Commit | Line | Data |
---|---|---|
ddbcc7e8 | 1 | /* |
ddbcc7e8 PM |
2 | * Generic process-grouping system. |
3 | * | |
4 | * Based originally on the cpuset system, extracted by Paul Menage | |
5 | * Copyright (C) 2006 Google, Inc | |
6 | * | |
0dea1168 KS |
7 | * Notifications support |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
ddbcc7e8 PM |
11 | * Copyright notices from the original cpuset code: |
12 | * -------------------------------------------------- | |
13 | * Copyright (C) 2003 BULL SA. | |
14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
15 | * | |
16 | * Portions derived from Patrick Mochel's sysfs code. | |
17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
18 | * | |
19 | * 2003-10-10 Written by Simon Derr. | |
20 | * 2003-10-22 Updates by Stephen Hemminger. | |
21 | * 2004 May-July Rework by Paul Jackson. | |
22 | * --------------------------------------------------- | |
23 | * | |
24 | * This file is subject to the terms and conditions of the GNU General Public | |
25 | * License. See the file COPYING in the main directory of the Linux | |
26 | * distribution for more details. | |
27 | */ | |
28 | ||
29 | #include <linux/cgroup.h> | |
2ce9738b | 30 | #include <linux/cred.h> |
c6d57f33 | 31 | #include <linux/ctype.h> |
ddbcc7e8 | 32 | #include <linux/errno.h> |
2ce9738b | 33 | #include <linux/init_task.h> |
ddbcc7e8 PM |
34 | #include <linux/kernel.h> |
35 | #include <linux/list.h> | |
36 | #include <linux/mm.h> | |
37 | #include <linux/mutex.h> | |
38 | #include <linux/mount.h> | |
39 | #include <linux/pagemap.h> | |
a424316c | 40 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
41 | #include <linux/rcupdate.h> |
42 | #include <linux/sched.h> | |
ddbcc7e8 | 43 | #include <linux/slab.h> |
ddbcc7e8 | 44 | #include <linux/spinlock.h> |
96d365e0 | 45 | #include <linux/rwsem.h> |
ddbcc7e8 | 46 | #include <linux/string.h> |
bbcb81d0 | 47 | #include <linux/sort.h> |
81a6a5cd | 48 | #include <linux/kmod.h> |
846c7bb0 BS |
49 | #include <linux/delayacct.h> |
50 | #include <linux/cgroupstats.h> | |
0ac801fe | 51 | #include <linux/hashtable.h> |
096b7fe0 | 52 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 53 | #include <linux/idr.h> |
d1d9fd33 | 54 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 55 | #include <linux/kthread.h> |
776f02fa | 56 | #include <linux/delay.h> |
846c7bb0 | 57 | |
60063497 | 58 | #include <linux/atomic.h> |
ddbcc7e8 | 59 | |
b1a21367 TH |
60 | /* |
61 | * pidlists linger the following amount before being destroyed. The goal | |
62 | * is avoiding frequent destruction in the middle of consecutive read calls | |
63 | * Expiring in the middle is a performance problem not a correctness one. | |
64 | * 1 sec should be enough. | |
65 | */ | |
66 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
67 | ||
8d7e6fb0 TH |
68 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
69 | MAX_CFTYPE_NAME + 2) | |
70 | ||
ace2bee8 TH |
71 | /* |
72 | * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file | |
73 | * creation/removal and hierarchy changing operations including cgroup | |
74 | * creation, removal, css association and controller rebinding. This outer | |
75 | * lock is needed mainly to resolve the circular dependency between kernfs | |
76 | * active ref and cgroup_mutex. cgroup_tree_mutex nests above both. | |
77 | */ | |
78 | static DEFINE_MUTEX(cgroup_tree_mutex); | |
79 | ||
e25e2cbb TH |
80 | /* |
81 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
82 | * hierarchy must be performed while holding it. | |
e25e2cbb | 83 | */ |
2219449a TH |
84 | #ifdef CONFIG_PROVE_RCU |
85 | DEFINE_MUTEX(cgroup_mutex); | |
8af01f56 | 86 | EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for lockdep */ |
2219449a | 87 | #else |
81a6a5cd | 88 | static DEFINE_MUTEX(cgroup_mutex); |
2219449a TH |
89 | #endif |
90 | ||
69e943b7 TH |
91 | /* |
92 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
93 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
94 | */ | |
95 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
96 | ||
ace2bee8 | 97 | #define cgroup_assert_mutexes_or_rcu_locked() \ |
87fb54f1 | 98 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
ace2bee8 | 99 | lockdep_is_held(&cgroup_tree_mutex) || \ |
87fb54f1 | 100 | lockdep_is_held(&cgroup_mutex), \ |
ace2bee8 | 101 | "cgroup_[tree_]mutex or RCU read lock required"); |
87fb54f1 | 102 | |
e5fca243 TH |
103 | /* |
104 | * cgroup destruction makes heavy use of work items and there can be a lot | |
105 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
106 | * destruction work items don't end up filling up max_active of system_wq | |
107 | * which may lead to deadlock. | |
108 | */ | |
109 | static struct workqueue_struct *cgroup_destroy_wq; | |
110 | ||
b1a21367 TH |
111 | /* |
112 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
113 | * separate workqueue as flush domain. | |
114 | */ | |
115 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
116 | ||
3ed80a62 | 117 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 118 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 119 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
120 | #include <linux/cgroup_subsys.h> |
121 | }; | |
073219e9 TH |
122 | #undef SUBSYS |
123 | ||
124 | /* array of cgroup subsystem names */ | |
125 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
126 | static const char *cgroup_subsys_name[] = { | |
127 | #include <linux/cgroup_subsys.h> | |
128 | }; | |
129 | #undef SUBSYS | |
ddbcc7e8 | 130 | |
ddbcc7e8 | 131 | /* |
9871bf95 TH |
132 | * The dummy hierarchy, reserved for the subsystems that are otherwise |
133 | * unattached - it never has more than a single cgroup, and all tasks are | |
134 | * part of that cgroup. | |
ddbcc7e8 | 135 | */ |
9871bf95 TH |
136 | static struct cgroupfs_root cgroup_dummy_root; |
137 | ||
138 | /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */ | |
139 | static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup; | |
ddbcc7e8 PM |
140 | |
141 | /* The list of hierarchy roots */ | |
142 | ||
9871bf95 TH |
143 | static LIST_HEAD(cgroup_roots); |
144 | static int cgroup_root_count; | |
ddbcc7e8 | 145 | |
3417ae1f | 146 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 147 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 148 | |
794611a1 LZ |
149 | /* |
150 | * Assign a monotonically increasing serial number to cgroups. It | |
151 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
152 | * numbers. Also, as cgroups are always appended to the parent's | |
153 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
154 | * the ascending serial number order on the list. Protected by |
155 | * cgroup_mutex. | |
794611a1 | 156 | */ |
00356bd5 | 157 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 158 | |
ddbcc7e8 | 159 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
160 | * check for fork/exit handlers to call. This avoids us having to do |
161 | * extra work in the fork/exit path if none of the subsystems need to | |
162 | * be called. | |
ddbcc7e8 | 163 | */ |
8947f9d5 | 164 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 165 | |
628f7cd4 TH |
166 | static struct cftype cgroup_base_files[]; |
167 | ||
59f5296b | 168 | static void cgroup_put(struct cgroup *cgrp); |
f2e85d57 TH |
169 | static int rebind_subsystems(struct cgroupfs_root *root, |
170 | unsigned long added_mask, unsigned removed_mask); | |
f20104de | 171 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 172 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
2bb566cb TH |
173 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
174 | bool is_add); | |
b1a21367 | 175 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 176 | |
95109b62 TH |
177 | /** |
178 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
179 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 180 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 181 | * |
ca8bdcaf TH |
182 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
183 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
184 | * the caller is responsible for pinning the returned css if it wants to | |
185 | * keep accessing it outside the said locks. This function may return | |
186 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
187 | */ |
188 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 189 | struct cgroup_subsys *ss) |
95109b62 | 190 | { |
ca8bdcaf | 191 | if (ss) |
aec25020 | 192 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 TH |
193 | lockdep_is_held(&cgroup_tree_mutex) || |
194 | lockdep_is_held(&cgroup_mutex)); | |
ca8bdcaf TH |
195 | else |
196 | return &cgrp->dummy_css; | |
95109b62 | 197 | } |
42809dd4 | 198 | |
ddbcc7e8 | 199 | /* convenient tests for these bits */ |
54766d4a | 200 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 201 | { |
54766d4a | 202 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
203 | } |
204 | ||
59f5296b TH |
205 | struct cgroup_subsys_state *seq_css(struct seq_file *seq) |
206 | { | |
2bd59d48 TH |
207 | struct kernfs_open_file *of = seq->private; |
208 | struct cgroup *cgrp = of->kn->parent->priv; | |
209 | struct cftype *cft = seq_cft(seq); | |
210 | ||
211 | /* | |
212 | * This is open and unprotected implementation of cgroup_css(). | |
213 | * seq_css() is only called from a kernfs file operation which has | |
214 | * an active reference on the file. Because all the subsystem | |
215 | * files are drained before a css is disassociated with a cgroup, | |
216 | * the matching css from the cgroup's subsys table is guaranteed to | |
217 | * be and stay valid until the enclosing operation is complete. | |
218 | */ | |
219 | if (cft->ss) | |
220 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
221 | else | |
222 | return &cgrp->dummy_css; | |
59f5296b TH |
223 | } |
224 | EXPORT_SYMBOL_GPL(seq_css); | |
225 | ||
78574cf9 LZ |
226 | /** |
227 | * cgroup_is_descendant - test ancestry | |
228 | * @cgrp: the cgroup to be tested | |
229 | * @ancestor: possible ancestor of @cgrp | |
230 | * | |
231 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
232 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
233 | * and @ancestor are accessible. | |
234 | */ | |
235 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
236 | { | |
237 | while (cgrp) { | |
238 | if (cgrp == ancestor) | |
239 | return true; | |
240 | cgrp = cgrp->parent; | |
241 | } | |
242 | return false; | |
243 | } | |
ddbcc7e8 | 244 | |
e9685a03 | 245 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
246 | { |
247 | const int bits = | |
bd89aabc PM |
248 | (1 << CGRP_RELEASABLE) | |
249 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
250 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
251 | } |
252 | ||
e9685a03 | 253 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 254 | { |
bd89aabc | 255 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
256 | } |
257 | ||
1c6727af TH |
258 | /** |
259 | * for_each_css - iterate all css's of a cgroup | |
260 | * @css: the iteration cursor | |
261 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
262 | * @cgrp: the target cgroup to iterate css's of | |
263 | * | |
264 | * Should be called under cgroup_mutex. | |
265 | */ | |
266 | #define for_each_css(css, ssid, cgrp) \ | |
267 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
268 | if (!((css) = rcu_dereference_check( \ | |
269 | (cgrp)->subsys[(ssid)], \ | |
ace2bee8 | 270 | lockdep_is_held(&cgroup_tree_mutex) || \ |
1c6727af TH |
271 | lockdep_is_held(&cgroup_mutex)))) { } \ |
272 | else | |
273 | ||
30159ec7 | 274 | /** |
3ed80a62 | 275 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 276 | * @ss: the iteration cursor |
780cd8b3 | 277 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 278 | */ |
780cd8b3 | 279 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
280 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
281 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 282 | |
5549c497 TH |
283 | /* iterate across the active hierarchies */ |
284 | #define for_each_active_root(root) \ | |
285 | list_for_each_entry((root), &cgroup_roots, root_list) | |
ddbcc7e8 | 286 | |
7ae1bad9 TH |
287 | /** |
288 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
289 | * @cgrp: the cgroup to be checked for liveness | |
290 | * | |
47cfcd09 TH |
291 | * On success, returns true; the mutex should be later unlocked. On |
292 | * failure returns false with no lock held. | |
7ae1bad9 | 293 | */ |
b9777cf8 | 294 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
295 | { |
296 | mutex_lock(&cgroup_mutex); | |
54766d4a | 297 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
298 | mutex_unlock(&cgroup_mutex); |
299 | return false; | |
300 | } | |
301 | return true; | |
302 | } | |
7ae1bad9 | 303 | |
81a6a5cd PM |
304 | /* the list of cgroups eligible for automatic release. Protected by |
305 | * release_list_lock */ | |
306 | static LIST_HEAD(release_list); | |
cdcc136f | 307 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
308 | static void cgroup_release_agent(struct work_struct *work); |
309 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 310 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 311 | |
69d0206c TH |
312 | /* |
313 | * A cgroup can be associated with multiple css_sets as different tasks may | |
314 | * belong to different cgroups on different hierarchies. In the other | |
315 | * direction, a css_set is naturally associated with multiple cgroups. | |
316 | * This M:N relationship is represented by the following link structure | |
317 | * which exists for each association and allows traversing the associations | |
318 | * from both sides. | |
319 | */ | |
320 | struct cgrp_cset_link { | |
321 | /* the cgroup and css_set this link associates */ | |
322 | struct cgroup *cgrp; | |
323 | struct css_set *cset; | |
324 | ||
325 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
326 | struct list_head cset_link; | |
327 | ||
328 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
329 | struct list_head cgrp_link; | |
817929ec PM |
330 | }; |
331 | ||
332 | /* The default css_set - used by init and its children prior to any | |
333 | * hierarchies being mounted. It contains a pointer to the root state | |
334 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
335 | * reference-counted, to improve performance when child cgroups | |
336 | * haven't been created. | |
337 | */ | |
338 | ||
339 | static struct css_set init_css_set; | |
69d0206c | 340 | static struct cgrp_cset_link init_cgrp_cset_link; |
817929ec | 341 | |
0942eeee | 342 | /* |
96d365e0 TH |
343 | * css_set_rwsem protects the list of css_set objects, and the chain of |
344 | * tasks off each css_set. | |
0942eeee | 345 | */ |
96d365e0 | 346 | static DECLARE_RWSEM(css_set_rwsem); |
817929ec PM |
347 | static int css_set_count; |
348 | ||
7717f7ba PM |
349 | /* |
350 | * hash table for cgroup groups. This improves the performance to find | |
351 | * an existing css_set. This hash doesn't (currently) take into | |
352 | * account cgroups in empty hierarchies. | |
353 | */ | |
472b1053 | 354 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 355 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 356 | |
0ac801fe | 357 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 358 | { |
0ac801fe | 359 | unsigned long key = 0UL; |
30159ec7 TH |
360 | struct cgroup_subsys *ss; |
361 | int i; | |
472b1053 | 362 | |
30159ec7 | 363 | for_each_subsys(ss, i) |
0ac801fe LZ |
364 | key += (unsigned long)css[i]; |
365 | key = (key >> 16) ^ key; | |
472b1053 | 366 | |
0ac801fe | 367 | return key; |
472b1053 LZ |
368 | } |
369 | ||
89c5509b | 370 | static void put_css_set_locked(struct css_set *cset, bool taskexit) |
b4f48b63 | 371 | { |
69d0206c | 372 | struct cgrp_cset_link *link, *tmp_link; |
5abb8855 | 373 | |
89c5509b TH |
374 | lockdep_assert_held(&css_set_rwsem); |
375 | ||
376 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 377 | return; |
81a6a5cd | 378 | |
2c6ab6d2 | 379 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
5abb8855 | 380 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
381 | css_set_count--; |
382 | ||
69d0206c | 383 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 384 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 385 | |
69d0206c TH |
386 | list_del(&link->cset_link); |
387 | list_del(&link->cgrp_link); | |
71b5707e | 388 | |
96d365e0 | 389 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
6f3d828f | 390 | if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { |
81a6a5cd | 391 | if (taskexit) |
bd89aabc PM |
392 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
393 | check_for_release(cgrp); | |
81a6a5cd | 394 | } |
2c6ab6d2 PM |
395 | |
396 | kfree(link); | |
81a6a5cd | 397 | } |
2c6ab6d2 | 398 | |
5abb8855 | 399 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
400 | } |
401 | ||
89c5509b TH |
402 | static void put_css_set(struct css_set *cset, bool taskexit) |
403 | { | |
404 | /* | |
405 | * Ensure that the refcount doesn't hit zero while any readers | |
406 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
407 | * rwlock | |
408 | */ | |
409 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
410 | return; | |
411 | ||
412 | down_write(&css_set_rwsem); | |
413 | put_css_set_locked(cset, taskexit); | |
414 | up_write(&css_set_rwsem); | |
415 | } | |
416 | ||
817929ec PM |
417 | /* |
418 | * refcounted get/put for css_set objects | |
419 | */ | |
5abb8855 | 420 | static inline void get_css_set(struct css_set *cset) |
817929ec | 421 | { |
5abb8855 | 422 | atomic_inc(&cset->refcount); |
817929ec PM |
423 | } |
424 | ||
b326f9d0 | 425 | /** |
7717f7ba | 426 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
427 | * @cset: candidate css_set being tested |
428 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
429 | * @new_cgrp: cgroup that's being entered by the task |
430 | * @template: desired set of css pointers in css_set (pre-calculated) | |
431 | * | |
6f4b7e63 | 432 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
433 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
434 | */ | |
5abb8855 TH |
435 | static bool compare_css_sets(struct css_set *cset, |
436 | struct css_set *old_cset, | |
7717f7ba PM |
437 | struct cgroup *new_cgrp, |
438 | struct cgroup_subsys_state *template[]) | |
439 | { | |
440 | struct list_head *l1, *l2; | |
441 | ||
5abb8855 | 442 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) { |
7717f7ba PM |
443 | /* Not all subsystems matched */ |
444 | return false; | |
445 | } | |
446 | ||
447 | /* | |
448 | * Compare cgroup pointers in order to distinguish between | |
449 | * different cgroups in heirarchies with no subsystems. We | |
450 | * could get by with just this check alone (and skip the | |
451 | * memcmp above) but on most setups the memcmp check will | |
452 | * avoid the need for this more expensive check on almost all | |
453 | * candidates. | |
454 | */ | |
455 | ||
69d0206c TH |
456 | l1 = &cset->cgrp_links; |
457 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 458 | while (1) { |
69d0206c | 459 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 460 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
461 | |
462 | l1 = l1->next; | |
463 | l2 = l2->next; | |
464 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
465 | if (l1 == &cset->cgrp_links) { |
466 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
467 | break; |
468 | } else { | |
69d0206c | 469 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
470 | } |
471 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
472 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
473 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
474 | cgrp1 = link1->cgrp; | |
475 | cgrp2 = link2->cgrp; | |
7717f7ba | 476 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 477 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
478 | |
479 | /* | |
480 | * If this hierarchy is the hierarchy of the cgroup | |
481 | * that's changing, then we need to check that this | |
482 | * css_set points to the new cgroup; if it's any other | |
483 | * hierarchy, then this css_set should point to the | |
484 | * same cgroup as the old css_set. | |
485 | */ | |
5abb8855 TH |
486 | if (cgrp1->root == new_cgrp->root) { |
487 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
488 | return false; |
489 | } else { | |
5abb8855 | 490 | if (cgrp1 != cgrp2) |
7717f7ba PM |
491 | return false; |
492 | } | |
493 | } | |
494 | return true; | |
495 | } | |
496 | ||
b326f9d0 TH |
497 | /** |
498 | * find_existing_css_set - init css array and find the matching css_set | |
499 | * @old_cset: the css_set that we're using before the cgroup transition | |
500 | * @cgrp: the cgroup that we're moving into | |
501 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 502 | */ |
5abb8855 TH |
503 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
504 | struct cgroup *cgrp, | |
505 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 506 | { |
bd89aabc | 507 | struct cgroupfs_root *root = cgrp->root; |
30159ec7 | 508 | struct cgroup_subsys *ss; |
5abb8855 | 509 | struct css_set *cset; |
0ac801fe | 510 | unsigned long key; |
b326f9d0 | 511 | int i; |
817929ec | 512 | |
aae8aab4 BB |
513 | /* |
514 | * Build the set of subsystem state objects that we want to see in the | |
515 | * new css_set. while subsystems can change globally, the entries here | |
516 | * won't change, so no need for locking. | |
517 | */ | |
30159ec7 | 518 | for_each_subsys(ss, i) { |
a1a71b45 | 519 | if (root->subsys_mask & (1UL << i)) { |
817929ec PM |
520 | /* Subsystem is in this hierarchy. So we want |
521 | * the subsystem state from the new | |
522 | * cgroup */ | |
ca8bdcaf | 523 | template[i] = cgroup_css(cgrp, ss); |
817929ec PM |
524 | } else { |
525 | /* Subsystem is not in this hierarchy, so we | |
526 | * don't want to change the subsystem state */ | |
5abb8855 | 527 | template[i] = old_cset->subsys[i]; |
817929ec PM |
528 | } |
529 | } | |
530 | ||
0ac801fe | 531 | key = css_set_hash(template); |
5abb8855 TH |
532 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
533 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
534 | continue; |
535 | ||
536 | /* This css_set matches what we need */ | |
5abb8855 | 537 | return cset; |
472b1053 | 538 | } |
817929ec PM |
539 | |
540 | /* No existing cgroup group matched */ | |
541 | return NULL; | |
542 | } | |
543 | ||
69d0206c | 544 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 545 | { |
69d0206c | 546 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 547 | |
69d0206c TH |
548 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
549 | list_del(&link->cset_link); | |
36553434 LZ |
550 | kfree(link); |
551 | } | |
552 | } | |
553 | ||
69d0206c TH |
554 | /** |
555 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
556 | * @count: the number of links to allocate | |
557 | * @tmp_links: list_head the allocated links are put on | |
558 | * | |
559 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
560 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 561 | */ |
69d0206c | 562 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 563 | { |
69d0206c | 564 | struct cgrp_cset_link *link; |
817929ec | 565 | int i; |
69d0206c TH |
566 | |
567 | INIT_LIST_HEAD(tmp_links); | |
568 | ||
817929ec | 569 | for (i = 0; i < count; i++) { |
f4f4be2b | 570 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 571 | if (!link) { |
69d0206c | 572 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
573 | return -ENOMEM; |
574 | } | |
69d0206c | 575 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
576 | } |
577 | return 0; | |
578 | } | |
579 | ||
c12f65d4 LZ |
580 | /** |
581 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 582 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 583 | * @cset: the css_set to be linked |
c12f65d4 LZ |
584 | * @cgrp: the destination cgroup |
585 | */ | |
69d0206c TH |
586 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
587 | struct cgroup *cgrp) | |
c12f65d4 | 588 | { |
69d0206c | 589 | struct cgrp_cset_link *link; |
c12f65d4 | 590 | |
69d0206c TH |
591 | BUG_ON(list_empty(tmp_links)); |
592 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); | |
593 | link->cset = cset; | |
7717f7ba | 594 | link->cgrp = cgrp; |
69d0206c | 595 | list_move(&link->cset_link, &cgrp->cset_links); |
7717f7ba PM |
596 | /* |
597 | * Always add links to the tail of the list so that the list | |
598 | * is sorted by order of hierarchy creation | |
599 | */ | |
69d0206c | 600 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
601 | } |
602 | ||
b326f9d0 TH |
603 | /** |
604 | * find_css_set - return a new css_set with one cgroup updated | |
605 | * @old_cset: the baseline css_set | |
606 | * @cgrp: the cgroup to be updated | |
607 | * | |
608 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
609 | * substituted into the appropriate hierarchy. | |
817929ec | 610 | */ |
5abb8855 TH |
611 | static struct css_set *find_css_set(struct css_set *old_cset, |
612 | struct cgroup *cgrp) | |
817929ec | 613 | { |
b326f9d0 | 614 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 615 | struct css_set *cset; |
69d0206c TH |
616 | struct list_head tmp_links; |
617 | struct cgrp_cset_link *link; | |
0ac801fe | 618 | unsigned long key; |
472b1053 | 619 | |
b326f9d0 TH |
620 | lockdep_assert_held(&cgroup_mutex); |
621 | ||
817929ec PM |
622 | /* First see if we already have a cgroup group that matches |
623 | * the desired set */ | |
96d365e0 | 624 | down_read(&css_set_rwsem); |
5abb8855 TH |
625 | cset = find_existing_css_set(old_cset, cgrp, template); |
626 | if (cset) | |
627 | get_css_set(cset); | |
96d365e0 | 628 | up_read(&css_set_rwsem); |
817929ec | 629 | |
5abb8855 TH |
630 | if (cset) |
631 | return cset; | |
817929ec | 632 | |
f4f4be2b | 633 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 634 | if (!cset) |
817929ec PM |
635 | return NULL; |
636 | ||
69d0206c | 637 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 638 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 639 | kfree(cset); |
817929ec PM |
640 | return NULL; |
641 | } | |
642 | ||
5abb8855 | 643 | atomic_set(&cset->refcount, 1); |
69d0206c | 644 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 645 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 646 | INIT_LIST_HEAD(&cset->mg_tasks); |
b3dc094e | 647 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 648 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
649 | |
650 | /* Copy the set of subsystem state objects generated in | |
651 | * find_existing_css_set() */ | |
5abb8855 | 652 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 653 | |
96d365e0 | 654 | down_write(&css_set_rwsem); |
817929ec | 655 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 656 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 657 | struct cgroup *c = link->cgrp; |
69d0206c | 658 | |
7717f7ba PM |
659 | if (c->root == cgrp->root) |
660 | c = cgrp; | |
69d0206c | 661 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 662 | } |
817929ec | 663 | |
69d0206c | 664 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 665 | |
817929ec | 666 | css_set_count++; |
472b1053 LZ |
667 | |
668 | /* Add this cgroup group to the hash table */ | |
5abb8855 TH |
669 | key = css_set_hash(cset->subsys); |
670 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 671 | |
96d365e0 | 672 | up_write(&css_set_rwsem); |
817929ec | 673 | |
5abb8855 | 674 | return cset; |
b4f48b63 PM |
675 | } |
676 | ||
2bd59d48 TH |
677 | static struct cgroupfs_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
678 | { | |
679 | struct cgroup *top_cgrp = kf_root->kn->priv; | |
680 | ||
681 | return top_cgrp->root; | |
682 | } | |
683 | ||
f2e85d57 TH |
684 | static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end) |
685 | { | |
686 | int id; | |
687 | ||
688 | lockdep_assert_held(&cgroup_mutex); | |
689 | ||
690 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end, | |
691 | GFP_KERNEL); | |
692 | if (id < 0) | |
693 | return id; | |
694 | ||
695 | root->hierarchy_id = id; | |
696 | return 0; | |
697 | } | |
698 | ||
699 | static void cgroup_exit_root_id(struct cgroupfs_root *root) | |
700 | { | |
701 | lockdep_assert_held(&cgroup_mutex); | |
702 | ||
703 | if (root->hierarchy_id) { | |
704 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
705 | root->hierarchy_id = 0; | |
706 | } | |
707 | } | |
708 | ||
709 | static void cgroup_free_root(struct cgroupfs_root *root) | |
710 | { | |
711 | if (root) { | |
712 | /* hierarhcy ID shoulid already have been released */ | |
713 | WARN_ON_ONCE(root->hierarchy_id); | |
714 | ||
715 | idr_destroy(&root->cgroup_idr); | |
716 | kfree(root); | |
717 | } | |
718 | } | |
719 | ||
776f02fa | 720 | static void cgroup_destroy_root(struct cgroupfs_root *root) |
59f5296b | 721 | { |
f2e85d57 TH |
722 | struct cgroup *cgrp = &root->top_cgroup; |
723 | struct cgrp_cset_link *link, *tmp_link; | |
f2e85d57 | 724 | |
2bd59d48 | 725 | mutex_lock(&cgroup_tree_mutex); |
2bd59d48 | 726 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 727 | |
776f02fa | 728 | BUG_ON(atomic_read(&root->nr_cgrps)); |
f2e85d57 TH |
729 | BUG_ON(!list_empty(&cgrp->children)); |
730 | ||
f2e85d57 | 731 | /* Rebind all subsystems back to the default hierarchy */ |
35585573 | 732 | WARN_ON(rebind_subsystems(root, 0, root->subsys_mask)); |
f2e85d57 TH |
733 | |
734 | /* | |
735 | * Release all the links from cset_links to this hierarchy's | |
736 | * root cgroup | |
737 | */ | |
96d365e0 | 738 | down_write(&css_set_rwsem); |
f2e85d57 TH |
739 | |
740 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
741 | list_del(&link->cset_link); | |
742 | list_del(&link->cgrp_link); | |
743 | kfree(link); | |
744 | } | |
96d365e0 | 745 | up_write(&css_set_rwsem); |
f2e85d57 TH |
746 | |
747 | if (!list_empty(&root->root_list)) { | |
748 | list_del(&root->root_list); | |
749 | cgroup_root_count--; | |
750 | } | |
751 | ||
752 | cgroup_exit_root_id(root); | |
753 | ||
754 | mutex_unlock(&cgroup_mutex); | |
755 | mutex_unlock(&cgroup_tree_mutex); | |
f2e85d57 | 756 | |
2bd59d48 | 757 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
758 | cgroup_free_root(root); |
759 | } | |
760 | ||
ceb6a081 TH |
761 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
762 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
7717f7ba PM |
763 | struct cgroupfs_root *root) |
764 | { | |
7717f7ba PM |
765 | struct cgroup *res = NULL; |
766 | ||
96d365e0 TH |
767 | lockdep_assert_held(&cgroup_mutex); |
768 | lockdep_assert_held(&css_set_rwsem); | |
769 | ||
5abb8855 | 770 | if (cset == &init_css_set) { |
7717f7ba PM |
771 | res = &root->top_cgroup; |
772 | } else { | |
69d0206c TH |
773 | struct cgrp_cset_link *link; |
774 | ||
775 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 776 | struct cgroup *c = link->cgrp; |
69d0206c | 777 | |
7717f7ba PM |
778 | if (c->root == root) { |
779 | res = c; | |
780 | break; | |
781 | } | |
782 | } | |
783 | } | |
96d365e0 | 784 | |
7717f7ba PM |
785 | BUG_ON(!res); |
786 | return res; | |
787 | } | |
788 | ||
ceb6a081 TH |
789 | /* |
790 | * Return the cgroup for "task" from the given hierarchy. Must be | |
791 | * called with cgroup_mutex and css_set_rwsem held. | |
792 | */ | |
793 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
794 | struct cgroupfs_root *root) | |
795 | { | |
796 | /* | |
797 | * No need to lock the task - since we hold cgroup_mutex the | |
798 | * task can't change groups, so the only thing that can happen | |
799 | * is that it exits and its css is set back to init_css_set. | |
800 | */ | |
801 | return cset_cgroup_from_root(task_css_set(task), root); | |
802 | } | |
803 | ||
ddbcc7e8 PM |
804 | /* |
805 | * There is one global cgroup mutex. We also require taking | |
806 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
807 | * See "The task_lock() exception", at the end of this comment. | |
808 | * | |
809 | * A task must hold cgroup_mutex to modify cgroups. | |
810 | * | |
811 | * Any task can increment and decrement the count field without lock. | |
812 | * So in general, code holding cgroup_mutex can't rely on the count | |
813 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 814 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
815 | * means that no tasks are currently attached, therefore there is no |
816 | * way a task attached to that cgroup can fork (the other way to | |
817 | * increment the count). So code holding cgroup_mutex can safely | |
818 | * assume that if the count is zero, it will stay zero. Similarly, if | |
819 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
820 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
821 | * needs that mutex. | |
822 | * | |
ddbcc7e8 PM |
823 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
824 | * (usually) take cgroup_mutex. These are the two most performance | |
825 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
826 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
827 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
828 | * to the release agent with the name of the cgroup (path relative to |
829 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
830 | * |
831 | * A cgroup can only be deleted if both its 'count' of using tasks | |
832 | * is zero, and its list of 'children' cgroups is empty. Since all | |
833 | * tasks in the system use _some_ cgroup, and since there is always at | |
834 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
835 | * always has either children cgroups and/or using tasks. So we don't | |
836 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
837 | * | |
838 | * The task_lock() exception | |
839 | * | |
840 | * The need for this exception arises from the action of | |
d0b2fdd2 | 841 | * cgroup_attach_task(), which overwrites one task's cgroup pointer with |
a043e3b2 | 842 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
843 | * several performance critical places that need to reference |
844 | * task->cgroup without the expense of grabbing a system global | |
845 | * mutex. Therefore except as noted below, when dereferencing or, as | |
d0b2fdd2 | 846 | * in cgroup_attach_task(), modifying a task's cgroup pointer we use |
ddbcc7e8 PM |
847 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
848 | * the task_struct routinely used for such matters. | |
849 | * | |
850 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 851 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
852 | */ |
853 | ||
628f7cd4 | 854 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
2bd59d48 | 855 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 856 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 857 | |
8d7e6fb0 TH |
858 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
859 | char *buf) | |
860 | { | |
861 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && | |
862 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
863 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
864 | cft->ss->name, cft->name); | |
865 | else | |
866 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
867 | return buf; | |
868 | } | |
869 | ||
f2e85d57 TH |
870 | /** |
871 | * cgroup_file_mode - deduce file mode of a control file | |
872 | * @cft: the control file in question | |
873 | * | |
874 | * returns cft->mode if ->mode is not 0 | |
875 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
876 | * returns S_IRUGO if it has only a read handler | |
877 | * returns S_IWUSR if it has only a write hander | |
878 | */ | |
879 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
880 | { | |
881 | umode_t mode = 0; | |
882 | ||
883 | if (cft->mode) | |
884 | return cft->mode; | |
885 | ||
886 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
887 | mode |= S_IRUGO; | |
888 | ||
889 | if (cft->write_u64 || cft->write_s64 || cft->write_string || | |
890 | cft->trigger) | |
891 | mode |= S_IWUSR; | |
892 | ||
893 | return mode; | |
894 | } | |
895 | ||
be445626 LZ |
896 | static void cgroup_free_fn(struct work_struct *work) |
897 | { | |
ea15f8cc | 898 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 | 899 | |
3c9c825b | 900 | atomic_dec(&cgrp->root->nr_cgrps); |
b1a21367 | 901 | cgroup_pidlist_destroy_all(cgrp); |
be445626 | 902 | |
776f02fa TH |
903 | if (cgrp->parent) { |
904 | /* | |
905 | * We get a ref to the parent, and put the ref when this | |
906 | * cgroup is being freed, so it's guaranteed that the | |
907 | * parent won't be destroyed before its children. | |
908 | */ | |
909 | cgroup_put(cgrp->parent); | |
910 | kernfs_put(cgrp->kn); | |
911 | kfree(cgrp); | |
912 | } else { | |
913 | /* | |
914 | * This is top cgroup's refcnt reaching zero, which | |
915 | * indicates that the root should be released. | |
916 | */ | |
917 | cgroup_destroy_root(cgrp->root); | |
918 | } | |
be445626 LZ |
919 | } |
920 | ||
921 | static void cgroup_free_rcu(struct rcu_head *head) | |
922 | { | |
923 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
924 | ||
ea15f8cc | 925 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 926 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
927 | } |
928 | ||
59f5296b TH |
929 | static void cgroup_get(struct cgroup *cgrp) |
930 | { | |
2bd59d48 TH |
931 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
932 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
933 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
934 | } |
935 | ||
59f5296b TH |
936 | static void cgroup_put(struct cgroup *cgrp) |
937 | { | |
2bd59d48 TH |
938 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
939 | return; | |
776f02fa | 940 | if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp))) |
2bd59d48 | 941 | return; |
59f5296b | 942 | |
2bd59d48 TH |
943 | /* |
944 | * XXX: cgrp->id is only used to look up css's. As cgroup and | |
945 | * css's lifetimes will be decoupled, it should be made | |
946 | * per-subsystem and moved to css->id so that lookups are | |
947 | * successful until the target css is released. | |
948 | */ | |
949 | mutex_lock(&cgroup_mutex); | |
950 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
951 | mutex_unlock(&cgroup_mutex); | |
952 | cgrp->id = -1; | |
ddbcc7e8 | 953 | |
2bd59d48 | 954 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 PM |
955 | } |
956 | ||
2739d3cc | 957 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 958 | { |
2bd59d48 | 959 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 960 | |
ace2bee8 | 961 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 962 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
963 | } |
964 | ||
13af07df | 965 | /** |
628f7cd4 | 966 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 967 | * @cgrp: target cgroup |
13af07df AR |
968 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
969 | */ | |
628f7cd4 | 970 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 971 | { |
13af07df | 972 | struct cgroup_subsys *ss; |
b420ba7d | 973 | int i; |
05ef1d7c | 974 | |
b420ba7d | 975 | for_each_subsys(ss, i) { |
0adb0704 | 976 | struct cftype *cfts; |
b420ba7d TH |
977 | |
978 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 979 | continue; |
0adb0704 TH |
980 | list_for_each_entry(cfts, &ss->cfts, node) |
981 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 982 | } |
ddbcc7e8 PM |
983 | } |
984 | ||
ddbcc7e8 | 985 | static int rebind_subsystems(struct cgroupfs_root *root, |
a8a648c4 | 986 | unsigned long added_mask, unsigned removed_mask) |
ddbcc7e8 | 987 | { |
bd89aabc | 988 | struct cgroup *cgrp = &root->top_cgroup; |
30159ec7 | 989 | struct cgroup_subsys *ss; |
3126121f | 990 | int i, ret; |
ddbcc7e8 | 991 | |
ace2bee8 TH |
992 | lockdep_assert_held(&cgroup_tree_mutex); |
993 | lockdep_assert_held(&cgroup_mutex); | |
aae8aab4 | 994 | |
ddbcc7e8 | 995 | /* Check that any added subsystems are currently free */ |
3ed80a62 TH |
996 | for_each_subsys(ss, i) |
997 | if ((added_mask & (1 << i)) && ss->root != &cgroup_dummy_root) | |
998 | return -EBUSY; | |
ddbcc7e8 | 999 | |
3126121f TH |
1000 | ret = cgroup_populate_dir(cgrp, added_mask); |
1001 | if (ret) | |
3ed80a62 | 1002 | return ret; |
3126121f TH |
1003 | |
1004 | /* | |
1005 | * Nothing can fail from this point on. Remove files for the | |
1006 | * removed subsystems and rebind each subsystem. | |
1007 | */ | |
4ac06017 | 1008 | mutex_unlock(&cgroup_mutex); |
3126121f | 1009 | cgroup_clear_dir(cgrp, removed_mask); |
4ac06017 | 1010 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1011 | |
30159ec7 | 1012 | for_each_subsys(ss, i) { |
ddbcc7e8 | 1013 | unsigned long bit = 1UL << i; |
30159ec7 | 1014 | |
a1a71b45 | 1015 | if (bit & added_mask) { |
ddbcc7e8 | 1016 | /* We're binding this subsystem to this hierarchy */ |
ca8bdcaf TH |
1017 | BUG_ON(cgroup_css(cgrp, ss)); |
1018 | BUG_ON(!cgroup_css(cgroup_dummy_top, ss)); | |
1019 | BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top); | |
a8a648c4 | 1020 | |
73e80ed8 | 1021 | rcu_assign_pointer(cgrp->subsys[i], |
ca8bdcaf TH |
1022 | cgroup_css(cgroup_dummy_top, ss)); |
1023 | cgroup_css(cgrp, ss)->cgroup = cgrp; | |
a8a648c4 | 1024 | |
b2aa30f7 | 1025 | ss->root = root; |
ddbcc7e8 | 1026 | if (ss->bind) |
ca8bdcaf | 1027 | ss->bind(cgroup_css(cgrp, ss)); |
a8a648c4 | 1028 | |
cf5d5941 | 1029 | /* refcount was already taken, and we're keeping it */ |
a8a648c4 | 1030 | root->subsys_mask |= bit; |
a1a71b45 | 1031 | } else if (bit & removed_mask) { |
ddbcc7e8 | 1032 | /* We're removing this subsystem */ |
ca8bdcaf TH |
1033 | BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss)); |
1034 | BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp); | |
a8a648c4 | 1035 | |
ddbcc7e8 | 1036 | if (ss->bind) |
ca8bdcaf | 1037 | ss->bind(cgroup_css(cgroup_dummy_top, ss)); |
73e80ed8 | 1038 | |
ca8bdcaf | 1039 | cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top; |
73e80ed8 TH |
1040 | RCU_INIT_POINTER(cgrp->subsys[i], NULL); |
1041 | ||
9871bf95 | 1042 | cgroup_subsys[i]->root = &cgroup_dummy_root; |
a8a648c4 | 1043 | root->subsys_mask &= ~bit; |
ddbcc7e8 PM |
1044 | } |
1045 | } | |
ddbcc7e8 | 1046 | |
2bd59d48 | 1047 | kernfs_activate(cgrp->kn); |
ddbcc7e8 PM |
1048 | return 0; |
1049 | } | |
1050 | ||
2bd59d48 TH |
1051 | static int cgroup_show_options(struct seq_file *seq, |
1052 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1053 | { |
2bd59d48 | 1054 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1055 | struct cgroup_subsys *ss; |
b85d2040 | 1056 | int ssid; |
ddbcc7e8 | 1057 | |
b85d2040 TH |
1058 | for_each_subsys(ss, ssid) |
1059 | if (root->subsys_mask & (1 << ssid)) | |
1060 | seq_printf(seq, ",%s", ss->name); | |
873fe09e TH |
1061 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1062 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1063 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1064 | seq_puts(seq, ",noprefix"); |
93438629 | 1065 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1066 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1067 | |
1068 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1069 | if (strlen(root->release_agent_path)) |
1070 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1071 | spin_unlock(&release_agent_path_lock); |
1072 | ||
2260e7fc | 1073 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags)) |
97978e6d | 1074 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1075 | if (strlen(root->name)) |
1076 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1077 | return 0; |
1078 | } | |
1079 | ||
1080 | struct cgroup_sb_opts { | |
a1a71b45 | 1081 | unsigned long subsys_mask; |
ddbcc7e8 | 1082 | unsigned long flags; |
81a6a5cd | 1083 | char *release_agent; |
2260e7fc | 1084 | bool cpuset_clone_children; |
c6d57f33 | 1085 | char *name; |
2c6ab6d2 PM |
1086 | /* User explicitly requested empty subsystem */ |
1087 | bool none; | |
ddbcc7e8 PM |
1088 | }; |
1089 | ||
aae8aab4 | 1090 | /* |
9871bf95 TH |
1091 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1092 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1093 | * array. This function takes refcounts on subsystems to be used, unless it | |
1094 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1095 | */ |
cf5d5941 | 1096 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1097 | { |
32a8cf23 DL |
1098 | char *token, *o = data; |
1099 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1100 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1101 | struct cgroup_subsys *ss; |
1102 | int i; | |
f9ab5b5b | 1103 | |
aae8aab4 BB |
1104 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1105 | ||
f9ab5b5b | 1106 | #ifdef CONFIG_CPUSETS |
073219e9 | 1107 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1108 | #endif |
ddbcc7e8 | 1109 | |
c6d57f33 | 1110 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1111 | |
1112 | while ((token = strsep(&o, ",")) != NULL) { | |
1113 | if (!*token) | |
1114 | return -EINVAL; | |
32a8cf23 | 1115 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1116 | /* Explicitly have no subsystems */ |
1117 | opts->none = true; | |
32a8cf23 DL |
1118 | continue; |
1119 | } | |
1120 | if (!strcmp(token, "all")) { | |
1121 | /* Mutually exclusive option 'all' + subsystem name */ | |
1122 | if (one_ss) | |
1123 | return -EINVAL; | |
1124 | all_ss = true; | |
1125 | continue; | |
1126 | } | |
873fe09e TH |
1127 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1128 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1129 | continue; | |
1130 | } | |
32a8cf23 | 1131 | if (!strcmp(token, "noprefix")) { |
93438629 | 1132 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1133 | continue; |
1134 | } | |
1135 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1136 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1137 | continue; |
1138 | } | |
03b1cde6 | 1139 | if (!strcmp(token, "xattr")) { |
93438629 | 1140 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1141 | continue; |
1142 | } | |
32a8cf23 | 1143 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1144 | /* Specifying two release agents is forbidden */ |
1145 | if (opts->release_agent) | |
1146 | return -EINVAL; | |
c6d57f33 | 1147 | opts->release_agent = |
e400c285 | 1148 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1149 | if (!opts->release_agent) |
1150 | return -ENOMEM; | |
32a8cf23 DL |
1151 | continue; |
1152 | } | |
1153 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1154 | const char *name = token + 5; |
1155 | /* Can't specify an empty name */ | |
1156 | if (!strlen(name)) | |
1157 | return -EINVAL; | |
1158 | /* Must match [\w.-]+ */ | |
1159 | for (i = 0; i < strlen(name); i++) { | |
1160 | char c = name[i]; | |
1161 | if (isalnum(c)) | |
1162 | continue; | |
1163 | if ((c == '.') || (c == '-') || (c == '_')) | |
1164 | continue; | |
1165 | return -EINVAL; | |
1166 | } | |
1167 | /* Specifying two names is forbidden */ | |
1168 | if (opts->name) | |
1169 | return -EINVAL; | |
1170 | opts->name = kstrndup(name, | |
e400c285 | 1171 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1172 | GFP_KERNEL); |
1173 | if (!opts->name) | |
1174 | return -ENOMEM; | |
32a8cf23 DL |
1175 | |
1176 | continue; | |
1177 | } | |
1178 | ||
30159ec7 | 1179 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1180 | if (strcmp(token, ss->name)) |
1181 | continue; | |
1182 | if (ss->disabled) | |
1183 | continue; | |
1184 | ||
1185 | /* Mutually exclusive option 'all' + subsystem name */ | |
1186 | if (all_ss) | |
1187 | return -EINVAL; | |
a1a71b45 | 1188 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1189 | one_ss = true; |
1190 | ||
1191 | break; | |
1192 | } | |
1193 | if (i == CGROUP_SUBSYS_COUNT) | |
1194 | return -ENOENT; | |
1195 | } | |
1196 | ||
1197 | /* | |
1198 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1199 | * otherwise if 'none', 'name=' and a subsystem name options |
1200 | * were not specified, let's default to 'all' | |
32a8cf23 | 1201 | */ |
30159ec7 TH |
1202 | if (all_ss || (!one_ss && !opts->none && !opts->name)) |
1203 | for_each_subsys(ss, i) | |
1204 | if (!ss->disabled) | |
1205 | set_bit(i, &opts->subsys_mask); | |
ddbcc7e8 | 1206 | |
2c6ab6d2 PM |
1207 | /* Consistency checks */ |
1208 | ||
873fe09e TH |
1209 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1210 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1211 | ||
d3ba07c3 TH |
1212 | if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) || |
1213 | opts->cpuset_clone_children || opts->release_agent || | |
1214 | opts->name) { | |
1215 | pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n"); | |
873fe09e TH |
1216 | return -EINVAL; |
1217 | } | |
873fe09e TH |
1218 | } |
1219 | ||
f9ab5b5b LZ |
1220 | /* |
1221 | * Option noprefix was introduced just for backward compatibility | |
1222 | * with the old cpuset, so we allow noprefix only if mounting just | |
1223 | * the cpuset subsystem. | |
1224 | */ | |
93438629 | 1225 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1226 | return -EINVAL; |
1227 | ||
2c6ab6d2 PM |
1228 | |
1229 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1230 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1231 | return -EINVAL; |
1232 | ||
1233 | /* | |
1234 | * We either have to specify by name or by subsystems. (So all | |
1235 | * empty hierarchies must have a name). | |
1236 | */ | |
a1a71b45 | 1237 | if (!opts->subsys_mask && !opts->name) |
ddbcc7e8 PM |
1238 | return -EINVAL; |
1239 | ||
1240 | return 0; | |
1241 | } | |
1242 | ||
2bd59d48 | 1243 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1244 | { |
1245 | int ret = 0; | |
2bd59d48 | 1246 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1247 | struct cgroup_sb_opts opts; |
a1a71b45 | 1248 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1249 | |
873fe09e TH |
1250 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1251 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1252 | return -EINVAL; | |
1253 | } | |
1254 | ||
ace2bee8 | 1255 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1256 | mutex_lock(&cgroup_mutex); |
1257 | ||
1258 | /* See what subsystems are wanted */ | |
1259 | ret = parse_cgroupfs_options(data, &opts); | |
1260 | if (ret) | |
1261 | goto out_unlock; | |
1262 | ||
a8a648c4 | 1263 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1264 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1265 | task_tgid_nr(current), current->comm); | |
1266 | ||
a1a71b45 AR |
1267 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1268 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1269 | |
cf5d5941 | 1270 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1271 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1272 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1273 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1274 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1275 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1276 | ret = -EINVAL; |
1277 | goto out_unlock; | |
1278 | } | |
1279 | ||
f172e67c | 1280 | /* remounting is not allowed for populated hierarchies */ |
3c9c825b | 1281 | if (!list_empty(&root->top_cgroup.children)) { |
f172e67c | 1282 | ret = -EBUSY; |
0670e08b | 1283 | goto out_unlock; |
cf5d5941 | 1284 | } |
ddbcc7e8 | 1285 | |
a8a648c4 | 1286 | ret = rebind_subsystems(root, added_mask, removed_mask); |
3126121f | 1287 | if (ret) |
0670e08b | 1288 | goto out_unlock; |
ddbcc7e8 | 1289 | |
69e943b7 TH |
1290 | if (opts.release_agent) { |
1291 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1292 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1293 | spin_unlock(&release_agent_path_lock); |
1294 | } | |
ddbcc7e8 | 1295 | out_unlock: |
66bdc9cf | 1296 | kfree(opts.release_agent); |
c6d57f33 | 1297 | kfree(opts.name); |
ddbcc7e8 | 1298 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1299 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1300 | return ret; |
1301 | } | |
1302 | ||
afeb0f9f TH |
1303 | /* |
1304 | * To reduce the fork() overhead for systems that are not actually using | |
1305 | * their cgroups capability, we don't maintain the lists running through | |
1306 | * each css_set to its tasks until we see the list actually used - in other | |
1307 | * words after the first mount. | |
1308 | */ | |
1309 | static bool use_task_css_set_links __read_mostly; | |
1310 | ||
1311 | static void cgroup_enable_task_cg_lists(void) | |
1312 | { | |
1313 | struct task_struct *p, *g; | |
1314 | ||
96d365e0 | 1315 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1316 | |
1317 | if (use_task_css_set_links) | |
1318 | goto out_unlock; | |
1319 | ||
1320 | use_task_css_set_links = true; | |
1321 | ||
1322 | /* | |
1323 | * We need tasklist_lock because RCU is not safe against | |
1324 | * while_each_thread(). Besides, a forking task that has passed | |
1325 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1326 | * is not guaranteed to have its child immediately visible in the | |
1327 | * tasklist if we walk through it with RCU. | |
1328 | */ | |
1329 | read_lock(&tasklist_lock); | |
1330 | do_each_thread(g, p) { | |
1331 | task_lock(p); | |
1332 | ||
1333 | WARN_ON_ONCE(!list_empty(&p->cg_list) || | |
1334 | task_css_set(p) != &init_css_set); | |
1335 | ||
1336 | /* | |
1337 | * We should check if the process is exiting, otherwise | |
1338 | * it will race with cgroup_exit() in that the list | |
1339 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1340 | * Do it while holding siglock so that we don't end up |
1341 | * racing against cgroup_exit(). | |
afeb0f9f | 1342 | */ |
f153ad11 | 1343 | spin_lock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1344 | if (!(p->flags & PF_EXITING)) |
1345 | list_add(&p->cg_list, &task_css_set(p)->tasks); | |
f153ad11 | 1346 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1347 | |
1348 | task_unlock(p); | |
1349 | } while_each_thread(g, p); | |
1350 | read_unlock(&tasklist_lock); | |
1351 | out_unlock: | |
96d365e0 | 1352 | up_write(&css_set_rwsem); |
afeb0f9f TH |
1353 | } |
1354 | ||
cc31edce PM |
1355 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1356 | { | |
2bd59d48 | 1357 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1358 | INIT_LIST_HEAD(&cgrp->sibling); |
1359 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1360 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1361 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1362 | INIT_LIST_HEAD(&cgrp->pidlists); |
1363 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1364 | cgrp->dummy_css.cgroup = cgrp; |
cc31edce | 1365 | } |
c6d57f33 | 1366 | |
ddbcc7e8 PM |
1367 | static void init_cgroup_root(struct cgroupfs_root *root) |
1368 | { | |
bd89aabc | 1369 | struct cgroup *cgrp = &root->top_cgroup; |
b0ca5a84 | 1370 | |
ddbcc7e8 | 1371 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1372 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1373 | cgrp->root = root; |
cc31edce | 1374 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1375 | idr_init(&root->cgroup_idr); |
ddbcc7e8 PM |
1376 | } |
1377 | ||
c6d57f33 PM |
1378 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1379 | { | |
1380 | struct cgroupfs_root *root; | |
1381 | ||
a1a71b45 | 1382 | if (!opts->subsys_mask && !opts->none) |
2bd59d48 | 1383 | return ERR_PTR(-EINVAL); |
c6d57f33 PM |
1384 | |
1385 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1386 | if (!root) | |
1387 | return ERR_PTR(-ENOMEM); | |
1388 | ||
1389 | init_cgroup_root(root); | |
2c6ab6d2 | 1390 | |
c6d57f33 PM |
1391 | root->flags = opts->flags; |
1392 | if (opts->release_agent) | |
1393 | strcpy(root->release_agent_path, opts->release_agent); | |
1394 | if (opts->name) | |
1395 | strcpy(root->name, opts->name); | |
2260e7fc TH |
1396 | if (opts->cpuset_clone_children) |
1397 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1398 | return root; |
1399 | } | |
1400 | ||
35585573 | 1401 | static int cgroup_setup_root(struct cgroupfs_root *root, unsigned long ss_mask) |
d427dfeb TH |
1402 | { |
1403 | LIST_HEAD(tmp_links); | |
d427dfeb | 1404 | struct cgroup *root_cgrp = &root->top_cgroup; |
d427dfeb | 1405 | struct css_set *cset; |
d427dfeb TH |
1406 | int i, ret; |
1407 | ||
1408 | lockdep_assert_held(&cgroup_tree_mutex); | |
1409 | lockdep_assert_held(&cgroup_mutex); | |
d427dfeb TH |
1410 | |
1411 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); | |
1412 | if (ret < 0) | |
2bd59d48 | 1413 | goto out; |
d427dfeb TH |
1414 | root_cgrp->id = ret; |
1415 | ||
d427dfeb | 1416 | /* |
96d365e0 | 1417 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1418 | * but that's OK - it can only be increased by someone holding |
1419 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1420 | * have some link structures left over | |
1421 | */ | |
1422 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1423 | if (ret) | |
2bd59d48 | 1424 | goto out; |
d427dfeb TH |
1425 | |
1426 | /* ID 0 is reserved for dummy root, 1 for unified hierarchy */ | |
1427 | ret = cgroup_init_root_id(root, 2, 0); | |
1428 | if (ret) | |
2bd59d48 | 1429 | goto out; |
d427dfeb | 1430 | |
2bd59d48 TH |
1431 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1432 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1433 | root_cgrp); | |
1434 | if (IS_ERR(root->kf_root)) { | |
1435 | ret = PTR_ERR(root->kf_root); | |
1436 | goto exit_root_id; | |
1437 | } | |
1438 | root_cgrp->kn = root->kf_root->kn; | |
d427dfeb TH |
1439 | |
1440 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); | |
1441 | if (ret) | |
2bd59d48 | 1442 | goto destroy_root; |
d427dfeb | 1443 | |
35585573 | 1444 | ret = rebind_subsystems(root, ss_mask, 0); |
d427dfeb | 1445 | if (ret) |
2bd59d48 | 1446 | goto destroy_root; |
d427dfeb TH |
1447 | |
1448 | /* | |
1449 | * There must be no failure case after here, since rebinding takes | |
1450 | * care of subsystems' refcounts, which are explicitly dropped in | |
1451 | * the failure exit path. | |
1452 | */ | |
1453 | list_add(&root->root_list, &cgroup_roots); | |
1454 | cgroup_root_count++; | |
1455 | ||
1456 | /* | |
1457 | * Link the top cgroup in this hierarchy into all the css_set | |
1458 | * objects. | |
1459 | */ | |
96d365e0 | 1460 | down_write(&css_set_rwsem); |
d427dfeb TH |
1461 | hash_for_each(css_set_table, i, cset, hlist) |
1462 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1463 | up_write(&css_set_rwsem); |
d427dfeb TH |
1464 | |
1465 | BUG_ON(!list_empty(&root_cgrp->children)); | |
3c9c825b | 1466 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
d427dfeb | 1467 | |
2bd59d48 | 1468 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1469 | ret = 0; |
2bd59d48 | 1470 | goto out; |
d427dfeb | 1471 | |
2bd59d48 TH |
1472 | destroy_root: |
1473 | kernfs_destroy_root(root->kf_root); | |
1474 | root->kf_root = NULL; | |
1475 | exit_root_id: | |
d427dfeb | 1476 | cgroup_exit_root_id(root); |
2bd59d48 | 1477 | out: |
d427dfeb TH |
1478 | free_cgrp_cset_links(&tmp_links); |
1479 | return ret; | |
1480 | } | |
1481 | ||
f7e83571 | 1482 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1483 | int flags, const char *unused_dev_name, |
f7e83571 | 1484 | void *data) |
ddbcc7e8 | 1485 | { |
2bd59d48 | 1486 | struct cgroupfs_root *root; |
ddbcc7e8 | 1487 | struct cgroup_sb_opts opts; |
2bd59d48 | 1488 | struct dentry *dentry; |
8e30e2b8 | 1489 | int ret; |
56fde9e0 TH |
1490 | |
1491 | /* | |
1492 | * The first time anyone tries to mount a cgroup, enable the list | |
1493 | * linking each css_set to its tasks and fix up all existing tasks. | |
1494 | */ | |
1495 | if (!use_task_css_set_links) | |
1496 | cgroup_enable_task_cg_lists(); | |
776f02fa | 1497 | retry: |
8e30e2b8 | 1498 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1499 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1500 | |
1501 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1502 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1503 | if (ret) |
8e30e2b8 | 1504 | goto out_unlock; |
ddbcc7e8 | 1505 | |
2bd59d48 TH |
1506 | /* look for a matching existing root */ |
1507 | for_each_active_root(root) { | |
1508 | bool name_match = false; | |
ddbcc7e8 | 1509 | |
2bd59d48 TH |
1510 | /* |
1511 | * If we asked for a name then it must match. Also, if | |
1512 | * name matches but sybsys_mask doesn't, we should fail. | |
1513 | * Remember whether name matched. | |
1514 | */ | |
1515 | if (opts.name) { | |
1516 | if (strcmp(opts.name, root->name)) | |
1517 | continue; | |
1518 | name_match = true; | |
1519 | } | |
ddbcc7e8 | 1520 | |
c6d57f33 | 1521 | /* |
2bd59d48 TH |
1522 | * If we asked for subsystems (or explicitly for no |
1523 | * subsystems) then they must match. | |
c6d57f33 | 1524 | */ |
2bd59d48 TH |
1525 | if ((opts.subsys_mask || opts.none) && |
1526 | (opts.subsys_mask != root->subsys_mask)) { | |
1527 | if (!name_match) | |
1528 | continue; | |
1529 | ret = -EBUSY; | |
1530 | goto out_unlock; | |
1531 | } | |
873fe09e | 1532 | |
c7ba8287 | 1533 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1534 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1535 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1536 | ret = -EINVAL; | |
8e30e2b8 | 1537 | goto out_unlock; |
2a0ff3fb JL |
1538 | } else { |
1539 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1540 | } | |
873fe09e | 1541 | } |
2bd59d48 | 1542 | |
776f02fa TH |
1543 | /* |
1544 | * A root's lifetime is governed by its top cgroup. Zero | |
1545 | * ref indicate that the root is being destroyed. Wait for | |
1546 | * destruction to complete so that the subsystems are free. | |
1547 | * We can use wait_queue for the wait but this path is | |
1548 | * super cold. Let's just sleep for a bit and retry. | |
1549 | */ | |
1550 | if (!atomic_inc_not_zero(&root->top_cgroup.refcnt)) { | |
1551 | mutex_unlock(&cgroup_mutex); | |
1552 | mutex_unlock(&cgroup_tree_mutex); | |
6534fd6c LZ |
1553 | kfree(opts.release_agent); |
1554 | kfree(opts.name); | |
776f02fa TH |
1555 | msleep(10); |
1556 | goto retry; | |
1557 | } | |
1558 | ||
1559 | ret = 0; | |
2bd59d48 | 1560 | goto out_unlock; |
ddbcc7e8 PM |
1561 | } |
1562 | ||
2bd59d48 TH |
1563 | /* no such thing, create a new one */ |
1564 | root = cgroup_root_from_opts(&opts); | |
1565 | if (IS_ERR(root)) { | |
1566 | ret = PTR_ERR(root); | |
1567 | goto out_unlock; | |
1568 | } | |
1569 | ||
35585573 | 1570 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1571 | if (ret) |
1572 | cgroup_free_root(root); | |
1573 | ||
8e30e2b8 | 1574 | out_unlock: |
e25e2cbb | 1575 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1576 | mutex_unlock(&cgroup_tree_mutex); |
8e30e2b8 | 1577 | |
c6d57f33 PM |
1578 | kfree(opts.release_agent); |
1579 | kfree(opts.name); | |
8e30e2b8 | 1580 | |
2bd59d48 | 1581 | if (ret) |
8e30e2b8 | 1582 | return ERR_PTR(ret); |
2bd59d48 TH |
1583 | |
1584 | dentry = kernfs_mount(fs_type, flags, root->kf_root); | |
1585 | if (IS_ERR(dentry)) | |
776f02fa | 1586 | cgroup_put(&root->top_cgroup); |
2bd59d48 TH |
1587 | return dentry; |
1588 | } | |
1589 | ||
1590 | static void cgroup_kill_sb(struct super_block *sb) | |
1591 | { | |
1592 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
1593 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); | |
1594 | ||
776f02fa | 1595 | cgroup_put(&root->top_cgroup); |
2bd59d48 | 1596 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1597 | } |
1598 | ||
ddbcc7e8 PM |
1599 | static struct file_system_type cgroup_fs_type = { |
1600 | .name = "cgroup", | |
f7e83571 | 1601 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1602 | .kill_sb = cgroup_kill_sb, |
1603 | }; | |
1604 | ||
676db4af GK |
1605 | static struct kobject *cgroup_kobj; |
1606 | ||
857a2beb | 1607 | /** |
913ffdb5 | 1608 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1609 | * @task: target task |
857a2beb TH |
1610 | * @buf: the buffer to write the path into |
1611 | * @buflen: the length of the buffer | |
1612 | * | |
913ffdb5 TH |
1613 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1614 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1615 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1616 | * cgroup controller callbacks. | |
1617 | * | |
e61734c5 | 1618 | * Return value is the same as kernfs_path(). |
857a2beb | 1619 | */ |
e61734c5 | 1620 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb TH |
1621 | { |
1622 | struct cgroupfs_root *root; | |
913ffdb5 | 1623 | struct cgroup *cgrp; |
e61734c5 TH |
1624 | int hierarchy_id = 1; |
1625 | char *path = NULL; | |
857a2beb TH |
1626 | |
1627 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1628 | down_read(&css_set_rwsem); |
857a2beb | 1629 | |
913ffdb5 TH |
1630 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1631 | ||
857a2beb TH |
1632 | if (root) { |
1633 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1634 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1635 | } else { |
1636 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1637 | if (strlcpy(buf, "/", buflen) < buflen) |
1638 | path = buf; | |
857a2beb TH |
1639 | } |
1640 | ||
96d365e0 | 1641 | up_read(&css_set_rwsem); |
857a2beb | 1642 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1643 | return path; |
857a2beb | 1644 | } |
913ffdb5 | 1645 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1646 | |
b3dc094e | 1647 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 1648 | struct cgroup_taskset { |
b3dc094e TH |
1649 | /* the src and dst cset list running through cset->mg_node */ |
1650 | struct list_head src_csets; | |
1651 | struct list_head dst_csets; | |
1652 | ||
1653 | /* | |
1654 | * Fields for cgroup_taskset_*() iteration. | |
1655 | * | |
1656 | * Before migration is committed, the target migration tasks are on | |
1657 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
1658 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
1659 | * or ->dst_csets depending on whether migration is committed. | |
1660 | * | |
1661 | * ->cur_csets and ->cur_task point to the current task position | |
1662 | * during iteration. | |
1663 | */ | |
1664 | struct list_head *csets; | |
1665 | struct css_set *cur_cset; | |
1666 | struct task_struct *cur_task; | |
2f7ee569 TH |
1667 | }; |
1668 | ||
1669 | /** | |
1670 | * cgroup_taskset_first - reset taskset and return the first task | |
1671 | * @tset: taskset of interest | |
1672 | * | |
1673 | * @tset iteration is initialized and the first task is returned. | |
1674 | */ | |
1675 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1676 | { | |
b3dc094e TH |
1677 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
1678 | tset->cur_task = NULL; | |
1679 | ||
1680 | return cgroup_taskset_next(tset); | |
2f7ee569 | 1681 | } |
2f7ee569 TH |
1682 | |
1683 | /** | |
1684 | * cgroup_taskset_next - iterate to the next task in taskset | |
1685 | * @tset: taskset of interest | |
1686 | * | |
1687 | * Return the next task in @tset. Iteration must have been initialized | |
1688 | * with cgroup_taskset_first(). | |
1689 | */ | |
1690 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1691 | { | |
b3dc094e TH |
1692 | struct css_set *cset = tset->cur_cset; |
1693 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 1694 | |
b3dc094e TH |
1695 | while (&cset->mg_node != tset->csets) { |
1696 | if (!task) | |
1697 | task = list_first_entry(&cset->mg_tasks, | |
1698 | struct task_struct, cg_list); | |
1699 | else | |
1700 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 1701 | |
b3dc094e TH |
1702 | if (&task->cg_list != &cset->mg_tasks) { |
1703 | tset->cur_cset = cset; | |
1704 | tset->cur_task = task; | |
1705 | return task; | |
1706 | } | |
1707 | ||
1708 | cset = list_next_entry(cset, mg_node); | |
1709 | task = NULL; | |
1710 | } | |
1711 | ||
1712 | return NULL; | |
2f7ee569 | 1713 | } |
2f7ee569 | 1714 | |
cb0f1fe9 | 1715 | /** |
74a1166d | 1716 | * cgroup_task_migrate - move a task from one cgroup to another. |
cb0f1fe9 TH |
1717 | * @old_cgrp; the cgroup @tsk is being migrated from |
1718 | * @tsk: the task being migrated | |
1719 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 1720 | * |
cb0f1fe9 | 1721 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 1722 | */ |
5abb8855 TH |
1723 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1724 | struct task_struct *tsk, | |
1725 | struct css_set *new_cset) | |
74a1166d | 1726 | { |
5abb8855 | 1727 | struct css_set *old_cset; |
74a1166d | 1728 | |
cb0f1fe9 TH |
1729 | lockdep_assert_held(&cgroup_mutex); |
1730 | lockdep_assert_held(&css_set_rwsem); | |
1731 | ||
74a1166d | 1732 | /* |
026085ef MSB |
1733 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1734 | * setting such that we can't race against cgroup_exit() changing the | |
1735 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1736 | */ |
c84cdf75 | 1737 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1738 | old_cset = task_css_set(tsk); |
74a1166d | 1739 | |
b3dc094e TH |
1740 | get_css_set(new_cset); |
1741 | ||
74a1166d | 1742 | task_lock(tsk); |
5abb8855 | 1743 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d BB |
1744 | task_unlock(tsk); |
1745 | ||
b3dc094e | 1746 | list_move(&tsk->cg_list, &new_cset->mg_tasks); |
74a1166d BB |
1747 | |
1748 | /* | |
5abb8855 TH |
1749 | * We just gained a reference on old_cset by taking it from the |
1750 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1751 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1752 | */ |
5abb8855 | 1753 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
cb0f1fe9 | 1754 | put_css_set_locked(old_cset, false); |
74a1166d BB |
1755 | } |
1756 | ||
a043e3b2 | 1757 | /** |
081aa458 | 1758 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup |
74a1166d | 1759 | * @cgrp: the cgroup to attach to |
9db8de37 | 1760 | * @leader: the task or the leader of the threadgroup to be attached |
081aa458 | 1761 | * @threadgroup: attach the whole threadgroup? |
74a1166d | 1762 | * |
257058ae | 1763 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take |
081aa458 | 1764 | * task_lock of @tsk or each thread in the threadgroup individually in turn. |
74a1166d | 1765 | */ |
9db8de37 | 1766 | static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *leader, |
47cfcd09 | 1767 | bool threadgroup) |
74a1166d | 1768 | { |
b3dc094e TH |
1769 | struct cgroup_taskset tset = { |
1770 | .src_csets = LIST_HEAD_INIT(tset.src_csets), | |
1771 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), | |
1772 | .csets = &tset.src_csets, | |
1773 | }; | |
1c6727af | 1774 | struct cgroup_subsys_state *css, *failed_css = NULL; |
b3dc094e TH |
1775 | struct css_set *cset, *tmp_cset; |
1776 | struct task_struct *task, *tmp_task; | |
1777 | int i, ret; | |
74a1166d | 1778 | |
fb5d2b4c MSB |
1779 | /* |
1780 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1781 | * already PF_EXITING could be freed from underneath us unless we | |
1782 | * take an rcu_read_lock. | |
1783 | */ | |
b3dc094e | 1784 | down_write(&css_set_rwsem); |
fb5d2b4c | 1785 | rcu_read_lock(); |
9db8de37 | 1786 | task = leader; |
74a1166d | 1787 | do { |
b3dc094e | 1788 | struct cgroup *src_cgrp; |
134d3373 | 1789 | |
9db8de37 TH |
1790 | /* @task either already exited or can't exit until the end */ |
1791 | if (task->flags & PF_EXITING) | |
ea84753c | 1792 | goto next; |
cd3d0952 | 1793 | |
b3dc094e TH |
1794 | cset = task_css_set(task); |
1795 | src_cgrp = task_cgroup_from_root(task, cgrp->root); | |
1796 | ||
892a2b90 | 1797 | /* nothing to do if this task is already in the cgroup */ |
b3dc094e | 1798 | if (src_cgrp == cgrp) |
ea84753c | 1799 | goto next; |
b3dc094e TH |
1800 | |
1801 | if (!cset->mg_src_cgrp) { | |
1802 | WARN_ON(!list_empty(&cset->mg_tasks)); | |
1803 | WARN_ON(!list_empty(&cset->mg_node)); | |
1804 | ||
1805 | cset->mg_src_cgrp = src_cgrp; | |
1806 | list_add(&cset->mg_node, &tset.src_csets); | |
1807 | get_css_set(cset); | |
1808 | } | |
1809 | ||
1810 | list_move(&task->cg_list, &cset->mg_tasks); | |
ea84753c | 1811 | next: |
081aa458 LZ |
1812 | if (!threadgroup) |
1813 | break; | |
9db8de37 | 1814 | } while_each_thread(leader, task); |
fb5d2b4c | 1815 | rcu_read_unlock(); |
b3dc094e | 1816 | up_write(&css_set_rwsem); |
74a1166d | 1817 | |
134d3373 | 1818 | /* methods shouldn't be called if no task is actually migrating */ |
b3dc094e TH |
1819 | if (list_empty(&tset.src_csets)) |
1820 | return 0; | |
134d3373 | 1821 | |
74a1166d BB |
1822 | /* |
1823 | * step 1: check that we can legitimately attach to the cgroup. | |
1824 | */ | |
1c6727af TH |
1825 | for_each_css(css, i, cgrp) { |
1826 | if (css->ss->can_attach) { | |
9db8de37 TH |
1827 | ret = css->ss->can_attach(css, &tset); |
1828 | if (ret) { | |
1c6727af | 1829 | failed_css = css; |
74a1166d BB |
1830 | goto out_cancel_attach; |
1831 | } | |
1832 | } | |
74a1166d BB |
1833 | } |
1834 | ||
1835 | /* | |
1836 | * step 2: make sure css_sets exist for all threads to be migrated. | |
1837 | * we use find_css_set, which allocates a new one if necessary. | |
1838 | */ | |
b3dc094e TH |
1839 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
1840 | struct css_set *dst_cset; | |
a8ad805c | 1841 | |
b3dc094e TH |
1842 | dst_cset = find_css_set(cset, cgrp); |
1843 | if (!dst_cset) { | |
9db8de37 | 1844 | ret = -ENOMEM; |
b3dc094e | 1845 | goto out_release_tset; |
74a1166d | 1846 | } |
b3dc094e TH |
1847 | |
1848 | if (list_empty(&dst_cset->mg_node)) | |
1849 | list_add(&dst_cset->mg_node, &tset.dst_csets); | |
1850 | else | |
1851 | put_css_set(dst_cset, false); | |
1852 | ||
1853 | cset->mg_dst_cset = dst_cset; | |
74a1166d BB |
1854 | } |
1855 | ||
1856 | /* | |
494c167c TH |
1857 | * step 3: now that we're guaranteed success wrt the css_sets, |
1858 | * proceed to move all tasks to the new cgroup. There are no | |
1859 | * failure cases after here, so this is the commit point. | |
74a1166d | 1860 | */ |
cb0f1fe9 | 1861 | down_write(&css_set_rwsem); |
b3dc094e TH |
1862 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
1863 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
1864 | cgroup_task_migrate(cset->mg_src_cgrp, task, | |
1865 | cset->mg_dst_cset); | |
74a1166d | 1866 | } |
cb0f1fe9 | 1867 | up_write(&css_set_rwsem); |
b3dc094e TH |
1868 | |
1869 | /* migration is committed, all target tasks are now on dst_csets */ | |
1870 | tset.csets = &tset.dst_csets; | |
1871 | ||
1872 | /* nothing is sensitive to fork() after this point */ | |
74a1166d BB |
1873 | |
1874 | /* | |
494c167c | 1875 | * step 4: do subsystem attach callbacks. |
74a1166d | 1876 | */ |
1c6727af TH |
1877 | for_each_css(css, i, cgrp) |
1878 | if (css->ss->attach) | |
1879 | css->ss->attach(css, &tset); | |
74a1166d | 1880 | |
9db8de37 | 1881 | ret = 0; |
b3dc094e TH |
1882 | goto out_release_tset; |
1883 | ||
74a1166d | 1884 | out_cancel_attach: |
b3dc094e TH |
1885 | for_each_css(css, i, cgrp) { |
1886 | if (css == failed_css) | |
1887 | break; | |
1888 | if (css->ss->cancel_attach) | |
1889 | css->ss->cancel_attach(css, &tset); | |
74a1166d | 1890 | } |
b3dc094e TH |
1891 | out_release_tset: |
1892 | down_write(&css_set_rwsem); | |
1893 | list_splice_init(&tset.dst_csets, &tset.src_csets); | |
1894 | list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { | |
1895 | list_splice_init(&cset->mg_tasks, &cset->tasks); | |
1896 | cset->mg_dst_cset = NULL; | |
1897 | cset->mg_src_cgrp = NULL; | |
1898 | list_del_init(&cset->mg_node); | |
1899 | put_css_set_locked(cset, false); | |
1900 | } | |
1901 | up_write(&css_set_rwsem); | |
9db8de37 | 1902 | return ret; |
74a1166d BB |
1903 | } |
1904 | ||
1905 | /* | |
1906 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
1907 | * function to attach either it or all tasks in its threadgroup. Will lock |
1908 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 1909 | */ |
74a1166d | 1910 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 1911 | { |
bbcb81d0 | 1912 | struct task_struct *tsk; |
c69e8d9c | 1913 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
1914 | int ret; |
1915 | ||
74a1166d BB |
1916 | if (!cgroup_lock_live_group(cgrp)) |
1917 | return -ENODEV; | |
1918 | ||
b78949eb MSB |
1919 | retry_find_task: |
1920 | rcu_read_lock(); | |
bbcb81d0 | 1921 | if (pid) { |
73507f33 | 1922 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
1923 | if (!tsk) { |
1924 | rcu_read_unlock(); | |
dd4b0a46 | 1925 | ret = -ESRCH; |
b78949eb | 1926 | goto out_unlock_cgroup; |
bbcb81d0 | 1927 | } |
74a1166d BB |
1928 | /* |
1929 | * even if we're attaching all tasks in the thread group, we | |
1930 | * only need to check permissions on one of them. | |
1931 | */ | |
c69e8d9c | 1932 | tcred = __task_cred(tsk); |
14a590c3 EB |
1933 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
1934 | !uid_eq(cred->euid, tcred->uid) && | |
1935 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 1936 | rcu_read_unlock(); |
b78949eb MSB |
1937 | ret = -EACCES; |
1938 | goto out_unlock_cgroup; | |
bbcb81d0 | 1939 | } |
b78949eb MSB |
1940 | } else |
1941 | tsk = current; | |
cd3d0952 TH |
1942 | |
1943 | if (threadgroup) | |
b78949eb | 1944 | tsk = tsk->group_leader; |
c4c27fbd MG |
1945 | |
1946 | /* | |
14a40ffc | 1947 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
1948 | * trapped in a cpuset, or RT worker may be born in a cgroup |
1949 | * with no rt_runtime allocated. Just say no. | |
1950 | */ | |
14a40ffc | 1951 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
1952 | ret = -EINVAL; |
1953 | rcu_read_unlock(); | |
1954 | goto out_unlock_cgroup; | |
1955 | } | |
1956 | ||
b78949eb MSB |
1957 | get_task_struct(tsk); |
1958 | rcu_read_unlock(); | |
1959 | ||
1960 | threadgroup_lock(tsk); | |
1961 | if (threadgroup) { | |
1962 | if (!thread_group_leader(tsk)) { | |
1963 | /* | |
1964 | * a race with de_thread from another thread's exec() | |
1965 | * may strip us of our leadership, if this happens, | |
1966 | * there is no choice but to throw this task away and | |
1967 | * try again; this is | |
1968 | * "double-double-toil-and-trouble-check locking". | |
1969 | */ | |
1970 | threadgroup_unlock(tsk); | |
1971 | put_task_struct(tsk); | |
1972 | goto retry_find_task; | |
1973 | } | |
081aa458 LZ |
1974 | } |
1975 | ||
1976 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
1977 | ||
cd3d0952 TH |
1978 | threadgroup_unlock(tsk); |
1979 | ||
bbcb81d0 | 1980 | put_task_struct(tsk); |
b78949eb | 1981 | out_unlock_cgroup: |
47cfcd09 | 1982 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
1983 | return ret; |
1984 | } | |
1985 | ||
7ae1bad9 TH |
1986 | /** |
1987 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
1988 | * @from: attach to all cgroups of a given task | |
1989 | * @tsk: the task to be attached | |
1990 | */ | |
1991 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
1992 | { | |
1993 | struct cgroupfs_root *root; | |
1994 | int retval = 0; | |
1995 | ||
47cfcd09 | 1996 | mutex_lock(&cgroup_mutex); |
7ae1bad9 | 1997 | for_each_active_root(root) { |
96d365e0 TH |
1998 | struct cgroup *from_cgrp; |
1999 | ||
2000 | down_read(&css_set_rwsem); | |
2001 | from_cgrp = task_cgroup_from_root(from, root); | |
2002 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2003 | |
6f4b7e63 | 2004 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2005 | if (retval) |
2006 | break; | |
2007 | } | |
47cfcd09 | 2008 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2009 | |
2010 | return retval; | |
2011 | } | |
2012 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2013 | ||
182446d0 TH |
2014 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2015 | struct cftype *cft, u64 pid) | |
74a1166d | 2016 | { |
182446d0 | 2017 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2018 | } |
2019 | ||
182446d0 TH |
2020 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2021 | struct cftype *cft, u64 tgid) | |
af351026 | 2022 | { |
182446d0 | 2023 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2024 | } |
2025 | ||
182446d0 TH |
2026 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
2027 | struct cftype *cft, const char *buffer) | |
e788e066 | 2028 | { |
5f469907 TH |
2029 | struct cgroupfs_root *root = css->cgroup->root; |
2030 | ||
2031 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
182446d0 | 2032 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2033 | return -ENODEV; |
69e943b7 | 2034 | spin_lock(&release_agent_path_lock); |
5f469907 TH |
2035 | strlcpy(root->release_agent_path, buffer, |
2036 | sizeof(root->release_agent_path)); | |
69e943b7 | 2037 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2038 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2039 | return 0; |
2040 | } | |
2041 | ||
2da8ca82 | 2042 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2043 | { |
2da8ca82 | 2044 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2045 | |
e788e066 PM |
2046 | if (!cgroup_lock_live_group(cgrp)) |
2047 | return -ENODEV; | |
2048 | seq_puts(seq, cgrp->root->release_agent_path); | |
2049 | seq_putc(seq, '\n'); | |
47cfcd09 | 2050 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2051 | return 0; |
2052 | } | |
2053 | ||
2da8ca82 | 2054 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2055 | { |
2da8ca82 TH |
2056 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2057 | ||
2058 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2059 | return 0; |
2060 | } | |
2061 | ||
2bd59d48 TH |
2062 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2063 | size_t nbytes, loff_t off) | |
355e0c48 | 2064 | { |
2bd59d48 TH |
2065 | struct cgroup *cgrp = of->kn->parent->priv; |
2066 | struct cftype *cft = of->kn->priv; | |
2067 | struct cgroup_subsys_state *css; | |
a742c59d | 2068 | int ret; |
355e0c48 | 2069 | |
2bd59d48 TH |
2070 | /* |
2071 | * kernfs guarantees that a file isn't deleted with operations in | |
2072 | * flight, which means that the matching css is and stays alive and | |
2073 | * doesn't need to be pinned. The RCU locking is not necessary | |
2074 | * either. It's just for the convenience of using cgroup_css(). | |
2075 | */ | |
2076 | rcu_read_lock(); | |
2077 | css = cgroup_css(cgrp, cft->ss); | |
2078 | rcu_read_unlock(); | |
a742c59d TH |
2079 | |
2080 | if (cft->write_string) { | |
2081 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2082 | } else if (cft->write_u64) { | |
2083 | unsigned long long v; | |
2084 | ret = kstrtoull(buf, 0, &v); | |
2085 | if (!ret) | |
2086 | ret = cft->write_u64(css, cft, v); | |
2087 | } else if (cft->write_s64) { | |
2088 | long long v; | |
2089 | ret = kstrtoll(buf, 0, &v); | |
2090 | if (!ret) | |
2091 | ret = cft->write_s64(css, cft, v); | |
2092 | } else if (cft->trigger) { | |
2093 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2094 | } else { |
a742c59d | 2095 | ret = -EINVAL; |
e73d2c61 | 2096 | } |
2bd59d48 | 2097 | |
a742c59d | 2098 | return ret ?: nbytes; |
355e0c48 PM |
2099 | } |
2100 | ||
6612f05b | 2101 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2102 | { |
2bd59d48 | 2103 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2104 | } |
2105 | ||
6612f05b | 2106 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2107 | { |
2bd59d48 | 2108 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2109 | } |
2110 | ||
6612f05b | 2111 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2112 | { |
2bd59d48 | 2113 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2114 | } |
2115 | ||
91796569 | 2116 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2117 | { |
7da11279 TH |
2118 | struct cftype *cft = seq_cft(m); |
2119 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2120 | |
2da8ca82 TH |
2121 | if (cft->seq_show) |
2122 | return cft->seq_show(m, arg); | |
e73d2c61 | 2123 | |
f4c753b7 | 2124 | if (cft->read_u64) |
896f5199 TH |
2125 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2126 | else if (cft->read_s64) | |
2127 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2128 | else | |
2129 | return -EINVAL; | |
2130 | return 0; | |
91796569 PM |
2131 | } |
2132 | ||
2bd59d48 TH |
2133 | static struct kernfs_ops cgroup_kf_single_ops = { |
2134 | .atomic_write_len = PAGE_SIZE, | |
2135 | .write = cgroup_file_write, | |
2136 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2137 | }; |
2138 | ||
2bd59d48 TH |
2139 | static struct kernfs_ops cgroup_kf_ops = { |
2140 | .atomic_write_len = PAGE_SIZE, | |
2141 | .write = cgroup_file_write, | |
2142 | .seq_start = cgroup_seqfile_start, | |
2143 | .seq_next = cgroup_seqfile_next, | |
2144 | .seq_stop = cgroup_seqfile_stop, | |
2145 | .seq_show = cgroup_seqfile_show, | |
2146 | }; | |
ddbcc7e8 PM |
2147 | |
2148 | /* | |
2149 | * cgroup_rename - Only allow simple rename of directories in place. | |
2150 | */ | |
2bd59d48 TH |
2151 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2152 | const char *new_name_str) | |
ddbcc7e8 | 2153 | { |
2bd59d48 | 2154 | struct cgroup *cgrp = kn->priv; |
2bd59d48 | 2155 | int ret; |
65dff759 | 2156 | |
2bd59d48 | 2157 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2158 | return -ENOTDIR; |
2bd59d48 | 2159 | if (kn->parent != new_parent) |
ddbcc7e8 | 2160 | return -EIO; |
65dff759 | 2161 | |
6db8e85c TH |
2162 | /* |
2163 | * This isn't a proper migration and its usefulness is very | |
2164 | * limited. Disallow if sane_behavior. | |
2165 | */ | |
2166 | if (cgroup_sane_behavior(cgrp)) | |
2167 | return -EPERM; | |
2168 | ||
2bd59d48 TH |
2169 | mutex_lock(&cgroup_tree_mutex); |
2170 | mutex_lock(&cgroup_mutex); | |
2171 | ||
2172 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
65dff759 | 2173 | |
2bd59d48 TH |
2174 | mutex_unlock(&cgroup_mutex); |
2175 | mutex_unlock(&cgroup_tree_mutex); | |
2bd59d48 | 2176 | return ret; |
ddbcc7e8 PM |
2177 | } |
2178 | ||
2bb566cb | 2179 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2180 | { |
8d7e6fb0 | 2181 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2182 | struct kernfs_node *kn; |
2183 | struct lock_class_key *key = NULL; | |
05ef1d7c | 2184 | |
2bd59d48 TH |
2185 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2186 | key = &cft->lockdep_key; | |
2187 | #endif | |
2188 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2189 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2190 | NULL, false, key); | |
430af8ad | 2191 | return PTR_ERR_OR_ZERO(kn); |
ddbcc7e8 PM |
2192 | } |
2193 | ||
b1f28d31 TH |
2194 | /** |
2195 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2196 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2197 | * @cfts: array of cftypes to be added |
2198 | * @is_add: whether to add or remove | |
2199 | * | |
2200 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2201 | * For removals, this function never fails. If addition fails, this |
2202 | * function doesn't remove files already added. The caller is responsible | |
2203 | * for cleaning up. | |
b1f28d31 | 2204 | */ |
2bb566cb TH |
2205 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2206 | bool is_add) | |
ddbcc7e8 | 2207 | { |
03b1cde6 | 2208 | struct cftype *cft; |
b1f28d31 TH |
2209 | int ret; |
2210 | ||
ace2bee8 | 2211 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2212 | |
2213 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2214 | /* does cft->flags tell us to skip this file on @cgrp? */ |
873fe09e TH |
2215 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2216 | continue; | |
f33fddc2 G |
2217 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2218 | continue; | |
2219 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2220 | continue; | |
2221 | ||
2739d3cc | 2222 | if (is_add) { |
2bb566cb | 2223 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2224 | if (ret) { |
2739d3cc | 2225 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2226 | cft->name, ret); |
2227 | return ret; | |
2228 | } | |
2739d3cc LZ |
2229 | } else { |
2230 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2231 | } |
ddbcc7e8 | 2232 | } |
b1f28d31 | 2233 | return 0; |
ddbcc7e8 PM |
2234 | } |
2235 | ||
21a2d343 | 2236 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2237 | { |
2238 | LIST_HEAD(pending); | |
2bb566cb | 2239 | struct cgroup_subsys *ss = cfts[0].ss; |
492eb21b | 2240 | struct cgroup *root = &ss->root->top_cgroup; |
492eb21b | 2241 | struct cgroup_subsys_state *css; |
9ccece80 | 2242 | int ret = 0; |
8e3f6541 | 2243 | |
21a2d343 | 2244 | lockdep_assert_held(&cgroup_tree_mutex); |
4ac06017 | 2245 | |
21a2d343 TH |
2246 | /* don't bother if @ss isn't attached */ |
2247 | if (ss->root == &cgroup_dummy_root) | |
9ccece80 | 2248 | return 0; |
e8c82d20 | 2249 | |
e8c82d20 | 2250 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2251 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2252 | struct cgroup *cgrp = css->cgroup; |
2253 | ||
e8c82d20 LZ |
2254 | if (cgroup_is_dead(cgrp)) |
2255 | continue; | |
2256 | ||
21a2d343 | 2257 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2258 | if (ret) |
2259 | break; | |
8e3f6541 | 2260 | } |
21a2d343 TH |
2261 | |
2262 | if (is_add && !ret) | |
2263 | kernfs_activate(root->kn); | |
9ccece80 | 2264 | return ret; |
8e3f6541 TH |
2265 | } |
2266 | ||
2da440a2 TH |
2267 | static void cgroup_exit_cftypes(struct cftype *cfts) |
2268 | { | |
2269 | struct cftype *cft; | |
2270 | ||
2bd59d48 TH |
2271 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2272 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2273 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2274 | kfree(cft->kf_ops); | |
2275 | cft->kf_ops = NULL; | |
2da440a2 | 2276 | cft->ss = NULL; |
2bd59d48 | 2277 | } |
2da440a2 TH |
2278 | } |
2279 | ||
2bd59d48 | 2280 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2281 | { |
2282 | struct cftype *cft; | |
2283 | ||
2bd59d48 TH |
2284 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2285 | struct kernfs_ops *kf_ops; | |
2286 | ||
0adb0704 TH |
2287 | WARN_ON(cft->ss || cft->kf_ops); |
2288 | ||
2bd59d48 TH |
2289 | if (cft->seq_start) |
2290 | kf_ops = &cgroup_kf_ops; | |
2291 | else | |
2292 | kf_ops = &cgroup_kf_single_ops; | |
2293 | ||
2294 | /* | |
2295 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2296 | * make a copy of kf_ops to set its atomic_write_len. | |
2297 | */ | |
2298 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2299 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2300 | if (!kf_ops) { | |
2301 | cgroup_exit_cftypes(cfts); | |
2302 | return -ENOMEM; | |
2303 | } | |
2304 | kf_ops->atomic_write_len = cft->max_write_len; | |
2305 | } | |
2306 | ||
2307 | cft->kf_ops = kf_ops; | |
2da440a2 | 2308 | cft->ss = ss; |
2bd59d48 TH |
2309 | } |
2310 | ||
2311 | return 0; | |
2da440a2 TH |
2312 | } |
2313 | ||
21a2d343 TH |
2314 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
2315 | { | |
2316 | lockdep_assert_held(&cgroup_tree_mutex); | |
2317 | ||
2318 | if (!cfts || !cfts[0].ss) | |
2319 | return -ENOENT; | |
2320 | ||
2321 | list_del(&cfts->node); | |
2322 | cgroup_apply_cftypes(cfts, false); | |
2323 | cgroup_exit_cftypes(cfts); | |
2324 | return 0; | |
2325 | } | |
2326 | ||
80b13586 TH |
2327 | /** |
2328 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
2329 | * @cfts: zero-length name terminated array of cftypes | |
2330 | * | |
2331 | * Unregister @cfts. Files described by @cfts are removed from all | |
2332 | * existing cgroups and all future cgroups won't have them either. This | |
2333 | * function can be called anytime whether @cfts' subsys is attached or not. | |
2334 | * | |
2335 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2336 | * registered. | |
2337 | */ | |
2338 | int cgroup_rm_cftypes(struct cftype *cfts) | |
2339 | { | |
21a2d343 | 2340 | int ret; |
80b13586 | 2341 | |
21a2d343 TH |
2342 | mutex_lock(&cgroup_tree_mutex); |
2343 | ret = cgroup_rm_cftypes_locked(cfts); | |
2344 | mutex_unlock(&cgroup_tree_mutex); | |
2345 | return ret; | |
80b13586 TH |
2346 | } |
2347 | ||
8e3f6541 TH |
2348 | /** |
2349 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2350 | * @ss: target cgroup subsystem | |
2351 | * @cfts: zero-length name terminated array of cftypes | |
2352 | * | |
2353 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2354 | * existing cgroups to which @ss is attached and all future cgroups will | |
2355 | * have them too. This function can be called anytime whether @ss is | |
2356 | * attached or not. | |
2357 | * | |
2358 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2359 | * function currently returns 0 as long as @cfts registration is successful | |
2360 | * even if some file creation attempts on existing cgroups fail. | |
2361 | */ | |
03b1cde6 | 2362 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 2363 | { |
9ccece80 | 2364 | int ret; |
8e3f6541 | 2365 | |
dc5736ed LZ |
2366 | if (!cfts || cfts[0].name[0] == '\0') |
2367 | return 0; | |
2368 | ||
2bd59d48 TH |
2369 | ret = cgroup_init_cftypes(ss, cfts); |
2370 | if (ret) | |
2371 | return ret; | |
2bb566cb | 2372 | |
21a2d343 TH |
2373 | mutex_lock(&cgroup_tree_mutex); |
2374 | ||
0adb0704 | 2375 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 2376 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 2377 | if (ret) |
21a2d343 TH |
2378 | cgroup_rm_cftypes_locked(cfts); |
2379 | ||
2380 | mutex_unlock(&cgroup_tree_mutex); | |
9ccece80 | 2381 | return ret; |
8e3f6541 | 2382 | } |
8e3f6541 | 2383 | |
a043e3b2 LZ |
2384 | /** |
2385 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2386 | * @cgrp: the cgroup in question | |
2387 | * | |
2388 | * Return the number of tasks in the cgroup. | |
2389 | */ | |
07bc356e | 2390 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2391 | { |
2392 | int count = 0; | |
69d0206c | 2393 | struct cgrp_cset_link *link; |
817929ec | 2394 | |
96d365e0 | 2395 | down_read(&css_set_rwsem); |
69d0206c TH |
2396 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2397 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 2398 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
2399 | return count; |
2400 | } | |
2401 | ||
53fa5261 | 2402 | /** |
492eb21b TH |
2403 | * css_next_child - find the next child of a given css |
2404 | * @pos_css: the current position (%NULL to initiate traversal) | |
2405 | * @parent_css: css whose children to walk | |
53fa5261 | 2406 | * |
492eb21b | 2407 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
2408 | * under either cgroup_mutex or RCU read lock. The only requirement is |
2409 | * that @parent_css and @pos_css are accessible. The next sibling is | |
2410 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 2411 | */ |
492eb21b TH |
2412 | struct cgroup_subsys_state * |
2413 | css_next_child(struct cgroup_subsys_state *pos_css, | |
2414 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 2415 | { |
492eb21b TH |
2416 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
2417 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
2418 | struct cgroup *next; |
2419 | ||
ace2bee8 | 2420 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
2421 | |
2422 | /* | |
2423 | * @pos could already have been removed. Once a cgroup is removed, | |
2424 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
2425 | * changes. As CGRP_DEAD assertion is serialized and happens |
2426 | * before the cgroup is taken off the ->sibling list, if we see it | |
2427 | * unasserted, it's guaranteed that the next sibling hasn't | |
2428 | * finished its grace period even if it's already removed, and thus | |
2429 | * safe to dereference from this RCU critical section. If | |
2430 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
2431 | * to be visible as %true here. | |
3b287a50 TH |
2432 | * |
2433 | * If @pos is dead, its next pointer can't be dereferenced; | |
2434 | * however, as each cgroup is given a monotonically increasing | |
2435 | * unique serial number and always appended to the sibling list, | |
2436 | * the next one can be found by walking the parent's children until | |
2437 | * we see a cgroup with higher serial number than @pos's. While | |
2438 | * this path can be slower, it's taken only when either the current | |
2439 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 2440 | */ |
3b287a50 TH |
2441 | if (!pos) { |
2442 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
2443 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 2444 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
2445 | } else { |
2446 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
2447 | if (next->serial_nr > pos->serial_nr) | |
2448 | break; | |
53fa5261 TH |
2449 | } |
2450 | ||
492eb21b TH |
2451 | if (&next->sibling == &cgrp->children) |
2452 | return NULL; | |
2453 | ||
ca8bdcaf | 2454 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 2455 | } |
53fa5261 | 2456 | |
574bd9f7 | 2457 | /** |
492eb21b | 2458 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 2459 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2460 | * @root: css whose descendants to walk |
574bd9f7 | 2461 | * |
492eb21b | 2462 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
2463 | * to visit for pre-order traversal of @root's descendants. @root is |
2464 | * included in the iteration and the first node to be visited. | |
75501a6d | 2465 | * |
87fb54f1 TH |
2466 | * While this function requires cgroup_mutex or RCU read locking, it |
2467 | * doesn't require the whole traversal to be contained in a single critical | |
2468 | * section. This function will return the correct next descendant as long | |
2469 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 2470 | */ |
492eb21b TH |
2471 | struct cgroup_subsys_state * |
2472 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
2473 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2474 | { |
492eb21b | 2475 | struct cgroup_subsys_state *next; |
574bd9f7 | 2476 | |
ace2bee8 | 2477 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2478 | |
bd8815a6 | 2479 | /* if first iteration, visit @root */ |
7805d000 | 2480 | if (!pos) |
bd8815a6 | 2481 | return root; |
574bd9f7 TH |
2482 | |
2483 | /* visit the first child if exists */ | |
492eb21b | 2484 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
2485 | if (next) |
2486 | return next; | |
2487 | ||
2488 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
2489 | while (pos != root) { |
2490 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 2491 | if (next) |
574bd9f7 | 2492 | return next; |
492eb21b | 2493 | pos = css_parent(pos); |
7805d000 | 2494 | } |
574bd9f7 TH |
2495 | |
2496 | return NULL; | |
2497 | } | |
574bd9f7 | 2498 | |
12a9d2fe | 2499 | /** |
492eb21b TH |
2500 | * css_rightmost_descendant - return the rightmost descendant of a css |
2501 | * @pos: css of interest | |
12a9d2fe | 2502 | * |
492eb21b TH |
2503 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
2504 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 2505 | * subtree of @pos. |
75501a6d | 2506 | * |
87fb54f1 TH |
2507 | * While this function requires cgroup_mutex or RCU read locking, it |
2508 | * doesn't require the whole traversal to be contained in a single critical | |
2509 | * section. This function will return the correct rightmost descendant as | |
2510 | * long as @pos is accessible. | |
12a9d2fe | 2511 | */ |
492eb21b TH |
2512 | struct cgroup_subsys_state * |
2513 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 2514 | { |
492eb21b | 2515 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 2516 | |
ace2bee8 | 2517 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
2518 | |
2519 | do { | |
2520 | last = pos; | |
2521 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
2522 | pos = NULL; | |
492eb21b | 2523 | css_for_each_child(tmp, last) |
12a9d2fe TH |
2524 | pos = tmp; |
2525 | } while (pos); | |
2526 | ||
2527 | return last; | |
2528 | } | |
12a9d2fe | 2529 | |
492eb21b TH |
2530 | static struct cgroup_subsys_state * |
2531 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 2532 | { |
492eb21b | 2533 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
2534 | |
2535 | do { | |
2536 | last = pos; | |
492eb21b | 2537 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
2538 | } while (pos); |
2539 | ||
2540 | return last; | |
2541 | } | |
2542 | ||
2543 | /** | |
492eb21b | 2544 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 2545 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2546 | * @root: css whose descendants to walk |
574bd9f7 | 2547 | * |
492eb21b | 2548 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
2549 | * to visit for post-order traversal of @root's descendants. @root is |
2550 | * included in the iteration and the last node to be visited. | |
75501a6d | 2551 | * |
87fb54f1 TH |
2552 | * While this function requires cgroup_mutex or RCU read locking, it |
2553 | * doesn't require the whole traversal to be contained in a single critical | |
2554 | * section. This function will return the correct next descendant as long | |
2555 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
2556 | * @cgroup. | |
574bd9f7 | 2557 | */ |
492eb21b TH |
2558 | struct cgroup_subsys_state * |
2559 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
2560 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2561 | { |
492eb21b | 2562 | struct cgroup_subsys_state *next; |
574bd9f7 | 2563 | |
ace2bee8 | 2564 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2565 | |
58b79a91 TH |
2566 | /* if first iteration, visit leftmost descendant which may be @root */ |
2567 | if (!pos) | |
2568 | return css_leftmost_descendant(root); | |
574bd9f7 | 2569 | |
bd8815a6 TH |
2570 | /* if we visited @root, we're done */ |
2571 | if (pos == root) | |
2572 | return NULL; | |
2573 | ||
574bd9f7 | 2574 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 2575 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 2576 | if (next) |
492eb21b | 2577 | return css_leftmost_descendant(next); |
574bd9f7 TH |
2578 | |
2579 | /* no sibling left, visit parent */ | |
bd8815a6 | 2580 | return css_parent(pos); |
574bd9f7 | 2581 | } |
574bd9f7 | 2582 | |
0942eeee | 2583 | /** |
72ec7029 | 2584 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
2585 | * @it: the iterator to advance |
2586 | * | |
2587 | * Advance @it to the next css_set to walk. | |
d515876e | 2588 | */ |
72ec7029 | 2589 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
2590 | { |
2591 | struct list_head *l = it->cset_link; | |
2592 | struct cgrp_cset_link *link; | |
2593 | struct css_set *cset; | |
2594 | ||
2595 | /* Advance to the next non-empty css_set */ | |
2596 | do { | |
2597 | l = l->next; | |
72ec7029 | 2598 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
2599 | it->cset_link = NULL; |
2600 | return; | |
2601 | } | |
2602 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
2603 | cset = link->cset; | |
c7561128 TH |
2604 | } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); |
2605 | ||
d515876e | 2606 | it->cset_link = l; |
c7561128 TH |
2607 | |
2608 | if (!list_empty(&cset->tasks)) | |
2609 | it->task = cset->tasks.next; | |
2610 | else | |
2611 | it->task = cset->mg_tasks.next; | |
d515876e TH |
2612 | } |
2613 | ||
0942eeee | 2614 | /** |
72ec7029 TH |
2615 | * css_task_iter_start - initiate task iteration |
2616 | * @css: the css to walk tasks of | |
0942eeee TH |
2617 | * @it: the task iterator to use |
2618 | * | |
72ec7029 TH |
2619 | * Initiate iteration through the tasks of @css. The caller can call |
2620 | * css_task_iter_next() to walk through the tasks until the function | |
2621 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
2622 | * called. | |
0942eeee TH |
2623 | * |
2624 | * Note that this function acquires a lock which is released when the | |
2625 | * iteration finishes. The caller can't sleep while iteration is in | |
2626 | * progress. | |
2627 | */ | |
72ec7029 TH |
2628 | void css_task_iter_start(struct cgroup_subsys_state *css, |
2629 | struct css_task_iter *it) | |
96d365e0 | 2630 | __acquires(css_set_rwsem) |
817929ec | 2631 | { |
56fde9e0 TH |
2632 | /* no one should try to iterate before mounting cgroups */ |
2633 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 2634 | |
96d365e0 | 2635 | down_read(&css_set_rwsem); |
c59cd3d8 | 2636 | |
72ec7029 TH |
2637 | it->origin_css = css; |
2638 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 2639 | |
72ec7029 | 2640 | css_advance_task_iter(it); |
817929ec PM |
2641 | } |
2642 | ||
0942eeee | 2643 | /** |
72ec7029 | 2644 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
2645 | * @it: the task iterator being iterated |
2646 | * | |
2647 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
2648 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
2649 | * reaches the end. | |
0942eeee | 2650 | */ |
72ec7029 | 2651 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
2652 | { |
2653 | struct task_struct *res; | |
2654 | struct list_head *l = it->task; | |
c7561128 TH |
2655 | struct cgrp_cset_link *link = list_entry(it->cset_link, |
2656 | struct cgrp_cset_link, cset_link); | |
817929ec PM |
2657 | |
2658 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 2659 | if (!it->cset_link) |
817929ec PM |
2660 | return NULL; |
2661 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
2662 | |
2663 | /* | |
2664 | * Advance iterator to find next entry. cset->tasks is consumed | |
2665 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
2666 | * next cset. | |
2667 | */ | |
817929ec | 2668 | l = l->next; |
c7561128 TH |
2669 | |
2670 | if (l == &link->cset->tasks) | |
2671 | l = link->cset->mg_tasks.next; | |
2672 | ||
2673 | if (l == &link->cset->mg_tasks) | |
72ec7029 | 2674 | css_advance_task_iter(it); |
c7561128 | 2675 | else |
817929ec | 2676 | it->task = l; |
c7561128 | 2677 | |
817929ec PM |
2678 | return res; |
2679 | } | |
2680 | ||
0942eeee | 2681 | /** |
72ec7029 | 2682 | * css_task_iter_end - finish task iteration |
0942eeee TH |
2683 | * @it: the task iterator to finish |
2684 | * | |
72ec7029 | 2685 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 2686 | */ |
72ec7029 | 2687 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 2688 | __releases(css_set_rwsem) |
817929ec | 2689 | { |
96d365e0 | 2690 | up_read(&css_set_rwsem); |
817929ec PM |
2691 | } |
2692 | ||
8cc99345 TH |
2693 | /** |
2694 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
2695 | * @to: cgroup to which the tasks will be moved | |
2696 | * @from: cgroup in which the tasks currently reside | |
2697 | */ | |
2698 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
2699 | { | |
e406d1cf TH |
2700 | struct css_task_iter it; |
2701 | struct task_struct *task; | |
2702 | int ret = 0; | |
2703 | ||
2704 | do { | |
2705 | css_task_iter_start(&from->dummy_css, &it); | |
2706 | task = css_task_iter_next(&it); | |
2707 | if (task) | |
2708 | get_task_struct(task); | |
2709 | css_task_iter_end(&it); | |
2710 | ||
2711 | if (task) { | |
2712 | mutex_lock(&cgroup_mutex); | |
2713 | ret = cgroup_attach_task(to, task, false); | |
2714 | mutex_unlock(&cgroup_mutex); | |
2715 | put_task_struct(task); | |
2716 | } | |
2717 | } while (task && !ret); | |
2718 | ||
2719 | return ret; | |
8cc99345 TH |
2720 | } |
2721 | ||
bbcb81d0 | 2722 | /* |
102a775e | 2723 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
2724 | * |
2725 | * Reading this file can return large amounts of data if a cgroup has | |
2726 | * *lots* of attached tasks. So it may need several calls to read(), | |
2727 | * but we cannot guarantee that the information we produce is correct | |
2728 | * unless we produce it entirely atomically. | |
2729 | * | |
bbcb81d0 | 2730 | */ |
bbcb81d0 | 2731 | |
24528255 LZ |
2732 | /* which pidlist file are we talking about? */ |
2733 | enum cgroup_filetype { | |
2734 | CGROUP_FILE_PROCS, | |
2735 | CGROUP_FILE_TASKS, | |
2736 | }; | |
2737 | ||
2738 | /* | |
2739 | * A pidlist is a list of pids that virtually represents the contents of one | |
2740 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
2741 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
2742 | * to the cgroup. | |
2743 | */ | |
2744 | struct cgroup_pidlist { | |
2745 | /* | |
2746 | * used to find which pidlist is wanted. doesn't change as long as | |
2747 | * this particular list stays in the list. | |
2748 | */ | |
2749 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
2750 | /* array of xids */ | |
2751 | pid_t *list; | |
2752 | /* how many elements the above list has */ | |
2753 | int length; | |
24528255 LZ |
2754 | /* each of these stored in a list by its cgroup */ |
2755 | struct list_head links; | |
2756 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
2757 | struct cgroup *owner; | |
b1a21367 TH |
2758 | /* for delayed destruction */ |
2759 | struct delayed_work destroy_dwork; | |
24528255 LZ |
2760 | }; |
2761 | ||
d1d9fd33 BB |
2762 | /* |
2763 | * The following two functions "fix" the issue where there are more pids | |
2764 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
2765 | * TODO: replace with a kernel-wide solution to this problem | |
2766 | */ | |
2767 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
2768 | static void *pidlist_allocate(int count) | |
2769 | { | |
2770 | if (PIDLIST_TOO_LARGE(count)) | |
2771 | return vmalloc(count * sizeof(pid_t)); | |
2772 | else | |
2773 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
2774 | } | |
b1a21367 | 2775 | |
d1d9fd33 BB |
2776 | static void pidlist_free(void *p) |
2777 | { | |
2778 | if (is_vmalloc_addr(p)) | |
2779 | vfree(p); | |
2780 | else | |
2781 | kfree(p); | |
2782 | } | |
d1d9fd33 | 2783 | |
b1a21367 TH |
2784 | /* |
2785 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
2786 | * should be left afterwards. | |
2787 | */ | |
2788 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
2789 | { | |
2790 | struct cgroup_pidlist *l, *tmp_l; | |
2791 | ||
2792 | mutex_lock(&cgrp->pidlist_mutex); | |
2793 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
2794 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
2795 | mutex_unlock(&cgrp->pidlist_mutex); | |
2796 | ||
2797 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
2798 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
2799 | } | |
2800 | ||
2801 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
2802 | { | |
2803 | struct delayed_work *dwork = to_delayed_work(work); | |
2804 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
2805 | destroy_dwork); | |
2806 | struct cgroup_pidlist *tofree = NULL; | |
2807 | ||
2808 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
2809 | |
2810 | /* | |
04502365 TH |
2811 | * Destroy iff we didn't get queued again. The state won't change |
2812 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 2813 | */ |
04502365 | 2814 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
2815 | list_del(&l->links); |
2816 | pidlist_free(l->list); | |
2817 | put_pid_ns(l->key.ns); | |
2818 | tofree = l; | |
2819 | } | |
2820 | ||
b1a21367 TH |
2821 | mutex_unlock(&l->owner->pidlist_mutex); |
2822 | kfree(tofree); | |
2823 | } | |
2824 | ||
bbcb81d0 | 2825 | /* |
102a775e | 2826 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 2827 | * Returns the number of unique elements. |
bbcb81d0 | 2828 | */ |
6ee211ad | 2829 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 2830 | { |
102a775e | 2831 | int src, dest = 1; |
102a775e BB |
2832 | |
2833 | /* | |
2834 | * we presume the 0th element is unique, so i starts at 1. trivial | |
2835 | * edge cases first; no work needs to be done for either | |
2836 | */ | |
2837 | if (length == 0 || length == 1) | |
2838 | return length; | |
2839 | /* src and dest walk down the list; dest counts unique elements */ | |
2840 | for (src = 1; src < length; src++) { | |
2841 | /* find next unique element */ | |
2842 | while (list[src] == list[src-1]) { | |
2843 | src++; | |
2844 | if (src == length) | |
2845 | goto after; | |
2846 | } | |
2847 | /* dest always points to where the next unique element goes */ | |
2848 | list[dest] = list[src]; | |
2849 | dest++; | |
2850 | } | |
2851 | after: | |
102a775e BB |
2852 | return dest; |
2853 | } | |
2854 | ||
afb2bc14 TH |
2855 | /* |
2856 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
2857 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
2858 | * different per namespace, each namespace needs differently sorted list, | |
2859 | * making it impossible to use, for example, single rbtree of member tasks | |
2860 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
2861 | * per open file is dangerous, so cgroup had to implement shared pool of | |
2862 | * pidlists keyed by cgroup and namespace. | |
2863 | * | |
2864 | * All this extra complexity was caused by the original implementation | |
2865 | * committing to an entirely unnecessary property. In the long term, we | |
2866 | * want to do away with it. Explicitly scramble sort order if | |
2867 | * sane_behavior so that no such expectation exists in the new interface. | |
2868 | * | |
2869 | * Scrambling is done by swapping every two consecutive bits, which is | |
2870 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
2871 | */ | |
2872 | static pid_t pid_fry(pid_t pid) | |
2873 | { | |
2874 | unsigned a = pid & 0x55555555; | |
2875 | unsigned b = pid & 0xAAAAAAAA; | |
2876 | ||
2877 | return (a << 1) | (b >> 1); | |
2878 | } | |
2879 | ||
2880 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
2881 | { | |
2882 | if (cgroup_sane_behavior(cgrp)) | |
2883 | return pid_fry(pid); | |
2884 | else | |
2885 | return pid; | |
2886 | } | |
2887 | ||
102a775e BB |
2888 | static int cmppid(const void *a, const void *b) |
2889 | { | |
2890 | return *(pid_t *)a - *(pid_t *)b; | |
2891 | } | |
2892 | ||
afb2bc14 TH |
2893 | static int fried_cmppid(const void *a, const void *b) |
2894 | { | |
2895 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
2896 | } | |
2897 | ||
e6b81710 TH |
2898 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
2899 | enum cgroup_filetype type) | |
2900 | { | |
2901 | struct cgroup_pidlist *l; | |
2902 | /* don't need task_nsproxy() if we're looking at ourself */ | |
2903 | struct pid_namespace *ns = task_active_pid_ns(current); | |
2904 | ||
2905 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
2906 | ||
2907 | list_for_each_entry(l, &cgrp->pidlists, links) | |
2908 | if (l->key.type == type && l->key.ns == ns) | |
2909 | return l; | |
2910 | return NULL; | |
2911 | } | |
2912 | ||
72a8cb30 BB |
2913 | /* |
2914 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
2915 | * returns with the lock on that pidlist already held, and takes care | |
2916 | * of the use count, or returns NULL with no locks held if we're out of | |
2917 | * memory. | |
2918 | */ | |
e6b81710 TH |
2919 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
2920 | enum cgroup_filetype type) | |
72a8cb30 BB |
2921 | { |
2922 | struct cgroup_pidlist *l; | |
b70cc5fd | 2923 | |
e6b81710 TH |
2924 | lockdep_assert_held(&cgrp->pidlist_mutex); |
2925 | ||
2926 | l = cgroup_pidlist_find(cgrp, type); | |
2927 | if (l) | |
2928 | return l; | |
2929 | ||
72a8cb30 | 2930 | /* entry not found; create a new one */ |
f4f4be2b | 2931 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 2932 | if (!l) |
72a8cb30 | 2933 | return l; |
e6b81710 | 2934 | |
b1a21367 | 2935 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 2936 | l->key.type = type; |
e6b81710 TH |
2937 | /* don't need task_nsproxy() if we're looking at ourself */ |
2938 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
2939 | l->owner = cgrp; |
2940 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
2941 | return l; |
2942 | } | |
2943 | ||
102a775e BB |
2944 | /* |
2945 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
2946 | */ | |
72a8cb30 BB |
2947 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
2948 | struct cgroup_pidlist **lp) | |
102a775e BB |
2949 | { |
2950 | pid_t *array; | |
2951 | int length; | |
2952 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 2953 | struct css_task_iter it; |
817929ec | 2954 | struct task_struct *tsk; |
102a775e BB |
2955 | struct cgroup_pidlist *l; |
2956 | ||
4bac00d1 TH |
2957 | lockdep_assert_held(&cgrp->pidlist_mutex); |
2958 | ||
102a775e BB |
2959 | /* |
2960 | * If cgroup gets more users after we read count, we won't have | |
2961 | * enough space - tough. This race is indistinguishable to the | |
2962 | * caller from the case that the additional cgroup users didn't | |
2963 | * show up until sometime later on. | |
2964 | */ | |
2965 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 2966 | array = pidlist_allocate(length); |
102a775e BB |
2967 | if (!array) |
2968 | return -ENOMEM; | |
2969 | /* now, populate the array */ | |
72ec7029 TH |
2970 | css_task_iter_start(&cgrp->dummy_css, &it); |
2971 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 2972 | if (unlikely(n == length)) |
817929ec | 2973 | break; |
102a775e | 2974 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
2975 | if (type == CGROUP_FILE_PROCS) |
2976 | pid = task_tgid_vnr(tsk); | |
2977 | else | |
2978 | pid = task_pid_vnr(tsk); | |
102a775e BB |
2979 | if (pid > 0) /* make sure to only use valid results */ |
2980 | array[n++] = pid; | |
817929ec | 2981 | } |
72ec7029 | 2982 | css_task_iter_end(&it); |
102a775e BB |
2983 | length = n; |
2984 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
2985 | if (cgroup_sane_behavior(cgrp)) |
2986 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
2987 | else | |
2988 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 2989 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 2990 | length = pidlist_uniq(array, length); |
e6b81710 | 2991 | |
e6b81710 | 2992 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 2993 | if (!l) { |
e6b81710 | 2994 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 2995 | pidlist_free(array); |
72a8cb30 | 2996 | return -ENOMEM; |
102a775e | 2997 | } |
e6b81710 TH |
2998 | |
2999 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3000 | pidlist_free(l->list); |
102a775e BB |
3001 | l->list = array; |
3002 | l->length = length; | |
72a8cb30 | 3003 | *lp = l; |
102a775e | 3004 | return 0; |
bbcb81d0 PM |
3005 | } |
3006 | ||
846c7bb0 | 3007 | /** |
a043e3b2 | 3008 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3009 | * @stats: cgroupstats to fill information into |
3010 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3011 | * been requested. | |
a043e3b2 LZ |
3012 | * |
3013 | * Build and fill cgroupstats so that taskstats can export it to user | |
3014 | * space. | |
846c7bb0 BS |
3015 | */ |
3016 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3017 | { | |
2bd59d48 | 3018 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3019 | struct cgroup *cgrp; |
72ec7029 | 3020 | struct css_task_iter it; |
846c7bb0 | 3021 | struct task_struct *tsk; |
33d283be | 3022 | |
2bd59d48 TH |
3023 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3024 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3025 | kernfs_type(kn) != KERNFS_DIR) | |
3026 | return -EINVAL; | |
3027 | ||
bad34660 LZ |
3028 | mutex_lock(&cgroup_mutex); |
3029 | ||
846c7bb0 | 3030 | /* |
2bd59d48 TH |
3031 | * We aren't being called from kernfs and there's no guarantee on |
3032 | * @kn->priv's validity. For this and css_tryget_from_dir(), | |
3033 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
846c7bb0 | 3034 | */ |
2bd59d48 TH |
3035 | rcu_read_lock(); |
3036 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 3037 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 3038 | rcu_read_unlock(); |
bad34660 | 3039 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
3040 | return -ENOENT; |
3041 | } | |
bad34660 | 3042 | rcu_read_unlock(); |
846c7bb0 | 3043 | |
72ec7029 TH |
3044 | css_task_iter_start(&cgrp->dummy_css, &it); |
3045 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3046 | switch (tsk->state) { |
3047 | case TASK_RUNNING: | |
3048 | stats->nr_running++; | |
3049 | break; | |
3050 | case TASK_INTERRUPTIBLE: | |
3051 | stats->nr_sleeping++; | |
3052 | break; | |
3053 | case TASK_UNINTERRUPTIBLE: | |
3054 | stats->nr_uninterruptible++; | |
3055 | break; | |
3056 | case TASK_STOPPED: | |
3057 | stats->nr_stopped++; | |
3058 | break; | |
3059 | default: | |
3060 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3061 | stats->nr_io_wait++; | |
3062 | break; | |
3063 | } | |
3064 | } | |
72ec7029 | 3065 | css_task_iter_end(&it); |
846c7bb0 | 3066 | |
bad34660 | 3067 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 3068 | return 0; |
846c7bb0 BS |
3069 | } |
3070 | ||
8f3ff208 | 3071 | |
bbcb81d0 | 3072 | /* |
102a775e | 3073 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3074 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3075 | * in the cgroup->l->list array. |
bbcb81d0 | 3076 | */ |
cc31edce | 3077 | |
102a775e | 3078 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3079 | { |
cc31edce PM |
3080 | /* |
3081 | * Initially we receive a position value that corresponds to | |
3082 | * one more than the last pid shown (or 0 on the first call or | |
3083 | * after a seek to the start). Use a binary-search to find the | |
3084 | * next pid to display, if any | |
3085 | */ | |
2bd59d48 | 3086 | struct kernfs_open_file *of = s->private; |
7da11279 | 3087 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3088 | struct cgroup_pidlist *l; |
7da11279 | 3089 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3090 | int index = 0, pid = *pos; |
4bac00d1 TH |
3091 | int *iter, ret; |
3092 | ||
3093 | mutex_lock(&cgrp->pidlist_mutex); | |
3094 | ||
3095 | /* | |
5d22444f | 3096 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3097 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3098 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3099 | * could already have been destroyed. |
3100 | */ | |
5d22444f TH |
3101 | if (of->priv) |
3102 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3103 | |
3104 | /* | |
3105 | * Either this is the first start() after open or the matching | |
3106 | * pidlist has been destroyed inbetween. Create a new one. | |
3107 | */ | |
5d22444f TH |
3108 | if (!of->priv) { |
3109 | ret = pidlist_array_load(cgrp, type, | |
3110 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3111 | if (ret) |
3112 | return ERR_PTR(ret); | |
3113 | } | |
5d22444f | 3114 | l = of->priv; |
cc31edce | 3115 | |
cc31edce | 3116 | if (pid) { |
102a775e | 3117 | int end = l->length; |
20777766 | 3118 | |
cc31edce PM |
3119 | while (index < end) { |
3120 | int mid = (index + end) / 2; | |
afb2bc14 | 3121 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3122 | index = mid; |
3123 | break; | |
afb2bc14 | 3124 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3125 | index = mid + 1; |
3126 | else | |
3127 | end = mid; | |
3128 | } | |
3129 | } | |
3130 | /* If we're off the end of the array, we're done */ | |
102a775e | 3131 | if (index >= l->length) |
cc31edce PM |
3132 | return NULL; |
3133 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3134 | iter = l->list + index; |
afb2bc14 | 3135 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3136 | return iter; |
3137 | } | |
3138 | ||
102a775e | 3139 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3140 | { |
2bd59d48 | 3141 | struct kernfs_open_file *of = s->private; |
5d22444f | 3142 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3143 | |
5d22444f TH |
3144 | if (l) |
3145 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3146 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3147 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3148 | } |
3149 | ||
102a775e | 3150 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3151 | { |
2bd59d48 | 3152 | struct kernfs_open_file *of = s->private; |
5d22444f | 3153 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3154 | pid_t *p = v; |
3155 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3156 | /* |
3157 | * Advance to the next pid in the array. If this goes off the | |
3158 | * end, we're done | |
3159 | */ | |
3160 | p++; | |
3161 | if (p >= end) { | |
3162 | return NULL; | |
3163 | } else { | |
7da11279 | 3164 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3165 | return p; |
3166 | } | |
3167 | } | |
3168 | ||
102a775e | 3169 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3170 | { |
3171 | return seq_printf(s, "%d\n", *(int *)v); | |
3172 | } | |
bbcb81d0 | 3173 | |
102a775e BB |
3174 | /* |
3175 | * seq_operations functions for iterating on pidlists through seq_file - | |
3176 | * independent of whether it's tasks or procs | |
3177 | */ | |
3178 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3179 | .start = cgroup_pidlist_start, | |
3180 | .stop = cgroup_pidlist_stop, | |
3181 | .next = cgroup_pidlist_next, | |
3182 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3183 | }; |
3184 | ||
182446d0 TH |
3185 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3186 | struct cftype *cft) | |
81a6a5cd | 3187 | { |
182446d0 | 3188 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3189 | } |
3190 | ||
182446d0 TH |
3191 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3192 | struct cftype *cft, u64 val) | |
6379c106 | 3193 | { |
182446d0 | 3194 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3195 | if (val) |
182446d0 | 3196 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3197 | else |
182446d0 | 3198 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3199 | return 0; |
3200 | } | |
3201 | ||
182446d0 TH |
3202 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3203 | struct cftype *cft) | |
97978e6d | 3204 | { |
182446d0 | 3205 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3206 | } |
3207 | ||
182446d0 TH |
3208 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3209 | struct cftype *cft, u64 val) | |
97978e6d DL |
3210 | { |
3211 | if (val) | |
182446d0 | 3212 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3213 | else |
182446d0 | 3214 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3215 | return 0; |
3216 | } | |
3217 | ||
d5c56ced | 3218 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3219 | { |
d5c56ced | 3220 | .name = "cgroup.procs", |
6612f05b TH |
3221 | .seq_start = cgroup_pidlist_start, |
3222 | .seq_next = cgroup_pidlist_next, | |
3223 | .seq_stop = cgroup_pidlist_stop, | |
3224 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3225 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3226 | .write_u64 = cgroup_procs_write, |
74a1166d | 3227 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3228 | }, |
97978e6d DL |
3229 | { |
3230 | .name = "cgroup.clone_children", | |
873fe09e | 3231 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3232 | .read_u64 = cgroup_clone_children_read, |
3233 | .write_u64 = cgroup_clone_children_write, | |
3234 | }, | |
873fe09e TH |
3235 | { |
3236 | .name = "cgroup.sane_behavior", | |
3237 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3238 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3239 | }, |
d5c56ced TH |
3240 | |
3241 | /* | |
3242 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3243 | * don't exist if sane_behavior. If you're depending on these, be | |
3244 | * prepared to be burned. | |
3245 | */ | |
3246 | { | |
3247 | .name = "tasks", | |
3248 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3249 | .seq_start = cgroup_pidlist_start, |
3250 | .seq_next = cgroup_pidlist_next, | |
3251 | .seq_stop = cgroup_pidlist_stop, | |
3252 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3253 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3254 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3255 | .mode = S_IRUGO | S_IWUSR, |
3256 | }, | |
3257 | { | |
3258 | .name = "notify_on_release", | |
3259 | .flags = CFTYPE_INSANE, | |
3260 | .read_u64 = cgroup_read_notify_on_release, | |
3261 | .write_u64 = cgroup_write_notify_on_release, | |
3262 | }, | |
6e6ff25b TH |
3263 | { |
3264 | .name = "release_agent", | |
cc5943a7 | 3265 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3266 | .seq_show = cgroup_release_agent_show, |
6e6ff25b | 3267 | .write_string = cgroup_release_agent_write, |
5f469907 | 3268 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 3269 | }, |
db0416b6 | 3270 | { } /* terminate */ |
bbcb81d0 PM |
3271 | }; |
3272 | ||
13af07df | 3273 | /** |
628f7cd4 | 3274 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3275 | * @cgrp: target cgroup |
13af07df | 3276 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3277 | * |
3278 | * On failure, no file is added. | |
13af07df | 3279 | */ |
628f7cd4 | 3280 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3281 | { |
ddbcc7e8 | 3282 | struct cgroup_subsys *ss; |
b420ba7d | 3283 | int i, ret = 0; |
bbcb81d0 | 3284 | |
8e3f6541 | 3285 | /* process cftsets of each subsystem */ |
b420ba7d | 3286 | for_each_subsys(ss, i) { |
0adb0704 | 3287 | struct cftype *cfts; |
b420ba7d TH |
3288 | |
3289 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3290 | continue; |
8e3f6541 | 3291 | |
0adb0704 TH |
3292 | list_for_each_entry(cfts, &ss->cfts, node) { |
3293 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
3294 | if (ret < 0) |
3295 | goto err; | |
3296 | } | |
ddbcc7e8 | 3297 | } |
ddbcc7e8 | 3298 | return 0; |
bee55099 TH |
3299 | err: |
3300 | cgroup_clear_dir(cgrp, subsys_mask); | |
3301 | return ret; | |
ddbcc7e8 PM |
3302 | } |
3303 | ||
0c21ead1 TH |
3304 | /* |
3305 | * css destruction is four-stage process. | |
3306 | * | |
3307 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3308 | * Implemented in kill_css(). | |
3309 | * | |
3310 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3311 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3312 | * by invoking offline_css(). After offlining, the base ref is put. | |
3313 | * Implemented in css_killed_work_fn(). | |
3314 | * | |
3315 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3316 | * accessors are inside RCU read sections. css_release() schedules the | |
3317 | * RCU callback. | |
3318 | * | |
3319 | * 4. After the grace period, the css can be freed. Implemented in | |
3320 | * css_free_work_fn(). | |
3321 | * | |
3322 | * It is actually hairier because both step 2 and 4 require process context | |
3323 | * and thus involve punting to css->destroy_work adding two additional | |
3324 | * steps to the already complex sequence. | |
3325 | */ | |
35ef10da | 3326 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3327 | { |
3328 | struct cgroup_subsys_state *css = | |
35ef10da | 3329 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 3330 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3331 | |
0ae78e0b TH |
3332 | if (css->parent) |
3333 | css_put(css->parent); | |
3334 | ||
0c21ead1 | 3335 | css->ss->css_free(css); |
2bd59d48 | 3336 | cgroup_put(cgrp); |
48ddbe19 TH |
3337 | } |
3338 | ||
0c21ead1 | 3339 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3340 | { |
3341 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3342 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3343 | |
35ef10da | 3344 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 3345 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
3346 | } |
3347 | ||
d3daf28d TH |
3348 | static void css_release(struct percpu_ref *ref) |
3349 | { | |
3350 | struct cgroup_subsys_state *css = | |
3351 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3352 | ||
aec25020 | 3353 | rcu_assign_pointer(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 3354 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
3355 | } |
3356 | ||
623f926b TH |
3357 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
3358 | struct cgroup *cgrp) | |
ddbcc7e8 | 3359 | { |
bd89aabc | 3360 | css->cgroup = cgrp; |
72c97e54 | 3361 | css->ss = ss; |
ddbcc7e8 | 3362 | css->flags = 0; |
0ae78e0b TH |
3363 | |
3364 | if (cgrp->parent) | |
ca8bdcaf | 3365 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 3366 | else |
38b53aba | 3367 | css->flags |= CSS_ROOT; |
48ddbe19 | 3368 | |
ca8bdcaf | 3369 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
3370 | } |
3371 | ||
2a4ac633 | 3372 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 3373 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3374 | { |
623f926b | 3375 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
3376 | int ret = 0; |
3377 | ||
ace2bee8 | 3378 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3379 | lockdep_assert_held(&cgroup_mutex); |
3380 | ||
92fb9748 | 3381 | if (ss->css_online) |
eb95419b | 3382 | ret = ss->css_online(css); |
ae7f164a | 3383 | if (!ret) { |
eb95419b | 3384 | css->flags |= CSS_ONLINE; |
f20104de | 3385 | css->cgroup->nr_css++; |
aec25020 | 3386 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 3387 | } |
b1929db4 | 3388 | return ret; |
a31f2d3f TH |
3389 | } |
3390 | ||
2a4ac633 | 3391 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 3392 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3393 | { |
623f926b | 3394 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 3395 | |
ace2bee8 | 3396 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3397 | lockdep_assert_held(&cgroup_mutex); |
3398 | ||
3399 | if (!(css->flags & CSS_ONLINE)) | |
3400 | return; | |
3401 | ||
d7eeac19 | 3402 | if (ss->css_offline) |
eb95419b | 3403 | ss->css_offline(css); |
a31f2d3f | 3404 | |
eb95419b | 3405 | css->flags &= ~CSS_ONLINE; |
09a503ea | 3406 | css->cgroup->nr_css--; |
aec25020 | 3407 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); |
a31f2d3f TH |
3408 | } |
3409 | ||
c81c925a TH |
3410 | /** |
3411 | * create_css - create a cgroup_subsys_state | |
3412 | * @cgrp: the cgroup new css will be associated with | |
3413 | * @ss: the subsys of new css | |
3414 | * | |
3415 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
3416 | * css is online and installed in @cgrp with all interface files created. | |
3417 | * Returns 0 on success, -errno on failure. | |
3418 | */ | |
3419 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
3420 | { | |
3421 | struct cgroup *parent = cgrp->parent; | |
3422 | struct cgroup_subsys_state *css; | |
3423 | int err; | |
3424 | ||
c81c925a TH |
3425 | lockdep_assert_held(&cgroup_mutex); |
3426 | ||
3427 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
3428 | if (IS_ERR(css)) | |
3429 | return PTR_ERR(css); | |
3430 | ||
3431 | err = percpu_ref_init(&css->refcnt, css_release); | |
3432 | if (err) | |
3433 | goto err_free; | |
3434 | ||
3435 | init_css(css, ss, cgrp); | |
3436 | ||
aec25020 | 3437 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a TH |
3438 | if (err) |
3439 | goto err_free; | |
3440 | ||
3441 | err = online_css(css); | |
3442 | if (err) | |
3443 | goto err_free; | |
3444 | ||
59f5296b | 3445 | cgroup_get(cgrp); |
c81c925a TH |
3446 | css_get(css->parent); |
3447 | ||
3448 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
3449 | parent->parent) { | |
3450 | 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", | |
3451 | current->comm, current->pid, ss->name); | |
3452 | if (!strcmp(ss->name, "memory")) | |
3453 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
3454 | ss->warned_broken_hierarchy = true; | |
3455 | } | |
3456 | ||
3457 | return 0; | |
3458 | ||
3459 | err_free: | |
3460 | percpu_ref_cancel_init(&css->refcnt); | |
3461 | ss->css_free(css); | |
3462 | return err; | |
3463 | } | |
3464 | ||
2bd59d48 | 3465 | /** |
a043e3b2 LZ |
3466 | * cgroup_create - create a cgroup |
3467 | * @parent: cgroup that will be parent of the new cgroup | |
e61734c5 | 3468 | * @name: name of the new cgroup |
2bd59d48 | 3469 | * @mode: mode to set on new cgroup |
ddbcc7e8 | 3470 | */ |
e61734c5 | 3471 | static long cgroup_create(struct cgroup *parent, const char *name, |
2bd59d48 | 3472 | umode_t mode) |
ddbcc7e8 | 3473 | { |
bd89aabc | 3474 | struct cgroup *cgrp; |
ddbcc7e8 | 3475 | struct cgroupfs_root *root = parent->root; |
b58c8998 | 3476 | int ssid, err; |
ddbcc7e8 | 3477 | struct cgroup_subsys *ss; |
2bd59d48 | 3478 | struct kernfs_node *kn; |
ddbcc7e8 | 3479 | |
0a950f65 | 3480 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
3481 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
3482 | if (!cgrp) | |
ddbcc7e8 PM |
3483 | return -ENOMEM; |
3484 | ||
ace2bee8 TH |
3485 | mutex_lock(&cgroup_tree_mutex); |
3486 | ||
976c06bc TH |
3487 | /* |
3488 | * Only live parents can have children. Note that the liveliness | |
3489 | * check isn't strictly necessary because cgroup_mkdir() and | |
3490 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
3491 | * anyway so that locking is contained inside cgroup proper and we | |
3492 | * don't get nasty surprises if we ever grow another caller. | |
3493 | */ | |
3494 | if (!cgroup_lock_live_group(parent)) { | |
3495 | err = -ENODEV; | |
ace2bee8 | 3496 | goto err_unlock_tree; |
0ab02ca8 LZ |
3497 | } |
3498 | ||
3499 | /* | |
3500 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
3501 | * a half-baked cgroup. | |
3502 | */ | |
3503 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
3504 | if (cgrp->id < 0) { | |
3505 | err = -ENOMEM; | |
3506 | goto err_unlock; | |
976c06bc TH |
3507 | } |
3508 | ||
cc31edce | 3509 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3510 | |
bd89aabc | 3511 | cgrp->parent = parent; |
0ae78e0b | 3512 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 3513 | cgrp->root = parent->root; |
ddbcc7e8 | 3514 | |
b6abdb0e LZ |
3515 | if (notify_on_release(parent)) |
3516 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3517 | ||
2260e7fc TH |
3518 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
3519 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 3520 | |
2bd59d48 | 3521 | /* create the directory */ |
e61734c5 | 3522 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 TH |
3523 | if (IS_ERR(kn)) { |
3524 | err = PTR_ERR(kn); | |
0ab02ca8 | 3525 | goto err_free_id; |
2bd59d48 TH |
3526 | } |
3527 | cgrp->kn = kn; | |
ddbcc7e8 | 3528 | |
6f30558f TH |
3529 | /* |
3530 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
3531 | * that @cgrp->kn is always accessible. | |
3532 | */ | |
3533 | kernfs_get(kn); | |
3534 | ||
00356bd5 | 3535 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 3536 | |
4e139afc | 3537 | /* allocation complete, commit to creation */ |
4e139afc | 3538 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3c9c825b | 3539 | atomic_inc(&root->nr_cgrps); |
59f5296b | 3540 | cgroup_get(parent); |
415cf07a | 3541 | |
0d80255e TH |
3542 | /* |
3543 | * @cgrp is now fully operational. If something fails after this | |
3544 | * point, it'll be released via the normal destruction path. | |
3545 | */ | |
4e96ee8e LZ |
3546 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
3547 | ||
2bb566cb | 3548 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
3549 | if (err) |
3550 | goto err_destroy; | |
3551 | ||
9d403e99 | 3552 | /* let's create and online css's */ |
b85d2040 TH |
3553 | for_each_subsys(ss, ssid) { |
3554 | if (root->subsys_mask & (1 << ssid)) { | |
3555 | err = create_css(cgrp, ss); | |
3556 | if (err) | |
3557 | goto err_destroy; | |
3558 | } | |
a8638030 | 3559 | } |
ddbcc7e8 | 3560 | |
2bd59d48 TH |
3561 | kernfs_activate(kn); |
3562 | ||
ddbcc7e8 | 3563 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3564 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
3565 | |
3566 | return 0; | |
3567 | ||
0a950f65 | 3568 | err_free_id: |
4e96ee8e | 3569 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
3570 | err_unlock: |
3571 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
3572 | err_unlock_tree: |
3573 | mutex_unlock(&cgroup_tree_mutex); | |
bd89aabc | 3574 | kfree(cgrp); |
ddbcc7e8 | 3575 | return err; |
4b8b47eb TH |
3576 | |
3577 | err_destroy: | |
3578 | cgroup_destroy_locked(cgrp); | |
3579 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3580 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 3581 | return err; |
ddbcc7e8 PM |
3582 | } |
3583 | ||
2bd59d48 TH |
3584 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
3585 | umode_t mode) | |
ddbcc7e8 | 3586 | { |
2bd59d48 | 3587 | struct cgroup *parent = parent_kn->priv; |
ddbcc7e8 | 3588 | |
2bd59d48 | 3589 | return cgroup_create(parent, name, mode); |
ddbcc7e8 PM |
3590 | } |
3591 | ||
223dbc38 TH |
3592 | /* |
3593 | * This is called when the refcnt of a css is confirmed to be killed. | |
3594 | * css_tryget() is now guaranteed to fail. | |
3595 | */ | |
3596 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 3597 | { |
223dbc38 TH |
3598 | struct cgroup_subsys_state *css = |
3599 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
3600 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 3601 | |
ace2bee8 | 3602 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
3603 | mutex_lock(&cgroup_mutex); |
3604 | ||
09a503ea TH |
3605 | /* |
3606 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
3607 | * initate destruction. | |
3608 | */ | |
3609 | offline_css(css); | |
3610 | ||
f20104de TH |
3611 | /* |
3612 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
3613 | * be disabled before proceeding to the second phase of cgroup | |
3614 | * destruction. If we are the last one, kick it off. | |
3615 | */ | |
09a503ea | 3616 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
3617 | cgroup_destroy_css_killed(cgrp); |
3618 | ||
3619 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3620 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
3621 | |
3622 | /* | |
3623 | * Put the css refs from kill_css(). Each css holds an extra | |
3624 | * reference to the cgroup's dentry and cgroup removal proceeds | |
3625 | * regardless of css refs. On the last put of each css, whenever | |
3626 | * that may be, the extra dentry ref is put so that dentry | |
3627 | * destruction happens only after all css's are released. | |
3628 | */ | |
3629 | css_put(css); | |
d3daf28d TH |
3630 | } |
3631 | ||
223dbc38 TH |
3632 | /* css kill confirmation processing requires process context, bounce */ |
3633 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
3634 | { |
3635 | struct cgroup_subsys_state *css = | |
3636 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3637 | ||
223dbc38 | 3638 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 3639 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
3640 | } |
3641 | ||
edae0c33 TH |
3642 | /** |
3643 | * kill_css - destroy a css | |
3644 | * @css: css to destroy | |
3645 | * | |
3c14f8b4 TH |
3646 | * This function initiates destruction of @css by removing cgroup interface |
3647 | * files and putting its base reference. ->css_offline() will be invoked | |
3648 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
3649 | * reference count reaches zero, @css will be released. | |
edae0c33 TH |
3650 | */ |
3651 | static void kill_css(struct cgroup_subsys_state *css) | |
3652 | { | |
2bd59d48 TH |
3653 | /* |
3654 | * This must happen before css is disassociated with its cgroup. | |
3655 | * See seq_css() for details. | |
3656 | */ | |
aec25020 | 3657 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 3658 | |
edae0c33 TH |
3659 | /* |
3660 | * Killing would put the base ref, but we need to keep it alive | |
3661 | * until after ->css_offline(). | |
3662 | */ | |
3663 | css_get(css); | |
3664 | ||
3665 | /* | |
3666 | * cgroup core guarantees that, by the time ->css_offline() is | |
3667 | * invoked, no new css reference will be given out via | |
3668 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
3669 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
3670 | * guarantee that the ref is seen as killed on all CPUs on return. | |
3671 | * | |
3672 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
3673 | * css is confirmed to be seen as killed on all CPUs. | |
3674 | */ | |
3675 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
3676 | } |
3677 | ||
3678 | /** | |
3679 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
3680 | * @cgrp: cgroup to be destroyed | |
3681 | * | |
3682 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
3683 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
3684 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
3685 | * invoked. To satisfy all the requirements, destruction is implemented in | |
3686 | * the following two steps. | |
3687 | * | |
3688 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
3689 | * userland visible parts and start killing the percpu refcnts of | |
3690 | * css's. Set up so that the next stage will be kicked off once all | |
3691 | * the percpu refcnts are confirmed to be killed. | |
3692 | * | |
3693 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
3694 | * rest of destruction. Once all cgroup references are gone, the | |
3695 | * cgroup is RCU-freed. | |
3696 | * | |
3697 | * This function implements s1. After this step, @cgrp is gone as far as | |
3698 | * the userland is concerned and a new cgroup with the same name may be | |
3699 | * created. As cgroup doesn't care about the names internally, this | |
3700 | * doesn't cause any problem. | |
3701 | */ | |
42809dd4 TH |
3702 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
3703 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 3704 | { |
bb78a92f | 3705 | struct cgroup *child; |
2bd59d48 | 3706 | struct cgroup_subsys_state *css; |
ddd69148 | 3707 | bool empty; |
1c6727af | 3708 | int ssid; |
ddbcc7e8 | 3709 | |
ace2bee8 | 3710 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
3711 | lockdep_assert_held(&cgroup_mutex); |
3712 | ||
ddd69148 | 3713 | /* |
96d365e0 | 3714 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 3715 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 3716 | */ |
96d365e0 | 3717 | down_read(&css_set_rwsem); |
bb78a92f | 3718 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 3719 | up_read(&css_set_rwsem); |
ddd69148 | 3720 | if (!empty) |
ddbcc7e8 | 3721 | return -EBUSY; |
a043e3b2 | 3722 | |
bb78a92f HD |
3723 | /* |
3724 | * Make sure there's no live children. We can't test ->children | |
3725 | * emptiness as dead children linger on it while being destroyed; | |
3726 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
3727 | */ | |
3728 | empty = true; | |
3729 | rcu_read_lock(); | |
3730 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
3731 | empty = cgroup_is_dead(child); | |
3732 | if (!empty) | |
3733 | break; | |
3734 | } | |
3735 | rcu_read_unlock(); | |
3736 | if (!empty) | |
3737 | return -EBUSY; | |
3738 | ||
88703267 | 3739 | /* |
edae0c33 TH |
3740 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
3741 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
3742 | * percpu refs of all css's are confirmed to be killed. This |
3743 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 3744 | */ |
4ac06017 | 3745 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
3746 | for_each_css(css, ssid, cgrp) |
3747 | kill_css(css); | |
4ac06017 | 3748 | mutex_lock(&cgroup_mutex); |
455050d2 TH |
3749 | |
3750 | /* | |
3751 | * Mark @cgrp dead. This prevents further task migration and child | |
3752 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 3753 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 3754 | * resume iteration after dropping RCU read lock. See |
492eb21b | 3755 | * css_next_child() for details. |
455050d2 | 3756 | */ |
54766d4a | 3757 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 3758 | |
455050d2 TH |
3759 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
3760 | raw_spin_lock(&release_list_lock); | |
3761 | if (!list_empty(&cgrp->release_list)) | |
3762 | list_del_init(&cgrp->release_list); | |
3763 | raw_spin_unlock(&release_list_lock); | |
3764 | ||
3765 | /* | |
f20104de TH |
3766 | * If @cgrp has css's attached, the second stage of cgroup |
3767 | * destruction is kicked off from css_killed_work_fn() after the | |
3768 | * refs of all attached css's are killed. If @cgrp doesn't have | |
3769 | * any css, we kick it off here. | |
3770 | */ | |
3771 | if (!cgrp->nr_css) | |
3772 | cgroup_destroy_css_killed(cgrp); | |
3773 | ||
2bd59d48 TH |
3774 | /* remove @cgrp directory along with the base files */ |
3775 | mutex_unlock(&cgroup_mutex); | |
3776 | ||
455050d2 | 3777 | /* |
2bd59d48 TH |
3778 | * There are two control paths which try to determine cgroup from |
3779 | * dentry without going through kernfs - cgroupstats_build() and | |
3780 | * css_tryget_from_dir(). Those are supported by RCU protecting | |
3781 | * clearing of cgrp->kn->priv backpointer, which should happen | |
3782 | * after all files under it have been removed. | |
455050d2 | 3783 | */ |
6f30558f | 3784 | kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ |
2bd59d48 | 3785 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); |
2bd59d48 | 3786 | |
4ac06017 | 3787 | mutex_lock(&cgroup_mutex); |
455050d2 | 3788 | |
ea15f8cc TH |
3789 | return 0; |
3790 | }; | |
3791 | ||
d3daf28d | 3792 | /** |
f20104de | 3793 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
3794 | * @work: cgroup->destroy_free_work |
3795 | * | |
3796 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
3797 | * destroyed after all css's are offlined and performs the rest of |
3798 | * destruction. This is the second step of destruction described in the | |
3799 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 3800 | */ |
f20104de | 3801 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 3802 | { |
ea15f8cc | 3803 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 3804 | |
ace2bee8 | 3805 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 3806 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 3807 | |
999cd8a4 | 3808 | /* delete this cgroup from parent->children */ |
eb6fd504 | 3809 | list_del_rcu(&cgrp->sibling); |
ed957793 | 3810 | |
59f5296b | 3811 | cgroup_put(cgrp); |
ddbcc7e8 | 3812 | |
bd89aabc | 3813 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 3814 | check_for_release(parent); |
ddbcc7e8 PM |
3815 | } |
3816 | ||
2bd59d48 | 3817 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 3818 | { |
2bd59d48 TH |
3819 | struct cgroup *cgrp = kn->priv; |
3820 | int ret = 0; | |
3821 | ||
3822 | /* | |
3823 | * This is self-destruction but @kn can't be removed while this | |
3824 | * callback is in progress. Let's break active protection. Once | |
3825 | * the protection is broken, @cgrp can be destroyed at any point. | |
3826 | * Pin it so that it stays accessible. | |
3827 | */ | |
3828 | cgroup_get(cgrp); | |
3829 | kernfs_break_active_protection(kn); | |
42809dd4 | 3830 | |
ace2bee8 | 3831 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 3832 | mutex_lock(&cgroup_mutex); |
2bd59d48 TH |
3833 | |
3834 | /* | |
3835 | * @cgrp might already have been destroyed while we're trying to | |
3836 | * grab the mutexes. | |
3837 | */ | |
3838 | if (!cgroup_is_dead(cgrp)) | |
3839 | ret = cgroup_destroy_locked(cgrp); | |
3840 | ||
42809dd4 | 3841 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3842 | mutex_unlock(&cgroup_tree_mutex); |
42809dd4 | 3843 | |
2bd59d48 TH |
3844 | kernfs_unbreak_active_protection(kn); |
3845 | cgroup_put(cgrp); | |
42809dd4 TH |
3846 | return ret; |
3847 | } | |
3848 | ||
2bd59d48 TH |
3849 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
3850 | .remount_fs = cgroup_remount, | |
3851 | .show_options = cgroup_show_options, | |
3852 | .mkdir = cgroup_mkdir, | |
3853 | .rmdir = cgroup_rmdir, | |
3854 | .rename = cgroup_rename, | |
3855 | }; | |
3856 | ||
06a11920 | 3857 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 3858 | { |
ddbcc7e8 | 3859 | struct cgroup_subsys_state *css; |
cfe36bde DC |
3860 | |
3861 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 3862 | |
ace2bee8 | 3863 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
3864 | mutex_lock(&cgroup_mutex); |
3865 | ||
0adb0704 | 3866 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 3867 | |
ddbcc7e8 | 3868 | /* Create the top cgroup state for this subsystem */ |
9871bf95 | 3869 | ss->root = &cgroup_dummy_root; |
ca8bdcaf | 3870 | css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss)); |
ddbcc7e8 PM |
3871 | /* We don't handle early failures gracefully */ |
3872 | BUG_ON(IS_ERR(css)); | |
623f926b | 3873 | init_css(css, ss, cgroup_dummy_top); |
ddbcc7e8 | 3874 | |
e8d55fde | 3875 | /* Update the init_css_set to contain a subsys |
817929ec | 3876 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
3877 | * newly registered, all tasks and hence the |
3878 | * init_css_set is in the subsystem's top cgroup. */ | |
aec25020 | 3879 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
3880 | |
3881 | need_forkexit_callback |= ss->fork || ss->exit; | |
3882 | ||
e8d55fde LZ |
3883 | /* At system boot, before all subsystems have been |
3884 | * registered, no tasks have been forked, so we don't | |
3885 | * need to invoke fork callbacks here. */ | |
3886 | BUG_ON(!list_empty(&init_task.tasks)); | |
3887 | ||
ae7f164a | 3888 | BUG_ON(online_css(css)); |
a8638030 | 3889 | |
648bb56d | 3890 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3891 | mutex_unlock(&cgroup_tree_mutex); |
e6a1105b BB |
3892 | } |
3893 | ||
ddbcc7e8 | 3894 | /** |
a043e3b2 LZ |
3895 | * cgroup_init_early - cgroup initialization at system boot |
3896 | * | |
3897 | * Initialize cgroups at system boot, and initialize any | |
3898 | * subsystems that request early init. | |
ddbcc7e8 PM |
3899 | */ |
3900 | int __init cgroup_init_early(void) | |
3901 | { | |
30159ec7 | 3902 | struct cgroup_subsys *ss; |
ddbcc7e8 | 3903 | int i; |
30159ec7 | 3904 | |
146aa1bd | 3905 | atomic_set(&init_css_set.refcount, 1); |
69d0206c | 3906 | INIT_LIST_HEAD(&init_css_set.cgrp_links); |
817929ec | 3907 | INIT_LIST_HEAD(&init_css_set.tasks); |
b3dc094e TH |
3908 | INIT_LIST_HEAD(&init_css_set.mg_tasks); |
3909 | INIT_LIST_HEAD(&init_css_set.mg_node); | |
472b1053 | 3910 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 3911 | css_set_count = 1; |
9871bf95 TH |
3912 | init_cgroup_root(&cgroup_dummy_root); |
3913 | cgroup_root_count = 1; | |
a4ea1cc9 | 3914 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 3915 | |
69d0206c | 3916 | init_cgrp_cset_link.cset = &init_css_set; |
9871bf95 TH |
3917 | init_cgrp_cset_link.cgrp = cgroup_dummy_top; |
3918 | list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links); | |
69d0206c | 3919 | list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links); |
ddbcc7e8 | 3920 | |
3ed80a62 | 3921 | for_each_subsys(ss, i) { |
aec25020 | 3922 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
3923 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
3924 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 3925 | ss->id, ss->name); |
073219e9 TH |
3926 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
3927 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
3928 | ||
aec25020 | 3929 | ss->id = i; |
073219e9 | 3930 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
3931 | |
3932 | if (ss->early_init) | |
3933 | cgroup_init_subsys(ss); | |
3934 | } | |
3935 | return 0; | |
3936 | } | |
3937 | ||
3938 | /** | |
a043e3b2 LZ |
3939 | * cgroup_init - cgroup initialization |
3940 | * | |
3941 | * Register cgroup filesystem and /proc file, and initialize | |
3942 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
3943 | */ |
3944 | int __init cgroup_init(void) | |
3945 | { | |
30159ec7 | 3946 | struct cgroup_subsys *ss; |
0ac801fe | 3947 | unsigned long key; |
30159ec7 | 3948 | int i, err; |
a424316c | 3949 | |
2bd59d48 | 3950 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
2da440a2 | 3951 | |
3ed80a62 | 3952 | for_each_subsys(ss, i) { |
ddbcc7e8 PM |
3953 | if (!ss->early_init) |
3954 | cgroup_init_subsys(ss); | |
de00ffa5 TH |
3955 | |
3956 | /* | |
3957 | * cftype registration needs kmalloc and can't be done | |
3958 | * during early_init. Register base cftypes separately. | |
3959 | */ | |
3960 | if (ss->base_cftypes) | |
3961 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
ddbcc7e8 PM |
3962 | } |
3963 | ||
fa3ca07e | 3964 | /* allocate id for the dummy hierarchy */ |
54e7b4eb | 3965 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 3966 | |
82fe9b0d TH |
3967 | /* Add init_css_set to the hash table */ |
3968 | key = css_set_hash(init_css_set.subsys); | |
3969 | hash_add(css_set_table, &init_css_set.hlist, key); | |
3970 | ||
fc76df70 | 3971 | BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1)); |
676db4af | 3972 | |
4e96ee8e LZ |
3973 | err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top, |
3974 | 0, 1, GFP_KERNEL); | |
3975 | BUG_ON(err < 0); | |
3976 | ||
54e7b4eb TH |
3977 | mutex_unlock(&cgroup_mutex); |
3978 | ||
676db4af | 3979 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
3980 | if (!cgroup_kobj) |
3981 | return -ENOMEM; | |
676db4af | 3982 | |
ddbcc7e8 | 3983 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
3984 | if (err < 0) { |
3985 | kobject_put(cgroup_kobj); | |
2bd59d48 | 3986 | return err; |
676db4af | 3987 | } |
ddbcc7e8 | 3988 | |
46ae220b | 3989 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 3990 | return 0; |
ddbcc7e8 | 3991 | } |
b4f48b63 | 3992 | |
e5fca243 TH |
3993 | static int __init cgroup_wq_init(void) |
3994 | { | |
3995 | /* | |
3996 | * There isn't much point in executing destruction path in | |
3997 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 3998 | * Use 1 for @max_active. |
e5fca243 TH |
3999 | * |
4000 | * We would prefer to do this in cgroup_init() above, but that | |
4001 | * is called before init_workqueues(): so leave this until after. | |
4002 | */ | |
1a11533f | 4003 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 4004 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4005 | |
4006 | /* | |
4007 | * Used to destroy pidlists and separate to serve as flush domain. | |
4008 | * Cap @max_active to 1 too. | |
4009 | */ | |
4010 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4011 | 0, 1); | |
4012 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4013 | ||
e5fca243 TH |
4014 | return 0; |
4015 | } | |
4016 | core_initcall(cgroup_wq_init); | |
4017 | ||
a424316c PM |
4018 | /* |
4019 | * proc_cgroup_show() | |
4020 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4021 | * - Used for /proc/<pid>/cgroup. | |
4022 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4023 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4024 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4025 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4026 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4027 | * cgroup to top_cgroup. | |
4028 | */ | |
4029 | ||
4030 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4031 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4032 | { |
4033 | struct pid *pid; | |
4034 | struct task_struct *tsk; | |
e61734c5 | 4035 | char *buf, *path; |
a424316c PM |
4036 | int retval; |
4037 | struct cgroupfs_root *root; | |
4038 | ||
4039 | retval = -ENOMEM; | |
e61734c5 | 4040 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4041 | if (!buf) |
4042 | goto out; | |
4043 | ||
4044 | retval = -ESRCH; | |
4045 | pid = m->private; | |
4046 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4047 | if (!tsk) | |
4048 | goto out_free; | |
4049 | ||
4050 | retval = 0; | |
4051 | ||
4052 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 4053 | down_read(&css_set_rwsem); |
a424316c | 4054 | |
e5f6a860 | 4055 | for_each_active_root(root) { |
a424316c | 4056 | struct cgroup_subsys *ss; |
bd89aabc | 4057 | struct cgroup *cgrp; |
b85d2040 | 4058 | int ssid, count = 0; |
a424316c | 4059 | |
2c6ab6d2 | 4060 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 TH |
4061 | for_each_subsys(ss, ssid) |
4062 | if (root->subsys_mask & (1 << ssid)) | |
4063 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4064 | if (strlen(root->name)) |
4065 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4066 | root->name); | |
a424316c | 4067 | seq_putc(m, ':'); |
7717f7ba | 4068 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
4069 | path = cgroup_path(cgrp, buf, PATH_MAX); |
4070 | if (!path) { | |
4071 | retval = -ENAMETOOLONG; | |
a424316c | 4072 | goto out_unlock; |
e61734c5 TH |
4073 | } |
4074 | seq_puts(m, path); | |
a424316c PM |
4075 | seq_putc(m, '\n'); |
4076 | } | |
4077 | ||
4078 | out_unlock: | |
96d365e0 | 4079 | up_read(&css_set_rwsem); |
a424316c PM |
4080 | mutex_unlock(&cgroup_mutex); |
4081 | put_task_struct(tsk); | |
4082 | out_free: | |
4083 | kfree(buf); | |
4084 | out: | |
4085 | return retval; | |
4086 | } | |
4087 | ||
a424316c PM |
4088 | /* Display information about each subsystem and each hierarchy */ |
4089 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4090 | { | |
30159ec7 | 4091 | struct cgroup_subsys *ss; |
a424316c | 4092 | int i; |
a424316c | 4093 | |
8bab8dde | 4094 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4095 | /* |
4096 | * ideally we don't want subsystems moving around while we do this. | |
4097 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4098 | * subsys/hierarchy state. | |
4099 | */ | |
a424316c | 4100 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4101 | |
4102 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4103 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4104 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 4105 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 4106 | |
a424316c PM |
4107 | mutex_unlock(&cgroup_mutex); |
4108 | return 0; | |
4109 | } | |
4110 | ||
4111 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4112 | { | |
9dce07f1 | 4113 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4114 | } |
4115 | ||
828c0950 | 4116 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4117 | .open = cgroupstats_open, |
4118 | .read = seq_read, | |
4119 | .llseek = seq_lseek, | |
4120 | .release = single_release, | |
4121 | }; | |
4122 | ||
b4f48b63 PM |
4123 | /** |
4124 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4125 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4126 | * |
4127 | * Description: A task inherits its parent's cgroup at fork(). | |
4128 | * | |
4129 | * A pointer to the shared css_set was automatically copied in | |
4130 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
9bb71308 TH |
4131 | * it was not made under the protection of RCU or cgroup_mutex, so |
4132 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
4133 | * have already changed current->cgroups, allowing the previously | |
4134 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
4135 | * |
4136 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4137 | * task, and the passed argument 'child' points to the child task. | |
4138 | */ | |
4139 | void cgroup_fork(struct task_struct *child) | |
4140 | { | |
9bb71308 | 4141 | task_lock(current); |
a8ad805c | 4142 | get_css_set(task_css_set(current)); |
817929ec | 4143 | child->cgroups = current->cgroups; |
9bb71308 | 4144 | task_unlock(current); |
817929ec | 4145 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4146 | } |
4147 | ||
817929ec | 4148 | /** |
a043e3b2 LZ |
4149 | * cgroup_post_fork - called on a new task after adding it to the task list |
4150 | * @child: the task in question | |
4151 | * | |
5edee61e TH |
4152 | * Adds the task to the list running through its css_set if necessary and |
4153 | * call the subsystem fork() callbacks. Has to be after the task is | |
4154 | * visible on the task list in case we race with the first call to | |
0942eeee | 4155 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4156 | * list. |
a043e3b2 | 4157 | */ |
817929ec PM |
4158 | void cgroup_post_fork(struct task_struct *child) |
4159 | { | |
30159ec7 | 4160 | struct cgroup_subsys *ss; |
5edee61e TH |
4161 | int i; |
4162 | ||
3ce3230a FW |
4163 | /* |
4164 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
4165 | * under the tasklist_lock and we read it here after we added the child | |
4166 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
4167 | * yet in the tasklist when we walked through it from | |
4168 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
4169 | * should be visible now due to the paired locking and barriers implied | |
4170 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
4171 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
4172 | * lock on fork. | |
4173 | */ | |
817929ec | 4174 | if (use_task_css_set_links) { |
96d365e0 | 4175 | down_write(&css_set_rwsem); |
d8783832 TH |
4176 | task_lock(child); |
4177 | if (list_empty(&child->cg_list)) | |
a8ad805c | 4178 | list_add(&child->cg_list, &task_css_set(child)->tasks); |
d8783832 | 4179 | task_unlock(child); |
96d365e0 | 4180 | up_write(&css_set_rwsem); |
817929ec | 4181 | } |
5edee61e TH |
4182 | |
4183 | /* | |
4184 | * Call ss->fork(). This must happen after @child is linked on | |
4185 | * css_set; otherwise, @child might change state between ->fork() | |
4186 | * and addition to css_set. | |
4187 | */ | |
4188 | if (need_forkexit_callback) { | |
3ed80a62 | 4189 | for_each_subsys(ss, i) |
5edee61e TH |
4190 | if (ss->fork) |
4191 | ss->fork(child); | |
5edee61e | 4192 | } |
817929ec | 4193 | } |
5edee61e | 4194 | |
b4f48b63 PM |
4195 | /** |
4196 | * cgroup_exit - detach cgroup from exiting task | |
4197 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4198 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4199 | * |
4200 | * Description: Detach cgroup from @tsk and release it. | |
4201 | * | |
4202 | * Note that cgroups marked notify_on_release force every task in | |
4203 | * them to take the global cgroup_mutex mutex when exiting. | |
4204 | * This could impact scaling on very large systems. Be reluctant to | |
4205 | * use notify_on_release cgroups where very high task exit scaling | |
4206 | * is required on large systems. | |
4207 | * | |
4208 | * the_top_cgroup_hack: | |
4209 | * | |
4210 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4211 | * | |
4212 | * We call cgroup_exit() while the task is still competent to | |
4213 | * handle notify_on_release(), then leave the task attached to the | |
4214 | * root cgroup in each hierarchy for the remainder of its exit. | |
4215 | * | |
4216 | * To do this properly, we would increment the reference count on | |
4217 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4218 | * code we would add a second cgroup function call, to drop that | |
4219 | * reference. This would just create an unnecessary hot spot on | |
4220 | * the top_cgroup reference count, to no avail. | |
4221 | * | |
4222 | * Normally, holding a reference to a cgroup without bumping its | |
4223 | * count is unsafe. The cgroup could go away, or someone could | |
4224 | * attach us to a different cgroup, decrementing the count on | |
4225 | * the first cgroup that we never incremented. But in this case, | |
4226 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4227 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4228 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4229 | */ |
4230 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4231 | { | |
30159ec7 | 4232 | struct cgroup_subsys *ss; |
5abb8855 | 4233 | struct css_set *cset; |
d41d5a01 | 4234 | int i; |
817929ec PM |
4235 | |
4236 | /* | |
96d365e0 TH |
4237 | * Unlink from the css_set task list if necessary. Optimistically |
4238 | * check cg_list before taking css_set_rwsem. | |
817929ec PM |
4239 | */ |
4240 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 4241 | down_write(&css_set_rwsem); |
817929ec | 4242 | if (!list_empty(&tsk->cg_list)) |
8d258797 | 4243 | list_del_init(&tsk->cg_list); |
96d365e0 | 4244 | up_write(&css_set_rwsem); |
817929ec PM |
4245 | } |
4246 | ||
b4f48b63 PM |
4247 | /* Reassign the task to the init_css_set. */ |
4248 | task_lock(tsk); | |
a8ad805c TH |
4249 | cset = task_css_set(tsk); |
4250 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 PZ |
4251 | |
4252 | if (run_callbacks && need_forkexit_callback) { | |
3ed80a62 TH |
4253 | /* see cgroup_post_fork() for details */ |
4254 | for_each_subsys(ss, i) { | |
d41d5a01 | 4255 | if (ss->exit) { |
eb95419b TH |
4256 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4257 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4258 | |
eb95419b | 4259 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4260 | } |
4261 | } | |
4262 | } | |
b4f48b63 | 4263 | task_unlock(tsk); |
d41d5a01 | 4264 | |
89c5509b | 4265 | put_css_set(cset, true); |
b4f48b63 | 4266 | } |
697f4161 | 4267 | |
bd89aabc | 4268 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4269 | { |
f50daa70 | 4270 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4271 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4272 | /* |
4273 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4274 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4275 | * it now |
4276 | */ | |
81a6a5cd | 4277 | int need_schedule_work = 0; |
f50daa70 | 4278 | |
cdcc136f | 4279 | raw_spin_lock(&release_list_lock); |
54766d4a | 4280 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4281 | list_empty(&cgrp->release_list)) { |
4282 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4283 | need_schedule_work = 1; |
4284 | } | |
cdcc136f | 4285 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4286 | if (need_schedule_work) |
4287 | schedule_work(&release_agent_work); | |
4288 | } | |
4289 | } | |
4290 | ||
81a6a5cd PM |
4291 | /* |
4292 | * Notify userspace when a cgroup is released, by running the | |
4293 | * configured release agent with the name of the cgroup (path | |
4294 | * relative to the root of cgroup file system) as the argument. | |
4295 | * | |
4296 | * Most likely, this user command will try to rmdir this cgroup. | |
4297 | * | |
4298 | * This races with the possibility that some other task will be | |
4299 | * attached to this cgroup before it is removed, or that some other | |
4300 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4301 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4302 | * unused, and this cgroup will be reprieved from its death sentence, | |
4303 | * to continue to serve a useful existence. Next time it's released, | |
4304 | * we will get notified again, if it still has 'notify_on_release' set. | |
4305 | * | |
4306 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4307 | * means only wait until the task is successfully execve()'d. The | |
4308 | * separate release agent task is forked by call_usermodehelper(), | |
4309 | * then control in this thread returns here, without waiting for the | |
4310 | * release agent task. We don't bother to wait because the caller of | |
4311 | * this routine has no use for the exit status of the release agent | |
4312 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4313 | */ |
81a6a5cd PM |
4314 | static void cgroup_release_agent(struct work_struct *work) |
4315 | { | |
4316 | BUG_ON(work != &release_agent_work); | |
4317 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4318 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4319 | while (!list_empty(&release_list)) { |
4320 | char *argv[3], *envp[3]; | |
4321 | int i; | |
e61734c5 | 4322 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 4323 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4324 | struct cgroup, |
4325 | release_list); | |
bd89aabc | 4326 | list_del_init(&cgrp->release_list); |
cdcc136f | 4327 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 4328 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
4329 | if (!pathbuf) |
4330 | goto continue_free; | |
e61734c5 TH |
4331 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
4332 | if (!path) | |
e788e066 PM |
4333 | goto continue_free; |
4334 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4335 | if (!agentbuf) | |
4336 | goto continue_free; | |
81a6a5cd PM |
4337 | |
4338 | i = 0; | |
e788e066 | 4339 | argv[i++] = agentbuf; |
e61734c5 | 4340 | argv[i++] = path; |
81a6a5cd PM |
4341 | argv[i] = NULL; |
4342 | ||
4343 | i = 0; | |
4344 | /* minimal command environment */ | |
4345 | envp[i++] = "HOME=/"; | |
4346 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4347 | envp[i] = NULL; | |
4348 | ||
4349 | /* Drop the lock while we invoke the usermode helper, | |
4350 | * since the exec could involve hitting disk and hence | |
4351 | * be a slow process */ | |
4352 | mutex_unlock(&cgroup_mutex); | |
4353 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4354 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4355 | continue_free: |
4356 | kfree(pathbuf); | |
4357 | kfree(agentbuf); | |
cdcc136f | 4358 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4359 | } |
cdcc136f | 4360 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4361 | mutex_unlock(&cgroup_mutex); |
4362 | } | |
8bab8dde PM |
4363 | |
4364 | static int __init cgroup_disable(char *str) | |
4365 | { | |
30159ec7 | 4366 | struct cgroup_subsys *ss; |
8bab8dde | 4367 | char *token; |
30159ec7 | 4368 | int i; |
8bab8dde PM |
4369 | |
4370 | while ((token = strsep(&str, ",")) != NULL) { | |
4371 | if (!*token) | |
4372 | continue; | |
be45c900 | 4373 | |
3ed80a62 | 4374 | for_each_subsys(ss, i) { |
8bab8dde PM |
4375 | if (!strcmp(token, ss->name)) { |
4376 | ss->disabled = 1; | |
4377 | printk(KERN_INFO "Disabling %s control group" | |
4378 | " subsystem\n", ss->name); | |
4379 | break; | |
4380 | } | |
4381 | } | |
4382 | } | |
4383 | return 1; | |
4384 | } | |
4385 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 4386 | |
b77d7b60 | 4387 | /** |
5a17f543 | 4388 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
4389 | * @dentry: directory dentry of interest |
4390 | * @ss: subsystem of interest | |
b77d7b60 | 4391 | * |
5a17f543 TH |
4392 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
4393 | * to get the corresponding css and return it. If such css doesn't exist | |
4394 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 4395 | */ |
5a17f543 TH |
4396 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
4397 | struct cgroup_subsys *ss) | |
e5d1367f | 4398 | { |
2bd59d48 TH |
4399 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
4400 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 4401 | struct cgroup *cgrp; |
b77d7b60 | 4402 | |
35cf0836 | 4403 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
4404 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
4405 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
4406 | return ERR_PTR(-EBADF); |
4407 | ||
5a17f543 TH |
4408 | rcu_read_lock(); |
4409 | ||
2bd59d48 TH |
4410 | /* |
4411 | * This path doesn't originate from kernfs and @kn could already | |
4412 | * have been or be removed at any point. @kn->priv is RCU | |
4413 | * protected for this access. See destroy_locked() for details. | |
4414 | */ | |
4415 | cgrp = rcu_dereference(kn->priv); | |
4416 | if (cgrp) | |
4417 | css = cgroup_css(cgrp, ss); | |
5a17f543 TH |
4418 | |
4419 | if (!css || !css_tryget(css)) | |
4420 | css = ERR_PTR(-ENOENT); | |
4421 | ||
4422 | rcu_read_unlock(); | |
4423 | return css; | |
e5d1367f | 4424 | } |
e5d1367f | 4425 | |
1cb650b9 LZ |
4426 | /** |
4427 | * css_from_id - lookup css by id | |
4428 | * @id: the cgroup id | |
4429 | * @ss: cgroup subsys to be looked into | |
4430 | * | |
4431 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
4432 | * Should be called under rcu_read_lock(). | |
4433 | */ | |
4434 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
4435 | { | |
4436 | struct cgroup *cgrp; | |
4437 | ||
ace2bee8 | 4438 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
4439 | |
4440 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
4441 | if (cgrp) | |
d1625964 | 4442 | return cgroup_css(cgrp, ss); |
1cb650b9 | 4443 | return NULL; |
e5d1367f SE |
4444 | } |
4445 | ||
fe693435 | 4446 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
4447 | static struct cgroup_subsys_state * |
4448 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
4449 | { |
4450 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
4451 | ||
4452 | if (!css) | |
4453 | return ERR_PTR(-ENOMEM); | |
4454 | ||
4455 | return css; | |
4456 | } | |
4457 | ||
eb95419b | 4458 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 4459 | { |
eb95419b | 4460 | kfree(css); |
fe693435 PM |
4461 | } |
4462 | ||
182446d0 TH |
4463 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
4464 | struct cftype *cft) | |
fe693435 | 4465 | { |
182446d0 | 4466 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
4467 | } |
4468 | ||
182446d0 TH |
4469 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
4470 | struct cftype *cft) | |
fe693435 PM |
4471 | { |
4472 | return (u64)(unsigned long)current->cgroups; | |
4473 | } | |
4474 | ||
182446d0 | 4475 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 4476 | struct cftype *cft) |
fe693435 PM |
4477 | { |
4478 | u64 count; | |
4479 | ||
4480 | rcu_read_lock(); | |
a8ad805c | 4481 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
4482 | rcu_read_unlock(); |
4483 | return count; | |
4484 | } | |
4485 | ||
2da8ca82 | 4486 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4487 | { |
69d0206c | 4488 | struct cgrp_cset_link *link; |
5abb8855 | 4489 | struct css_set *cset; |
e61734c5 TH |
4490 | char *name_buf; |
4491 | ||
4492 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
4493 | if (!name_buf) | |
4494 | return -ENOMEM; | |
7717f7ba | 4495 | |
96d365e0 | 4496 | down_read(&css_set_rwsem); |
7717f7ba | 4497 | rcu_read_lock(); |
5abb8855 | 4498 | cset = rcu_dereference(current->cgroups); |
69d0206c | 4499 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 4500 | struct cgroup *c = link->cgrp; |
59f5296b TH |
4501 | const char *name = "?"; |
4502 | ||
e61734c5 TH |
4503 | if (c != cgroup_dummy_top) { |
4504 | cgroup_name(c, name_buf, NAME_MAX + 1); | |
4505 | name = name_buf; | |
4506 | } | |
7717f7ba | 4507 | |
2c6ab6d2 PM |
4508 | seq_printf(seq, "Root %d group %s\n", |
4509 | c->root->hierarchy_id, name); | |
7717f7ba PM |
4510 | } |
4511 | rcu_read_unlock(); | |
96d365e0 | 4512 | up_read(&css_set_rwsem); |
e61734c5 | 4513 | kfree(name_buf); |
7717f7ba PM |
4514 | return 0; |
4515 | } | |
4516 | ||
4517 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 4518 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4519 | { |
2da8ca82 | 4520 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 4521 | struct cgrp_cset_link *link; |
7717f7ba | 4522 | |
96d365e0 | 4523 | down_read(&css_set_rwsem); |
182446d0 | 4524 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 4525 | struct css_set *cset = link->cset; |
7717f7ba PM |
4526 | struct task_struct *task; |
4527 | int count = 0; | |
c7561128 | 4528 | |
5abb8855 | 4529 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 4530 | |
5abb8855 | 4531 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
4532 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
4533 | goto overflow; | |
4534 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
4535 | } | |
4536 | ||
4537 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
4538 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
4539 | goto overflow; | |
4540 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 4541 | } |
c7561128 TH |
4542 | continue; |
4543 | overflow: | |
4544 | seq_puts(seq, " ...\n"); | |
7717f7ba | 4545 | } |
96d365e0 | 4546 | up_read(&css_set_rwsem); |
7717f7ba PM |
4547 | return 0; |
4548 | } | |
4549 | ||
182446d0 | 4550 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 4551 | { |
182446d0 | 4552 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
4553 | } |
4554 | ||
4555 | static struct cftype debug_files[] = { | |
fe693435 PM |
4556 | { |
4557 | .name = "taskcount", | |
4558 | .read_u64 = debug_taskcount_read, | |
4559 | }, | |
4560 | ||
4561 | { | |
4562 | .name = "current_css_set", | |
4563 | .read_u64 = current_css_set_read, | |
4564 | }, | |
4565 | ||
4566 | { | |
4567 | .name = "current_css_set_refcount", | |
4568 | .read_u64 = current_css_set_refcount_read, | |
4569 | }, | |
4570 | ||
7717f7ba PM |
4571 | { |
4572 | .name = "current_css_set_cg_links", | |
2da8ca82 | 4573 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
4574 | }, |
4575 | ||
4576 | { | |
4577 | .name = "cgroup_css_links", | |
2da8ca82 | 4578 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
4579 | }, |
4580 | ||
fe693435 PM |
4581 | { |
4582 | .name = "releasable", | |
4583 | .read_u64 = releasable_read, | |
4584 | }, | |
fe693435 | 4585 | |
4baf6e33 TH |
4586 | { } /* terminate */ |
4587 | }; | |
fe693435 | 4588 | |
073219e9 | 4589 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
4590 | .css_alloc = debug_css_alloc, |
4591 | .css_free = debug_css_free, | |
4baf6e33 | 4592 | .base_cftypes = debug_files, |
fe693435 PM |
4593 | }; |
4594 | #endif /* CONFIG_CGROUP_DEBUG */ |