup_seqfile_show, | |
2159 | }; | |
ddbcc7e8 PM |
2160 | |
2161 | /* | |
2162 | * cgroup_rename - Only allow simple rename of directories in place. | |
2163 | */ | |
2bd59d48 TH |
2164 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2165 | const char *new_name_str) | |
ddbcc7e8 | 2166 | { |
2bd59d48 | 2167 | struct cgroup *cgrp = kn->priv; |
2bd59d48 | 2168 | int ret; |
65dff759 | 2169 | |
2bd59d48 | 2170 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2171 | return -ENOTDIR; |
2bd59d48 | 2172 | if (kn->parent != new_parent) |
ddbcc7e8 | 2173 | return -EIO; |
65dff759 | 2174 | |
6db8e85c TH |
2175 | /* |
2176 | * This isn't a proper migration and its usefulness is very | |
2177 | * limited. Disallow if sane_behavior. | |
2178 | */ | |
2179 | if (cgroup_sane_behavior(cgrp)) | |
2180 | return -EPERM; | |
2181 | ||
2bd59d48 TH |
2182 | mutex_lock(&cgroup_tree_mutex); |
2183 | mutex_lock(&cgroup_mutex); | |
2184 | ||
2185 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
65dff759 | 2186 | |
2bd59d48 TH |
2187 | mutex_unlock(&cgroup_mutex); |
2188 | mutex_unlock(&cgroup_tree_mutex); | |
2bd59d48 | 2189 | return ret; |
ddbcc7e8 PM |
2190 | } |
2191 | ||
2bb566cb | 2192 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2193 | { |
8d7e6fb0 | 2194 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2195 | struct kernfs_node *kn; |
2196 | struct lock_class_key *key = NULL; | |
05ef1d7c | 2197 | |
2bd59d48 TH |
2198 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2199 | key = &cft->lockdep_key; | |
2200 | #endif | |
2201 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2202 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2203 | NULL, false, key); | |
2204 | if (IS_ERR(kn)) | |
2205 | return PTR_ERR(kn); | |
2206 | return 0; | |
ddbcc7e8 PM |
2207 | } |
2208 | ||
b1f28d31 TH |
2209 | /** |
2210 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2211 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2212 | * @cfts: array of cftypes to be added |
2213 | * @is_add: whether to add or remove | |
2214 | * | |
2215 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2216 | * For removals, this function never fails. If addition fails, this |
2217 | * function doesn't remove files already added. The caller is responsible | |
2218 | * for cleaning up. | |
b1f28d31 | 2219 | */ |
2bb566cb TH |
2220 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2221 | bool is_add) | |
ddbcc7e8 | 2222 | { |
03b1cde6 | 2223 | struct cftype *cft; |
b1f28d31 TH |
2224 | int ret; |
2225 | ||
ace2bee8 | 2226 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2227 | |
2228 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2229 | /* does cft->flags tell us to skip this file on @cgrp? */ |
873fe09e TH |
2230 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2231 | continue; | |
f33fddc2 G |
2232 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2233 | continue; | |
2234 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2235 | continue; | |
2236 | ||
2739d3cc | 2237 | if (is_add) { |
2bb566cb | 2238 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2239 | if (ret) { |
2739d3cc | 2240 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2241 | cft->name, ret); |
2242 | return ret; | |
2243 | } | |
2739d3cc LZ |
2244 | } else { |
2245 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2246 | } |
ddbcc7e8 | 2247 | } |
b1f28d31 | 2248 | return 0; |
ddbcc7e8 PM |
2249 | } |
2250 | ||
21a2d343 | 2251 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2252 | { |
2253 | LIST_HEAD(pending); | |
2bb566cb | 2254 | struct cgroup_subsys *ss = cfts[0].ss; |
492eb21b | 2255 | struct cgroup *root = &ss->root->top_cgroup; |
492eb21b | 2256 | struct cgroup_subsys_state *css; |
9ccece80 | 2257 | int ret = 0; |
8e3f6541 | 2258 | |
21a2d343 | 2259 | lockdep_assert_held(&cgroup_tree_mutex); |
4ac06017 | 2260 | |
21a2d343 TH |
2261 | /* don't bother if @ss isn't attached */ |
2262 | if (ss->root == &cgroup_dummy_root) | |
9ccece80 | 2263 | return 0; |
e8c82d20 | 2264 | |
e8c82d20 | 2265 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2266 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2267 | struct cgroup *cgrp = css->cgroup; |
2268 | ||
e8c82d20 LZ |
2269 | if (cgroup_is_dead(cgrp)) |
2270 | continue; | |
2271 | ||
21a2d343 | 2272 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2273 | if (ret) |
2274 | break; | |
8e3f6541 | 2275 | } |
21a2d343 TH |
2276 | |
2277 | if (is_add && !ret) | |
2278 | kernfs_activate(root->kn); | |
9ccece80 | 2279 | return ret; |
8e3f6541 TH |
2280 | } |
2281 | ||
2da440a2 TH |
2282 | static void cgroup_exit_cftypes(struct cftype *cfts) |
2283 | { | |
2284 | struct cftype *cft; | |
2285 | ||
2bd59d48 TH |
2286 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2287 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2288 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2289 | kfree(cft->kf_ops); | |
2290 | cft->kf_ops = NULL; | |
2da440a2 | 2291 | cft->ss = NULL; |
2bd59d48 | 2292 | } |
2da440a2 TH |
2293 | } |
2294 | ||
2bd59d48 | 2295 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2296 | { |
2297 | struct cftype *cft; | |
2298 | ||
2bd59d48 TH |
2299 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2300 | struct kernfs_ops *kf_ops; | |
2301 | ||
0adb0704 TH |
2302 | WARN_ON(cft->ss || cft->kf_ops); |
2303 | ||
2bd59d48 TH |
2304 | if (cft->seq_start) |
2305 | kf_ops = &cgroup_kf_ops; | |
2306 | else | |
2307 | kf_ops = &cgroup_kf_single_ops; | |
2308 | ||
2309 | /* | |
2310 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2311 | * make a copy of kf_ops to set its atomic_write_len. | |
2312 | */ | |
2313 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2314 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2315 | if (!kf_ops) { | |
2316 | cgroup_exit_cftypes(cfts); | |
2317 | return -ENOMEM; | |
2318 | } | |
2319 | kf_ops->atomic_write_len = cft->max_write_len; | |
2320 | } | |
2321 | ||
2322 | cft->kf_ops = kf_ops; | |
2da440a2 | 2323 | cft->ss = ss; |
2bd59d48 TH |
2324 | } |
2325 | ||
2326 | return 0; | |
2da440a2 TH |
2327 | } |
2328 | ||
21a2d343 TH |
2329 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
2330 | { | |
2331 | lockdep_assert_held(&cgroup_tree_mutex); | |
2332 | ||
2333 | if (!cfts || !cfts[0].ss) | |
2334 | return -ENOENT; | |
2335 | ||
2336 | list_del(&cfts->node); | |
2337 | cgroup_apply_cftypes(cfts, false); | |
2338 | cgroup_exit_cftypes(cfts); | |
2339 | return 0; | |
2340 | } | |
2341 | ||
80b13586 TH |
2342 | /** |
2343 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
2344 | * @cfts: zero-length name terminated array of cftypes | |
2345 | * | |
2346 | * Unregister @cfts. Files described by @cfts are removed from all | |
2347 | * existing cgroups and all future cgroups won't have them either. This | |
2348 | * function can be called anytime whether @cfts' subsys is attached or not. | |
2349 | * | |
2350 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2351 | * registered. | |
2352 | */ | |
2353 | int cgroup_rm_cftypes(struct cftype *cfts) | |
2354 | { | |
21a2d343 | 2355 | int ret; |
80b13586 | 2356 | |
21a2d343 TH |
2357 | mutex_lock(&cgroup_tree_mutex); |
2358 | ret = cgroup_rm_cftypes_locked(cfts); | |
2359 | mutex_unlock(&cgroup_tree_mutex); | |
2360 | return ret; | |
80b13586 TH |
2361 | } |
2362 | ||
8e3f6541 TH |
2363 | /** |
2364 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2365 | * @ss: target cgroup subsystem | |
2366 | * @cfts: zero-length name terminated array of cftypes | |
2367 | * | |
2368 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2369 | * existing cgroups to which @ss is attached and all future cgroups will | |
2370 | * have them too. This function can be called anytime whether @ss is | |
2371 | * attached or not. | |
2372 | * | |
2373 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2374 | * function currently returns 0 as long as @cfts registration is successful | |
2375 | * even if some file creation attempts on existing cgroups fail. | |
2376 | */ | |
03b1cde6 | 2377 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 2378 | { |
9ccece80 | 2379 | int ret; |
8e3f6541 | 2380 | |
2bd59d48 TH |
2381 | ret = cgroup_init_cftypes(ss, cfts); |
2382 | if (ret) | |
2383 | return ret; | |
2bb566cb | 2384 | |
21a2d343 TH |
2385 | mutex_lock(&cgroup_tree_mutex); |
2386 | ||
0adb0704 | 2387 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 2388 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 2389 | if (ret) |
21a2d343 TH |
2390 | cgroup_rm_cftypes_locked(cfts); |
2391 | ||
2392 | mutex_unlock(&cgroup_tree_mutex); | |
9ccece80 | 2393 | return ret; |
8e3f6541 TH |
2394 | } |
2395 | EXPORT_SYMBOL_GPL(cgroup_add_cftypes); | |
2396 | ||
a043e3b2 LZ |
2397 | /** |
2398 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2399 | * @cgrp: the cgroup in question | |
2400 | * | |
2401 | * Return the number of tasks in the cgroup. | |
2402 | */ | |
07bc356e | 2403 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2404 | { |
2405 | int count = 0; | |
69d0206c | 2406 | struct cgrp_cset_link *link; |
817929ec | 2407 | |
96d365e0 | 2408 | down_read(&css_set_rwsem); |
69d0206c TH |
2409 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2410 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 2411 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
2412 | return count; |
2413 | } | |
2414 | ||
53fa5261 | 2415 | /** |
492eb21b TH |
2416 | * css_next_child - find the next child of a given css |
2417 | * @pos_css: the current position (%NULL to initiate traversal) | |
2418 | * @parent_css: css whose children to walk | |
53fa5261 | 2419 | * |
492eb21b | 2420 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
2421 | * under either cgroup_mutex or RCU read lock. The only requirement is |
2422 | * that @parent_css and @pos_css are accessible. The next sibling is | |
2423 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 2424 | */ |
492eb21b TH |
2425 | struct cgroup_subsys_state * |
2426 | css_next_child(struct cgroup_subsys_state *pos_css, | |
2427 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 2428 | { |
492eb21b TH |
2429 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
2430 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
2431 | struct cgroup *next; |
2432 | ||
ace2bee8 | 2433 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
2434 | |
2435 | /* | |
2436 | * @pos could already have been removed. Once a cgroup is removed, | |
2437 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
2438 | * changes. As CGRP_DEAD assertion is serialized and happens |
2439 | * before the cgroup is taken off the ->sibling list, if we see it | |
2440 | * unasserted, it's guaranteed that the next sibling hasn't | |
2441 | * finished its grace period even if it's already removed, and thus | |
2442 | * safe to dereference from this RCU critical section. If | |
2443 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
2444 | * to be visible as %true here. | |
3b287a50 TH |
2445 | * |
2446 | * If @pos is dead, its next pointer can't be dereferenced; | |
2447 | * however, as each cgroup is given a monotonically increasing | |
2448 | * unique serial number and always appended to the sibling list, | |
2449 | * the next one can be found by walking the parent's children until | |
2450 | * we see a cgroup with higher serial number than @pos's. While | |
2451 | * this path can be slower, it's taken only when either the current | |
2452 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 2453 | */ |
3b287a50 TH |
2454 | if (!pos) { |
2455 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
2456 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 2457 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
2458 | } else { |
2459 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
2460 | if (next->serial_nr > pos->serial_nr) | |
2461 | break; | |
53fa5261 TH |
2462 | } |
2463 | ||
492eb21b TH |
2464 | if (&next->sibling == &cgrp->children) |
2465 | return NULL; | |
2466 | ||
ca8bdcaf | 2467 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 2468 | } |
492eb21b | 2469 | EXPORT_SYMBOL_GPL(css_next_child); |
53fa5261 | 2470 | |
574bd9f7 | 2471 | /** |
492eb21b | 2472 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 2473 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2474 | * @root: css whose descendants to walk |
574bd9f7 | 2475 | * |
492eb21b | 2476 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
2477 | * to visit for pre-order traversal of @root's descendants. @root is |
2478 | * included in the iteration and the first node to be visited. | |
75501a6d | 2479 | * |
87fb54f1 TH |
2480 | * While this function requires cgroup_mutex or RCU read locking, it |
2481 | * doesn't require the whole traversal to be contained in a single critical | |
2482 | * section. This function will return the correct next descendant as long | |
2483 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 2484 | */ |
492eb21b TH |
2485 | struct cgroup_subsys_state * |
2486 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
2487 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2488 | { |
492eb21b | 2489 | struct cgroup_subsys_state *next; |
574bd9f7 | 2490 | |
ace2bee8 | 2491 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2492 | |
bd8815a6 | 2493 | /* if first iteration, visit @root */ |
7805d000 | 2494 | if (!pos) |
bd8815a6 | 2495 | return root; |
574bd9f7 TH |
2496 | |
2497 | /* visit the first child if exists */ | |
492eb21b | 2498 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
2499 | if (next) |
2500 | return next; | |
2501 | ||
2502 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
2503 | while (pos != root) { |
2504 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 2505 | if (next) |
574bd9f7 | 2506 | return next; |
492eb21b | 2507 | pos = css_parent(pos); |
7805d000 | 2508 | } |
574bd9f7 TH |
2509 | |
2510 | return NULL; | |
2511 | } | |
492eb21b | 2512 | EXPORT_SYMBOL_GPL(css_next_descendant_pre); |
574bd9f7 | 2513 | |
12a9d2fe | 2514 | /** |
492eb21b TH |
2515 | * css_rightmost_descendant - return the rightmost descendant of a css |
2516 | * @pos: css of interest | |
12a9d2fe | 2517 | * |
492eb21b TH |
2518 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
2519 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 2520 | * subtree of @pos. |
75501a6d | 2521 | * |
87fb54f1 TH |
2522 | * While this function requires cgroup_mutex or RCU read locking, it |
2523 | * doesn't require the whole traversal to be contained in a single critical | |
2524 | * section. This function will return the correct rightmost descendant as | |
2525 | * long as @pos is accessible. | |
12a9d2fe | 2526 | */ |
492eb21b TH |
2527 | struct cgroup_subsys_state * |
2528 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 2529 | { |
492eb21b | 2530 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 2531 | |
ace2bee8 | 2532 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
2533 | |
2534 | do { | |
2535 | last = pos; | |
2536 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
2537 | pos = NULL; | |
492eb21b | 2538 | css_for_each_child(tmp, last) |
12a9d2fe TH |
2539 | pos = tmp; |
2540 | } while (pos); | |
2541 | ||
2542 | return last; | |
2543 | } | |
492eb21b | 2544 | EXPORT_SYMBOL_GPL(css_rightmost_descendant); |
12a9d2fe | 2545 | |
492eb21b TH |
2546 | static struct cgroup_subsys_state * |
2547 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 2548 | { |
492eb21b | 2549 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
2550 | |
2551 | do { | |
2552 | last = pos; | |
492eb21b | 2553 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
2554 | } while (pos); |
2555 | ||
2556 | return last; | |
2557 | } | |
2558 | ||
2559 | /** | |
492eb21b | 2560 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 2561 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2562 | * @root: css whose descendants to walk |
574bd9f7 | 2563 | * |
492eb21b | 2564 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
2565 | * to visit for post-order traversal of @root's descendants. @root is |
2566 | * included in the iteration and the last node to be visited. | |
75501a6d | 2567 | * |
87fb54f1 TH |
2568 | * While this function requires cgroup_mutex or RCU read locking, it |
2569 | * doesn't require the whole traversal to be contained in a single critical | |
2570 | * section. This function will return the correct next descendant as long | |
2571 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
2572 | * @cgroup. | |
574bd9f7 | 2573 | */ |
492eb21b TH |
2574 | struct cgroup_subsys_state * |
2575 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
2576 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2577 | { |
492eb21b | 2578 | struct cgroup_subsys_state *next; |
574bd9f7 | 2579 | |
ace2bee8 | 2580 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2581 | |
58b79a91 TH |
2582 | /* if first iteration, visit leftmost descendant which may be @root */ |
2583 | if (!pos) | |
2584 | return css_leftmost_descendant(root); | |
574bd9f7 | 2585 | |
bd8815a6 TH |
2586 | /* if we visited @root, we're done */ |
2587 | if (pos == root) | |
2588 | return NULL; | |
2589 | ||
574bd9f7 | 2590 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 2591 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 2592 | if (next) |
492eb21b | 2593 | return css_leftmost_descendant(next); |
574bd9f7 TH |
2594 | |
2595 | /* no sibling left, visit parent */ | |
bd8815a6 | 2596 | return css_parent(pos); |
574bd9f7 | 2597 | } |
492eb21b | 2598 | EXPORT_SYMBOL_GPL(css_next_descendant_post); |
574bd9f7 | 2599 | |
0942eeee | 2600 | /** |
72ec7029 | 2601 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
2602 | * @it: the iterator to advance |
2603 | * | |
2604 | * Advance @it to the next css_set to walk. | |
d515876e | 2605 | */ |
72ec7029 | 2606 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
2607 | { |
2608 | struct list_head *l = it->cset_link; | |
2609 | struct cgrp_cset_link *link; | |
2610 | struct css_set *cset; | |
2611 | ||
2612 | /* Advance to the next non-empty css_set */ | |
2613 | do { | |
2614 | l = l->next; | |
72ec7029 | 2615 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
2616 | it->cset_link = NULL; |
2617 | return; | |
2618 | } | |
2619 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
2620 | cset = link->cset; | |
2621 | } while (list_empty(&cset->tasks)); | |
2622 | it->cset_link = l; | |
2623 | it->task = cset->tasks.next; | |
2624 | } | |
2625 | ||
0942eeee | 2626 | /** |
72ec7029 TH |
2627 | * css_task_iter_start - initiate task iteration |
2628 | * @css: the css to walk tasks of | |
0942eeee TH |
2629 | * @it: the task iterator to use |
2630 | * | |
72ec7029 TH |
2631 | * Initiate iteration through the tasks of @css. The caller can call |
2632 | * css_task_iter_next() to walk through the tasks until the function | |
2633 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
2634 | * called. | |
0942eeee TH |
2635 | * |
2636 | * Note that this function acquires a lock which is released when the | |
2637 | * iteration finishes. The caller can't sleep while iteration is in | |
2638 | * progress. | |
2639 | */ | |
72ec7029 TH |
2640 | void css_task_iter_start(struct cgroup_subsys_state *css, |
2641 | struct css_task_iter *it) | |
96d365e0 | 2642 | __acquires(css_set_rwsem) |
817929ec | 2643 | { |
56fde9e0 TH |
2644 | /* no one should try to iterate before mounting cgroups */ |
2645 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 2646 | |
96d365e0 | 2647 | down_read(&css_set_rwsem); |
c59cd3d8 | 2648 | |
72ec7029 TH |
2649 | it->origin_css = css; |
2650 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 2651 | |
72ec7029 | 2652 | css_advance_task_iter(it); |
817929ec PM |
2653 | } |
2654 | ||
0942eeee | 2655 | /** |
72ec7029 | 2656 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
2657 | * @it: the task iterator being iterated |
2658 | * | |
2659 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
2660 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
2661 | * reaches the end. | |
0942eeee | 2662 | */ |
72ec7029 | 2663 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
2664 | { |
2665 | struct task_struct *res; | |
2666 | struct list_head *l = it->task; | |
69d0206c | 2667 | struct cgrp_cset_link *link; |
817929ec PM |
2668 | |
2669 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 2670 | if (!it->cset_link) |
817929ec PM |
2671 | return NULL; |
2672 | res = list_entry(l, struct task_struct, cg_list); | |
2673 | /* Advance iterator to find next entry */ | |
2674 | l = l->next; | |
69d0206c TH |
2675 | link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link); |
2676 | if (l == &link->cset->tasks) { | |
0942eeee TH |
2677 | /* |
2678 | * We reached the end of this task list - move on to the | |
2679 | * next cgrp_cset_link. | |
2680 | */ | |
72ec7029 | 2681 | css_advance_task_iter(it); |
817929ec PM |
2682 | } else { |
2683 | it->task = l; | |
2684 | } | |
2685 | return res; | |
2686 | } | |
2687 | ||
0942eeee | 2688 | /** |
72ec7029 | 2689 | * css_task_iter_end - finish task iteration |
0942eeee TH |
2690 | * @it: the task iterator to finish |
2691 | * | |
72ec7029 | 2692 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 2693 | */ |
72ec7029 | 2694 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 2695 | __releases(css_set_rwsem) |
817929ec | 2696 | { |
96d365e0 | 2697 | up_read(&css_set_rwsem); |
817929ec PM |
2698 | } |
2699 | ||
8cc99345 TH |
2700 | /** |
2701 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
2702 | * @to: cgroup to which the tasks will be moved | |
2703 | * @from: cgroup in which the tasks currently reside | |
2704 | */ | |
2705 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
2706 | { | |
e406d1cf TH |
2707 | struct css_task_iter it; |
2708 | struct task_struct *task; | |
2709 | int ret = 0; | |
2710 | ||
2711 | do { | |
2712 | css_task_iter_start(&from->dummy_css, &it); | |
2713 | task = css_task_iter_next(&it); | |
2714 | if (task) | |
2715 | get_task_struct(task); | |
2716 | css_task_iter_end(&it); | |
2717 | ||
2718 | if (task) { | |
2719 | mutex_lock(&cgroup_mutex); | |
2720 | ret = cgroup_attach_task(to, task, false); | |
2721 | mutex_unlock(&cgroup_mutex); | |
2722 | put_task_struct(task); | |
2723 | } | |
2724 | } while (task && !ret); | |
2725 | ||
2726 | return ret; | |
8cc99345 TH |
2727 | } |
2728 | ||
bbcb81d0 | 2729 | /* |
102a775e | 2730 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
2731 | * |
2732 | * Reading this file can return large amounts of data if a cgroup has | |
2733 | * *lots* of attached tasks. So it may need several calls to read(), | |
2734 | * but we cannot guarantee that the information we produce is correct | |
2735 | * unless we produce it entirely atomically. | |
2736 | * | |
bbcb81d0 | 2737 | */ |
bbcb81d0 | 2738 | |
24528255 LZ |
2739 | /* which pidlist file are we talking about? */ |
2740 | enum cgroup_filetype { | |
2741 | CGROUP_FILE_PROCS, | |
2742 | CGROUP_FILE_TASKS, | |
2743 | }; | |
2744 | ||
2745 | /* | |
2746 | * A pidlist is a list of pids that virtually represents the contents of one | |
2747 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
2748 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
2749 | * to the cgroup. | |
2750 | */ | |
2751 | struct cgroup_pidlist { | |
2752 | /* | |
2753 | * used to find which pidlist is wanted. doesn't change as long as | |
2754 | * this particular list stays in the list. | |
2755 | */ | |
2756 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
2757 | /* array of xids */ | |
2758 | pid_t *list; | |
2759 | /* how many elements the above list has */ | |
2760 | int length; | |
24528255 LZ |
2761 | /* each of these stored in a list by its cgroup */ |
2762 | struct list_head links; | |
2763 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
2764 | struct cgroup *owner; | |
b1a21367 TH |
2765 | /* for delayed destruction */ |
2766 | struct delayed_work destroy_dwork; | |
24528255 LZ |
2767 | }; |
2768 | ||
d1d9fd33 BB |
2769 | /* |
2770 | * The following two functions "fix" the issue where there are more pids | |
2771 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
2772 | * TODO: replace with a kernel-wide solution to this problem | |
2773 | */ | |
2774 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
2775 | static void *pidlist_allocate(int count) | |
2776 | { | |
2777 | if (PIDLIST_TOO_LARGE(count)) | |
2778 | return vmalloc(count * sizeof(pid_t)); | |
2779 | else | |
2780 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
2781 | } | |
b1a21367 | 2782 | |
d1d9fd33 BB |
2783 | static void pidlist_free(void *p) |
2784 | { | |
2785 | if (is_vmalloc_addr(p)) | |
2786 | vfree(p); | |
2787 | else | |
2788 | kfree(p); | |
2789 | } | |
d1d9fd33 | 2790 | |
b1a21367 TH |
2791 | /* |
2792 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
2793 | * should be left afterwards. | |
2794 | */ | |
2795 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
2796 | { | |
2797 | struct cgroup_pidlist *l, *tmp_l; | |
2798 | ||
2799 | mutex_lock(&cgrp->pidlist_mutex); | |
2800 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
2801 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
2802 | mutex_unlock(&cgrp->pidlist_mutex); | |
2803 | ||
2804 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
2805 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
2806 | } | |
2807 | ||
2808 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
2809 | { | |
2810 | struct delayed_work *dwork = to_delayed_work(work); | |
2811 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
2812 | destroy_dwork); | |
2813 | struct cgroup_pidlist *tofree = NULL; | |
2814 | ||
2815 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
2816 | |
2817 | /* | |
04502365 TH |
2818 | * Destroy iff we didn't get queued again. The state won't change |
2819 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 2820 | */ |
04502365 | 2821 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
2822 | list_del(&l->links); |
2823 | pidlist_free(l->list); | |
2824 | put_pid_ns(l->key.ns); | |
2825 | tofree = l; | |
2826 | } | |
2827 | ||
b1a21367 TH |
2828 | mutex_unlock(&l->owner->pidlist_mutex); |
2829 | kfree(tofree); | |
2830 | } | |
2831 | ||
bbcb81d0 | 2832 | /* |
102a775e | 2833 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 2834 | * Returns the number of unique elements. |
bbcb81d0 | 2835 | */ |
6ee211ad | 2836 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 2837 | { |
102a775e | 2838 | int src, dest = 1; |
102a775e BB |
2839 | |
2840 | /* | |
2841 | * we presume the 0th element is unique, so i starts at 1. trivial | |
2842 | * edge cases first; no work needs to be done for either | |
2843 | */ | |
2844 | if (length == 0 || length == 1) | |
2845 | return length; | |
2846 | /* src and dest walk down the list; dest counts unique elements */ | |
2847 | for (src = 1; src < length; src++) { | |
2848 | /* find next unique element */ | |
2849 | while (list[src] == list[src-1]) { | |
2850 | src++; | |
2851 | if (src == length) | |
2852 | goto after; | |
2853 | } | |
2854 | /* dest always points to where the next unique element goes */ | |
2855 | list[dest] = list[src]; | |
2856 | dest++; | |
2857 | } | |
2858 | after: | |
102a775e BB |
2859 | return dest; |
2860 | } | |
2861 | ||
afb2bc14 TH |
2862 | /* |
2863 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
2864 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
2865 | * different per namespace, each namespace needs differently sorted list, | |
2866 | * making it impossible to use, for example, single rbtree of member tasks | |
2867 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
2868 | * per open file is dangerous, so cgroup had to implement shared pool of | |
2869 | * pidlists keyed by cgroup and namespace. | |
2870 | * | |
2871 | * All this extra complexity was caused by the original implementation | |
2872 | * committing to an entirely unnecessary property. In the long term, we | |
2873 | * want to do away with it. Explicitly scramble sort order if | |
2874 | * sane_behavior so that no such expectation exists in the new interface. | |
2875 | * | |
2876 | * Scrambling is done by swapping every two consecutive bits, which is | |
2877 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
2878 | */ | |
2879 | static pid_t pid_fry(pid_t pid) | |
2880 | { | |
2881 | unsigned a = pid & 0x55555555; | |
2882 | unsigned b = pid & 0xAAAAAAAA; | |
2883 | ||
2884 | return (a << 1) | (b >> 1); | |
2885 | } | |
2886 | ||
2887 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
2888 | { | |
2889 | if (cgroup_sane_behavior(cgrp)) | |
2890 | return pid_fry(pid); | |
2891 | else | |
2892 | return pid; | |
2893 | } | |
2894 | ||
102a775e BB |
2895 | static int cmppid(const void *a, const void *b) |
2896 | { | |
2897 | return *(pid_t *)a - *(pid_t *)b; | |
2898 | } | |
2899 | ||
afb2bc14 TH |
2900 | static int fried_cmppid(const void *a, const void *b) |
2901 | { | |
2902 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
2903 | } | |
2904 | ||
e6b81710 TH |
2905 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
2906 | enum cgroup_filetype type) | |
2907 | { | |
2908 | struct cgroup_pidlist *l; | |
2909 | /* don't need task_nsproxy() if we're looking at ourself */ | |
2910 | struct pid_namespace *ns = task_active_pid_ns(current); | |
2911 | ||
2912 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
2913 | ||
2914 | list_for_each_entry(l, &cgrp->pidlists, links) | |
2915 | if (l->key.type == type && l->key.ns == ns) | |
2916 | return l; | |
2917 | return NULL; | |
2918 | } | |
2919 | ||
72a8cb30 BB |
2920 | /* |
2921 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
2922 | * returns with the lock on that pidlist already held, and takes care | |
2923 | * of the use count, or returns NULL with no locks held if we're out of | |
2924 | * memory. | |
2925 | */ | |
e6b81710 TH |
2926 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
2927 | enum cgroup_filetype type) | |
72a8cb30 BB |
2928 | { |
2929 | struct cgroup_pidlist *l; | |
b70cc5fd | 2930 | |
e6b81710 TH |
2931 | lockdep_assert_held(&cgrp->pidlist_mutex); |
2932 | ||
2933 | l = cgroup_pidlist_find(cgrp, type); | |
2934 | if (l) | |
2935 | return l; | |
2936 | ||
72a8cb30 | 2937 | /* entry not found; create a new one */ |
f4f4be2b | 2938 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 2939 | if (!l) |
72a8cb30 | 2940 | return l; |
e6b81710 | 2941 | |
b1a21367 | 2942 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 2943 | l->key.type = type; |
e6b81710 TH |
2944 | /* don't need task_nsproxy() if we're looking at ourself */ |
2945 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
2946 | l->owner = cgrp; |
2947 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
2948 | return l; |
2949 | } | |
2950 | ||
102a775e BB |
2951 | /* |
2952 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
2953 | */ | |
72a8cb30 BB |
2954 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
2955 | struct cgroup_pidlist **lp) | |
102a775e BB |
2956 | { |
2957 | pid_t *array; | |
2958 | int length; | |
2959 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 2960 | struct css_task_iter it; |
817929ec | 2961 | struct task_struct *tsk; |
102a775e BB |
2962 | struct cgroup_pidlist *l; |
2963 | ||
4bac00d1 TH |
2964 | lockdep_assert_held(&cgrp->pidlist_mutex); |
2965 | ||
102a775e BB |
2966 | /* |
2967 | * If cgroup gets more users after we read count, we won't have | |
2968 | * enough space - tough. This race is indistinguishable to the | |
2969 | * caller from the case that the additional cgroup users didn't | |
2970 | * show up until sometime later on. | |
2971 | */ | |
2972 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 2973 | array = pidlist_allocate(length); |
102a775e BB |
2974 | if (!array) |
2975 | return -ENOMEM; | |
2976 | /* now, populate the array */ | |
72ec7029 TH |
2977 | css_task_iter_start(&cgrp->dummy_css, &it); |
2978 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 2979 | if (unlikely(n == length)) |
817929ec | 2980 | break; |
102a775e | 2981 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
2982 | if (type == CGROUP_FILE_PROCS) |
2983 | pid = task_tgid_vnr(tsk); | |
2984 | else | |
2985 | pid = task_pid_vnr(tsk); | |
102a775e BB |
2986 | if (pid > 0) /* make sure to only use valid results */ |
2987 | array[n++] = pid; | |
817929ec | 2988 | } |
72ec7029 | 2989 | css_task_iter_end(&it); |
102a775e BB |
2990 | length = n; |
2991 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
2992 | if (cgroup_sane_behavior(cgrp)) |
2993 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
2994 | else | |
2995 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 2996 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 2997 | length = pidlist_uniq(array, length); |
e6b81710 | 2998 | |
e6b81710 | 2999 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3000 | if (!l) { |
e6b81710 | 3001 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3002 | pidlist_free(array); |
72a8cb30 | 3003 | return -ENOMEM; |
102a775e | 3004 | } |
e6b81710 TH |
3005 | |
3006 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3007 | pidlist_free(l->list); |
102a775e BB |
3008 | l->list = array; |
3009 | l->length = length; | |
72a8cb30 | 3010 | *lp = l; |
102a775e | 3011 | return 0; |
bbcb81d0 PM |
3012 | } |
3013 | ||
846c7bb0 | 3014 | /** |
a043e3b2 | 3015 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3016 | * @stats: cgroupstats to fill information into |
3017 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3018 | * been requested. | |
a043e3b2 LZ |
3019 | * |
3020 | * Build and fill cgroupstats so that taskstats can export it to user | |
3021 | * space. | |
846c7bb0 BS |
3022 | */ |
3023 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3024 | { | |
2bd59d48 | 3025 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3026 | struct cgroup *cgrp; |
72ec7029 | 3027 | struct css_task_iter it; |
846c7bb0 | 3028 | struct task_struct *tsk; |
33d283be | 3029 | |
2bd59d48 TH |
3030 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3031 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3032 | kernfs_type(kn) != KERNFS_DIR) | |
3033 | return -EINVAL; | |
3034 | ||
846c7bb0 | 3035 | /* |
2bd59d48 TH |
3036 | * We aren't being called from kernfs and there's no guarantee on |
3037 | * @kn->priv's validity. For this and css_tryget_from_dir(), | |
3038 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
846c7bb0 | 3039 | */ |
2bd59d48 TH |
3040 | rcu_read_lock(); |
3041 | cgrp = rcu_dereference(kn->priv); | |
3042 | if (!cgrp) { | |
3043 | rcu_read_unlock(); | |
3044 | return -ENOENT; | |
3045 | } | |
846c7bb0 | 3046 | |
72ec7029 TH |
3047 | css_task_iter_start(&cgrp->dummy_css, &it); |
3048 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3049 | switch (tsk->state) { |
3050 | case TASK_RUNNING: | |
3051 | stats->nr_running++; | |
3052 | break; | |
3053 | case TASK_INTERRUPTIBLE: | |
3054 | stats->nr_sleeping++; | |
3055 | break; | |
3056 | case TASK_UNINTERRUPTIBLE: | |
3057 | stats->nr_uninterruptible++; | |
3058 | break; | |
3059 | case TASK_STOPPED: | |
3060 | stats->nr_stopped++; | |
3061 | break; | |
3062 | default: | |
3063 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3064 | stats->nr_io_wait++; | |
3065 | break; | |
3066 | } | |
3067 | } | |
72ec7029 | 3068 | css_task_iter_end(&it); |
846c7bb0 | 3069 | |
2bd59d48 TH |
3070 | rcu_read_unlock(); |
3071 | return 0; | |
846c7bb0 BS |
3072 | } |
3073 | ||
8f3ff208 | 3074 | |
bbcb81d0 | 3075 | /* |
102a775e | 3076 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3077 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3078 | * in the cgroup->l->list array. |
bbcb81d0 | 3079 | */ |
cc31edce | 3080 | |
102a775e | 3081 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3082 | { |
cc31edce PM |
3083 | /* |
3084 | * Initially we receive a position value that corresponds to | |
3085 | * one more than the last pid shown (or 0 on the first call or | |
3086 | * after a seek to the start). Use a binary-search to find the | |
3087 | * next pid to display, if any | |
3088 | */ | |
2bd59d48 | 3089 | struct kernfs_open_file *of = s->private; |
7da11279 | 3090 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3091 | struct cgroup_pidlist *l; |
7da11279 | 3092 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3093 | int index = 0, pid = *pos; |
4bac00d1 TH |
3094 | int *iter, ret; |
3095 | ||
3096 | mutex_lock(&cgrp->pidlist_mutex); | |
3097 | ||
3098 | /* | |
5d22444f | 3099 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3100 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3101 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3102 | * could already have been destroyed. |
3103 | */ | |
5d22444f TH |
3104 | if (of->priv) |
3105 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3106 | |
3107 | /* | |
3108 | * Either this is the first start() after open or the matching | |
3109 | * pidlist has been destroyed inbetween. Create a new one. | |
3110 | */ | |
5d22444f TH |
3111 | if (!of->priv) { |
3112 | ret = pidlist_array_load(cgrp, type, | |
3113 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3114 | if (ret) |
3115 | return ERR_PTR(ret); | |
3116 | } | |
5d22444f | 3117 | l = of->priv; |
cc31edce | 3118 | |
cc31edce | 3119 | if (pid) { |
102a775e | 3120 | int end = l->length; |
20777766 | 3121 | |
cc31edce PM |
3122 | while (index < end) { |
3123 | int mid = (index + end) / 2; | |
afb2bc14 | 3124 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3125 | index = mid; |
3126 | break; | |
afb2bc14 | 3127 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3128 | index = mid + 1; |
3129 | else | |
3130 | end = mid; | |
3131 | } | |
3132 | } | |
3133 | /* If we're off the end of the array, we're done */ | |
102a775e | 3134 | if (index >= l->length) |
cc31edce PM |
3135 | return NULL; |
3136 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3137 | iter = l->list + index; |
afb2bc14 | 3138 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3139 | return iter; |
3140 | } | |
3141 | ||
102a775e | 3142 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3143 | { |
2bd59d48 | 3144 | struct kernfs_open_file *of = s->private; |
5d22444f | 3145 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3146 | |
5d22444f TH |
3147 | if (l) |
3148 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3149 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3150 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3151 | } |
3152 | ||
102a775e | 3153 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3154 | { |
2bd59d48 | 3155 | struct kernfs_open_file *of = s->private; |
5d22444f | 3156 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3157 | pid_t *p = v; |
3158 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3159 | /* |
3160 | * Advance to the next pid in the array. If this goes off the | |
3161 | * end, we're done | |
3162 | */ | |
3163 | p++; | |
3164 | if (p >= end) { | |
3165 | return NULL; | |
3166 | } else { | |
7da11279 | 3167 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3168 | return p; |
3169 | } | |
3170 | } | |
3171 | ||
102a775e | 3172 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3173 | { |
3174 | return seq_printf(s, "%d\n", *(int *)v); | |
3175 | } | |
bbcb81d0 | 3176 | |
102a775e BB |
3177 | /* |
3178 | * seq_operations functions for iterating on pidlists through seq_file - | |
3179 | * independent of whether it's tasks or procs | |
3180 | */ | |
3181 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3182 | .start = cgroup_pidlist_start, | |
3183 | .stop = cgroup_pidlist_stop, | |
3184 | .next = cgroup_pidlist_next, | |
3185 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3186 | }; |
3187 | ||
182446d0 TH |
3188 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3189 | struct cftype *cft) | |
81a6a5cd | 3190 | { |
182446d0 | 3191 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3192 | } |
3193 | ||
182446d0 TH |
3194 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3195 | struct cftype *cft, u64 val) | |
6379c106 | 3196 | { |
182446d0 | 3197 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3198 | if (val) |
182446d0 | 3199 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3200 | else |
182446d0 | 3201 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3202 | return 0; |
3203 | } | |
3204 | ||
182446d0 TH |
3205 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3206 | struct cftype *cft) | |
97978e6d | 3207 | { |
182446d0 | 3208 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3209 | } |
3210 | ||
182446d0 TH |
3211 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3212 | struct cftype *cft, u64 val) | |
97978e6d DL |
3213 | { |
3214 | if (val) | |
182446d0 | 3215 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3216 | else |
182446d0 | 3217 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3218 | return 0; |
3219 | } | |
3220 | ||
d5c56ced | 3221 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3222 | { |
d5c56ced | 3223 | .name = "cgroup.procs", |
6612f05b TH |
3224 | .seq_start = cgroup_pidlist_start, |
3225 | .seq_next = cgroup_pidlist_next, | |
3226 | .seq_stop = cgroup_pidlist_stop, | |
3227 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3228 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3229 | .write_u64 = cgroup_procs_write, |
74a1166d | 3230 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3231 | }, |
97978e6d DL |
3232 | { |
3233 | .name = "cgroup.clone_children", | |
873fe09e | 3234 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3235 | .read_u64 = cgroup_clone_children_read, |
3236 | .write_u64 = cgroup_clone_children_write, | |
3237 | }, | |
873fe09e TH |
3238 | { |
3239 | .name = "cgroup.sane_behavior", | |
3240 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3241 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3242 | }, |
d5c56ced TH |
3243 | |
3244 | /* | |
3245 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3246 | * don't exist if sane_behavior. If you're depending on these, be | |
3247 | * prepared to be burned. | |
3248 | */ | |
3249 | { | |
3250 | .name = "tasks", | |
3251 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3252 | .seq_start = cgroup_pidlist_start, |
3253 | .seq_next = cgroup_pidlist_next, | |
3254 | .seq_stop = cgroup_pidlist_stop, | |
3255 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3256 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3257 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3258 | .mode = S_IRUGO | S_IWUSR, |
3259 | }, | |
3260 | { | |
3261 | .name = "notify_on_release", | |
3262 | .flags = CFTYPE_INSANE, | |
3263 | .read_u64 = cgroup_read_notify_on_release, | |
3264 | .write_u64 = cgroup_write_notify_on_release, | |
3265 | }, | |
6e6ff25b TH |
3266 | { |
3267 | .name = "release_agent", | |
cc5943a7 | 3268 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3269 | .seq_show = cgroup_release_agent_show, |
6e6ff25b | 3270 | .write_string = cgroup_release_agent_write, |
5f469907 | 3271 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 3272 | }, |
db0416b6 | 3273 | { } /* terminate */ |
bbcb81d0 PM |
3274 | }; |
3275 | ||
13af07df | 3276 | /** |
628f7cd4 | 3277 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3278 | * @cgrp: target cgroup |
13af07df | 3279 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3280 | * |
3281 | * On failure, no file is added. | |
13af07df | 3282 | */ |
628f7cd4 | 3283 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3284 | { |
ddbcc7e8 | 3285 | struct cgroup_subsys *ss; |
b420ba7d | 3286 | int i, ret = 0; |
bbcb81d0 | 3287 | |
8e3f6541 | 3288 | /* process cftsets of each subsystem */ |
b420ba7d | 3289 | for_each_subsys(ss, i) { |
0adb0704 | 3290 | struct cftype *cfts; |
b420ba7d TH |
3291 | |
3292 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3293 | continue; |
8e3f6541 | 3294 | |
0adb0704 TH |
3295 | list_for_each_entry(cfts, &ss->cfts, node) { |
3296 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
3297 | if (ret < 0) |
3298 | goto err; | |
3299 | } | |
ddbcc7e8 | 3300 | } |
ddbcc7e8 | 3301 | return 0; |
bee55099 TH |
3302 | err: |
3303 | cgroup_clear_dir(cgrp, subsys_mask); | |
3304 | return ret; | |
ddbcc7e8 PM |
3305 | } |
3306 | ||
0c21ead1 TH |
3307 | /* |
3308 | * css destruction is four-stage process. | |
3309 | * | |
3310 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3311 | * Implemented in kill_css(). | |
3312 | * | |
3313 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3314 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3315 | * by invoking offline_css(). After offlining, the base ref is put. | |
3316 | * Implemented in css_killed_work_fn(). | |
3317 | * | |
3318 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3319 | * accessors are inside RCU read sections. css_release() schedules the | |
3320 | * RCU callback. | |
3321 | * | |
3322 | * 4. After the grace period, the css can be freed. Implemented in | |
3323 | * css_free_work_fn(). | |
3324 | * | |
3325 | * It is actually hairier because both step 2 and 4 require process context | |
3326 | * and thus involve punting to css->destroy_work adding two additional | |
3327 | * steps to the already complex sequence. | |
3328 | */ | |
35ef10da | 3329 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3330 | { |
3331 | struct cgroup_subsys_state *css = | |
35ef10da | 3332 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 3333 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3334 | |
0ae78e0b TH |
3335 | if (css->parent) |
3336 | css_put(css->parent); | |
3337 | ||
0c21ead1 | 3338 | css->ss->css_free(css); |
2bd59d48 | 3339 | cgroup_put(cgrp); |
48ddbe19 TH |
3340 | } |
3341 | ||
0c21ead1 | 3342 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3343 | { |
3344 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3345 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3346 | |
35ef10da | 3347 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 3348 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
3349 | } |
3350 | ||
d3daf28d TH |
3351 | static void css_release(struct percpu_ref *ref) |
3352 | { | |
3353 | struct cgroup_subsys_state *css = | |
3354 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3355 | ||
aec25020 | 3356 | rcu_assign_pointer(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 3357 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
3358 | } |
3359 | ||
623f926b TH |
3360 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
3361 | struct cgroup *cgrp) | |
ddbcc7e8 | 3362 | { |
bd89aabc | 3363 | css->cgroup = cgrp; |
72c97e54 | 3364 | css->ss = ss; |
ddbcc7e8 | 3365 | css->flags = 0; |
0ae78e0b TH |
3366 | |
3367 | if (cgrp->parent) | |
ca8bdcaf | 3368 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 3369 | else |
38b53aba | 3370 | css->flags |= CSS_ROOT; |
48ddbe19 | 3371 | |
ca8bdcaf | 3372 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
3373 | } |
3374 | ||
2a4ac633 | 3375 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 3376 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3377 | { |
623f926b | 3378 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
3379 | int ret = 0; |
3380 | ||
ace2bee8 | 3381 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3382 | lockdep_assert_held(&cgroup_mutex); |
3383 | ||
92fb9748 | 3384 | if (ss->css_online) |
eb95419b | 3385 | ret = ss->css_online(css); |
ae7f164a | 3386 | if (!ret) { |
eb95419b | 3387 | css->flags |= CSS_ONLINE; |
f20104de | 3388 | css->cgroup->nr_css++; |
aec25020 | 3389 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 3390 | } |
b1929db4 | 3391 | return ret; |
a31f2d3f TH |
3392 | } |
3393 | ||
2a4ac633 | 3394 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 3395 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3396 | { |
623f926b | 3397 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 3398 | |
ace2bee8 | 3399 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3400 | lockdep_assert_held(&cgroup_mutex); |
3401 | ||
3402 | if (!(css->flags & CSS_ONLINE)) | |
3403 | return; | |
3404 | ||
d7eeac19 | 3405 | if (ss->css_offline) |
eb95419b | 3406 | ss->css_offline(css); |
a31f2d3f | 3407 | |
eb95419b | 3408 | css->flags &= ~CSS_ONLINE; |
09a503ea | 3409 | css->cgroup->nr_css--; |
aec25020 | 3410 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); |
a31f2d3f TH |
3411 | } |
3412 | ||
c81c925a TH |
3413 | /** |
3414 | * create_css - create a cgroup_subsys_state | |
3415 | * @cgrp: the cgroup new css will be associated with | |
3416 | * @ss: the subsys of new css | |
3417 | * | |
3418 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
3419 | * css is online and installed in @cgrp with all interface files created. | |
3420 | * Returns 0 on success, -errno on failure. | |
3421 | */ | |
3422 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
3423 | { | |
3424 | struct cgroup *parent = cgrp->parent; | |
3425 | struct cgroup_subsys_state *css; | |
3426 | int err; | |
3427 | ||
c81c925a TH |
3428 | lockdep_assert_held(&cgroup_mutex); |
3429 | ||
3430 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
3431 | if (IS_ERR(css)) | |
3432 | return PTR_ERR(css); | |
3433 | ||
3434 | err = percpu_ref_init(&css->refcnt, css_release); | |
3435 | if (err) | |
3436 | goto err_free; | |
3437 | ||
3438 | init_css(css, ss, cgrp); | |
3439 | ||
aec25020 | 3440 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a TH |
3441 | if (err) |
3442 | goto err_free; | |
3443 | ||
3444 | err = online_css(css); | |
3445 | if (err) | |
3446 | goto err_free; | |
3447 | ||
59f5296b | 3448 | cgroup_get(cgrp); |
c81c925a TH |
3449 | css_get(css->parent); |
3450 | ||
3451 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
3452 | parent->parent) { | |
3453 | 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", | |
3454 | current->comm, current->pid, ss->name); | |
3455 | if (!strcmp(ss->name, "memory")) | |
3456 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
3457 | ss->warned_broken_hierarchy = true; | |
3458 | } | |
3459 | ||
3460 | return 0; | |
3461 | ||
3462 | err_free: | |
3463 | percpu_ref_cancel_init(&css->refcnt); | |
3464 | ss->css_free(css); | |
3465 | return err; | |
3466 | } | |
3467 | ||
2bd59d48 | 3468 | /** |
a043e3b2 LZ |
3469 | * cgroup_create - create a cgroup |
3470 | * @parent: cgroup that will be parent of the new cgroup | |
e61734c5 | 3471 | * @name: name of the new cgroup |
2bd59d48 | 3472 | * @mode: mode to set on new cgroup |
ddbcc7e8 | 3473 | */ |
e61734c5 | 3474 | static long cgroup_create(struct cgroup *parent, const char *name, |
2bd59d48 | 3475 | umode_t mode) |
ddbcc7e8 | 3476 | { |
bd89aabc | 3477 | struct cgroup *cgrp; |
ddbcc7e8 | 3478 | struct cgroupfs_root *root = parent->root; |
b58c8998 | 3479 | int ssid, err; |
ddbcc7e8 | 3480 | struct cgroup_subsys *ss; |
2bd59d48 | 3481 | struct kernfs_node *kn; |
ddbcc7e8 | 3482 | |
0a950f65 | 3483 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
3484 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
3485 | if (!cgrp) | |
ddbcc7e8 PM |
3486 | return -ENOMEM; |
3487 | ||
ace2bee8 TH |
3488 | mutex_lock(&cgroup_tree_mutex); |
3489 | ||
976c06bc TH |
3490 | /* |
3491 | * Only live parents can have children. Note that the liveliness | |
3492 | * check isn't strictly necessary because cgroup_mkdir() and | |
3493 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
3494 | * anyway so that locking is contained inside cgroup proper and we | |
3495 | * don't get nasty surprises if we ever grow another caller. | |
3496 | */ | |
3497 | if (!cgroup_lock_live_group(parent)) { | |
3498 | err = -ENODEV; | |
ace2bee8 | 3499 | goto err_unlock_tree; |
0ab02ca8 LZ |
3500 | } |
3501 | ||
3502 | /* | |
3503 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
3504 | * a half-baked cgroup. | |
3505 | */ | |
3506 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
3507 | if (cgrp->id < 0) { | |
3508 | err = -ENOMEM; | |
3509 | goto err_unlock; | |
976c06bc TH |
3510 | } |
3511 | ||
cc31edce | 3512 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3513 | |
bd89aabc | 3514 | cgrp->parent = parent; |
0ae78e0b | 3515 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 3516 | cgrp->root = parent->root; |
ddbcc7e8 | 3517 | |
b6abdb0e LZ |
3518 | if (notify_on_release(parent)) |
3519 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3520 | ||
2260e7fc TH |
3521 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
3522 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 3523 | |
2bd59d48 | 3524 | /* create the directory */ |
e61734c5 | 3525 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 TH |
3526 | if (IS_ERR(kn)) { |
3527 | err = PTR_ERR(kn); | |
0ab02ca8 | 3528 | goto err_free_id; |
2bd59d48 TH |
3529 | } |
3530 | cgrp->kn = kn; | |
ddbcc7e8 | 3531 | |
6f30558f TH |
3532 | /* |
3533 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
3534 | * that @cgrp->kn is always accessible. | |
3535 | */ | |
3536 | kernfs_get(kn); | |
3537 | ||
00356bd5 | 3538 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 3539 | |
4e139afc | 3540 | /* allocation complete, commit to creation */ |
4e139afc | 3541 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3c9c825b | 3542 | atomic_inc(&root->nr_cgrps); |
59f5296b | 3543 | cgroup_get(parent); |
415cf07a | 3544 | |
0d80255e TH |
3545 | /* |
3546 | * @cgrp is now fully operational. If something fails after this | |
3547 | * point, it'll be released via the normal destruction path. | |
3548 | */ | |
4e96ee8e LZ |
3549 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
3550 | ||
2bb566cb | 3551 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
3552 | if (err) |
3553 | goto err_destroy; | |
3554 | ||
9d403e99 | 3555 | /* let's create and online css's */ |
b85d2040 TH |
3556 | for_each_subsys(ss, ssid) { |
3557 | if (root->subsys_mask & (1 << ssid)) { | |
3558 | err = create_css(cgrp, ss); | |
3559 | if (err) | |
3560 | goto err_destroy; | |
3561 | } | |
a8638030 | 3562 | } |
ddbcc7e8 | 3563 | |
2bd59d48 TH |
3564 | kernfs_activate(kn); |
3565 | ||
ddbcc7e8 | 3566 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3567 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
3568 | |
3569 | return 0; | |
3570 | ||
0a950f65 | 3571 | err_free_id: |
4e96ee8e | 3572 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
3573 | err_unlock: |
3574 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
3575 | err_unlock_tree: |
3576 | mutex_unlock(&cgroup_tree_mutex); | |
bd89aabc | 3577 | kfree(cgrp); |
ddbcc7e8 | 3578 | return err; |
4b8b47eb TH |
3579 | |
3580 | err_destroy: | |
3581 | cgroup_destroy_locked(cgrp); | |
3582 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3583 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 3584 | return err; |
ddbcc7e8 PM |
3585 | } |
3586 | ||
2bd59d48 TH |
3587 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
3588 | umode_t mode) | |
ddbcc7e8 | 3589 | { |
2bd59d48 | 3590 | struct cgroup *parent = parent_kn->priv; |
ddbcc7e8 | 3591 | |
2bd59d48 | 3592 | return cgroup_create(parent, name, mode); |
ddbcc7e8 PM |
3593 | } |
3594 | ||
223dbc38 TH |
3595 | /* |
3596 | * This is called when the refcnt of a css is confirmed to be killed. | |
3597 | * css_tryget() is now guaranteed to fail. | |
3598 | */ | |
3599 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 3600 | { |
223dbc38 TH |
3601 | struct cgroup_subsys_state *css = |
3602 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
3603 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 3604 | |
ace2bee8 | 3605 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
3606 | mutex_lock(&cgroup_mutex); |
3607 | ||
09a503ea TH |
3608 | /* |
3609 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
3610 | * initate destruction. | |
3611 | */ | |
3612 | offline_css(css); | |
3613 | ||
f20104de TH |
3614 | /* |
3615 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
3616 | * be disabled before proceeding to the second phase of cgroup | |
3617 | * destruction. If we are the last one, kick it off. | |
3618 | */ | |
09a503ea | 3619 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
3620 | cgroup_destroy_css_killed(cgrp); |
3621 | ||
3622 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3623 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
3624 | |
3625 | /* | |
3626 | * Put the css refs from kill_css(). Each css holds an extra | |
3627 | * reference to the cgroup's dentry and cgroup removal proceeds | |
3628 | * regardless of css refs. On the last put of each css, whenever | |
3629 | * that may be, the extra dentry ref is put so that dentry | |
3630 | * destruction happens only after all css's are released. | |
3631 | */ | |
3632 | css_put(css); | |
d3daf28d TH |
3633 | } |
3634 | ||
223dbc38 TH |
3635 | /* css kill confirmation processing requires process context, bounce */ |
3636 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
3637 | { |
3638 | struct cgroup_subsys_state *css = | |
3639 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3640 | ||
223dbc38 | 3641 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 3642 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
3643 | } |
3644 | ||
edae0c33 TH |
3645 | /** |
3646 | * kill_css - destroy a css | |
3647 | * @css: css to destroy | |
3648 | * | |
3c14f8b4 TH |
3649 | * This function initiates destruction of @css by removing cgroup interface |
3650 | * files and putting its base reference. ->css_offline() will be invoked | |
3651 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
3652 | * reference count reaches zero, @css will be released. | |
edae0c33 TH |
3653 | */ |
3654 | static void kill_css(struct cgroup_subsys_state *css) | |
3655 | { | |
2bd59d48 TH |
3656 | /* |
3657 | * This must happen before css is disassociated with its cgroup. | |
3658 | * See seq_css() for details. | |
3659 | */ | |
aec25020 | 3660 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 3661 | |
edae0c33 TH |
3662 | /* |
3663 | * Killing would put the base ref, but we need to keep it alive | |
3664 | * until after ->css_offline(). | |
3665 | */ | |
3666 | css_get(css); | |
3667 | ||
3668 | /* | |
3669 | * cgroup core guarantees that, by the time ->css_offline() is | |
3670 | * invoked, no new css reference will be given out via | |
3671 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
3672 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
3673 | * guarantee that the ref is seen as killed on all CPUs on return. | |
3674 | * | |
3675 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
3676 | * css is confirmed to be seen as killed on all CPUs. | |
3677 | */ | |
3678 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
3679 | } |
3680 | ||
3681 | /** | |
3682 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
3683 | * @cgrp: cgroup to be destroyed | |
3684 | * | |
3685 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
3686 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
3687 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
3688 | * invoked. To satisfy all the requirements, destruction is implemented in | |
3689 | * the following two steps. | |
3690 | * | |
3691 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
3692 | * userland visible parts and start killing the percpu refcnts of | |
3693 | * css's. Set up so that the next stage will be kicked off once all | |
3694 | * the percpu refcnts are confirmed to be killed. | |
3695 | * | |
3696 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
3697 | * rest of destruction. Once all cgroup references are gone, the | |
3698 | * cgroup is RCU-freed. | |
3699 | * | |
3700 | * This function implements s1. After this step, @cgrp is gone as far as | |
3701 | * the userland is concerned and a new cgroup with the same name may be | |
3702 | * created. As cgroup doesn't care about the names internally, this | |
3703 | * doesn't cause any problem. | |
3704 | */ | |
42809dd4 TH |
3705 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
3706 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 3707 | { |
bb78a92f | 3708 | struct cgroup *child; |
2bd59d48 | 3709 | struct cgroup_subsys_state *css; |
ddd69148 | 3710 | bool empty; |
1c6727af | 3711 | int ssid; |
ddbcc7e8 | 3712 | |
ace2bee8 | 3713 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
3714 | lockdep_assert_held(&cgroup_mutex); |
3715 | ||
ddd69148 | 3716 | /* |
96d365e0 | 3717 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
6f3d828f | 3718 | * @cgrp from being removed while __put_css_set() is in progress. |
ddd69148 | 3719 | */ |
96d365e0 | 3720 | down_read(&css_set_rwsem); |
bb78a92f | 3721 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 3722 | up_read(&css_set_rwsem); |
ddd69148 | 3723 | if (!empty) |
ddbcc7e8 | 3724 | return -EBUSY; |
a043e3b2 | 3725 | |
bb78a92f HD |
3726 | /* |
3727 | * Make sure there's no live children. We can't test ->children | |
3728 | * emptiness as dead children linger on it while being destroyed; | |
3729 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
3730 | */ | |
3731 | empty = true; | |
3732 | rcu_read_lock(); | |
3733 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
3734 | empty = cgroup_is_dead(child); | |
3735 | if (!empty) | |
3736 | break; | |
3737 | } | |
3738 | rcu_read_unlock(); | |
3739 | if (!empty) | |
3740 | return -EBUSY; | |
3741 | ||
88703267 | 3742 | /* |
edae0c33 TH |
3743 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
3744 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
3745 | * percpu refs of all css's are confirmed to be killed. This |
3746 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 3747 | */ |
4ac06017 | 3748 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
3749 | for_each_css(css, ssid, cgrp) |
3750 | kill_css(css); | |
4ac06017 | 3751 | mutex_lock(&cgroup_mutex); |
455050d2 TH |
3752 | |
3753 | /* | |
3754 | * Mark @cgrp dead. This prevents further task migration and child | |
3755 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 3756 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 3757 | * resume iteration after dropping RCU read lock. See |
492eb21b | 3758 | * css_next_child() for details. |
455050d2 | 3759 | */ |
54766d4a | 3760 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 3761 | |
455050d2 TH |
3762 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
3763 | raw_spin_lock(&release_list_lock); | |
3764 | if (!list_empty(&cgrp->release_list)) | |
3765 | list_del_init(&cgrp->release_list); | |
3766 | raw_spin_unlock(&release_list_lock); | |
3767 | ||
3768 | /* | |
f20104de TH |
3769 | * If @cgrp has css's attached, the second stage of cgroup |
3770 | * destruction is kicked off from css_killed_work_fn() after the | |
3771 | * refs of all attached css's are killed. If @cgrp doesn't have | |
3772 | * any css, we kick it off here. | |
3773 | */ | |
3774 | if (!cgrp->nr_css) | |
3775 | cgroup_destroy_css_killed(cgrp); | |
3776 | ||
2bd59d48 TH |
3777 | /* remove @cgrp directory along with the base files */ |
3778 | mutex_unlock(&cgroup_mutex); | |
3779 | ||
455050d2 | 3780 | /* |
2bd59d48 TH |
3781 | * There are two control paths which try to determine cgroup from |
3782 | * dentry without going through kernfs - cgroupstats_build() and | |
3783 | * css_tryget_from_dir(). Those are supported by RCU protecting | |
3784 | * clearing of cgrp->kn->priv backpointer, which should happen | |
3785 | * after all files under it have been removed. | |
455050d2 | 3786 | */ |
6f30558f | 3787 | kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ |
2bd59d48 | 3788 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); |
2bd59d48 | 3789 | |
4ac06017 | 3790 | mutex_lock(&cgroup_mutex); |
455050d2 | 3791 | |
ea15f8cc TH |
3792 | return 0; |
3793 | }; | |
3794 | ||
d3daf28d | 3795 | /** |
f20104de | 3796 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
3797 | * @work: cgroup->destroy_free_work |
3798 | * | |
3799 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
3800 | * destroyed after all css's are offlined and performs the rest of |
3801 | * destruction. This is the second step of destruction described in the | |
3802 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 3803 | */ |
f20104de | 3804 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 3805 | { |
ea15f8cc | 3806 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 3807 | |
ace2bee8 | 3808 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 3809 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 3810 | |
999cd8a4 | 3811 | /* delete this cgroup from parent->children */ |
eb6fd504 | 3812 | list_del_rcu(&cgrp->sibling); |
ed957793 | 3813 | |
59f5296b | 3814 | cgroup_put(cgrp); |
ddbcc7e8 | 3815 | |
bd89aabc | 3816 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 3817 | check_for_release(parent); |
ddbcc7e8 PM |
3818 | } |
3819 | ||
2bd59d48 | 3820 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 3821 | { |
2bd59d48 TH |
3822 | struct cgroup *cgrp = kn->priv; |
3823 | int ret = 0; | |
3824 | ||
3825 | /* | |
3826 | * This is self-destruction but @kn can't be removed while this | |
3827 | * callback is in progress. Let's break active protection. Once | |
3828 | * the protection is broken, @cgrp can be destroyed at any point. | |
3829 | * Pin it so that it stays accessible. | |
3830 | */ | |
3831 | cgroup_get(cgrp); | |
3832 | kernfs_break_active_protection(kn); | |
42809dd4 | 3833 | |
ace2bee8 | 3834 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 3835 | mutex_lock(&cgroup_mutex); |
2bd59d48 TH |
3836 | |
3837 | /* | |
3838 | * @cgrp might already have been destroyed while we're trying to | |
3839 | * grab the mutexes. | |
3840 | */ | |
3841 | if (!cgroup_is_dead(cgrp)) | |
3842 | ret = cgroup_destroy_locked(cgrp); | |
3843 | ||
42809dd4 | 3844 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3845 | mutex_unlock(&cgroup_tree_mutex); |
42809dd4 | 3846 | |
2bd59d48 TH |
3847 | kernfs_unbreak_active_protection(kn); |
3848 | cgroup_put(cgrp); | |
42809dd4 TH |
3849 | return ret; |
3850 | } | |
3851 | ||
2bd59d48 TH |
3852 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
3853 | .remount_fs = cgroup_remount, | |
3854 | .show_options = cgroup_show_options, | |
3855 | .mkdir = cgroup_mkdir, | |
3856 | .rmdir = cgroup_rmdir, | |
3857 | .rename = cgroup_rename, | |
3858 | }; | |
3859 | ||
06a11920 | 3860 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 3861 | { |
ddbcc7e8 | 3862 | struct cgroup_subsys_state *css; |
cfe36bde DC |
3863 | |
3864 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 3865 | |
ace2bee8 | 3866 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
3867 | mutex_lock(&cgroup_mutex); |
3868 | ||
0adb0704 | 3869 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 3870 | |
ddbcc7e8 | 3871 | /* Create the top cgroup state for this subsystem */ |
9871bf95 | 3872 | ss->root = &cgroup_dummy_root; |
ca8bdcaf | 3873 | css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss)); |
ddbcc7e8 PM |
3874 | /* We don't handle early failures gracefully */ |
3875 | BUG_ON(IS_ERR(css)); | |
623f926b | 3876 | init_css(css, ss, cgroup_dummy_top); |
ddbcc7e8 | 3877 | |
e8d55fde | 3878 | /* Update the init_css_set to contain a subsys |
817929ec | 3879 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
3880 | * newly registered, all tasks and hence the |
3881 | * init_css_set is in the subsystem's top cgroup. */ | |
aec25020 | 3882 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
3883 | |
3884 | need_forkexit_callback |= ss->fork || ss->exit; | |
3885 | ||
e8d55fde LZ |
3886 | /* At system boot, before all subsystems have been |
3887 | * registered, no tasks have been forked, so we don't | |
3888 | * need to invoke fork callbacks here. */ | |
3889 | BUG_ON(!list_empty(&init_task.tasks)); | |
3890 | ||
ae7f164a | 3891 | BUG_ON(online_css(css)); |
a8638030 | 3892 | |
648bb56d | 3893 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3894 | mutex_unlock(&cgroup_tree_mutex); |
e6a1105b BB |
3895 | } |
3896 | ||
ddbcc7e8 | 3897 | /** |
a043e3b2 LZ |
3898 | * cgroup_init_early - cgroup initialization at system boot |
3899 | * | |
3900 | * Initialize cgroups at system boot, and initialize any | |
3901 | * subsystems that request early init. | |
ddbcc7e8 PM |
3902 | */ |
3903 | int __init cgroup_init_early(void) | |
3904 | { | |
30159ec7 | 3905 | struct cgroup_subsys *ss; |
ddbcc7e8 | 3906 | int i; |
30159ec7 | 3907 | |
146aa1bd | 3908 | atomic_set(&init_css_set.refcount, 1); |
69d0206c | 3909 | INIT_LIST_HEAD(&init_css_set.cgrp_links); |
817929ec | 3910 | INIT_LIST_HEAD(&init_css_set.tasks); |
472b1053 | 3911 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 3912 | css_set_count = 1; |
9871bf95 TH |
3913 | init_cgroup_root(&cgroup_dummy_root); |
3914 | cgroup_root_count = 1; | |
a4ea1cc9 | 3915 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 3916 | |
69d0206c | 3917 | init_cgrp_cset_link.cset = &init_css_set; |
9871bf95 TH |
3918 | init_cgrp_cset_link.cgrp = cgroup_dummy_top; |
3919 | list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links); | |
69d0206c | 3920 | list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links); |
ddbcc7e8 | 3921 | |
3ed80a62 | 3922 | for_each_subsys(ss, i) { |
aec25020 | 3923 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
3924 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
3925 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 3926 | ss->id, ss->name); |
073219e9 TH |
3927 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
3928 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
3929 | ||
aec25020 | 3930 | ss->id = i; |
073219e9 | 3931 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
3932 | |
3933 | if (ss->early_init) | |
3934 | cgroup_init_subsys(ss); | |
3935 | } | |
3936 | return 0; | |
3937 | } | |
3938 | ||
3939 | /** | |
a043e3b2 LZ |
3940 | * cgroup_init - cgroup initialization |
3941 | * | |
3942 | * Register cgroup filesystem and /proc file, and initialize | |
3943 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
3944 | */ |
3945 | int __init cgroup_init(void) | |
3946 | { | |
30159ec7 | 3947 | struct cgroup_subsys *ss; |
0ac801fe | 3948 | unsigned long key; |
30159ec7 | 3949 | int i, err; |
a424316c | 3950 | |
2bd59d48 | 3951 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
2da440a2 | 3952 | |
3ed80a62 | 3953 | for_each_subsys(ss, i) { |
ddbcc7e8 PM |
3954 | if (!ss->early_init) |
3955 | cgroup_init_subsys(ss); | |
de00ffa5 TH |
3956 | |
3957 | /* | |
3958 | * cftype registration needs kmalloc and can't be done | |
3959 | * during early_init. Register base cftypes separately. | |
3960 | */ | |
3961 | if (ss->base_cftypes) | |
3962 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
ddbcc7e8 PM |
3963 | } |
3964 | ||
fa3ca07e | 3965 | /* allocate id for the dummy hierarchy */ |
54e7b4eb | 3966 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 3967 | |
82fe9b0d TH |
3968 | /* Add init_css_set to the hash table */ |
3969 | key = css_set_hash(init_css_set.subsys); | |
3970 | hash_add(css_set_table, &init_css_set.hlist, key); | |
3971 | ||
fc76df70 | 3972 | BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1)); |
676db4af | 3973 | |
4e96ee8e LZ |
3974 | err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top, |
3975 | 0, 1, GFP_KERNEL); | |
3976 | BUG_ON(err < 0); | |
3977 | ||
54e7b4eb TH |
3978 | mutex_unlock(&cgroup_mutex); |
3979 | ||
676db4af | 3980 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
3981 | if (!cgroup_kobj) |
3982 | return -ENOMEM; | |
676db4af | 3983 | |
ddbcc7e8 | 3984 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
3985 | if (err < 0) { |
3986 | kobject_put(cgroup_kobj); | |
2bd59d48 | 3987 | return err; |
676db4af | 3988 | } |
ddbcc7e8 | 3989 | |
46ae220b | 3990 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 3991 | return 0; |
ddbcc7e8 | 3992 | } |
b4f48b63 | 3993 | |
e5fca243 TH |
3994 | static int __init cgroup_wq_init(void) |
3995 | { | |
3996 | /* | |
3997 | * There isn't much point in executing destruction path in | |
3998 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
ab3f5faa HD |
3999 | * |
4000 | * XXX: Must be ordered to make sure parent is offlined after | |
4001 | * children. The ordering requirement is for memcg where a | |
4002 | * parent's offline may wait for a child's leading to deadlock. In | |
4003 | * the long term, this should be fixed from memcg side. | |
e5fca243 TH |
4004 | * |
4005 | * We would prefer to do this in cgroup_init() above, but that | |
4006 | * is called before init_workqueues(): so leave this until after. | |
4007 | */ | |
ab3f5faa | 4008 | cgroup_destroy_wq = alloc_ordered_workqueue("cgroup_destroy", 0); |
e5fca243 | 4009 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4010 | |
4011 | /* | |
4012 | * Used to destroy pidlists and separate to serve as flush domain. | |
4013 | * Cap @max_active to 1 too. | |
4014 | */ | |
4015 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4016 | 0, 1); | |
4017 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4018 | ||
e5fca243 TH |
4019 | return 0; |
4020 | } | |
4021 | core_initcall(cgroup_wq_init); | |
4022 | ||
a424316c PM |
4023 | /* |
4024 | * proc_cgroup_show() | |
4025 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4026 | * - Used for /proc/<pid>/cgroup. | |
4027 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4028 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4029 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4030 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4031 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4032 | * cgroup to top_cgroup. | |
4033 | */ | |
4034 | ||
4035 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4036 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4037 | { |
4038 | struct pid *pid; | |
4039 | struct task_struct *tsk; | |
e61734c5 | 4040 | char *buf, *path; |
a424316c PM |
4041 | int retval; |
4042 | struct cgroupfs_root *root; | |
4043 | ||
4044 | retval = -ENOMEM; | |
e61734c5 | 4045 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4046 | if (!buf) |
4047 | goto out; | |
4048 | ||
4049 | retval = -ESRCH; | |
4050 | pid = m->private; | |
4051 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4052 | if (!tsk) | |
4053 | goto out_free; | |
4054 | ||
4055 | retval = 0; | |
4056 | ||
4057 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 4058 | down_read(&css_set_rwsem); |
a424316c | 4059 | |
e5f6a860 | 4060 | for_each_active_root(root) { |
a424316c | 4061 | struct cgroup_subsys *ss; |
bd89aabc | 4062 | struct cgroup *cgrp; |
b85d2040 | 4063 | int ssid, count = 0; |
a424316c | 4064 | |
2c6ab6d2 | 4065 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 TH |
4066 | for_each_subsys(ss, ssid) |
4067 | if (root->subsys_mask & (1 << ssid)) | |
4068 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4069 | if (strlen(root->name)) |
4070 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4071 | root->name); | |
a424316c | 4072 | seq_putc(m, ':'); |
7717f7ba | 4073 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
4074 | path = cgroup_path(cgrp, buf, PATH_MAX); |
4075 | if (!path) { | |
4076 | retval = -ENAMETOOLONG; | |
a424316c | 4077 | goto out_unlock; |
e61734c5 TH |
4078 | } |
4079 | seq_puts(m, path); | |
a424316c PM |
4080 | seq_putc(m, '\n'); |
4081 | } | |
4082 | ||
4083 | out_unlock: | |
96d365e0 | 4084 | up_read(&css_set_rwsem); |
a424316c PM |
4085 | mutex_unlock(&cgroup_mutex); |
4086 | put_task_struct(tsk); | |
4087 | out_free: | |
4088 | kfree(buf); | |
4089 | out: | |
4090 | return retval; | |
4091 | } | |
4092 | ||
a424316c PM |
4093 | /* Display information about each subsystem and each hierarchy */ |
4094 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4095 | { | |
30159ec7 | 4096 | struct cgroup_subsys *ss; |
a424316c | 4097 | int i; |
a424316c | 4098 | |
8bab8dde | 4099 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4100 | /* |
4101 | * ideally we don't want subsystems moving around while we do this. | |
4102 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4103 | * subsys/hierarchy state. | |
4104 | */ | |
a424316c | 4105 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4106 | |
4107 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4108 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4109 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 4110 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 4111 | |
a424316c PM |
4112 | mutex_unlock(&cgroup_mutex); |
4113 | return 0; | |
4114 | } | |
4115 | ||
4116 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4117 | { | |
9dce07f1 | 4118 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4119 | } |
4120 | ||
828c0950 | 4121 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4122 | .open = cgroupstats_open, |
4123 | .read = seq_read, | |
4124 | .llseek = seq_lseek, | |
4125 | .release = single_release, | |
4126 | }; | |
4127 | ||
b4f48b63 PM |
4128 | /** |
4129 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4130 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4131 | * |
4132 | * Description: A task inherits its parent's cgroup at fork(). | |
4133 | * | |
4134 | * A pointer to the shared css_set was automatically copied in | |
4135 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
9bb71308 TH |
4136 | * it was not made under the protection of RCU or cgroup_mutex, so |
4137 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
4138 | * have already changed current->cgroups, allowing the previously | |
4139 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
4140 | * |
4141 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4142 | * task, and the passed argument 'child' points to the child task. | |
4143 | */ | |
4144 | void cgroup_fork(struct task_struct *child) | |
4145 | { | |
9bb71308 | 4146 | task_lock(current); |
a8ad805c | 4147 | get_css_set(task_css_set(current)); |
817929ec | 4148 | child->cgroups = current->cgroups; |
9bb71308 | 4149 | task_unlock(current); |
817929ec | 4150 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4151 | } |
4152 | ||
817929ec | 4153 | /** |
a043e3b2 LZ |
4154 | * cgroup_post_fork - called on a new task after adding it to the task list |
4155 | * @child: the task in question | |
4156 | * | |
5edee61e TH |
4157 | * Adds the task to the list running through its css_set if necessary and |
4158 | * call the subsystem fork() callbacks. Has to be after the task is | |
4159 | * visible on the task list in case we race with the first call to | |
0942eeee | 4160 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4161 | * list. |
a043e3b2 | 4162 | */ |
817929ec PM |
4163 | void cgroup_post_fork(struct task_struct *child) |
4164 | { | |
30159ec7 | 4165 | struct cgroup_subsys *ss; |
5edee61e TH |
4166 | int i; |
4167 | ||
3ce3230a FW |
4168 | /* |
4169 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
4170 | * under the tasklist_lock and we read it here after we added the child | |
4171 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
4172 | * yet in the tasklist when we walked through it from | |
4173 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
4174 | * should be visible now due to the paired locking and barriers implied | |
4175 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
4176 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
4177 | * lock on fork. | |
4178 | */ | |
817929ec | 4179 | if (use_task_css_set_links) { |
96d365e0 | 4180 | down_write(&css_set_rwsem); |
d8783832 TH |
4181 | task_lock(child); |
4182 | if (list_empty(&child->cg_list)) | |
a8ad805c | 4183 | list_add(&child->cg_list, &task_css_set(child)->tasks); |
d8783832 | 4184 | task_unlock(child); |
96d365e0 | 4185 | up_write(&css_set_rwsem); |
817929ec | 4186 | } |
5edee61e TH |
4187 | |
4188 | /* | |
4189 | * Call ss->fork(). This must happen after @child is linked on | |
4190 | * css_set; otherwise, @child might change state between ->fork() | |
4191 | * and addition to css_set. | |
4192 | */ | |
4193 | if (need_forkexit_callback) { | |
3ed80a62 | 4194 | for_each_subsys(ss, i) |
5edee61e TH |
4195 | if (ss->fork) |
4196 | ss->fork(child); | |
5edee61e | 4197 | } |
817929ec | 4198 | } |
5edee61e | 4199 | |
b4f48b63 PM |
4200 | /** |
4201 | * cgroup_exit - detach cgroup from exiting task | |
4202 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4203 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4204 | * |
4205 | * Description: Detach cgroup from @tsk and release it. | |
4206 | * | |
4207 | * Note that cgroups marked notify_on_release force every task in | |
4208 | * them to take the global cgroup_mutex mutex when exiting. | |
4209 | * This could impact scaling on very large systems. Be reluctant to | |
4210 | * use notify_on_release cgroups where very high task exit scaling | |
4211 | * is required on large systems. | |
4212 | * | |
4213 | * the_top_cgroup_hack: | |
4214 | * | |
4215 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4216 | * | |
4217 | * We call cgroup_exit() while the task is still competent to | |
4218 | * handle notify_on_release(), then leave the task attached to the | |
4219 | * root cgroup in each hierarchy for the remainder of its exit. | |
4220 | * | |
4221 | * To do this properly, we would increment the reference count on | |
4222 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4223 | * code we would add a second cgroup function call, to drop that | |
4224 | * reference. This would just create an unnecessary hot spot on | |
4225 | * the top_cgroup reference count, to no avail. | |
4226 | * | |
4227 | * Normally, holding a reference to a cgroup without bumping its | |
4228 | * count is unsafe. The cgroup could go away, or someone could | |
4229 | * attach us to a different cgroup, decrementing the count on | |
4230 | * the first cgroup that we never incremented. But in this case, | |
4231 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4232 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4233 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4234 | */ |
4235 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4236 | { | |
30159ec7 | 4237 | struct cgroup_subsys *ss; |
5abb8855 | 4238 | struct css_set *cset; |
d41d5a01 | 4239 | int i; |
817929ec PM |
4240 | |
4241 | /* | |
96d365e0 TH |
4242 | * Unlink from the css_set task list if necessary. Optimistically |
4243 | * check cg_list before taking css_set_rwsem. | |
817929ec PM |
4244 | */ |
4245 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 4246 | down_write(&css_set_rwsem); |
817929ec | 4247 | if (!list_empty(&tsk->cg_list)) |
8d258797 | 4248 | list_del_init(&tsk->cg_list); |
96d365e0 | 4249 | up_write(&css_set_rwsem); |
817929ec PM |
4250 | } |
4251 | ||
b4f48b63 PM |
4252 | /* Reassign the task to the init_css_set. */ |
4253 | task_lock(tsk); | |
a8ad805c TH |
4254 | cset = task_css_set(tsk); |
4255 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 PZ |
4256 | |
4257 | if (run_callbacks && need_forkexit_callback) { | |
3ed80a62 TH |
4258 | /* see cgroup_post_fork() for details */ |
4259 | for_each_subsys(ss, i) { | |
d41d5a01 | 4260 | if (ss->exit) { |
eb95419b TH |
4261 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4262 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4263 | |
eb95419b | 4264 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4265 | } |
4266 | } | |
4267 | } | |
b4f48b63 | 4268 | task_unlock(tsk); |
d41d5a01 | 4269 | |
5abb8855 | 4270 | put_css_set_taskexit(cset); |
b4f48b63 | 4271 | } |
697f4161 | 4272 | |
bd89aabc | 4273 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4274 | { |
f50daa70 | 4275 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4276 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4277 | /* |
4278 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4279 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4280 | * it now |
4281 | */ | |
81a6a5cd | 4282 | int need_schedule_work = 0; |
f50daa70 | 4283 | |
cdcc136f | 4284 | raw_spin_lock(&release_list_lock); |
54766d4a | 4285 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4286 | list_empty(&cgrp->release_list)) { |
4287 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4288 | need_schedule_work = 1; |
4289 | } | |
cdcc136f | 4290 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4291 | if (need_schedule_work) |
4292 | schedule_work(&release_agent_work); | |
4293 | } | |
4294 | } | |
4295 | ||
81a6a5cd PM |
4296 | /* |
4297 | * Notify userspace when a cgroup is released, by running the | |
4298 | * configured release agent with the name of the cgroup (path | |
4299 | * relative to the root of cgroup file system) as the argument. | |
4300 | * | |
4301 | * Most likely, this user command will try to rmdir this cgroup. | |
4302 | * | |
4303 | * This races with the possibility that some other task will be | |
4304 | * attached to this cgroup before it is removed, or that some other | |
4305 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4306 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4307 | * unused, and this cgroup will be reprieved from its death sentence, | |
4308 | * to continue to serve a useful existence. Next time it's released, | |
4309 | * we will get notified again, if it still has 'notify_on_release' set. | |
4310 | * | |
4311 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4312 | * means only wait until the task is successfully execve()'d. The | |
4313 | * separate release agent task is forked by call_usermodehelper(), | |
4314 | * then control in this thread returns here, without waiting for the | |
4315 | * release agent task. We don't bother to wait because the caller of | |
4316 | * this routine has no use for the exit status of the release agent | |
4317 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4318 | */ |
81a6a5cd PM |
4319 | static void cgroup_release_agent(struct work_struct *work) |
4320 | { | |
4321 | BUG_ON(work != &release_agent_work); | |
4322 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4323 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4324 | while (!list_empty(&release_list)) { |
4325 | char *argv[3], *envp[3]; | |
4326 | int i; | |
e61734c5 | 4327 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 4328 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4329 | struct cgroup, |
4330 | release_list); | |
bd89aabc | 4331 | list_del_init(&cgrp->release_list); |
cdcc136f | 4332 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 4333 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
4334 | if (!pathbuf) |
4335 | goto continue_free; | |
e61734c5 TH |
4336 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
4337 | if (!path) | |
e788e066 PM |
4338 | goto continue_free; |
4339 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4340 | if (!agentbuf) | |
4341 | goto continue_free; | |
81a6a5cd PM |
4342 | |
4343 | i = 0; | |
e788e066 | 4344 | argv[i++] = agentbuf; |
e61734c5 | 4345 | argv[i++] = path; |
81a6a5cd PM |
4346 | argv[i] = NULL; |
4347 | ||
4348 | i = 0; | |
4349 | /* minimal command environment */ | |
4350 | envp[i++] = "HOME=/"; | |
4351 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4352 | envp[i] = NULL; | |
4353 | ||
4354 | /* Drop the lock while we invoke the usermode helper, | |
4355 | * since the exec could involve hitting disk and hence | |
4356 | * be a slow process */ | |
4357 | mutex_unlock(&cgroup_mutex); | |
4358 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4359 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4360 | continue_free: |
4361 | kfree(pathbuf); | |
4362 | kfree(agentbuf); | |
cdcc136f | 4363 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4364 | } |
cdcc136f | 4365 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4366 | mutex_unlock(&cgroup_mutex); |
4367 | } | |
8bab8dde PM |
4368 | |
4369 | static int __init cgroup_disable(char *str) | |
4370 | { | |
30159ec7 | 4371 | struct cgroup_subsys *ss; |
8bab8dde | 4372 | char *token; |
30159ec7 | 4373 | int i; |
8bab8dde PM |
4374 | |
4375 | while ((token = strsep(&str, ",")) != NULL) { | |
4376 | if (!*token) | |
4377 | continue; | |
be45c900 | 4378 | |
3ed80a62 | 4379 | for_each_subsys(ss, i) { |
8bab8dde PM |
4380 | if (!strcmp(token, ss->name)) { |
4381 | ss->disabled = 1; | |
4382 | printk(KERN_INFO "Disabling %s control group" | |
4383 | " subsystem\n", ss->name); | |
4384 | break; | |
4385 | } | |
4386 | } | |
4387 | } | |
4388 | return 1; | |
4389 | } | |
4390 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 4391 | |
b77d7b60 | 4392 | /** |
5a17f543 | 4393 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
4394 | * @dentry: directory dentry of interest |
4395 | * @ss: subsystem of interest | |
b77d7b60 | 4396 | * |
5a17f543 TH |
4397 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
4398 | * to get the corresponding css and return it. If such css doesn't exist | |
4399 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 4400 | */ |
5a17f543 TH |
4401 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
4402 | struct cgroup_subsys *ss) | |
e5d1367f | 4403 | { |
2bd59d48 TH |
4404 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
4405 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 4406 | struct cgroup *cgrp; |
b77d7b60 | 4407 | |
35cf0836 | 4408 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
4409 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
4410 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
4411 | return ERR_PTR(-EBADF); |
4412 | ||
5a17f543 TH |
4413 | rcu_read_lock(); |
4414 | ||
2bd59d48 TH |
4415 | /* |
4416 | * This path doesn't originate from kernfs and @kn could already | |
4417 | * have been or be removed at any point. @kn->priv is RCU | |
4418 | * protected for this access. See destroy_locked() for details. | |
4419 | */ | |
4420 | cgrp = rcu_dereference(kn->priv); | |
4421 | if (cgrp) | |
4422 | css = cgroup_css(cgrp, ss); | |
5a17f543 TH |
4423 | |
4424 | if (!css || !css_tryget(css)) | |
4425 | css = ERR_PTR(-ENOENT); | |
4426 | ||
4427 | rcu_read_unlock(); | |
4428 | return css; | |
e5d1367f | 4429 | } |
e5d1367f | 4430 | |
1cb650b9 LZ |
4431 | /** |
4432 | * css_from_id - lookup css by id | |
4433 | * @id: the cgroup id | |
4434 | * @ss: cgroup subsys to be looked into | |
4435 | * | |
4436 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
4437 | * Should be called under rcu_read_lock(). | |
4438 | */ | |
4439 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
4440 | { | |
4441 | struct cgroup *cgrp; | |
4442 | ||
ace2bee8 | 4443 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
4444 | |
4445 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
4446 | if (cgrp) | |
d1625964 | 4447 | return cgroup_css(cgrp, ss); |
1cb650b9 | 4448 | return NULL; |
e5d1367f SE |
4449 | } |
4450 | ||
fe693435 | 4451 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
4452 | static struct cgroup_subsys_state * |
4453 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
4454 | { |
4455 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
4456 | ||
4457 | if (!css) | |
4458 | return ERR_PTR(-ENOMEM); | |
4459 | ||
4460 | return css; | |
4461 | } | |
4462 | ||
eb95419b | 4463 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 4464 | { |
eb95419b | 4465 | kfree(css); |
fe693435 PM |
4466 | } |
4467 | ||
182446d0 TH |
4468 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
4469 | struct cftype *cft) | |
fe693435 | 4470 | { |
182446d0 | 4471 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
4472 | } |
4473 | ||
182446d0 TH |
4474 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
4475 | struct cftype *cft) | |
fe693435 PM |
4476 | { |
4477 | return (u64)(unsigned long)current->cgroups; | |
4478 | } | |
4479 | ||
182446d0 | 4480 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 4481 | struct cftype *cft) |
fe693435 PM |
4482 | { |
4483 | u64 count; | |
4484 | ||
4485 | rcu_read_lock(); | |
a8ad805c | 4486 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
4487 | rcu_read_unlock(); |
4488 | return count; | |
4489 | } | |
4490 | ||
2da8ca82 | 4491 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4492 | { |
69d0206c | 4493 | struct cgrp_cset_link *link; |
5abb8855 | 4494 | struct css_set *cset; |
e61734c5 TH |
4495 | char *name_buf; |
4496 | ||
4497 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
4498 | if (!name_buf) | |
4499 | return -ENOMEM; | |
7717f7ba | 4500 | |
96d365e0 | 4501 | down_read(&css_set_rwsem); |
7717f7ba | 4502 | rcu_read_lock(); |
5abb8855 | 4503 | cset = rcu_dereference(current->cgroups); |
69d0206c | 4504 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 4505 | struct cgroup *c = link->cgrp; |
59f5296b TH |
4506 | const char *name = "?"; |
4507 | ||
e61734c5 TH |
4508 | if (c != cgroup_dummy_top) { |
4509 | cgroup_name(c, name_buf, NAME_MAX + 1); | |
4510 | name = name_buf; | |
4511 | } | |
7717f7ba | 4512 | |
2c6ab6d2 PM |
4513 | seq_printf(seq, "Root %d group %s\n", |
4514 | c->root->hierarchy_id, name); | |
7717f7ba PM |
4515 | } |
4516 | rcu_read_unlock(); | |
96d365e0 | 4517 | up_read(&css_set_rwsem); |
e61734c5 | 4518 | kfree(name_buf); |
7717f7ba PM |
4519 | return 0; |
4520 | } | |
4521 | ||
4522 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 4523 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4524 | { |
2da8ca82 | 4525 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 4526 | struct cgrp_cset_link *link; |
7717f7ba | 4527 | |
96d365e0 | 4528 | down_read(&css_set_rwsem); |
182446d0 | 4529 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 4530 | struct css_set *cset = link->cset; |
7717f7ba PM |
4531 | struct task_struct *task; |
4532 | int count = 0; | |
5abb8855 TH |
4533 | seq_printf(seq, "css_set %p\n", cset); |
4534 | list_for_each_entry(task, &cset->tasks, cg_list) { | |
7717f7ba PM |
4535 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) { |
4536 | seq_puts(seq, " ...\n"); | |
4537 | break; | |
4538 | } else { | |
4539 | seq_printf(seq, " task %d\n", | |
4540 | task_pid_vnr(task)); | |
4541 | } | |
4542 | } | |
4543 | } | |
96d365e0 | 4544 | up_read(&css_set_rwsem); |
7717f7ba PM |
4545 | return 0; |
4546 | } | |
4547 | ||
182446d0 | 4548 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 4549 | { |
182446d0 | 4550 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
4551 | } |
4552 | ||
4553 | static struct cftype debug_files[] = { | |
fe693435 PM |
4554 | { |
4555 | .name = "taskcount", | |
4556 | .read_u64 = debug_taskcount_read, | |
4557 | }, | |
4558 | ||
4559 | { | |
4560 | .name = "current_css_set", | |
4561 | .read_u64 = current_css_set_read, | |
4562 | }, | |
4563 | ||
4564 | { | |
4565 | .name = "current_css_set_refcount", | |
4566 | .read_u64 = current_css_set_refcount_read, | |
4567 | }, | |
4568 | ||
7717f7ba PM |
4569 | { |
4570 | .name = "current_css_set_cg_links", | |
2da8ca82 | 4571 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
4572 | }, |
4573 | ||
4574 | { | |
4575 | .name = "cgroup_css_links", | |
2da8ca82 | 4576 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
4577 | }, |
4578 | ||
fe693435 PM |
4579 | { |
4580 | .name = "releasable", | |
4581 | .read_u64 = releasable_read, | |
4582 | }, | |
fe693435 | 4583 | |
4baf6e33 TH |
4584 | { } /* terminate */ |
4585 | }; | |
fe693435 | 4586 | |
073219e9 | 4587 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
4588 | .css_alloc = debug_css_alloc, |
4589 | .css_free = debug_css_free, | |
4baf6e33 | 4590 | .base_cftypes = debug_files, |
fe693435 PM |
4591 | }; |
4592 | #endif /* CONFIG_CGROUP_DEBUG */ |