Commit | Line | Data |
---|---|---|
ddbcc7e8 | 1 | /* |
ddbcc7e8 PM |
2 | * Generic process-grouping system. |
3 | * | |
4 | * Based originally on the cpuset system, extracted by Paul Menage | |
5 | * Copyright (C) 2006 Google, Inc | |
6 | * | |
0dea1168 KS |
7 | * Notifications support |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
ddbcc7e8 PM |
11 | * Copyright notices from the original cpuset code: |
12 | * -------------------------------------------------- | |
13 | * Copyright (C) 2003 BULL SA. | |
14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
15 | * | |
16 | * Portions derived from Patrick Mochel's sysfs code. | |
17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
18 | * | |
19 | * 2003-10-10 Written by Simon Derr. | |
20 | * 2003-10-22 Updates by Stephen Hemminger. | |
21 | * 2004 May-July Rework by Paul Jackson. | |
22 | * --------------------------------------------------- | |
23 | * | |
24 | * This file is subject to the terms and conditions of the GNU General Public | |
25 | * License. See the file COPYING in the main directory of the Linux | |
26 | * distribution for more details. | |
27 | */ | |
28 | ||
29 | #include <linux/cgroup.h> | |
2ce9738b | 30 | #include <linux/cred.h> |
c6d57f33 | 31 | #include <linux/ctype.h> |
ddbcc7e8 | 32 | #include <linux/errno.h> |
2ce9738b | 33 | #include <linux/init_task.h> |
ddbcc7e8 PM |
34 | #include <linux/kernel.h> |
35 | #include <linux/list.h> | |
36 | #include <linux/mm.h> | |
37 | #include <linux/mutex.h> | |
38 | #include <linux/mount.h> | |
39 | #include <linux/pagemap.h> | |
a424316c | 40 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
41 | #include <linux/rcupdate.h> |
42 | #include <linux/sched.h> | |
ddbcc7e8 | 43 | #include <linux/slab.h> |
ddbcc7e8 | 44 | #include <linux/spinlock.h> |
96d365e0 | 45 | #include <linux/rwsem.h> |
ddbcc7e8 | 46 | #include <linux/string.h> |
bbcb81d0 | 47 | #include <linux/sort.h> |
81a6a5cd | 48 | #include <linux/kmod.h> |
846c7bb0 BS |
49 | #include <linux/delayacct.h> |
50 | #include <linux/cgroupstats.h> | |
0ac801fe | 51 | #include <linux/hashtable.h> |
096b7fe0 | 52 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 53 | #include <linux/idr.h> |
d1d9fd33 | 54 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 55 | #include <linux/kthread.h> |
776f02fa | 56 | #include <linux/delay.h> |
846c7bb0 | 57 | |
60063497 | 58 | #include <linux/atomic.h> |
ddbcc7e8 | 59 | |
b1a21367 TH |
60 | /* |
61 | * pidlists linger the following amount before being destroyed. The goal | |
62 | * is avoiding frequent destruction in the middle of consecutive read calls | |
63 | * Expiring in the middle is a performance problem not a correctness one. | |
64 | * 1 sec should be enough. | |
65 | */ | |
66 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
67 | ||
8d7e6fb0 TH |
68 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
69 | MAX_CFTYPE_NAME + 2) | |
70 | ||
ace2bee8 TH |
71 | /* |
72 | * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file | |
73 | * creation/removal and hierarchy changing operations including cgroup | |
74 | * creation, removal, css association and controller rebinding. This outer | |
75 | * lock is needed mainly to resolve the circular dependency between kernfs | |
76 | * active ref and cgroup_mutex. cgroup_tree_mutex nests above both. | |
77 | */ | |
78 | static DEFINE_MUTEX(cgroup_tree_mutex); | |
79 | ||
e25e2cbb TH |
80 | /* |
81 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
82 | * hierarchy must be performed while holding it. | |
83 | * | |
0e1d768f TH |
84 | * css_set_rwsem protects task->cgroups pointer, the list of css_set |
85 | * objects, and the chain of tasks off each css_set. | |
e25e2cbb | 86 | * |
0e1d768f TH |
87 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
88 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 89 | */ |
2219449a TH |
90 | #ifdef CONFIG_PROVE_RCU |
91 | DEFINE_MUTEX(cgroup_mutex); | |
0e1d768f TH |
92 | DECLARE_RWSEM(css_set_rwsem); |
93 | EXPORT_SYMBOL_GPL(cgroup_mutex); | |
94 | EXPORT_SYMBOL_GPL(css_set_rwsem); | |
2219449a | 95 | #else |
81a6a5cd | 96 | static DEFINE_MUTEX(cgroup_mutex); |
0e1d768f | 97 | static DECLARE_RWSEM(css_set_rwsem); |
2219449a TH |
98 | #endif |
99 | ||
69e943b7 TH |
100 | /* |
101 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
102 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
103 | */ | |
104 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 105 | |
ace2bee8 | 106 | #define cgroup_assert_mutexes_or_rcu_locked() \ |
87fb54f1 | 107 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
ace2bee8 | 108 | lockdep_is_held(&cgroup_tree_mutex) || \ |
87fb54f1 | 109 | lockdep_is_held(&cgroup_mutex), \ |
ace2bee8 | 110 | "cgroup_[tree_]mutex or RCU read lock required"); |
780cd8b3 | 111 | |
e5fca243 TH |
112 | /* |
113 | * cgroup destruction makes heavy use of work items and there can be a lot | |
114 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
115 | * destruction work items don't end up filling up max_active of system_wq | |
116 | * which may lead to deadlock. | |
117 | */ | |
118 | static struct workqueue_struct *cgroup_destroy_wq; | |
119 | ||
b1a21367 TH |
120 | /* |
121 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
122 | * separate workqueue as flush domain. | |
123 | */ | |
124 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
125 | ||
3ed80a62 | 126 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 127 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 128 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
129 | #include <linux/cgroup_subsys.h> |
130 | }; | |
073219e9 TH |
131 | #undef SUBSYS |
132 | ||
133 | /* array of cgroup subsystem names */ | |
134 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
135 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
136 | #include <linux/cgroup_subsys.h> |
137 | }; | |
073219e9 | 138 | #undef SUBSYS |
ddbcc7e8 | 139 | |
ddbcc7e8 | 140 | /* |
3dd06ffa | 141 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
142 | * unattached - it never has more than a single cgroup, and all tasks are |
143 | * part of that cgroup. | |
ddbcc7e8 | 144 | */ |
a2dd4247 | 145 | struct cgroup_root cgrp_dfl_root; |
9871bf95 | 146 | |
a2dd4247 TH |
147 | /* |
148 | * The default hierarchy always exists but is hidden until mounted for the | |
149 | * first time. This is for backward compatibility. | |
150 | */ | |
151 | static bool cgrp_dfl_root_visible; | |
ddbcc7e8 PM |
152 | |
153 | /* The list of hierarchy roots */ | |
154 | ||
9871bf95 TH |
155 | static LIST_HEAD(cgroup_roots); |
156 | static int cgroup_root_count; | |
ddbcc7e8 | 157 | |
3417ae1f | 158 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 159 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 160 | |
794611a1 LZ |
161 | /* |
162 | * Assign a monotonically increasing serial number to cgroups. It | |
163 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
164 | * numbers. Also, as cgroups are always appended to the parent's | |
165 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
166 | * the ascending serial number order on the list. Protected by |
167 | * cgroup_mutex. | |
794611a1 | 168 | */ |
00356bd5 | 169 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 170 | |
ddbcc7e8 | 171 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
172 | * check for fork/exit handlers to call. This avoids us having to do |
173 | * extra work in the fork/exit path if none of the subsystems need to | |
174 | * be called. | |
ddbcc7e8 | 175 | */ |
8947f9d5 | 176 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 177 | |
628f7cd4 TH |
178 | static struct cftype cgroup_base_files[]; |
179 | ||
59f5296b | 180 | static void cgroup_put(struct cgroup *cgrp); |
3dd06ffa | 181 | static int rebind_subsystems(struct cgroup_root *dst_root, |
5df36032 | 182 | unsigned long ss_mask); |
f20104de | 183 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 184 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
f8f22e53 TH |
185 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss); |
186 | static void kill_css(struct cgroup_subsys_state *css); | |
2bb566cb TH |
187 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
188 | bool is_add); | |
b1a21367 | 189 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 190 | |
95109b62 TH |
191 | /** |
192 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
193 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 194 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 195 | * |
ca8bdcaf TH |
196 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
197 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
198 | * the caller is responsible for pinning the returned css if it wants to | |
199 | * keep accessing it outside the said locks. This function may return | |
200 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
201 | */ |
202 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 203 | struct cgroup_subsys *ss) |
95109b62 | 204 | { |
ca8bdcaf | 205 | if (ss) |
aec25020 | 206 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 TH |
207 | lockdep_is_held(&cgroup_tree_mutex) || |
208 | lockdep_is_held(&cgroup_mutex)); | |
ca8bdcaf TH |
209 | else |
210 | return &cgrp->dummy_css; | |
95109b62 | 211 | } |
42809dd4 | 212 | |
aec3dfcb TH |
213 | /** |
214 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
215 | * @cgrp: the cgroup of interest | |
216 | * @ss: the subsystem of interest (%NULL returns the dummy_css) | |
217 | * | |
218 | * Similar to cgroup_css() but returns the effctive css, which is defined | |
219 | * as the matching css of the nearest ancestor including self which has @ss | |
220 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
221 | * function is guaranteed to return non-NULL css. | |
222 | */ | |
223 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
224 | struct cgroup_subsys *ss) | |
225 | { | |
226 | lockdep_assert_held(&cgroup_mutex); | |
227 | ||
228 | if (!ss) | |
229 | return &cgrp->dummy_css; | |
230 | ||
231 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
232 | return NULL; | |
233 | ||
234 | while (cgrp->parent && | |
235 | !(cgrp->parent->child_subsys_mask & (1 << ss->id))) | |
236 | cgrp = cgrp->parent; | |
237 | ||
238 | return cgroup_css(cgrp, ss); | |
239 | } | |
240 | ||
ddbcc7e8 | 241 | /* convenient tests for these bits */ |
54766d4a | 242 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 243 | { |
54766d4a | 244 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
245 | } |
246 | ||
59f5296b TH |
247 | struct cgroup_subsys_state *seq_css(struct seq_file *seq) |
248 | { | |
2bd59d48 TH |
249 | struct kernfs_open_file *of = seq->private; |
250 | struct cgroup *cgrp = of->kn->parent->priv; | |
251 | struct cftype *cft = seq_cft(seq); | |
252 | ||
253 | /* | |
254 | * This is open and unprotected implementation of cgroup_css(). | |
255 | * seq_css() is only called from a kernfs file operation which has | |
256 | * an active reference on the file. Because all the subsystem | |
257 | * files are drained before a css is disassociated with a cgroup, | |
258 | * the matching css from the cgroup's subsys table is guaranteed to | |
259 | * be and stay valid until the enclosing operation is complete. | |
260 | */ | |
261 | if (cft->ss) | |
262 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
263 | else | |
264 | return &cgrp->dummy_css; | |
59f5296b TH |
265 | } |
266 | EXPORT_SYMBOL_GPL(seq_css); | |
267 | ||
78574cf9 LZ |
268 | /** |
269 | * cgroup_is_descendant - test ancestry | |
270 | * @cgrp: the cgroup to be tested | |
271 | * @ancestor: possible ancestor of @cgrp | |
272 | * | |
273 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
274 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
275 | * and @ancestor are accessible. | |
276 | */ | |
277 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
278 | { | |
279 | while (cgrp) { | |
280 | if (cgrp == ancestor) | |
281 | return true; | |
282 | cgrp = cgrp->parent; | |
283 | } | |
284 | return false; | |
285 | } | |
ddbcc7e8 | 286 | |
e9685a03 | 287 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
288 | { |
289 | const int bits = | |
bd89aabc PM |
290 | (1 << CGRP_RELEASABLE) | |
291 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
292 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
293 | } |
294 | ||
e9685a03 | 295 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 296 | { |
bd89aabc | 297 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
298 | } |
299 | ||
1c6727af TH |
300 | /** |
301 | * for_each_css - iterate all css's of a cgroup | |
302 | * @css: the iteration cursor | |
303 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
304 | * @cgrp: the target cgroup to iterate css's of | |
305 | * | |
aec3dfcb | 306 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
307 | */ |
308 | #define for_each_css(css, ssid, cgrp) \ | |
309 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
310 | if (!((css) = rcu_dereference_check( \ | |
311 | (cgrp)->subsys[(ssid)], \ | |
ace2bee8 | 312 | lockdep_is_held(&cgroup_tree_mutex) || \ |
1c6727af TH |
313 | lockdep_is_held(&cgroup_mutex)))) { } \ |
314 | else | |
315 | ||
aec3dfcb TH |
316 | /** |
317 | * for_each_e_css - iterate all effective css's of a cgroup | |
318 | * @css: the iteration cursor | |
319 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
320 | * @cgrp: the target cgroup to iterate css's of | |
321 | * | |
322 | * Should be called under cgroup_[tree_]mutex. | |
323 | */ | |
324 | #define for_each_e_css(css, ssid, cgrp) \ | |
325 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
326 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
327 | ; \ | |
328 | else | |
329 | ||
30159ec7 | 330 | /** |
3ed80a62 | 331 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 332 | * @ss: the iteration cursor |
780cd8b3 | 333 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 334 | */ |
780cd8b3 | 335 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
336 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
337 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 338 | |
985ed670 TH |
339 | /* iterate across the hierarchies */ |
340 | #define for_each_root(root) \ | |
5549c497 | 341 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 342 | |
f8f22e53 TH |
343 | /* iterate over child cgrps, lock should be held throughout iteration */ |
344 | #define cgroup_for_each_live_child(child, cgrp) \ | |
345 | list_for_each_entry((child), &(cgrp)->children, sibling) \ | |
346 | if (({ lockdep_assert_held(&cgroup_tree_mutex); \ | |
347 | cgroup_is_dead(child); })) \ | |
348 | ; \ | |
349 | else | |
350 | ||
7ae1bad9 TH |
351 | /** |
352 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
353 | * @cgrp: the cgroup to be checked for liveness | |
354 | * | |
47cfcd09 TH |
355 | * On success, returns true; the mutex should be later unlocked. On |
356 | * failure returns false with no lock held. | |
7ae1bad9 | 357 | */ |
b9777cf8 | 358 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
359 | { |
360 | mutex_lock(&cgroup_mutex); | |
54766d4a | 361 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
362 | mutex_unlock(&cgroup_mutex); |
363 | return false; | |
364 | } | |
365 | return true; | |
366 | } | |
7ae1bad9 | 367 | |
81a6a5cd PM |
368 | /* the list of cgroups eligible for automatic release. Protected by |
369 | * release_list_lock */ | |
370 | static LIST_HEAD(release_list); | |
cdcc136f | 371 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
372 | static void cgroup_release_agent(struct work_struct *work); |
373 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 374 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 375 | |
69d0206c TH |
376 | /* |
377 | * A cgroup can be associated with multiple css_sets as different tasks may | |
378 | * belong to different cgroups on different hierarchies. In the other | |
379 | * direction, a css_set is naturally associated with multiple cgroups. | |
380 | * This M:N relationship is represented by the following link structure | |
381 | * which exists for each association and allows traversing the associations | |
382 | * from both sides. | |
383 | */ | |
384 | struct cgrp_cset_link { | |
385 | /* the cgroup and css_set this link associates */ | |
386 | struct cgroup *cgrp; | |
387 | struct css_set *cset; | |
388 | ||
389 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
390 | struct list_head cset_link; | |
391 | ||
392 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
393 | struct list_head cgrp_link; | |
817929ec PM |
394 | }; |
395 | ||
172a2c06 TH |
396 | /* |
397 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
398 | * hierarchies being mounted. It contains a pointer to the root state |
399 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
400 | * reference-counted, to improve performance when child cgroups | |
401 | * haven't been created. | |
402 | */ | |
172a2c06 TH |
403 | static struct css_set init_css_set = { |
404 | .refcount = ATOMIC_INIT(1), | |
405 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
406 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
407 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
408 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
409 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
410 | }; | |
817929ec | 411 | |
172a2c06 | 412 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 413 | |
7717f7ba PM |
414 | /* |
415 | * hash table for cgroup groups. This improves the performance to find | |
416 | * an existing css_set. This hash doesn't (currently) take into | |
417 | * account cgroups in empty hierarchies. | |
418 | */ | |
472b1053 | 419 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 420 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 421 | |
0ac801fe | 422 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 423 | { |
0ac801fe | 424 | unsigned long key = 0UL; |
30159ec7 TH |
425 | struct cgroup_subsys *ss; |
426 | int i; | |
472b1053 | 427 | |
30159ec7 | 428 | for_each_subsys(ss, i) |
0ac801fe LZ |
429 | key += (unsigned long)css[i]; |
430 | key = (key >> 16) ^ key; | |
472b1053 | 431 | |
0ac801fe | 432 | return key; |
472b1053 LZ |
433 | } |
434 | ||
89c5509b | 435 | static void put_css_set_locked(struct css_set *cset, bool taskexit) |
b4f48b63 | 436 | { |
69d0206c | 437 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
438 | struct cgroup_subsys *ss; |
439 | int ssid; | |
5abb8855 | 440 | |
89c5509b TH |
441 | lockdep_assert_held(&css_set_rwsem); |
442 | ||
443 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 444 | return; |
81a6a5cd | 445 | |
2c6ab6d2 | 446 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
2d8f243a TH |
447 | for_each_subsys(ss, ssid) |
448 | list_del(&cset->e_cset_node[ssid]); | |
5abb8855 | 449 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
450 | css_set_count--; |
451 | ||
69d0206c | 452 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 453 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 454 | |
69d0206c TH |
455 | list_del(&link->cset_link); |
456 | list_del(&link->cgrp_link); | |
71b5707e | 457 | |
96d365e0 | 458 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
6f3d828f | 459 | if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { |
81a6a5cd | 460 | if (taskexit) |
bd89aabc PM |
461 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
462 | check_for_release(cgrp); | |
81a6a5cd | 463 | } |
2c6ab6d2 PM |
464 | |
465 | kfree(link); | |
81a6a5cd | 466 | } |
2c6ab6d2 | 467 | |
5abb8855 | 468 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
469 | } |
470 | ||
89c5509b TH |
471 | static void put_css_set(struct css_set *cset, bool taskexit) |
472 | { | |
473 | /* | |
474 | * Ensure that the refcount doesn't hit zero while any readers | |
475 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
476 | * rwlock | |
477 | */ | |
478 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
479 | return; | |
480 | ||
481 | down_write(&css_set_rwsem); | |
482 | put_css_set_locked(cset, taskexit); | |
483 | up_write(&css_set_rwsem); | |
484 | } | |
485 | ||
817929ec PM |
486 | /* |
487 | * refcounted get/put for css_set objects | |
488 | */ | |
5abb8855 | 489 | static inline void get_css_set(struct css_set *cset) |
817929ec | 490 | { |
5abb8855 | 491 | atomic_inc(&cset->refcount); |
817929ec PM |
492 | } |
493 | ||
b326f9d0 | 494 | /** |
7717f7ba | 495 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
496 | * @cset: candidate css_set being tested |
497 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
498 | * @new_cgrp: cgroup that's being entered by the task |
499 | * @template: desired set of css pointers in css_set (pre-calculated) | |
500 | * | |
6f4b7e63 | 501 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
502 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
503 | */ | |
5abb8855 TH |
504 | static bool compare_css_sets(struct css_set *cset, |
505 | struct css_set *old_cset, | |
7717f7ba PM |
506 | struct cgroup *new_cgrp, |
507 | struct cgroup_subsys_state *template[]) | |
508 | { | |
509 | struct list_head *l1, *l2; | |
510 | ||
aec3dfcb TH |
511 | /* |
512 | * On the default hierarchy, there can be csets which are | |
513 | * associated with the same set of cgroups but different csses. | |
514 | * Let's first ensure that csses match. | |
515 | */ | |
516 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 517 | return false; |
7717f7ba PM |
518 | |
519 | /* | |
520 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
521 | * different cgroups in hierarchies. As different cgroups may |
522 | * share the same effective css, this comparison is always | |
523 | * necessary. | |
7717f7ba | 524 | */ |
69d0206c TH |
525 | l1 = &cset->cgrp_links; |
526 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 527 | while (1) { |
69d0206c | 528 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 529 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
530 | |
531 | l1 = l1->next; | |
532 | l2 = l2->next; | |
533 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
534 | if (l1 == &cset->cgrp_links) { |
535 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
536 | break; |
537 | } else { | |
69d0206c | 538 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
539 | } |
540 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
541 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
542 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
543 | cgrp1 = link1->cgrp; | |
544 | cgrp2 = link2->cgrp; | |
7717f7ba | 545 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 546 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
547 | |
548 | /* | |
549 | * If this hierarchy is the hierarchy of the cgroup | |
550 | * that's changing, then we need to check that this | |
551 | * css_set points to the new cgroup; if it's any other | |
552 | * hierarchy, then this css_set should point to the | |
553 | * same cgroup as the old css_set. | |
554 | */ | |
5abb8855 TH |
555 | if (cgrp1->root == new_cgrp->root) { |
556 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
557 | return false; |
558 | } else { | |
5abb8855 | 559 | if (cgrp1 != cgrp2) |
7717f7ba PM |
560 | return false; |
561 | } | |
562 | } | |
563 | return true; | |
564 | } | |
565 | ||
b326f9d0 TH |
566 | /** |
567 | * find_existing_css_set - init css array and find the matching css_set | |
568 | * @old_cset: the css_set that we're using before the cgroup transition | |
569 | * @cgrp: the cgroup that we're moving into | |
570 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 571 | */ |
5abb8855 TH |
572 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
573 | struct cgroup *cgrp, | |
574 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 575 | { |
3dd06ffa | 576 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 577 | struct cgroup_subsys *ss; |
5abb8855 | 578 | struct css_set *cset; |
0ac801fe | 579 | unsigned long key; |
b326f9d0 | 580 | int i; |
817929ec | 581 | |
aae8aab4 BB |
582 | /* |
583 | * Build the set of subsystem state objects that we want to see in the | |
584 | * new css_set. while subsystems can change globally, the entries here | |
585 | * won't change, so no need for locking. | |
586 | */ | |
30159ec7 | 587 | for_each_subsys(ss, i) { |
f392e51c | 588 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
589 | /* |
590 | * @ss is in this hierarchy, so we want the | |
591 | * effective css from @cgrp. | |
592 | */ | |
593 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 594 | } else { |
aec3dfcb TH |
595 | /* |
596 | * @ss is not in this hierarchy, so we don't want | |
597 | * to change the css. | |
598 | */ | |
5abb8855 | 599 | template[i] = old_cset->subsys[i]; |
817929ec PM |
600 | } |
601 | } | |
602 | ||
0ac801fe | 603 | key = css_set_hash(template); |
5abb8855 TH |
604 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
605 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
606 | continue; |
607 | ||
608 | /* This css_set matches what we need */ | |
5abb8855 | 609 | return cset; |
472b1053 | 610 | } |
817929ec PM |
611 | |
612 | /* No existing cgroup group matched */ | |
613 | return NULL; | |
614 | } | |
615 | ||
69d0206c | 616 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 617 | { |
69d0206c | 618 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 619 | |
69d0206c TH |
620 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
621 | list_del(&link->cset_link); | |
36553434 LZ |
622 | kfree(link); |
623 | } | |
624 | } | |
625 | ||
69d0206c TH |
626 | /** |
627 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
628 | * @count: the number of links to allocate | |
629 | * @tmp_links: list_head the allocated links are put on | |
630 | * | |
631 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
632 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 633 | */ |
69d0206c | 634 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 635 | { |
69d0206c | 636 | struct cgrp_cset_link *link; |
817929ec | 637 | int i; |
69d0206c TH |
638 | |
639 | INIT_LIST_HEAD(tmp_links); | |
640 | ||
817929ec | 641 | for (i = 0; i < count; i++) { |
f4f4be2b | 642 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 643 | if (!link) { |
69d0206c | 644 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
645 | return -ENOMEM; |
646 | } | |
69d0206c | 647 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
648 | } |
649 | return 0; | |
650 | } | |
651 | ||
c12f65d4 LZ |
652 | /** |
653 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 654 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 655 | * @cset: the css_set to be linked |
c12f65d4 LZ |
656 | * @cgrp: the destination cgroup |
657 | */ | |
69d0206c TH |
658 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
659 | struct cgroup *cgrp) | |
c12f65d4 | 660 | { |
69d0206c | 661 | struct cgrp_cset_link *link; |
c12f65d4 | 662 | |
69d0206c | 663 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
664 | |
665 | if (cgroup_on_dfl(cgrp)) | |
666 | cset->dfl_cgrp = cgrp; | |
667 | ||
69d0206c TH |
668 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
669 | link->cset = cset; | |
7717f7ba | 670 | link->cgrp = cgrp; |
69d0206c | 671 | list_move(&link->cset_link, &cgrp->cset_links); |
7717f7ba PM |
672 | /* |
673 | * Always add links to the tail of the list so that the list | |
674 | * is sorted by order of hierarchy creation | |
675 | */ | |
69d0206c | 676 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
677 | } |
678 | ||
b326f9d0 TH |
679 | /** |
680 | * find_css_set - return a new css_set with one cgroup updated | |
681 | * @old_cset: the baseline css_set | |
682 | * @cgrp: the cgroup to be updated | |
683 | * | |
684 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
685 | * substituted into the appropriate hierarchy. | |
817929ec | 686 | */ |
5abb8855 TH |
687 | static struct css_set *find_css_set(struct css_set *old_cset, |
688 | struct cgroup *cgrp) | |
817929ec | 689 | { |
b326f9d0 | 690 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 691 | struct css_set *cset; |
69d0206c TH |
692 | struct list_head tmp_links; |
693 | struct cgrp_cset_link *link; | |
2d8f243a | 694 | struct cgroup_subsys *ss; |
0ac801fe | 695 | unsigned long key; |
2d8f243a | 696 | int ssid; |
472b1053 | 697 | |
b326f9d0 TH |
698 | lockdep_assert_held(&cgroup_mutex); |
699 | ||
817929ec PM |
700 | /* First see if we already have a cgroup group that matches |
701 | * the desired set */ | |
96d365e0 | 702 | down_read(&css_set_rwsem); |
5abb8855 TH |
703 | cset = find_existing_css_set(old_cset, cgrp, template); |
704 | if (cset) | |
705 | get_css_set(cset); | |
96d365e0 | 706 | up_read(&css_set_rwsem); |
817929ec | 707 | |
5abb8855 TH |
708 | if (cset) |
709 | return cset; | |
817929ec | 710 | |
f4f4be2b | 711 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 712 | if (!cset) |
817929ec PM |
713 | return NULL; |
714 | ||
69d0206c | 715 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 716 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 717 | kfree(cset); |
817929ec PM |
718 | return NULL; |
719 | } | |
720 | ||
5abb8855 | 721 | atomic_set(&cset->refcount, 1); |
69d0206c | 722 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 723 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 724 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 725 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 726 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 727 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
728 | |
729 | /* Copy the set of subsystem state objects generated in | |
730 | * find_existing_css_set() */ | |
5abb8855 | 731 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 732 | |
96d365e0 | 733 | down_write(&css_set_rwsem); |
817929ec | 734 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 735 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 736 | struct cgroup *c = link->cgrp; |
69d0206c | 737 | |
7717f7ba PM |
738 | if (c->root == cgrp->root) |
739 | c = cgrp; | |
69d0206c | 740 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 741 | } |
817929ec | 742 | |
69d0206c | 743 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 744 | |
817929ec | 745 | css_set_count++; |
472b1053 | 746 | |
2d8f243a | 747 | /* Add @cset to the hash table */ |
5abb8855 TH |
748 | key = css_set_hash(cset->subsys); |
749 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 750 | |
2d8f243a TH |
751 | for_each_subsys(ss, ssid) |
752 | list_add_tail(&cset->e_cset_node[ssid], | |
753 | &cset->subsys[ssid]->cgroup->e_csets[ssid]); | |
754 | ||
96d365e0 | 755 | up_write(&css_set_rwsem); |
817929ec | 756 | |
5abb8855 | 757 | return cset; |
b4f48b63 PM |
758 | } |
759 | ||
3dd06ffa | 760 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 761 | { |
3dd06ffa | 762 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 763 | |
3dd06ffa | 764 | return root_cgrp->root; |
2bd59d48 TH |
765 | } |
766 | ||
3dd06ffa | 767 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
768 | { |
769 | int id; | |
770 | ||
771 | lockdep_assert_held(&cgroup_mutex); | |
772 | ||
985ed670 | 773 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
774 | if (id < 0) |
775 | return id; | |
776 | ||
777 | root->hierarchy_id = id; | |
778 | return 0; | |
779 | } | |
780 | ||
3dd06ffa | 781 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
782 | { |
783 | lockdep_assert_held(&cgroup_mutex); | |
784 | ||
785 | if (root->hierarchy_id) { | |
786 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
787 | root->hierarchy_id = 0; | |
788 | } | |
789 | } | |
790 | ||
3dd06ffa | 791 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
792 | { |
793 | if (root) { | |
794 | /* hierarhcy ID shoulid already have been released */ | |
795 | WARN_ON_ONCE(root->hierarchy_id); | |
796 | ||
797 | idr_destroy(&root->cgroup_idr); | |
798 | kfree(root); | |
799 | } | |
800 | } | |
801 | ||
3dd06ffa | 802 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 803 | { |
3dd06ffa | 804 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 805 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 806 | |
2bd59d48 | 807 | mutex_lock(&cgroup_tree_mutex); |
2bd59d48 | 808 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 809 | |
776f02fa | 810 | BUG_ON(atomic_read(&root->nr_cgrps)); |
f2e85d57 TH |
811 | BUG_ON(!list_empty(&cgrp->children)); |
812 | ||
f2e85d57 | 813 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 814 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 815 | |
7717f7ba | 816 | /* |
f2e85d57 TH |
817 | * Release all the links from cset_links to this hierarchy's |
818 | * root cgroup | |
7717f7ba | 819 | */ |
96d365e0 | 820 | down_write(&css_set_rwsem); |
f2e85d57 TH |
821 | |
822 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
823 | list_del(&link->cset_link); | |
824 | list_del(&link->cgrp_link); | |
825 | kfree(link); | |
826 | } | |
96d365e0 | 827 | up_write(&css_set_rwsem); |
f2e85d57 TH |
828 | |
829 | if (!list_empty(&root->root_list)) { | |
830 | list_del(&root->root_list); | |
831 | cgroup_root_count--; | |
832 | } | |
833 | ||
834 | cgroup_exit_root_id(root); | |
835 | ||
836 | mutex_unlock(&cgroup_mutex); | |
837 | mutex_unlock(&cgroup_tree_mutex); | |
f2e85d57 | 838 | |
2bd59d48 | 839 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
840 | cgroup_free_root(root); |
841 | } | |
842 | ||
ceb6a081 TH |
843 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
844 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 845 | struct cgroup_root *root) |
7717f7ba | 846 | { |
7717f7ba PM |
847 | struct cgroup *res = NULL; |
848 | ||
96d365e0 TH |
849 | lockdep_assert_held(&cgroup_mutex); |
850 | lockdep_assert_held(&css_set_rwsem); | |
851 | ||
5abb8855 | 852 | if (cset == &init_css_set) { |
3dd06ffa | 853 | res = &root->cgrp; |
7717f7ba | 854 | } else { |
69d0206c TH |
855 | struct cgrp_cset_link *link; |
856 | ||
857 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 858 | struct cgroup *c = link->cgrp; |
69d0206c | 859 | |
7717f7ba PM |
860 | if (c->root == root) { |
861 | res = c; | |
862 | break; | |
863 | } | |
864 | } | |
865 | } | |
96d365e0 | 866 | |
7717f7ba PM |
867 | BUG_ON(!res); |
868 | return res; | |
869 | } | |
870 | ||
ddbcc7e8 | 871 | /* |
ceb6a081 TH |
872 | * Return the cgroup for "task" from the given hierarchy. Must be |
873 | * called with cgroup_mutex and css_set_rwsem held. | |
874 | */ | |
875 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 876 | struct cgroup_root *root) |
ceb6a081 TH |
877 | { |
878 | /* | |
879 | * No need to lock the task - since we hold cgroup_mutex the | |
880 | * task can't change groups, so the only thing that can happen | |
881 | * is that it exits and its css is set back to init_css_set. | |
882 | */ | |
883 | return cset_cgroup_from_root(task_css_set(task), root); | |
884 | } | |
885 | ||
ddbcc7e8 | 886 | /* |
ddbcc7e8 PM |
887 | * A task must hold cgroup_mutex to modify cgroups. |
888 | * | |
889 | * Any task can increment and decrement the count field without lock. | |
890 | * So in general, code holding cgroup_mutex can't rely on the count | |
891 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 892 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
893 | * means that no tasks are currently attached, therefore there is no |
894 | * way a task attached to that cgroup can fork (the other way to | |
895 | * increment the count). So code holding cgroup_mutex can safely | |
896 | * assume that if the count is zero, it will stay zero. Similarly, if | |
897 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
898 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
899 | * needs that mutex. | |
900 | * | |
ddbcc7e8 PM |
901 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
902 | * (usually) take cgroup_mutex. These are the two most performance | |
903 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
904 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
905 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
906 | * to the release agent with the name of the cgroup (path relative to |
907 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
908 | * |
909 | * A cgroup can only be deleted if both its 'count' of using tasks | |
910 | * is zero, and its list of 'children' cgroups is empty. Since all | |
911 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 912 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 913 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 914 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
915 | * |
916 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 917 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
918 | */ |
919 | ||
628f7cd4 | 920 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
2bd59d48 | 921 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 922 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 923 | |
8d7e6fb0 TH |
924 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
925 | char *buf) | |
ddbcc7e8 | 926 | { |
8d7e6fb0 TH |
927 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
928 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
929 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
930 | cft->ss->name, cft->name); | |
931 | else | |
932 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
933 | return buf; | |
ddbcc7e8 PM |
934 | } |
935 | ||
f2e85d57 TH |
936 | /** |
937 | * cgroup_file_mode - deduce file mode of a control file | |
938 | * @cft: the control file in question | |
939 | * | |
940 | * returns cft->mode if ->mode is not 0 | |
941 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
942 | * returns S_IRUGO if it has only a read handler | |
943 | * returns S_IWUSR if it has only a write hander | |
944 | */ | |
945 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 946 | { |
f2e85d57 | 947 | umode_t mode = 0; |
65dff759 | 948 | |
f2e85d57 TH |
949 | if (cft->mode) |
950 | return cft->mode; | |
951 | ||
952 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
953 | mode |= S_IRUGO; | |
954 | ||
955 | if (cft->write_u64 || cft->write_s64 || cft->write_string || | |
956 | cft->trigger) | |
957 | mode |= S_IWUSR; | |
958 | ||
959 | return mode; | |
65dff759 LZ |
960 | } |
961 | ||
be445626 LZ |
962 | static void cgroup_free_fn(struct work_struct *work) |
963 | { | |
ea15f8cc | 964 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 | 965 | |
3c9c825b | 966 | atomic_dec(&cgrp->root->nr_cgrps); |
b1a21367 | 967 | cgroup_pidlist_destroy_all(cgrp); |
be445626 | 968 | |
776f02fa TH |
969 | if (cgrp->parent) { |
970 | /* | |
971 | * We get a ref to the parent, and put the ref when this | |
972 | * cgroup is being freed, so it's guaranteed that the | |
973 | * parent won't be destroyed before its children. | |
974 | */ | |
975 | cgroup_put(cgrp->parent); | |
976 | kernfs_put(cgrp->kn); | |
977 | kfree(cgrp); | |
978 | } else { | |
979 | /* | |
3dd06ffa | 980 | * This is root cgroup's refcnt reaching zero, which |
776f02fa TH |
981 | * indicates that the root should be released. |
982 | */ | |
983 | cgroup_destroy_root(cgrp->root); | |
984 | } | |
be445626 LZ |
985 | } |
986 | ||
987 | static void cgroup_free_rcu(struct rcu_head *head) | |
988 | { | |
989 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
990 | ||
ea15f8cc | 991 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 992 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
993 | } |
994 | ||
59f5296b | 995 | static void cgroup_get(struct cgroup *cgrp) |
ddbcc7e8 | 996 | { |
2bd59d48 TH |
997 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
998 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
999 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
1000 | } |
1001 | ||
59f5296b | 1002 | static void cgroup_put(struct cgroup *cgrp) |
05ef1d7c | 1003 | { |
2bd59d48 TH |
1004 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
1005 | return; | |
776f02fa | 1006 | if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp))) |
2bd59d48 | 1007 | return; |
05ef1d7c | 1008 | |
2739d3cc | 1009 | /* |
2bd59d48 TH |
1010 | * XXX: cgrp->id is only used to look up css's. As cgroup and |
1011 | * css's lifetimes will be decoupled, it should be made | |
1012 | * per-subsystem and moved to css->id so that lookups are | |
1013 | * successful until the target css is released. | |
2739d3cc | 1014 | */ |
2bd59d48 TH |
1015 | mutex_lock(&cgroup_mutex); |
1016 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
1017 | mutex_unlock(&cgroup_mutex); | |
1018 | cgrp->id = -1; | |
05ef1d7c | 1019 | |
2bd59d48 | 1020 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 | 1021 | } |
05ef1d7c | 1022 | |
2739d3cc | 1023 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1024 | { |
2bd59d48 | 1025 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1026 | |
ace2bee8 | 1027 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 1028 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1029 | } |
1030 | ||
13af07df | 1031 | /** |
628f7cd4 | 1032 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 1033 | * @cgrp: target cgroup |
13af07df AR |
1034 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
1035 | */ | |
628f7cd4 | 1036 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 1037 | { |
13af07df | 1038 | struct cgroup_subsys *ss; |
b420ba7d | 1039 | int i; |
05ef1d7c | 1040 | |
b420ba7d | 1041 | for_each_subsys(ss, i) { |
0adb0704 | 1042 | struct cftype *cfts; |
b420ba7d TH |
1043 | |
1044 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 1045 | continue; |
0adb0704 TH |
1046 | list_for_each_entry(cfts, &ss->cfts, node) |
1047 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 1048 | } |
ddbcc7e8 PM |
1049 | } |
1050 | ||
3dd06ffa | 1051 | static int rebind_subsystems(struct cgroup_root *dst_root, |
5df36032 | 1052 | unsigned long ss_mask) |
ddbcc7e8 | 1053 | { |
30159ec7 | 1054 | struct cgroup_subsys *ss; |
2d8f243a | 1055 | int ssid, i, ret; |
ddbcc7e8 | 1056 | |
ace2bee8 TH |
1057 | lockdep_assert_held(&cgroup_tree_mutex); |
1058 | lockdep_assert_held(&cgroup_mutex); | |
ddbcc7e8 | 1059 | |
5df36032 TH |
1060 | for_each_subsys(ss, ssid) { |
1061 | if (!(ss_mask & (1 << ssid))) | |
1062 | continue; | |
aae8aab4 | 1063 | |
7fd8c565 TH |
1064 | /* if @ss has non-root csses attached to it, can't move */ |
1065 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1066 | return -EBUSY; |
1d5be6b2 | 1067 | |
5df36032 | 1068 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1069 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1070 | return -EBUSY; |
ddbcc7e8 PM |
1071 | } |
1072 | ||
a2dd4247 TH |
1073 | ret = cgroup_populate_dir(&dst_root->cgrp, ss_mask); |
1074 | if (ret) { | |
1075 | if (dst_root != &cgrp_dfl_root) | |
5df36032 | 1076 | return ret; |
ddbcc7e8 | 1077 | |
a2dd4247 TH |
1078 | /* |
1079 | * Rebinding back to the default root is not allowed to | |
1080 | * fail. Using both default and non-default roots should | |
1081 | * be rare. Moving subsystems back and forth even more so. | |
1082 | * Just warn about it and continue. | |
1083 | */ | |
1084 | if (cgrp_dfl_root_visible) { | |
1085 | pr_warning("cgroup: failed to create files (%d) while rebinding 0x%lx to default root\n", | |
1086 | ret, ss_mask); | |
1087 | pr_warning("cgroup: you may retry by moving them to a different hierarchy and unbinding\n"); | |
1088 | } | |
5df36032 | 1089 | } |
3126121f TH |
1090 | |
1091 | /* | |
1092 | * Nothing can fail from this point on. Remove files for the | |
1093 | * removed subsystems and rebind each subsystem. | |
1094 | */ | |
4ac06017 | 1095 | mutex_unlock(&cgroup_mutex); |
5df36032 | 1096 | for_each_subsys(ss, ssid) |
a2dd4247 | 1097 | if (ss_mask & (1 << ssid)) |
3dd06ffa | 1098 | cgroup_clear_dir(&ss->root->cgrp, 1 << ssid); |
4ac06017 | 1099 | mutex_lock(&cgroup_mutex); |
a8a648c4 | 1100 | |
5df36032 | 1101 | for_each_subsys(ss, ssid) { |
3dd06ffa | 1102 | struct cgroup_root *src_root; |
5df36032 | 1103 | struct cgroup_subsys_state *css; |
2d8f243a | 1104 | struct css_set *cset; |
a8a648c4 | 1105 | |
5df36032 TH |
1106 | if (!(ss_mask & (1 << ssid))) |
1107 | continue; | |
a8a648c4 | 1108 | |
5df36032 | 1109 | src_root = ss->root; |
3dd06ffa | 1110 | css = cgroup_css(&src_root->cgrp, ss); |
a8a648c4 | 1111 | |
3dd06ffa | 1112 | WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss)); |
73e80ed8 | 1113 | |
3dd06ffa TH |
1114 | RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL); |
1115 | rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css); | |
5df36032 | 1116 | ss->root = dst_root; |
3dd06ffa | 1117 | css->cgroup = &dst_root->cgrp; |
73e80ed8 | 1118 | |
2d8f243a TH |
1119 | down_write(&css_set_rwsem); |
1120 | hash_for_each(css_set_table, i, cset, hlist) | |
1121 | list_move_tail(&cset->e_cset_node[ss->id], | |
1122 | &dst_root->cgrp.e_csets[ss->id]); | |
1123 | up_write(&css_set_rwsem); | |
1124 | ||
f392e51c TH |
1125 | src_root->subsys_mask &= ~(1 << ssid); |
1126 | src_root->cgrp.child_subsys_mask &= ~(1 << ssid); | |
1127 | ||
bd53d617 | 1128 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1129 | dst_root->subsys_mask |= 1 << ssid; |
bd53d617 TH |
1130 | if (dst_root != &cgrp_dfl_root) |
1131 | dst_root->cgrp.child_subsys_mask |= 1 << ssid; | |
a8a648c4 | 1132 | |
5df36032 TH |
1133 | if (ss->bind) |
1134 | ss->bind(css); | |
ddbcc7e8 | 1135 | } |
ddbcc7e8 | 1136 | |
a2dd4247 | 1137 | kernfs_activate(dst_root->cgrp.kn); |
ddbcc7e8 PM |
1138 | return 0; |
1139 | } | |
1140 | ||
2bd59d48 TH |
1141 | static int cgroup_show_options(struct seq_file *seq, |
1142 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1143 | { |
3dd06ffa | 1144 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1145 | struct cgroup_subsys *ss; |
b85d2040 | 1146 | int ssid; |
ddbcc7e8 | 1147 | |
b85d2040 | 1148 | for_each_subsys(ss, ssid) |
f392e51c | 1149 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 1150 | seq_printf(seq, ",%s", ss->name); |
873fe09e TH |
1151 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1152 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1153 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1154 | seq_puts(seq, ",noprefix"); |
93438629 | 1155 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1156 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1157 | |
1158 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1159 | if (strlen(root->release_agent_path)) |
1160 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1161 | spin_unlock(&release_agent_path_lock); |
1162 | ||
3dd06ffa | 1163 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1164 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1165 | if (strlen(root->name)) |
1166 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1167 | return 0; |
1168 | } | |
1169 | ||
1170 | struct cgroup_sb_opts { | |
a1a71b45 | 1171 | unsigned long subsys_mask; |
ddbcc7e8 | 1172 | unsigned long flags; |
81a6a5cd | 1173 | char *release_agent; |
2260e7fc | 1174 | bool cpuset_clone_children; |
c6d57f33 | 1175 | char *name; |
2c6ab6d2 PM |
1176 | /* User explicitly requested empty subsystem */ |
1177 | bool none; | |
ddbcc7e8 PM |
1178 | }; |
1179 | ||
aae8aab4 | 1180 | /* |
9871bf95 TH |
1181 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1182 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1183 | * array. This function takes refcounts on subsystems to be used, unless it | |
1184 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1185 | */ |
cf5d5941 | 1186 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1187 | { |
32a8cf23 DL |
1188 | char *token, *o = data; |
1189 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1190 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1191 | struct cgroup_subsys *ss; |
1192 | int i; | |
f9ab5b5b | 1193 | |
aae8aab4 BB |
1194 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1195 | ||
f9ab5b5b | 1196 | #ifdef CONFIG_CPUSETS |
073219e9 | 1197 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1198 | #endif |
ddbcc7e8 | 1199 | |
c6d57f33 | 1200 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1201 | |
1202 | while ((token = strsep(&o, ",")) != NULL) { | |
1203 | if (!*token) | |
1204 | return -EINVAL; | |
32a8cf23 | 1205 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1206 | /* Explicitly have no subsystems */ |
1207 | opts->none = true; | |
32a8cf23 DL |
1208 | continue; |
1209 | } | |
1210 | if (!strcmp(token, "all")) { | |
1211 | /* Mutually exclusive option 'all' + subsystem name */ | |
1212 | if (one_ss) | |
1213 | return -EINVAL; | |
1214 | all_ss = true; | |
1215 | continue; | |
1216 | } | |
873fe09e TH |
1217 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1218 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1219 | continue; | |
1220 | } | |
32a8cf23 | 1221 | if (!strcmp(token, "noprefix")) { |
93438629 | 1222 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1223 | continue; |
1224 | } | |
1225 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1226 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1227 | continue; |
1228 | } | |
03b1cde6 | 1229 | if (!strcmp(token, "xattr")) { |
93438629 | 1230 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1231 | continue; |
1232 | } | |
32a8cf23 | 1233 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1234 | /* Specifying two release agents is forbidden */ |
1235 | if (opts->release_agent) | |
1236 | return -EINVAL; | |
c6d57f33 | 1237 | opts->release_agent = |
e400c285 | 1238 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1239 | if (!opts->release_agent) |
1240 | return -ENOMEM; | |
32a8cf23 DL |
1241 | continue; |
1242 | } | |
1243 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1244 | const char *name = token + 5; |
1245 | /* Can't specify an empty name */ | |
1246 | if (!strlen(name)) | |
1247 | return -EINVAL; | |
1248 | /* Must match [\w.-]+ */ | |
1249 | for (i = 0; i < strlen(name); i++) { | |
1250 | char c = name[i]; | |
1251 | if (isalnum(c)) | |
1252 | continue; | |
1253 | if ((c == '.') || (c == '-') || (c == '_')) | |
1254 | continue; | |
1255 | return -EINVAL; | |
1256 | } | |
1257 | /* Specifying two names is forbidden */ | |
1258 | if (opts->name) | |
1259 | return -EINVAL; | |
1260 | opts->name = kstrndup(name, | |
e400c285 | 1261 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1262 | GFP_KERNEL); |
1263 | if (!opts->name) | |
1264 | return -ENOMEM; | |
32a8cf23 DL |
1265 | |
1266 | continue; | |
1267 | } | |
1268 | ||
30159ec7 | 1269 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1270 | if (strcmp(token, ss->name)) |
1271 | continue; | |
1272 | if (ss->disabled) | |
1273 | continue; | |
1274 | ||
1275 | /* Mutually exclusive option 'all' + subsystem name */ | |
1276 | if (all_ss) | |
1277 | return -EINVAL; | |
a1a71b45 | 1278 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1279 | one_ss = true; |
1280 | ||
1281 | break; | |
1282 | } | |
1283 | if (i == CGROUP_SUBSYS_COUNT) | |
1284 | return -ENOENT; | |
1285 | } | |
1286 | ||
2c6ab6d2 PM |
1287 | /* Consistency checks */ |
1288 | ||
873fe09e TH |
1289 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1290 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1291 | ||
d3ba07c3 TH |
1292 | if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) || |
1293 | opts->cpuset_clone_children || opts->release_agent || | |
1294 | opts->name) { | |
1295 | pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n"); | |
873fe09e TH |
1296 | return -EINVAL; |
1297 | } | |
a2dd4247 TH |
1298 | } else { |
1299 | /* | |
1300 | * If the 'all' option was specified select all the | |
1301 | * subsystems, otherwise if 'none', 'name=' and a subsystem | |
1302 | * name options were not specified, let's default to 'all' | |
1303 | */ | |
1304 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1305 | for_each_subsys(ss, i) | |
1306 | if (!ss->disabled) | |
1307 | set_bit(i, &opts->subsys_mask); | |
873fe09e | 1308 | |
a2dd4247 TH |
1309 | /* |
1310 | * We either have to specify by name or by subsystems. (So | |
1311 | * all empty hierarchies must have a name). | |
1312 | */ | |
1313 | if (!opts->subsys_mask && !opts->name) | |
873fe09e | 1314 | return -EINVAL; |
873fe09e TH |
1315 | } |
1316 | ||
f9ab5b5b LZ |
1317 | /* |
1318 | * Option noprefix was introduced just for backward compatibility | |
1319 | * with the old cpuset, so we allow noprefix only if mounting just | |
1320 | * the cpuset subsystem. | |
1321 | */ | |
93438629 | 1322 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1323 | return -EINVAL; |
1324 | ||
2c6ab6d2 PM |
1325 | |
1326 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1327 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1328 | return -EINVAL; |
1329 | ||
ddbcc7e8 PM |
1330 | return 0; |
1331 | } | |
1332 | ||
2bd59d48 | 1333 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1334 | { |
1335 | int ret = 0; | |
3dd06ffa | 1336 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1337 | struct cgroup_sb_opts opts; |
a1a71b45 | 1338 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1339 | |
873fe09e TH |
1340 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1341 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1342 | return -EINVAL; | |
1343 | } | |
1344 | ||
ace2bee8 | 1345 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1346 | mutex_lock(&cgroup_mutex); |
1347 | ||
1348 | /* See what subsystems are wanted */ | |
1349 | ret = parse_cgroupfs_options(data, &opts); | |
1350 | if (ret) | |
1351 | goto out_unlock; | |
1352 | ||
f392e51c | 1353 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1354 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1355 | task_tgid_nr(current), current->comm); | |
1356 | ||
f392e51c TH |
1357 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1358 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1359 | |
cf5d5941 | 1360 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1361 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1362 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1363 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1364 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1365 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1366 | ret = -EINVAL; |
1367 | goto out_unlock; | |
1368 | } | |
1369 | ||
f172e67c | 1370 | /* remounting is not allowed for populated hierarchies */ |
3dd06ffa | 1371 | if (!list_empty(&root->cgrp.children)) { |
f172e67c | 1372 | ret = -EBUSY; |
0670e08b | 1373 | goto out_unlock; |
cf5d5941 | 1374 | } |
ddbcc7e8 | 1375 | |
5df36032 | 1376 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1377 | if (ret) |
0670e08b | 1378 | goto out_unlock; |
ddbcc7e8 | 1379 | |
3dd06ffa | 1380 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1381 | |
69e943b7 TH |
1382 | if (opts.release_agent) { |
1383 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1384 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1385 | spin_unlock(&release_agent_path_lock); |
1386 | } | |
ddbcc7e8 | 1387 | out_unlock: |
66bdc9cf | 1388 | kfree(opts.release_agent); |
c6d57f33 | 1389 | kfree(opts.name); |
ddbcc7e8 | 1390 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1391 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1392 | return ret; |
1393 | } | |
1394 | ||
afeb0f9f TH |
1395 | /* |
1396 | * To reduce the fork() overhead for systems that are not actually using | |
1397 | * their cgroups capability, we don't maintain the lists running through | |
1398 | * each css_set to its tasks until we see the list actually used - in other | |
1399 | * words after the first mount. | |
1400 | */ | |
1401 | static bool use_task_css_set_links __read_mostly; | |
1402 | ||
1403 | static void cgroup_enable_task_cg_lists(void) | |
1404 | { | |
1405 | struct task_struct *p, *g; | |
1406 | ||
96d365e0 | 1407 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1408 | |
1409 | if (use_task_css_set_links) | |
1410 | goto out_unlock; | |
1411 | ||
1412 | use_task_css_set_links = true; | |
1413 | ||
1414 | /* | |
1415 | * We need tasklist_lock because RCU is not safe against | |
1416 | * while_each_thread(). Besides, a forking task that has passed | |
1417 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1418 | * is not guaranteed to have its child immediately visible in the | |
1419 | * tasklist if we walk through it with RCU. | |
1420 | */ | |
1421 | read_lock(&tasklist_lock); | |
1422 | do_each_thread(g, p) { | |
afeb0f9f TH |
1423 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1424 | task_css_set(p) != &init_css_set); | |
1425 | ||
1426 | /* | |
1427 | * We should check if the process is exiting, otherwise | |
1428 | * it will race with cgroup_exit() in that the list | |
1429 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1430 | * Do it while holding siglock so that we don't end up |
1431 | * racing against cgroup_exit(). | |
afeb0f9f | 1432 | */ |
f153ad11 | 1433 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1434 | if (!(p->flags & PF_EXITING)) { |
1435 | struct css_set *cset = task_css_set(p); | |
1436 | ||
1437 | list_add(&p->cg_list, &cset->tasks); | |
1438 | get_css_set(cset); | |
1439 | } | |
f153ad11 | 1440 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1441 | } while_each_thread(g, p); |
1442 | read_unlock(&tasklist_lock); | |
1443 | out_unlock: | |
96d365e0 | 1444 | up_write(&css_set_rwsem); |
afeb0f9f | 1445 | } |
ddbcc7e8 | 1446 | |
cc31edce PM |
1447 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1448 | { | |
2d8f243a TH |
1449 | struct cgroup_subsys *ss; |
1450 | int ssid; | |
1451 | ||
2bd59d48 | 1452 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1453 | INIT_LIST_HEAD(&cgrp->sibling); |
1454 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1455 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1456 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1457 | INIT_LIST_HEAD(&cgrp->pidlists); |
1458 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1459 | cgrp->dummy_css.cgroup = cgrp; |
2d8f243a TH |
1460 | |
1461 | for_each_subsys(ss, ssid) | |
1462 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1463 | |
1464 | init_waitqueue_head(&cgrp->offline_waitq); | |
cc31edce | 1465 | } |
c6d57f33 | 1466 | |
3dd06ffa | 1467 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1468 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1469 | { |
3dd06ffa | 1470 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1471 | |
ddbcc7e8 | 1472 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1473 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1474 | cgrp->root = root; |
cc31edce | 1475 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1476 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1477 | |
c6d57f33 PM |
1478 | root->flags = opts->flags; |
1479 | if (opts->release_agent) | |
1480 | strcpy(root->release_agent_path, opts->release_agent); | |
1481 | if (opts->name) | |
1482 | strcpy(root->name, opts->name); | |
2260e7fc | 1483 | if (opts->cpuset_clone_children) |
3dd06ffa | 1484 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1485 | } |
1486 | ||
3dd06ffa | 1487 | static int cgroup_setup_root(struct cgroup_root *root, unsigned long ss_mask) |
2c6ab6d2 | 1488 | { |
d427dfeb | 1489 | LIST_HEAD(tmp_links); |
3dd06ffa | 1490 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1491 | struct css_set *cset; |
d427dfeb | 1492 | int i, ret; |
2c6ab6d2 | 1493 | |
d427dfeb TH |
1494 | lockdep_assert_held(&cgroup_tree_mutex); |
1495 | lockdep_assert_held(&cgroup_mutex); | |
c6d57f33 | 1496 | |
d427dfeb TH |
1497 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); |
1498 | if (ret < 0) | |
2bd59d48 | 1499 | goto out; |
d427dfeb | 1500 | root_cgrp->id = ret; |
c6d57f33 | 1501 | |
d427dfeb | 1502 | /* |
96d365e0 | 1503 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1504 | * but that's OK - it can only be increased by someone holding |
1505 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1506 | * have some link structures left over | |
1507 | */ | |
1508 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1509 | if (ret) | |
2bd59d48 | 1510 | goto out; |
ddbcc7e8 | 1511 | |
985ed670 | 1512 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1513 | if (ret) |
2bd59d48 | 1514 | goto out; |
ddbcc7e8 | 1515 | |
2bd59d48 TH |
1516 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1517 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1518 | root_cgrp); | |
1519 | if (IS_ERR(root->kf_root)) { | |
1520 | ret = PTR_ERR(root->kf_root); | |
1521 | goto exit_root_id; | |
1522 | } | |
1523 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1524 | |
d427dfeb TH |
1525 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); |
1526 | if (ret) | |
2bd59d48 | 1527 | goto destroy_root; |
ddbcc7e8 | 1528 | |
5df36032 | 1529 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1530 | if (ret) |
2bd59d48 | 1531 | goto destroy_root; |
ddbcc7e8 | 1532 | |
d427dfeb TH |
1533 | /* |
1534 | * There must be no failure case after here, since rebinding takes | |
1535 | * care of subsystems' refcounts, which are explicitly dropped in | |
1536 | * the failure exit path. | |
1537 | */ | |
1538 | list_add(&root->root_list, &cgroup_roots); | |
1539 | cgroup_root_count++; | |
0df6a63f | 1540 | |
d427dfeb | 1541 | /* |
3dd06ffa | 1542 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1543 | * objects. |
1544 | */ | |
96d365e0 | 1545 | down_write(&css_set_rwsem); |
d427dfeb TH |
1546 | hash_for_each(css_set_table, i, cset, hlist) |
1547 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1548 | up_write(&css_set_rwsem); |
ddbcc7e8 | 1549 | |
d427dfeb | 1550 | BUG_ON(!list_empty(&root_cgrp->children)); |
3c9c825b | 1551 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1552 | |
2bd59d48 | 1553 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1554 | ret = 0; |
2bd59d48 | 1555 | goto out; |
d427dfeb | 1556 | |
2bd59d48 TH |
1557 | destroy_root: |
1558 | kernfs_destroy_root(root->kf_root); | |
1559 | root->kf_root = NULL; | |
1560 | exit_root_id: | |
d427dfeb | 1561 | cgroup_exit_root_id(root); |
2bd59d48 | 1562 | out: |
d427dfeb TH |
1563 | free_cgrp_cset_links(&tmp_links); |
1564 | return ret; | |
ddbcc7e8 PM |
1565 | } |
1566 | ||
f7e83571 | 1567 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1568 | int flags, const char *unused_dev_name, |
f7e83571 | 1569 | void *data) |
ddbcc7e8 | 1570 | { |
3dd06ffa | 1571 | struct cgroup_root *root; |
ddbcc7e8 | 1572 | struct cgroup_sb_opts opts; |
2bd59d48 | 1573 | struct dentry *dentry; |
8e30e2b8 | 1574 | int ret; |
c6b3d5bc | 1575 | bool new_sb; |
ddbcc7e8 | 1576 | |
56fde9e0 TH |
1577 | /* |
1578 | * The first time anyone tries to mount a cgroup, enable the list | |
1579 | * linking each css_set to its tasks and fix up all existing tasks. | |
1580 | */ | |
1581 | if (!use_task_css_set_links) | |
1582 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 1583 | |
8e30e2b8 | 1584 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1585 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1586 | |
1587 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1588 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1589 | if (ret) |
8e30e2b8 | 1590 | goto out_unlock; |
e37a06f1 | 1591 | retry: |
2bd59d48 | 1592 | /* look for a matching existing root */ |
a2dd4247 TH |
1593 | if (!opts.subsys_mask && !opts.none && !opts.name) { |
1594 | cgrp_dfl_root_visible = true; | |
1595 | root = &cgrp_dfl_root; | |
1596 | cgroup_get(&root->cgrp); | |
1597 | ret = 0; | |
1598 | goto out_unlock; | |
ddbcc7e8 PM |
1599 | } |
1600 | ||
985ed670 | 1601 | for_each_root(root) { |
2bd59d48 | 1602 | bool name_match = false; |
3126121f | 1603 | |
3dd06ffa | 1604 | if (root == &cgrp_dfl_root) |
985ed670 | 1605 | continue; |
3126121f | 1606 | |
cf5d5941 | 1607 | /* |
2bd59d48 TH |
1608 | * If we asked for a name then it must match. Also, if |
1609 | * name matches but sybsys_mask doesn't, we should fail. | |
1610 | * Remember whether name matched. | |
cf5d5941 | 1611 | */ |
2bd59d48 TH |
1612 | if (opts.name) { |
1613 | if (strcmp(opts.name, root->name)) | |
1614 | continue; | |
1615 | name_match = true; | |
1616 | } | |
ddbcc7e8 | 1617 | |
c6d57f33 | 1618 | /* |
2bd59d48 TH |
1619 | * If we asked for subsystems (or explicitly for no |
1620 | * subsystems) then they must match. | |
c6d57f33 | 1621 | */ |
2bd59d48 | 1622 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 1623 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
1624 | if (!name_match) |
1625 | continue; | |
1626 | ret = -EBUSY; | |
1627 | goto out_unlock; | |
1628 | } | |
873fe09e | 1629 | |
c7ba8287 | 1630 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1631 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1632 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1633 | ret = -EINVAL; | |
8e30e2b8 | 1634 | goto out_unlock; |
2a0ff3fb JL |
1635 | } else { |
1636 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1637 | } | |
873fe09e | 1638 | } |
ddbcc7e8 | 1639 | |
776f02fa | 1640 | /* |
3dd06ffa | 1641 | * A root's lifetime is governed by its root cgroup. Zero |
776f02fa TH |
1642 | * ref indicate that the root is being destroyed. Wait for |
1643 | * destruction to complete so that the subsystems are free. | |
1644 | * We can use wait_queue for the wait but this path is | |
1645 | * super cold. Let's just sleep for a bit and retry. | |
1646 | */ | |
3dd06ffa | 1647 | if (!atomic_inc_not_zero(&root->cgrp.refcnt)) { |
776f02fa TH |
1648 | mutex_unlock(&cgroup_mutex); |
1649 | mutex_unlock(&cgroup_tree_mutex); | |
1650 | msleep(10); | |
e37a06f1 LZ |
1651 | mutex_lock(&cgroup_tree_mutex); |
1652 | mutex_lock(&cgroup_mutex); | |
776f02fa TH |
1653 | goto retry; |
1654 | } | |
ddbcc7e8 | 1655 | |
776f02fa | 1656 | ret = 0; |
2bd59d48 | 1657 | goto out_unlock; |
ddbcc7e8 | 1658 | } |
ddbcc7e8 | 1659 | |
817929ec | 1660 | /* |
172a2c06 TH |
1661 | * No such thing, create a new one. name= matching without subsys |
1662 | * specification is allowed for already existing hierarchies but we | |
1663 | * can't create new one without subsys specification. | |
817929ec | 1664 | */ |
172a2c06 TH |
1665 | if (!opts.subsys_mask && !opts.none) { |
1666 | ret = -EINVAL; | |
1667 | goto out_unlock; | |
817929ec | 1668 | } |
817929ec | 1669 | |
172a2c06 TH |
1670 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
1671 | if (!root) { | |
1672 | ret = -ENOMEM; | |
2bd59d48 | 1673 | goto out_unlock; |
839ec545 | 1674 | } |
e5f6a860 | 1675 | |
172a2c06 TH |
1676 | init_cgroup_root(root, &opts); |
1677 | ||
35585573 | 1678 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1679 | if (ret) |
1680 | cgroup_free_root(root); | |
fa3ca07e | 1681 | |
8e30e2b8 | 1682 | out_unlock: |
ddbcc7e8 | 1683 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1684 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 | 1685 | |
c6d57f33 PM |
1686 | kfree(opts.release_agent); |
1687 | kfree(opts.name); | |
03b1cde6 | 1688 | |
2bd59d48 | 1689 | if (ret) |
8e30e2b8 | 1690 | return ERR_PTR(ret); |
2bd59d48 | 1691 | |
c6b3d5bc LZ |
1692 | dentry = kernfs_mount(fs_type, flags, root->kf_root, &new_sb); |
1693 | if (IS_ERR(dentry) || !new_sb) | |
3dd06ffa | 1694 | cgroup_put(&root->cgrp); |
2bd59d48 TH |
1695 | return dentry; |
1696 | } | |
1697 | ||
1698 | static void cgroup_kill_sb(struct super_block *sb) | |
1699 | { | |
1700 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 1701 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 1702 | |
3dd06ffa | 1703 | cgroup_put(&root->cgrp); |
2bd59d48 | 1704 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1705 | } |
1706 | ||
1707 | static struct file_system_type cgroup_fs_type = { | |
1708 | .name = "cgroup", | |
f7e83571 | 1709 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1710 | .kill_sb = cgroup_kill_sb, |
1711 | }; | |
1712 | ||
676db4af GK |
1713 | static struct kobject *cgroup_kobj; |
1714 | ||
857a2beb | 1715 | /** |
913ffdb5 | 1716 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1717 | * @task: target task |
857a2beb TH |
1718 | * @buf: the buffer to write the path into |
1719 | * @buflen: the length of the buffer | |
1720 | * | |
913ffdb5 TH |
1721 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1722 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1723 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1724 | * cgroup controller callbacks. | |
1725 | * | |
e61734c5 | 1726 | * Return value is the same as kernfs_path(). |
857a2beb | 1727 | */ |
e61734c5 | 1728 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 1729 | { |
3dd06ffa | 1730 | struct cgroup_root *root; |
913ffdb5 | 1731 | struct cgroup *cgrp; |
e61734c5 TH |
1732 | int hierarchy_id = 1; |
1733 | char *path = NULL; | |
857a2beb TH |
1734 | |
1735 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1736 | down_read(&css_set_rwsem); |
857a2beb | 1737 | |
913ffdb5 TH |
1738 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1739 | ||
857a2beb TH |
1740 | if (root) { |
1741 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1742 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1743 | } else { |
1744 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1745 | if (strlcpy(buf, "/", buflen) < buflen) |
1746 | path = buf; | |
857a2beb TH |
1747 | } |
1748 | ||
96d365e0 | 1749 | up_read(&css_set_rwsem); |
857a2beb | 1750 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1751 | return path; |
857a2beb | 1752 | } |
913ffdb5 | 1753 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1754 | |
b3dc094e | 1755 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 1756 | struct cgroup_taskset { |
b3dc094e TH |
1757 | /* the src and dst cset list running through cset->mg_node */ |
1758 | struct list_head src_csets; | |
1759 | struct list_head dst_csets; | |
1760 | ||
1761 | /* | |
1762 | * Fields for cgroup_taskset_*() iteration. | |
1763 | * | |
1764 | * Before migration is committed, the target migration tasks are on | |
1765 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
1766 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
1767 | * or ->dst_csets depending on whether migration is committed. | |
1768 | * | |
1769 | * ->cur_csets and ->cur_task point to the current task position | |
1770 | * during iteration. | |
1771 | */ | |
1772 | struct list_head *csets; | |
1773 | struct css_set *cur_cset; | |
1774 | struct task_struct *cur_task; | |
2f7ee569 TH |
1775 | }; |
1776 | ||
1777 | /** | |
1778 | * cgroup_taskset_first - reset taskset and return the first task | |
1779 | * @tset: taskset of interest | |
1780 | * | |
1781 | * @tset iteration is initialized and the first task is returned. | |
1782 | */ | |
1783 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1784 | { | |
b3dc094e TH |
1785 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
1786 | tset->cur_task = NULL; | |
1787 | ||
1788 | return cgroup_taskset_next(tset); | |
2f7ee569 | 1789 | } |
2f7ee569 TH |
1790 | |
1791 | /** | |
1792 | * cgroup_taskset_next - iterate to the next task in taskset | |
1793 | * @tset: taskset of interest | |
1794 | * | |
1795 | * Return the next task in @tset. Iteration must have been initialized | |
1796 | * with cgroup_taskset_first(). | |
1797 | */ | |
1798 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1799 | { | |
b3dc094e TH |
1800 | struct css_set *cset = tset->cur_cset; |
1801 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 1802 | |
b3dc094e TH |
1803 | while (&cset->mg_node != tset->csets) { |
1804 | if (!task) | |
1805 | task = list_first_entry(&cset->mg_tasks, | |
1806 | struct task_struct, cg_list); | |
1807 | else | |
1808 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 1809 | |
b3dc094e TH |
1810 | if (&task->cg_list != &cset->mg_tasks) { |
1811 | tset->cur_cset = cset; | |
1812 | tset->cur_task = task; | |
1813 | return task; | |
1814 | } | |
2f7ee569 | 1815 | |
b3dc094e TH |
1816 | cset = list_next_entry(cset, mg_node); |
1817 | task = NULL; | |
1818 | } | |
2f7ee569 | 1819 | |
b3dc094e | 1820 | return NULL; |
2f7ee569 | 1821 | } |
2f7ee569 | 1822 | |
cb0f1fe9 | 1823 | /** |
74a1166d | 1824 | * cgroup_task_migrate - move a task from one cgroup to another. |
cb0f1fe9 TH |
1825 | * @old_cgrp; the cgroup @tsk is being migrated from |
1826 | * @tsk: the task being migrated | |
1827 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 1828 | * |
cb0f1fe9 | 1829 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 1830 | */ |
5abb8855 TH |
1831 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1832 | struct task_struct *tsk, | |
1833 | struct css_set *new_cset) | |
74a1166d | 1834 | { |
5abb8855 | 1835 | struct css_set *old_cset; |
74a1166d | 1836 | |
cb0f1fe9 TH |
1837 | lockdep_assert_held(&cgroup_mutex); |
1838 | lockdep_assert_held(&css_set_rwsem); | |
1839 | ||
74a1166d | 1840 | /* |
026085ef MSB |
1841 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1842 | * setting such that we can't race against cgroup_exit() changing the | |
1843 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1844 | */ |
c84cdf75 | 1845 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1846 | old_cset = task_css_set(tsk); |
74a1166d | 1847 | |
b3dc094e | 1848 | get_css_set(new_cset); |
5abb8855 | 1849 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d | 1850 | |
1b9aba49 TH |
1851 | /* |
1852 | * Use move_tail so that cgroup_taskset_first() still returns the | |
1853 | * leader after migration. This works because cgroup_migrate() | |
1854 | * ensures that the dst_cset of the leader is the first on the | |
1855 | * tset's dst_csets list. | |
1856 | */ | |
1857 | list_move_tail(&tsk->cg_list, &new_cset->mg_tasks); | |
74a1166d BB |
1858 | |
1859 | /* | |
5abb8855 TH |
1860 | * We just gained a reference on old_cset by taking it from the |
1861 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1862 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1863 | */ |
5abb8855 | 1864 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
cb0f1fe9 | 1865 | put_css_set_locked(old_cset, false); |
74a1166d BB |
1866 | } |
1867 | ||
a043e3b2 | 1868 | /** |
1958d2d5 TH |
1869 | * cgroup_migrate_finish - cleanup after attach |
1870 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 1871 | * |
1958d2d5 TH |
1872 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
1873 | * those functions for details. | |
74a1166d | 1874 | */ |
1958d2d5 | 1875 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 1876 | { |
1958d2d5 | 1877 | struct css_set *cset, *tmp_cset; |
74a1166d | 1878 | |
1958d2d5 TH |
1879 | lockdep_assert_held(&cgroup_mutex); |
1880 | ||
1881 | down_write(&css_set_rwsem); | |
1882 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
1883 | cset->mg_src_cgrp = NULL; | |
1884 | cset->mg_dst_cset = NULL; | |
1885 | list_del_init(&cset->mg_preload_node); | |
1886 | put_css_set_locked(cset, false); | |
1887 | } | |
1888 | up_write(&css_set_rwsem); | |
1889 | } | |
1890 | ||
1891 | /** | |
1892 | * cgroup_migrate_add_src - add a migration source css_set | |
1893 | * @src_cset: the source css_set to add | |
1894 | * @dst_cgrp: the destination cgroup | |
1895 | * @preloaded_csets: list of preloaded css_sets | |
1896 | * | |
1897 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
1898 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
1899 | * up by cgroup_migrate_finish(). | |
1900 | * | |
1901 | * This function may be called without holding threadgroup_lock even if the | |
1902 | * target is a process. Threads may be created and destroyed but as long | |
1903 | * as cgroup_mutex is not dropped, no new css_set can be put into play and | |
1904 | * the preloaded css_sets are guaranteed to cover all migrations. | |
1905 | */ | |
1906 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
1907 | struct cgroup *dst_cgrp, | |
1908 | struct list_head *preloaded_csets) | |
1909 | { | |
1910 | struct cgroup *src_cgrp; | |
1911 | ||
1912 | lockdep_assert_held(&cgroup_mutex); | |
1913 | lockdep_assert_held(&css_set_rwsem); | |
1914 | ||
1915 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
1916 | ||
1958d2d5 TH |
1917 | if (!list_empty(&src_cset->mg_preload_node)) |
1918 | return; | |
1919 | ||
1920 | WARN_ON(src_cset->mg_src_cgrp); | |
1921 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
1922 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
1923 | ||
1924 | src_cset->mg_src_cgrp = src_cgrp; | |
1925 | get_css_set(src_cset); | |
1926 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
1927 | } | |
1928 | ||
1929 | /** | |
1930 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 1931 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
1932 | * @preloaded_csets: list of preloaded source css_sets |
1933 | * | |
1934 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
1935 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
1936 | * pins all destination css_sets, links each to its source, and append them |
1937 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
1938 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
1939 | * |
1940 | * This function must be called after cgroup_migrate_add_src() has been | |
1941 | * called on each migration source css_set. After migration is performed | |
1942 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
1943 | * @preloaded_csets. | |
1944 | */ | |
1945 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
1946 | struct list_head *preloaded_csets) | |
1947 | { | |
1948 | LIST_HEAD(csets); | |
f817de98 | 1949 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
1950 | |
1951 | lockdep_assert_held(&cgroup_mutex); | |
1952 | ||
f8f22e53 TH |
1953 | /* |
1954 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
1955 | * with tasks so that child cgroups don't compete against tasks. | |
1956 | */ | |
1957 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && dst_cgrp->parent && | |
1958 | dst_cgrp->child_subsys_mask) | |
1959 | return -EBUSY; | |
1960 | ||
1958d2d5 | 1961 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 1962 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
1963 | struct css_set *dst_cset; |
1964 | ||
f817de98 TH |
1965 | dst_cset = find_css_set(src_cset, |
1966 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
1967 | if (!dst_cset) |
1968 | goto err; | |
1969 | ||
1970 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
1971 | |
1972 | /* | |
1973 | * If src cset equals dst, it's noop. Drop the src. | |
1974 | * cgroup_migrate() will skip the cset too. Note that we | |
1975 | * can't handle src == dst as some nodes are used by both. | |
1976 | */ | |
1977 | if (src_cset == dst_cset) { | |
1978 | src_cset->mg_src_cgrp = NULL; | |
1979 | list_del_init(&src_cset->mg_preload_node); | |
1980 | put_css_set(src_cset, false); | |
1981 | put_css_set(dst_cset, false); | |
1982 | continue; | |
1983 | } | |
1984 | ||
1958d2d5 TH |
1985 | src_cset->mg_dst_cset = dst_cset; |
1986 | ||
1987 | if (list_empty(&dst_cset->mg_preload_node)) | |
1988 | list_add(&dst_cset->mg_preload_node, &csets); | |
1989 | else | |
1990 | put_css_set(dst_cset, false); | |
1991 | } | |
1992 | ||
f817de98 | 1993 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
1994 | return 0; |
1995 | err: | |
1996 | cgroup_migrate_finish(&csets); | |
1997 | return -ENOMEM; | |
1998 | } | |
1999 | ||
2000 | /** | |
2001 | * cgroup_migrate - migrate a process or task to a cgroup | |
2002 | * @cgrp: the destination cgroup | |
2003 | * @leader: the leader of the process or the task to migrate | |
2004 | * @threadgroup: whether @leader points to the whole process or a single task | |
2005 | * | |
2006 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
2007 | * process, the caller must be holding threadgroup_lock of @leader. The | |
2008 | * caller is also responsible for invoking cgroup_migrate_add_src() and | |
2009 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2010 | * function and following up with cgroup_migrate_finish(). | |
2011 | * | |
2012 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2013 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2014 | * failure, when migrating multiple targets, the success or failure can be | |
2015 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2016 | * actually starting migrating. | |
2017 | */ | |
2018 | static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader, | |
2019 | bool threadgroup) | |
74a1166d | 2020 | { |
b3dc094e TH |
2021 | struct cgroup_taskset tset = { |
2022 | .src_csets = LIST_HEAD_INIT(tset.src_csets), | |
2023 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), | |
2024 | .csets = &tset.src_csets, | |
2025 | }; | |
1c6727af | 2026 | struct cgroup_subsys_state *css, *failed_css = NULL; |
b3dc094e TH |
2027 | struct css_set *cset, *tmp_cset; |
2028 | struct task_struct *task, *tmp_task; | |
2029 | int i, ret; | |
74a1166d | 2030 | |
fb5d2b4c MSB |
2031 | /* |
2032 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2033 | * already PF_EXITING could be freed from underneath us unless we | |
2034 | * take an rcu_read_lock. | |
2035 | */ | |
b3dc094e | 2036 | down_write(&css_set_rwsem); |
fb5d2b4c | 2037 | rcu_read_lock(); |
9db8de37 | 2038 | task = leader; |
74a1166d | 2039 | do { |
9db8de37 TH |
2040 | /* @task either already exited or can't exit until the end */ |
2041 | if (task->flags & PF_EXITING) | |
ea84753c | 2042 | goto next; |
134d3373 | 2043 | |
eaf797ab TH |
2044 | /* leave @task alone if post_fork() hasn't linked it yet */ |
2045 | if (list_empty(&task->cg_list)) | |
ea84753c | 2046 | goto next; |
cd3d0952 | 2047 | |
b3dc094e | 2048 | cset = task_css_set(task); |
1958d2d5 | 2049 | if (!cset->mg_src_cgrp) |
ea84753c | 2050 | goto next; |
b3dc094e | 2051 | |
61d1d219 | 2052 | /* |
1b9aba49 TH |
2053 | * cgroup_taskset_first() must always return the leader. |
2054 | * Take care to avoid disturbing the ordering. | |
61d1d219 | 2055 | */ |
1b9aba49 TH |
2056 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2057 | if (list_empty(&cset->mg_node)) | |
2058 | list_add_tail(&cset->mg_node, &tset.src_csets); | |
2059 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2060 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2061 | &tset.dst_csets); | |
ea84753c | 2062 | next: |
081aa458 LZ |
2063 | if (!threadgroup) |
2064 | break; | |
9db8de37 | 2065 | } while_each_thread(leader, task); |
fb5d2b4c | 2066 | rcu_read_unlock(); |
b3dc094e | 2067 | up_write(&css_set_rwsem); |
74a1166d | 2068 | |
134d3373 | 2069 | /* methods shouldn't be called if no task is actually migrating */ |
b3dc094e TH |
2070 | if (list_empty(&tset.src_csets)) |
2071 | return 0; | |
134d3373 | 2072 | |
1958d2d5 | 2073 | /* check that we can legitimately attach to the cgroup */ |
aec3dfcb | 2074 | for_each_e_css(css, i, cgrp) { |
1c6727af | 2075 | if (css->ss->can_attach) { |
9db8de37 TH |
2076 | ret = css->ss->can_attach(css, &tset); |
2077 | if (ret) { | |
1c6727af | 2078 | failed_css = css; |
74a1166d BB |
2079 | goto out_cancel_attach; |
2080 | } | |
2081 | } | |
74a1166d BB |
2082 | } |
2083 | ||
2084 | /* | |
1958d2d5 TH |
2085 | * Now that we're guaranteed success, proceed to move all tasks to |
2086 | * the new cgroup. There are no failure cases after here, so this | |
2087 | * is the commit point. | |
74a1166d | 2088 | */ |
cb0f1fe9 | 2089 | down_write(&css_set_rwsem); |
b3dc094e TH |
2090 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
2091 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
2092 | cgroup_task_migrate(cset->mg_src_cgrp, task, | |
2093 | cset->mg_dst_cset); | |
74a1166d | 2094 | } |
cb0f1fe9 | 2095 | up_write(&css_set_rwsem); |
74a1166d BB |
2096 | |
2097 | /* | |
1958d2d5 TH |
2098 | * Migration is committed, all target tasks are now on dst_csets. |
2099 | * Nothing is sensitive to fork() after this point. Notify | |
2100 | * controllers that migration is complete. | |
74a1166d | 2101 | */ |
1958d2d5 | 2102 | tset.csets = &tset.dst_csets; |
74a1166d | 2103 | |
aec3dfcb | 2104 | for_each_e_css(css, i, cgrp) |
1c6727af TH |
2105 | if (css->ss->attach) |
2106 | css->ss->attach(css, &tset); | |
74a1166d | 2107 | |
9db8de37 | 2108 | ret = 0; |
b3dc094e TH |
2109 | goto out_release_tset; |
2110 | ||
74a1166d | 2111 | out_cancel_attach: |
aec3dfcb | 2112 | for_each_e_css(css, i, cgrp) { |
b3dc094e TH |
2113 | if (css == failed_css) |
2114 | break; | |
2115 | if (css->ss->cancel_attach) | |
2116 | css->ss->cancel_attach(css, &tset); | |
74a1166d | 2117 | } |
b3dc094e TH |
2118 | out_release_tset: |
2119 | down_write(&css_set_rwsem); | |
2120 | list_splice_init(&tset.dst_csets, &tset.src_csets); | |
2121 | list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { | |
1b9aba49 | 2122 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); |
b3dc094e | 2123 | list_del_init(&cset->mg_node); |
b3dc094e TH |
2124 | } |
2125 | up_write(&css_set_rwsem); | |
9db8de37 | 2126 | return ret; |
74a1166d BB |
2127 | } |
2128 | ||
1958d2d5 TH |
2129 | /** |
2130 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2131 | * @dst_cgrp: the cgroup to attach to | |
2132 | * @leader: the task or the leader of the threadgroup to be attached | |
2133 | * @threadgroup: attach the whole threadgroup? | |
2134 | * | |
0e1d768f | 2135 | * Call holding cgroup_mutex and threadgroup_lock of @leader. |
1958d2d5 TH |
2136 | */ |
2137 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2138 | struct task_struct *leader, bool threadgroup) | |
2139 | { | |
2140 | LIST_HEAD(preloaded_csets); | |
2141 | struct task_struct *task; | |
2142 | int ret; | |
2143 | ||
2144 | /* look up all src csets */ | |
2145 | down_read(&css_set_rwsem); | |
2146 | rcu_read_lock(); | |
2147 | task = leader; | |
2148 | do { | |
2149 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2150 | &preloaded_csets); | |
2151 | if (!threadgroup) | |
2152 | break; | |
2153 | } while_each_thread(leader, task); | |
2154 | rcu_read_unlock(); | |
2155 | up_read(&css_set_rwsem); | |
2156 | ||
2157 | /* prepare dst csets and commit */ | |
2158 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2159 | if (!ret) | |
2160 | ret = cgroup_migrate(dst_cgrp, leader, threadgroup); | |
2161 | ||
2162 | cgroup_migrate_finish(&preloaded_csets); | |
2163 | return ret; | |
74a1166d BB |
2164 | } |
2165 | ||
2166 | /* | |
2167 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2168 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2169 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2170 | */ |
74a1166d | 2171 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2172 | { |
bbcb81d0 | 2173 | struct task_struct *tsk; |
c69e8d9c | 2174 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2175 | int ret; |
2176 | ||
74a1166d BB |
2177 | if (!cgroup_lock_live_group(cgrp)) |
2178 | return -ENODEV; | |
2179 | ||
b78949eb MSB |
2180 | retry_find_task: |
2181 | rcu_read_lock(); | |
bbcb81d0 | 2182 | if (pid) { |
73507f33 | 2183 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2184 | if (!tsk) { |
2185 | rcu_read_unlock(); | |
dd4b0a46 | 2186 | ret = -ESRCH; |
b78949eb | 2187 | goto out_unlock_cgroup; |
bbcb81d0 | 2188 | } |
74a1166d BB |
2189 | /* |
2190 | * even if we're attaching all tasks in the thread group, we | |
2191 | * only need to check permissions on one of them. | |
2192 | */ | |
c69e8d9c | 2193 | tcred = __task_cred(tsk); |
14a590c3 EB |
2194 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2195 | !uid_eq(cred->euid, tcred->uid) && | |
2196 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2197 | rcu_read_unlock(); |
b78949eb MSB |
2198 | ret = -EACCES; |
2199 | goto out_unlock_cgroup; | |
bbcb81d0 | 2200 | } |
b78949eb MSB |
2201 | } else |
2202 | tsk = current; | |
cd3d0952 TH |
2203 | |
2204 | if (threadgroup) | |
b78949eb | 2205 | tsk = tsk->group_leader; |
c4c27fbd MG |
2206 | |
2207 | /* | |
14a40ffc | 2208 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2209 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2210 | * with no rt_runtime allocated. Just say no. | |
2211 | */ | |
14a40ffc | 2212 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2213 | ret = -EINVAL; |
2214 | rcu_read_unlock(); | |
2215 | goto out_unlock_cgroup; | |
2216 | } | |
2217 | ||
b78949eb MSB |
2218 | get_task_struct(tsk); |
2219 | rcu_read_unlock(); | |
2220 | ||
2221 | threadgroup_lock(tsk); | |
2222 | if (threadgroup) { | |
2223 | if (!thread_group_leader(tsk)) { | |
2224 | /* | |
2225 | * a race with de_thread from another thread's exec() | |
2226 | * may strip us of our leadership, if this happens, | |
2227 | * there is no choice but to throw this task away and | |
2228 | * try again; this is | |
2229 | * "double-double-toil-and-trouble-check locking". | |
2230 | */ | |
2231 | threadgroup_unlock(tsk); | |
2232 | put_task_struct(tsk); | |
2233 | goto retry_find_task; | |
2234 | } | |
081aa458 LZ |
2235 | } |
2236 | ||
2237 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2238 | ||
cd3d0952 TH |
2239 | threadgroup_unlock(tsk); |
2240 | ||
bbcb81d0 | 2241 | put_task_struct(tsk); |
b78949eb | 2242 | out_unlock_cgroup: |
47cfcd09 | 2243 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
2244 | return ret; |
2245 | } | |
2246 | ||
7ae1bad9 TH |
2247 | /** |
2248 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2249 | * @from: attach to all cgroups of a given task | |
2250 | * @tsk: the task to be attached | |
2251 | */ | |
2252 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2253 | { | |
3dd06ffa | 2254 | struct cgroup_root *root; |
7ae1bad9 TH |
2255 | int retval = 0; |
2256 | ||
47cfcd09 | 2257 | mutex_lock(&cgroup_mutex); |
985ed670 | 2258 | for_each_root(root) { |
96d365e0 TH |
2259 | struct cgroup *from_cgrp; |
2260 | ||
3dd06ffa | 2261 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2262 | continue; |
2263 | ||
96d365e0 TH |
2264 | down_read(&css_set_rwsem); |
2265 | from_cgrp = task_cgroup_from_root(from, root); | |
2266 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2267 | |
6f4b7e63 | 2268 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2269 | if (retval) |
2270 | break; | |
2271 | } | |
47cfcd09 | 2272 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2273 | |
2274 | return retval; | |
2275 | } | |
2276 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2277 | ||
182446d0 TH |
2278 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2279 | struct cftype *cft, u64 pid) | |
74a1166d | 2280 | { |
182446d0 | 2281 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2282 | } |
2283 | ||
182446d0 TH |
2284 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2285 | struct cftype *cft, u64 tgid) | |
af351026 | 2286 | { |
182446d0 | 2287 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2288 | } |
2289 | ||
182446d0 | 2290 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
4d3bb511 | 2291 | struct cftype *cft, char *buffer) |
e788e066 | 2292 | { |
3dd06ffa | 2293 | struct cgroup_root *root = css->cgroup->root; |
5f469907 TH |
2294 | |
2295 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
182446d0 | 2296 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2297 | return -ENODEV; |
69e943b7 | 2298 | spin_lock(&release_agent_path_lock); |
5f469907 TH |
2299 | strlcpy(root->release_agent_path, buffer, |
2300 | sizeof(root->release_agent_path)); | |
69e943b7 | 2301 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2302 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2303 | return 0; |
2304 | } | |
2305 | ||
2da8ca82 | 2306 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2307 | { |
2da8ca82 | 2308 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2309 | |
e788e066 PM |
2310 | if (!cgroup_lock_live_group(cgrp)) |
2311 | return -ENODEV; | |
2312 | seq_puts(seq, cgrp->root->release_agent_path); | |
2313 | seq_putc(seq, '\n'); | |
47cfcd09 | 2314 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2315 | return 0; |
2316 | } | |
2317 | ||
2da8ca82 | 2318 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2319 | { |
2da8ca82 TH |
2320 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2321 | ||
2322 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2323 | return 0; |
2324 | } | |
2325 | ||
f8f22e53 TH |
2326 | static void cgroup_print_ss_mask(struct seq_file *seq, unsigned int ss_mask) |
2327 | { | |
2328 | struct cgroup_subsys *ss; | |
2329 | bool printed = false; | |
2330 | int ssid; | |
2331 | ||
2332 | for_each_subsys(ss, ssid) { | |
2333 | if (ss_mask & (1 << ssid)) { | |
2334 | if (printed) | |
2335 | seq_putc(seq, ' '); | |
2336 | seq_printf(seq, "%s", ss->name); | |
2337 | printed = true; | |
2338 | } | |
2339 | } | |
2340 | if (printed) | |
2341 | seq_putc(seq, '\n'); | |
2342 | } | |
2343 | ||
2344 | /* show controllers which are currently attached to the default hierarchy */ | |
2345 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
2346 | { | |
2347 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2348 | ||
2349 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask); | |
2350 | return 0; | |
2351 | } | |
2352 | ||
2353 | /* show controllers which are enabled from the parent */ | |
2354 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
2355 | { | |
2356 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2357 | ||
2358 | cgroup_print_ss_mask(seq, cgrp->parent->child_subsys_mask); | |
2359 | return 0; | |
2360 | } | |
2361 | ||
2362 | /* show controllers which are enabled for a given cgroup's children */ | |
2363 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
2364 | { | |
2365 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2366 | ||
2367 | cgroup_print_ss_mask(seq, cgrp->child_subsys_mask); | |
2368 | return 0; | |
2369 | } | |
2370 | ||
2371 | /** | |
2372 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2373 | * @cgrp: root of the subtree to update csses for | |
2374 | * | |
2375 | * @cgrp's child_subsys_mask has changed and its subtree's (self excluded) | |
2376 | * css associations need to be updated accordingly. This function looks up | |
2377 | * all css_sets which are attached to the subtree, creates the matching | |
2378 | * updated css_sets and migrates the tasks to the new ones. | |
2379 | */ | |
2380 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2381 | { | |
2382 | LIST_HEAD(preloaded_csets); | |
2383 | struct cgroup_subsys_state *css; | |
2384 | struct css_set *src_cset; | |
2385 | int ret; | |
2386 | ||
2387 | lockdep_assert_held(&cgroup_tree_mutex); | |
2388 | lockdep_assert_held(&cgroup_mutex); | |
2389 | ||
2390 | /* look up all csses currently attached to @cgrp's subtree */ | |
2391 | down_read(&css_set_rwsem); | |
2392 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { | |
2393 | struct cgrp_cset_link *link; | |
2394 | ||
2395 | /* self is not affected by child_subsys_mask change */ | |
2396 | if (css->cgroup == cgrp) | |
2397 | continue; | |
2398 | ||
2399 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2400 | cgroup_migrate_add_src(link->cset, cgrp, | |
2401 | &preloaded_csets); | |
2402 | } | |
2403 | up_read(&css_set_rwsem); | |
2404 | ||
2405 | /* NULL dst indicates self on default hierarchy */ | |
2406 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2407 | if (ret) | |
2408 | goto out_finish; | |
2409 | ||
2410 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { | |
2411 | struct task_struct *last_task = NULL, *task; | |
2412 | ||
2413 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2414 | if (!src_cset->mg_src_cgrp) | |
2415 | break; | |
2416 | ||
2417 | /* | |
2418 | * All tasks in src_cset need to be migrated to the | |
2419 | * matching dst_cset. Empty it process by process. We | |
2420 | * walk tasks but migrate processes. The leader might even | |
2421 | * belong to a different cset but such src_cset would also | |
2422 | * be among the target src_csets because the default | |
2423 | * hierarchy enforces per-process membership. | |
2424 | */ | |
2425 | while (true) { | |
2426 | down_read(&css_set_rwsem); | |
2427 | task = list_first_entry_or_null(&src_cset->tasks, | |
2428 | struct task_struct, cg_list); | |
2429 | if (task) { | |
2430 | task = task->group_leader; | |
2431 | WARN_ON_ONCE(!task_css_set(task)->mg_src_cgrp); | |
2432 | get_task_struct(task); | |
2433 | } | |
2434 | up_read(&css_set_rwsem); | |
2435 | ||
2436 | if (!task) | |
2437 | break; | |
2438 | ||
2439 | /* guard against possible infinite loop */ | |
2440 | if (WARN(last_task == task, | |
2441 | "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n")) | |
2442 | goto out_finish; | |
2443 | last_task = task; | |
2444 | ||
2445 | threadgroup_lock(task); | |
2446 | /* raced against de_thread() from another thread? */ | |
2447 | if (!thread_group_leader(task)) { | |
2448 | threadgroup_unlock(task); | |
2449 | put_task_struct(task); | |
2450 | continue; | |
2451 | } | |
2452 | ||
2453 | ret = cgroup_migrate(src_cset->dfl_cgrp, task, true); | |
2454 | ||
2455 | threadgroup_unlock(task); | |
2456 | put_task_struct(task); | |
2457 | ||
2458 | if (WARN(ret, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret)) | |
2459 | goto out_finish; | |
2460 | } | |
2461 | } | |
2462 | ||
2463 | out_finish: | |
2464 | cgroup_migrate_finish(&preloaded_csets); | |
2465 | return ret; | |
2466 | } | |
2467 | ||
2468 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
2469 | static int cgroup_subtree_control_write(struct cgroup_subsys_state *dummy_css, | |
2470 | struct cftype *cft, char *buffer) | |
2471 | { | |
2472 | unsigned long enable_req = 0, disable_req = 0, enable, disable; | |
2473 | struct cgroup *cgrp = dummy_css->cgroup, *child; | |
2474 | struct cgroup_subsys *ss; | |
2475 | char *tok, *p; | |
2476 | int ssid, ret; | |
2477 | ||
2478 | /* | |
2479 | * Parse input - white space separated list of subsystem names | |
2480 | * prefixed with either + or -. | |
2481 | */ | |
2482 | p = buffer; | |
2483 | while ((tok = strsep(&p, " \t\n"))) { | |
2484 | for_each_subsys(ss, ssid) { | |
2485 | if (ss->disabled || strcmp(tok + 1, ss->name)) | |
2486 | continue; | |
2487 | ||
2488 | if (*tok == '+') { | |
2489 | enable_req |= 1 << ssid; | |
2490 | disable_req &= ~(1 << ssid); | |
2491 | } else if (*tok == '-') { | |
2492 | disable_req |= 1 << ssid; | |
2493 | enable_req &= ~(1 << ssid); | |
2494 | } else { | |
2495 | return -EINVAL; | |
2496 | } | |
2497 | break; | |
2498 | } | |
2499 | if (ssid == CGROUP_SUBSYS_COUNT) | |
2500 | return -EINVAL; | |
2501 | } | |
2502 | ||
2503 | /* | |
2504 | * We're gonna grab cgroup_tree_mutex which nests outside kernfs | |
2505 | * active_ref. cgroup_lock_live_group() already provides enough | |
2506 | * protection. Ensure @cgrp stays accessible and break the | |
2507 | * active_ref protection. | |
2508 | */ | |
2509 | cgroup_get(cgrp); | |
2510 | kernfs_break_active_protection(cgrp->control_kn); | |
2511 | retry: | |
2512 | enable = enable_req; | |
2513 | disable = disable_req; | |
2514 | ||
2515 | mutex_lock(&cgroup_tree_mutex); | |
2516 | ||
2517 | for_each_subsys(ss, ssid) { | |
2518 | if (enable & (1 << ssid)) { | |
2519 | if (cgrp->child_subsys_mask & (1 << ssid)) { | |
2520 | enable &= ~(1 << ssid); | |
2521 | continue; | |
2522 | } | |
2523 | ||
2524 | /* | |
2525 | * Because css offlining is asynchronous, userland | |
2526 | * might try to re-enable the same controller while | |
2527 | * the previous instance is still around. In such | |
2528 | * cases, wait till it's gone using offline_waitq. | |
2529 | */ | |
2530 | cgroup_for_each_live_child(child, cgrp) { | |
2531 | wait_queue_t wait; | |
2532 | ||
2533 | if (!cgroup_css(child, ss)) | |
2534 | continue; | |
2535 | ||
2536 | prepare_to_wait(&child->offline_waitq, &wait, | |
2537 | TASK_UNINTERRUPTIBLE); | |
2538 | mutex_unlock(&cgroup_tree_mutex); | |
2539 | schedule(); | |
2540 | finish_wait(&child->offline_waitq, &wait); | |
2541 | goto retry; | |
2542 | } | |
2543 | ||
2544 | /* unavailable or not enabled on the parent? */ | |
2545 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
2546 | (cgrp->parent && | |
2547 | !(cgrp->parent->child_subsys_mask & (1 << ssid)))) { | |
2548 | ret = -ENOENT; | |
2549 | goto out_unlock_tree; | |
2550 | } | |
2551 | } else if (disable & (1 << ssid)) { | |
2552 | if (!(cgrp->child_subsys_mask & (1 << ssid))) { | |
2553 | disable &= ~(1 << ssid); | |
2554 | continue; | |
2555 | } | |
2556 | ||
2557 | /* a child has it enabled? */ | |
2558 | cgroup_for_each_live_child(child, cgrp) { | |
2559 | if (child->child_subsys_mask & (1 << ssid)) { | |
2560 | ret = -EBUSY; | |
2561 | goto out_unlock_tree; | |
2562 | } | |
2563 | } | |
2564 | } | |
2565 | } | |
2566 | ||
2567 | if (!enable && !disable) { | |
2568 | ret = 0; | |
2569 | goto out_unlock_tree; | |
2570 | } | |
2571 | ||
2572 | if (!cgroup_lock_live_group(cgrp)) { | |
2573 | ret = -ENODEV; | |
2574 | goto out_unlock_tree; | |
2575 | } | |
2576 | ||
2577 | /* | |
2578 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2579 | * with tasks so that child cgroups don't compete against tasks. | |
2580 | */ | |
2581 | if (enable && cgrp->parent && !list_empty(&cgrp->cset_links)) { | |
2582 | ret = -EBUSY; | |
2583 | goto out_unlock; | |
2584 | } | |
2585 | ||
2586 | /* | |
2587 | * Create csses for enables and update child_subsys_mask. This | |
2588 | * changes cgroup_e_css() results which in turn makes the | |
2589 | * subsequent cgroup_update_dfl_csses() associate all tasks in the | |
2590 | * subtree to the updated csses. | |
2591 | */ | |
2592 | for_each_subsys(ss, ssid) { | |
2593 | if (!(enable & (1 << ssid))) | |
2594 | continue; | |
2595 | ||
2596 | cgroup_for_each_live_child(child, cgrp) { | |
2597 | ret = create_css(child, ss); | |
2598 | if (ret) | |
2599 | goto err_undo_css; | |
2600 | } | |
2601 | } | |
2602 | ||
2603 | cgrp->child_subsys_mask |= enable; | |
2604 | cgrp->child_subsys_mask &= ~disable; | |
2605 | ||
2606 | ret = cgroup_update_dfl_csses(cgrp); | |
2607 | if (ret) | |
2608 | goto err_undo_css; | |
2609 | ||
2610 | /* all tasks are now migrated away from the old csses, kill them */ | |
2611 | for_each_subsys(ss, ssid) { | |
2612 | if (!(disable & (1 << ssid))) | |
2613 | continue; | |
2614 | ||
2615 | cgroup_for_each_live_child(child, cgrp) | |
2616 | kill_css(cgroup_css(child, ss)); | |
2617 | } | |
2618 | ||
2619 | kernfs_activate(cgrp->kn); | |
2620 | ret = 0; | |
2621 | out_unlock: | |
2622 | mutex_unlock(&cgroup_mutex); | |
2623 | out_unlock_tree: | |
2624 | mutex_unlock(&cgroup_tree_mutex); | |
2625 | kernfs_unbreak_active_protection(cgrp->control_kn); | |
2626 | cgroup_put(cgrp); | |
2627 | return ret; | |
2628 | ||
2629 | err_undo_css: | |
2630 | cgrp->child_subsys_mask &= ~enable; | |
2631 | cgrp->child_subsys_mask |= disable; | |
2632 | ||
2633 | for_each_subsys(ss, ssid) { | |
2634 | if (!(enable & (1 << ssid))) | |
2635 | continue; | |
2636 | ||
2637 | cgroup_for_each_live_child(child, cgrp) { | |
2638 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
2639 | if (css) | |
2640 | kill_css(css); | |
2641 | } | |
2642 | } | |
2643 | goto out_unlock; | |
2644 | } | |
2645 | ||
2bd59d48 TH |
2646 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2647 | size_t nbytes, loff_t off) | |
355e0c48 | 2648 | { |
2bd59d48 TH |
2649 | struct cgroup *cgrp = of->kn->parent->priv; |
2650 | struct cftype *cft = of->kn->priv; | |
2651 | struct cgroup_subsys_state *css; | |
a742c59d | 2652 | int ret; |
355e0c48 | 2653 | |
2bd59d48 TH |
2654 | /* |
2655 | * kernfs guarantees that a file isn't deleted with operations in | |
2656 | * flight, which means that the matching css is and stays alive and | |
2657 | * doesn't need to be pinned. The RCU locking is not necessary | |
2658 | * either. It's just for the convenience of using cgroup_css(). | |
2659 | */ | |
2660 | rcu_read_lock(); | |
2661 | css = cgroup_css(cgrp, cft->ss); | |
2662 | rcu_read_unlock(); | |
a742c59d TH |
2663 | |
2664 | if (cft->write_string) { | |
2665 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2666 | } else if (cft->write_u64) { | |
2667 | unsigned long long v; | |
2668 | ret = kstrtoull(buf, 0, &v); | |
2669 | if (!ret) | |
2670 | ret = cft->write_u64(css, cft, v); | |
2671 | } else if (cft->write_s64) { | |
2672 | long long v; | |
2673 | ret = kstrtoll(buf, 0, &v); | |
2674 | if (!ret) | |
2675 | ret = cft->write_s64(css, cft, v); | |
2676 | } else if (cft->trigger) { | |
2677 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2678 | } else { |
a742c59d | 2679 | ret = -EINVAL; |
e73d2c61 | 2680 | } |
2bd59d48 | 2681 | |
a742c59d | 2682 | return ret ?: nbytes; |
355e0c48 PM |
2683 | } |
2684 | ||
6612f05b | 2685 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2686 | { |
2bd59d48 | 2687 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2688 | } |
2689 | ||
6612f05b | 2690 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2691 | { |
2bd59d48 | 2692 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2693 | } |
2694 | ||
6612f05b | 2695 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2696 | { |
2bd59d48 | 2697 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2698 | } |
2699 | ||
91796569 | 2700 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2701 | { |
7da11279 TH |
2702 | struct cftype *cft = seq_cft(m); |
2703 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2704 | |
2da8ca82 TH |
2705 | if (cft->seq_show) |
2706 | return cft->seq_show(m, arg); | |
e73d2c61 | 2707 | |
f4c753b7 | 2708 | if (cft->read_u64) |
896f5199 TH |
2709 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2710 | else if (cft->read_s64) | |
2711 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2712 | else | |
2713 | return -EINVAL; | |
2714 | return 0; | |
91796569 PM |
2715 | } |
2716 | ||
2bd59d48 TH |
2717 | static struct kernfs_ops cgroup_kf_single_ops = { |
2718 | .atomic_write_len = PAGE_SIZE, | |
2719 | .write = cgroup_file_write, | |
2720 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2721 | }; |
2722 | ||
2bd59d48 TH |
2723 | static struct kernfs_ops cgroup_kf_ops = { |
2724 | .atomic_write_len = PAGE_SIZE, | |
2725 | .write = cgroup_file_write, | |
2726 | .seq_start = cgroup_seqfile_start, | |
2727 | .seq_next = cgroup_seqfile_next, | |
2728 | .seq_stop = cgroup_seqfile_stop, | |
2729 | .seq_show = cgroup_seqfile_show, | |
2730 | }; | |
ddbcc7e8 PM |
2731 | |
2732 | /* | |
2733 | * cgroup_rename - Only allow simple rename of directories in place. | |
2734 | */ | |
2bd59d48 TH |
2735 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2736 | const char *new_name_str) | |
ddbcc7e8 | 2737 | { |
2bd59d48 | 2738 | struct cgroup *cgrp = kn->priv; |
65dff759 | 2739 | int ret; |
65dff759 | 2740 | |
2bd59d48 | 2741 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2742 | return -ENOTDIR; |
2bd59d48 | 2743 | if (kn->parent != new_parent) |
ddbcc7e8 | 2744 | return -EIO; |
65dff759 | 2745 | |
6db8e85c TH |
2746 | /* |
2747 | * This isn't a proper migration and its usefulness is very | |
2748 | * limited. Disallow if sane_behavior. | |
2749 | */ | |
2750 | if (cgroup_sane_behavior(cgrp)) | |
2751 | return -EPERM; | |
099fca32 | 2752 | |
e1b2dc17 TH |
2753 | /* |
2754 | * We're gonna grab cgroup_tree_mutex which nests outside kernfs | |
2755 | * active_ref. kernfs_rename() doesn't require active_ref | |
2756 | * protection. Break them before grabbing cgroup_tree_mutex. | |
2757 | */ | |
2758 | kernfs_break_active_protection(new_parent); | |
2759 | kernfs_break_active_protection(kn); | |
099fca32 | 2760 | |
2bd59d48 TH |
2761 | mutex_lock(&cgroup_tree_mutex); |
2762 | mutex_lock(&cgroup_mutex); | |
099fca32 | 2763 | |
2bd59d48 | 2764 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 2765 | |
2bd59d48 TH |
2766 | mutex_unlock(&cgroup_mutex); |
2767 | mutex_unlock(&cgroup_tree_mutex); | |
e1b2dc17 TH |
2768 | |
2769 | kernfs_unbreak_active_protection(kn); | |
2770 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 2771 | return ret; |
099fca32 LZ |
2772 | } |
2773 | ||
49957f8e TH |
2774 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
2775 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
2776 | { | |
2777 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
2778 | .ia_uid = current_fsuid(), | |
2779 | .ia_gid = current_fsgid(), }; | |
2780 | ||
2781 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
2782 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
2783 | return 0; | |
2784 | ||
2785 | return kernfs_setattr(kn, &iattr); | |
2786 | } | |
2787 | ||
2bb566cb | 2788 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2789 | { |
8d7e6fb0 | 2790 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2791 | struct kernfs_node *kn; |
2792 | struct lock_class_key *key = NULL; | |
49957f8e | 2793 | int ret; |
05ef1d7c | 2794 | |
2bd59d48 TH |
2795 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2796 | key = &cft->lockdep_key; | |
2797 | #endif | |
2798 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2799 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2800 | NULL, false, key); | |
49957f8e TH |
2801 | if (IS_ERR(kn)) |
2802 | return PTR_ERR(kn); | |
2803 | ||
2804 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 2805 | if (ret) { |
49957f8e | 2806 | kernfs_remove(kn); |
f8f22e53 TH |
2807 | return ret; |
2808 | } | |
2809 | ||
2810 | if (cft->seq_show == cgroup_subtree_control_show) | |
2811 | cgrp->control_kn = kn; | |
2812 | return 0; | |
ddbcc7e8 PM |
2813 | } |
2814 | ||
b1f28d31 TH |
2815 | /** |
2816 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2817 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2818 | * @cfts: array of cftypes to be added |
2819 | * @is_add: whether to add or remove | |
2820 | * | |
2821 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2822 | * For removals, this function never fails. If addition fails, this |
2823 | * function doesn't remove files already added. The caller is responsible | |
2824 | * for cleaning up. | |
b1f28d31 | 2825 | */ |
2bb566cb TH |
2826 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2827 | bool is_add) | |
ddbcc7e8 | 2828 | { |
03b1cde6 | 2829 | struct cftype *cft; |
b1f28d31 TH |
2830 | int ret; |
2831 | ||
ace2bee8 | 2832 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2833 | |
2834 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2835 | /* does cft->flags tell us to skip this file on @cgrp? */ |
8cbbf2c9 TH |
2836 | if ((cft->flags & CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
2837 | continue; | |
873fe09e TH |
2838 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2839 | continue; | |
f33fddc2 G |
2840 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2841 | continue; | |
2842 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2843 | continue; | |
2844 | ||
2739d3cc | 2845 | if (is_add) { |
2bb566cb | 2846 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2847 | if (ret) { |
2739d3cc | 2848 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2849 | cft->name, ret); |
2850 | return ret; | |
2851 | } | |
2739d3cc LZ |
2852 | } else { |
2853 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2854 | } |
ddbcc7e8 | 2855 | } |
b1f28d31 | 2856 | return 0; |
ddbcc7e8 PM |
2857 | } |
2858 | ||
21a2d343 | 2859 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2860 | { |
2861 | LIST_HEAD(pending); | |
2bb566cb | 2862 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 2863 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 2864 | struct cgroup_subsys_state *css; |
9ccece80 | 2865 | int ret = 0; |
8e3f6541 | 2866 | |
21a2d343 | 2867 | lockdep_assert_held(&cgroup_tree_mutex); |
8e3f6541 | 2868 | |
e8c82d20 | 2869 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2870 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2871 | struct cgroup *cgrp = css->cgroup; |
2872 | ||
e8c82d20 LZ |
2873 | if (cgroup_is_dead(cgrp)) |
2874 | continue; | |
2875 | ||
21a2d343 | 2876 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2877 | if (ret) |
2878 | break; | |
8e3f6541 | 2879 | } |
21a2d343 TH |
2880 | |
2881 | if (is_add && !ret) | |
2882 | kernfs_activate(root->kn); | |
9ccece80 | 2883 | return ret; |
8e3f6541 TH |
2884 | } |
2885 | ||
2da440a2 | 2886 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 2887 | { |
2bb566cb | 2888 | struct cftype *cft; |
8e3f6541 | 2889 | |
2bd59d48 TH |
2890 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2891 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2892 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2893 | kfree(cft->kf_ops); | |
2894 | cft->kf_ops = NULL; | |
2da440a2 | 2895 | cft->ss = NULL; |
2bd59d48 | 2896 | } |
2da440a2 TH |
2897 | } |
2898 | ||
2bd59d48 | 2899 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2900 | { |
2901 | struct cftype *cft; | |
2902 | ||
2bd59d48 TH |
2903 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2904 | struct kernfs_ops *kf_ops; | |
2905 | ||
0adb0704 TH |
2906 | WARN_ON(cft->ss || cft->kf_ops); |
2907 | ||
2bd59d48 TH |
2908 | if (cft->seq_start) |
2909 | kf_ops = &cgroup_kf_ops; | |
2910 | else | |
2911 | kf_ops = &cgroup_kf_single_ops; | |
2912 | ||
2913 | /* | |
2914 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2915 | * make a copy of kf_ops to set its atomic_write_len. | |
2916 | */ | |
2917 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2918 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2919 | if (!kf_ops) { | |
2920 | cgroup_exit_cftypes(cfts); | |
2921 | return -ENOMEM; | |
2922 | } | |
2923 | kf_ops->atomic_write_len = cft->max_write_len; | |
2924 | } | |
8e3f6541 | 2925 | |
2bd59d48 | 2926 | cft->kf_ops = kf_ops; |
2bb566cb | 2927 | cft->ss = ss; |
2bd59d48 | 2928 | } |
2bb566cb | 2929 | |
2bd59d48 | 2930 | return 0; |
2da440a2 TH |
2931 | } |
2932 | ||
21a2d343 TH |
2933 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
2934 | { | |
2935 | lockdep_assert_held(&cgroup_tree_mutex); | |
2936 | ||
2937 | if (!cfts || !cfts[0].ss) | |
2938 | return -ENOENT; | |
2939 | ||
2940 | list_del(&cfts->node); | |
2941 | cgroup_apply_cftypes(cfts, false); | |
2942 | cgroup_exit_cftypes(cfts); | |
2943 | return 0; | |
8e3f6541 | 2944 | } |
8e3f6541 | 2945 | |
79578621 TH |
2946 | /** |
2947 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
2948 | * @cfts: zero-length name terminated array of cftypes |
2949 | * | |
2bb566cb TH |
2950 | * Unregister @cfts. Files described by @cfts are removed from all |
2951 | * existing cgroups and all future cgroups won't have them either. This | |
2952 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
2953 | * |
2954 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 2955 | * registered. |
79578621 | 2956 | */ |
2bb566cb | 2957 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 2958 | { |
21a2d343 | 2959 | int ret; |
79578621 | 2960 | |
21a2d343 TH |
2961 | mutex_lock(&cgroup_tree_mutex); |
2962 | ret = cgroup_rm_cftypes_locked(cfts); | |
2963 | mutex_unlock(&cgroup_tree_mutex); | |
2964 | return ret; | |
80b13586 TH |
2965 | } |
2966 | ||
8e3f6541 TH |
2967 | /** |
2968 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2969 | * @ss: target cgroup subsystem | |
2970 | * @cfts: zero-length name terminated array of cftypes | |
2971 | * | |
2972 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2973 | * existing cgroups to which @ss is attached and all future cgroups will | |
2974 | * have them too. This function can be called anytime whether @ss is | |
2975 | * attached or not. | |
2976 | * | |
2977 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2978 | * function currently returns 0 as long as @cfts registration is successful | |
2979 | * even if some file creation attempts on existing cgroups fail. | |
2980 | */ | |
03b1cde6 | 2981 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 2982 | { |
9ccece80 | 2983 | int ret; |
8e3f6541 | 2984 | |
dc5736ed LZ |
2985 | if (!cfts || cfts[0].name[0] == '\0') |
2986 | return 0; | |
2bb566cb | 2987 | |
2bd59d48 TH |
2988 | ret = cgroup_init_cftypes(ss, cfts); |
2989 | if (ret) | |
2990 | return ret; | |
79578621 | 2991 | |
21a2d343 TH |
2992 | mutex_lock(&cgroup_tree_mutex); |
2993 | ||
0adb0704 | 2994 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 2995 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 2996 | if (ret) |
21a2d343 | 2997 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 2998 | |
21a2d343 | 2999 | mutex_unlock(&cgroup_tree_mutex); |
9ccece80 | 3000 | return ret; |
79578621 TH |
3001 | } |
3002 | ||
a043e3b2 LZ |
3003 | /** |
3004 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3005 | * @cgrp: the cgroup in question | |
3006 | * | |
3007 | * Return the number of tasks in the cgroup. | |
3008 | */ | |
07bc356e | 3009 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3010 | { |
3011 | int count = 0; | |
69d0206c | 3012 | struct cgrp_cset_link *link; |
817929ec | 3013 | |
96d365e0 | 3014 | down_read(&css_set_rwsem); |
69d0206c TH |
3015 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3016 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 3017 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
3018 | return count; |
3019 | } | |
3020 | ||
53fa5261 | 3021 | /** |
492eb21b TH |
3022 | * css_next_child - find the next child of a given css |
3023 | * @pos_css: the current position (%NULL to initiate traversal) | |
3024 | * @parent_css: css whose children to walk | |
53fa5261 | 3025 | * |
492eb21b | 3026 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
3027 | * under either cgroup_mutex or RCU read lock. The only requirement is |
3028 | * that @parent_css and @pos_css are accessible. The next sibling is | |
3029 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 3030 | */ |
492eb21b TH |
3031 | struct cgroup_subsys_state * |
3032 | css_next_child(struct cgroup_subsys_state *pos_css, | |
3033 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 3034 | { |
492eb21b TH |
3035 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
3036 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
3037 | struct cgroup *next; |
3038 | ||
ace2bee8 | 3039 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
3040 | |
3041 | /* | |
3042 | * @pos could already have been removed. Once a cgroup is removed, | |
3043 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
3044 | * changes. As CGRP_DEAD assertion is serialized and happens |
3045 | * before the cgroup is taken off the ->sibling list, if we see it | |
3046 | * unasserted, it's guaranteed that the next sibling hasn't | |
3047 | * finished its grace period even if it's already removed, and thus | |
3048 | * safe to dereference from this RCU critical section. If | |
3049 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
3050 | * to be visible as %true here. | |
3b287a50 TH |
3051 | * |
3052 | * If @pos is dead, its next pointer can't be dereferenced; | |
3053 | * however, as each cgroup is given a monotonically increasing | |
3054 | * unique serial number and always appended to the sibling list, | |
3055 | * the next one can be found by walking the parent's children until | |
3056 | * we see a cgroup with higher serial number than @pos's. While | |
3057 | * this path can be slower, it's taken only when either the current | |
3058 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 3059 | */ |
3b287a50 TH |
3060 | if (!pos) { |
3061 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
3062 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 3063 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
3064 | } else { |
3065 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
3066 | if (next->serial_nr > pos->serial_nr) | |
3067 | break; | |
53fa5261 TH |
3068 | } |
3069 | ||
3b281afb TH |
3070 | /* |
3071 | * @next, if not pointing to the head, can be dereferenced and is | |
3072 | * the next sibling; however, it might have @ss disabled. If so, | |
3073 | * fast-forward to the next enabled one. | |
3074 | */ | |
3075 | while (&next->sibling != &cgrp->children) { | |
3076 | struct cgroup_subsys_state *next_css = cgroup_css(next, parent_css->ss); | |
492eb21b | 3077 | |
3b281afb TH |
3078 | if (next_css) |
3079 | return next_css; | |
3080 | next = list_entry_rcu(next->sibling.next, struct cgroup, sibling); | |
3081 | } | |
3082 | return NULL; | |
53fa5261 | 3083 | } |
53fa5261 | 3084 | |
574bd9f7 | 3085 | /** |
492eb21b | 3086 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3087 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3088 | * @root: css whose descendants to walk |
574bd9f7 | 3089 | * |
492eb21b | 3090 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3091 | * to visit for pre-order traversal of @root's descendants. @root is |
3092 | * included in the iteration and the first node to be visited. | |
75501a6d | 3093 | * |
87fb54f1 TH |
3094 | * While this function requires cgroup_mutex or RCU read locking, it |
3095 | * doesn't require the whole traversal to be contained in a single critical | |
3096 | * section. This function will return the correct next descendant as long | |
3097 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 3098 | */ |
492eb21b TH |
3099 | struct cgroup_subsys_state * |
3100 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3101 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3102 | { |
492eb21b | 3103 | struct cgroup_subsys_state *next; |
574bd9f7 | 3104 | |
ace2bee8 | 3105 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 3106 | |
bd8815a6 | 3107 | /* if first iteration, visit @root */ |
7805d000 | 3108 | if (!pos) |
bd8815a6 | 3109 | return root; |
574bd9f7 TH |
3110 | |
3111 | /* visit the first child if exists */ | |
492eb21b | 3112 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3113 | if (next) |
3114 | return next; | |
3115 | ||
3116 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
3117 | while (pos != root) { |
3118 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 3119 | if (next) |
574bd9f7 | 3120 | return next; |
492eb21b | 3121 | pos = css_parent(pos); |
7805d000 | 3122 | } |
574bd9f7 TH |
3123 | |
3124 | return NULL; | |
3125 | } | |
574bd9f7 | 3126 | |
12a9d2fe | 3127 | /** |
492eb21b TH |
3128 | * css_rightmost_descendant - return the rightmost descendant of a css |
3129 | * @pos: css of interest | |
12a9d2fe | 3130 | * |
492eb21b TH |
3131 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3132 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3133 | * subtree of @pos. |
75501a6d | 3134 | * |
87fb54f1 TH |
3135 | * While this function requires cgroup_mutex or RCU read locking, it |
3136 | * doesn't require the whole traversal to be contained in a single critical | |
3137 | * section. This function will return the correct rightmost descendant as | |
3138 | * long as @pos is accessible. | |
12a9d2fe | 3139 | */ |
492eb21b TH |
3140 | struct cgroup_subsys_state * |
3141 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3142 | { |
492eb21b | 3143 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3144 | |
ace2bee8 | 3145 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
3146 | |
3147 | do { | |
3148 | last = pos; | |
3149 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3150 | pos = NULL; | |
492eb21b | 3151 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3152 | pos = tmp; |
3153 | } while (pos); | |
3154 | ||
3155 | return last; | |
3156 | } | |
12a9d2fe | 3157 | |
492eb21b TH |
3158 | static struct cgroup_subsys_state * |
3159 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3160 | { |
492eb21b | 3161 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3162 | |
3163 | do { | |
3164 | last = pos; | |
492eb21b | 3165 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3166 | } while (pos); |
3167 | ||
3168 | return last; | |
3169 | } | |
3170 | ||
3171 | /** | |
492eb21b | 3172 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3173 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3174 | * @root: css whose descendants to walk |
574bd9f7 | 3175 | * |
492eb21b | 3176 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3177 | * to visit for post-order traversal of @root's descendants. @root is |
3178 | * included in the iteration and the last node to be visited. | |
75501a6d | 3179 | * |
87fb54f1 TH |
3180 | * While this function requires cgroup_mutex or RCU read locking, it |
3181 | * doesn't require the whole traversal to be contained in a single critical | |
3182 | * section. This function will return the correct next descendant as long | |
3183 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3184 | * @cgroup. | |
574bd9f7 | 3185 | */ |
492eb21b TH |
3186 | struct cgroup_subsys_state * |
3187 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3188 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3189 | { |
492eb21b | 3190 | struct cgroup_subsys_state *next; |
574bd9f7 | 3191 | |
ace2bee8 | 3192 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 3193 | |
58b79a91 TH |
3194 | /* if first iteration, visit leftmost descendant which may be @root */ |
3195 | if (!pos) | |
3196 | return css_leftmost_descendant(root); | |
574bd9f7 | 3197 | |
bd8815a6 TH |
3198 | /* if we visited @root, we're done */ |
3199 | if (pos == root) | |
3200 | return NULL; | |
3201 | ||
574bd9f7 | 3202 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 3203 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 3204 | if (next) |
492eb21b | 3205 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3206 | |
3207 | /* no sibling left, visit parent */ | |
bd8815a6 | 3208 | return css_parent(pos); |
574bd9f7 | 3209 | } |
574bd9f7 | 3210 | |
0942eeee | 3211 | /** |
72ec7029 | 3212 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
3213 | * @it: the iterator to advance |
3214 | * | |
3215 | * Advance @it to the next css_set to walk. | |
d515876e | 3216 | */ |
72ec7029 | 3217 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e | 3218 | { |
0f0a2b4f | 3219 | struct list_head *l = it->cset_pos; |
d515876e TH |
3220 | struct cgrp_cset_link *link; |
3221 | struct css_set *cset; | |
3222 | ||
3223 | /* Advance to the next non-empty css_set */ | |
3224 | do { | |
3225 | l = l->next; | |
0f0a2b4f TH |
3226 | if (l == it->cset_head) { |
3227 | it->cset_pos = NULL; | |
d515876e TH |
3228 | return; |
3229 | } | |
3ebb2b6e TH |
3230 | |
3231 | if (it->ss) { | |
3232 | cset = container_of(l, struct css_set, | |
3233 | e_cset_node[it->ss->id]); | |
3234 | } else { | |
3235 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3236 | cset = link->cset; | |
3237 | } | |
c7561128 TH |
3238 | } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); |
3239 | ||
0f0a2b4f | 3240 | it->cset_pos = l; |
c7561128 TH |
3241 | |
3242 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3243 | it->task_pos = cset->tasks.next; |
c7561128 | 3244 | else |
0f0a2b4f TH |
3245 | it->task_pos = cset->mg_tasks.next; |
3246 | ||
3247 | it->tasks_head = &cset->tasks; | |
3248 | it->mg_tasks_head = &cset->mg_tasks; | |
d515876e TH |
3249 | } |
3250 | ||
0942eeee | 3251 | /** |
72ec7029 TH |
3252 | * css_task_iter_start - initiate task iteration |
3253 | * @css: the css to walk tasks of | |
0942eeee TH |
3254 | * @it: the task iterator to use |
3255 | * | |
72ec7029 TH |
3256 | * Initiate iteration through the tasks of @css. The caller can call |
3257 | * css_task_iter_next() to walk through the tasks until the function | |
3258 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3259 | * called. | |
0942eeee TH |
3260 | * |
3261 | * Note that this function acquires a lock which is released when the | |
3262 | * iteration finishes. The caller can't sleep while iteration is in | |
3263 | * progress. | |
3264 | */ | |
72ec7029 TH |
3265 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3266 | struct css_task_iter *it) | |
96d365e0 | 3267 | __acquires(css_set_rwsem) |
817929ec | 3268 | { |
56fde9e0 TH |
3269 | /* no one should try to iterate before mounting cgroups */ |
3270 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3271 | |
96d365e0 | 3272 | down_read(&css_set_rwsem); |
c59cd3d8 | 3273 | |
3ebb2b6e TH |
3274 | it->ss = css->ss; |
3275 | ||
3276 | if (it->ss) | |
3277 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3278 | else | |
3279 | it->cset_pos = &css->cgroup->cset_links; | |
3280 | ||
0f0a2b4f | 3281 | it->cset_head = it->cset_pos; |
c59cd3d8 | 3282 | |
72ec7029 | 3283 | css_advance_task_iter(it); |
817929ec PM |
3284 | } |
3285 | ||
0942eeee | 3286 | /** |
72ec7029 | 3287 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3288 | * @it: the task iterator being iterated |
3289 | * | |
3290 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3291 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3292 | * reaches the end. | |
0942eeee | 3293 | */ |
72ec7029 | 3294 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
3295 | { |
3296 | struct task_struct *res; | |
0f0a2b4f | 3297 | struct list_head *l = it->task_pos; |
817929ec PM |
3298 | |
3299 | /* If the iterator cg is NULL, we have no tasks */ | |
0f0a2b4f | 3300 | if (!it->cset_pos) |
817929ec PM |
3301 | return NULL; |
3302 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
3303 | |
3304 | /* | |
3305 | * Advance iterator to find next entry. cset->tasks is consumed | |
3306 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3307 | * next cset. | |
3308 | */ | |
817929ec | 3309 | l = l->next; |
c7561128 | 3310 | |
0f0a2b4f TH |
3311 | if (l == it->tasks_head) |
3312 | l = it->mg_tasks_head->next; | |
c7561128 | 3313 | |
0f0a2b4f | 3314 | if (l == it->mg_tasks_head) |
72ec7029 | 3315 | css_advance_task_iter(it); |
c7561128 | 3316 | else |
0f0a2b4f | 3317 | it->task_pos = l; |
c7561128 | 3318 | |
817929ec PM |
3319 | return res; |
3320 | } | |
3321 | ||
0942eeee | 3322 | /** |
72ec7029 | 3323 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3324 | * @it: the task iterator to finish |
3325 | * | |
72ec7029 | 3326 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3327 | */ |
72ec7029 | 3328 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 3329 | __releases(css_set_rwsem) |
31a7df01 | 3330 | { |
96d365e0 | 3331 | up_read(&css_set_rwsem); |
31a7df01 CW |
3332 | } |
3333 | ||
3334 | /** | |
8cc99345 TH |
3335 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
3336 | * @to: cgroup to which the tasks will be moved | |
3337 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 3338 | * |
eaf797ab TH |
3339 | * Locking rules between cgroup_post_fork() and the migration path |
3340 | * guarantee that, if a task is forking while being migrated, the new child | |
3341 | * is guaranteed to be either visible in the source cgroup after the | |
3342 | * parent's migration is complete or put into the target cgroup. No task | |
3343 | * can slip out of migration through forking. | |
31a7df01 | 3344 | */ |
8cc99345 | 3345 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 3346 | { |
952aaa12 TH |
3347 | LIST_HEAD(preloaded_csets); |
3348 | struct cgrp_cset_link *link; | |
72ec7029 | 3349 | struct css_task_iter it; |
e406d1cf | 3350 | struct task_struct *task; |
952aaa12 | 3351 | int ret; |
31a7df01 | 3352 | |
952aaa12 | 3353 | mutex_lock(&cgroup_mutex); |
31a7df01 | 3354 | |
952aaa12 TH |
3355 | /* all tasks in @from are being moved, all csets are source */ |
3356 | down_read(&css_set_rwsem); | |
3357 | list_for_each_entry(link, &from->cset_links, cset_link) | |
3358 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
3359 | up_read(&css_set_rwsem); | |
31a7df01 | 3360 | |
952aaa12 TH |
3361 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
3362 | if (ret) | |
3363 | goto out_err; | |
8cc99345 | 3364 | |
952aaa12 TH |
3365 | /* |
3366 | * Migrate tasks one-by-one until @form is empty. This fails iff | |
3367 | * ->can_attach() fails. | |
3368 | */ | |
e406d1cf TH |
3369 | do { |
3370 | css_task_iter_start(&from->dummy_css, &it); | |
3371 | task = css_task_iter_next(&it); | |
3372 | if (task) | |
3373 | get_task_struct(task); | |
3374 | css_task_iter_end(&it); | |
3375 | ||
3376 | if (task) { | |
952aaa12 | 3377 | ret = cgroup_migrate(to, task, false); |
e406d1cf TH |
3378 | put_task_struct(task); |
3379 | } | |
3380 | } while (task && !ret); | |
952aaa12 TH |
3381 | out_err: |
3382 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 3383 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 3384 | return ret; |
8cc99345 TH |
3385 | } |
3386 | ||
bbcb81d0 | 3387 | /* |
102a775e | 3388 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3389 | * |
3390 | * Reading this file can return large amounts of data if a cgroup has | |
3391 | * *lots* of attached tasks. So it may need several calls to read(), | |
3392 | * but we cannot guarantee that the information we produce is correct | |
3393 | * unless we produce it entirely atomically. | |
3394 | * | |
bbcb81d0 | 3395 | */ |
bbcb81d0 | 3396 | |
24528255 LZ |
3397 | /* which pidlist file are we talking about? */ |
3398 | enum cgroup_filetype { | |
3399 | CGROUP_FILE_PROCS, | |
3400 | CGROUP_FILE_TASKS, | |
3401 | }; | |
3402 | ||
3403 | /* | |
3404 | * A pidlist is a list of pids that virtually represents the contents of one | |
3405 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3406 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3407 | * to the cgroup. | |
3408 | */ | |
3409 | struct cgroup_pidlist { | |
3410 | /* | |
3411 | * used to find which pidlist is wanted. doesn't change as long as | |
3412 | * this particular list stays in the list. | |
3413 | */ | |
3414 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3415 | /* array of xids */ | |
3416 | pid_t *list; | |
3417 | /* how many elements the above list has */ | |
3418 | int length; | |
24528255 LZ |
3419 | /* each of these stored in a list by its cgroup */ |
3420 | struct list_head links; | |
3421 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3422 | struct cgroup *owner; | |
b1a21367 TH |
3423 | /* for delayed destruction */ |
3424 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3425 | }; |
3426 | ||
d1d9fd33 BB |
3427 | /* |
3428 | * The following two functions "fix" the issue where there are more pids | |
3429 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3430 | * TODO: replace with a kernel-wide solution to this problem | |
3431 | */ | |
3432 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3433 | static void *pidlist_allocate(int count) | |
3434 | { | |
3435 | if (PIDLIST_TOO_LARGE(count)) | |
3436 | return vmalloc(count * sizeof(pid_t)); | |
3437 | else | |
3438 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3439 | } | |
b1a21367 | 3440 | |
d1d9fd33 BB |
3441 | static void pidlist_free(void *p) |
3442 | { | |
3443 | if (is_vmalloc_addr(p)) | |
3444 | vfree(p); | |
3445 | else | |
3446 | kfree(p); | |
3447 | } | |
d1d9fd33 | 3448 | |
b1a21367 TH |
3449 | /* |
3450 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3451 | * should be left afterwards. | |
3452 | */ | |
3453 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3454 | { | |
3455 | struct cgroup_pidlist *l, *tmp_l; | |
3456 | ||
3457 | mutex_lock(&cgrp->pidlist_mutex); | |
3458 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3459 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3460 | mutex_unlock(&cgrp->pidlist_mutex); | |
3461 | ||
3462 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3463 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3464 | } | |
3465 | ||
3466 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3467 | { | |
3468 | struct delayed_work *dwork = to_delayed_work(work); | |
3469 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3470 | destroy_dwork); | |
3471 | struct cgroup_pidlist *tofree = NULL; | |
3472 | ||
3473 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3474 | |
3475 | /* | |
04502365 TH |
3476 | * Destroy iff we didn't get queued again. The state won't change |
3477 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3478 | */ |
04502365 | 3479 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3480 | list_del(&l->links); |
3481 | pidlist_free(l->list); | |
3482 | put_pid_ns(l->key.ns); | |
3483 | tofree = l; | |
3484 | } | |
3485 | ||
b1a21367 TH |
3486 | mutex_unlock(&l->owner->pidlist_mutex); |
3487 | kfree(tofree); | |
3488 | } | |
3489 | ||
bbcb81d0 | 3490 | /* |
102a775e | 3491 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3492 | * Returns the number of unique elements. |
bbcb81d0 | 3493 | */ |
6ee211ad | 3494 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3495 | { |
102a775e | 3496 | int src, dest = 1; |
102a775e BB |
3497 | |
3498 | /* | |
3499 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3500 | * edge cases first; no work needs to be done for either | |
3501 | */ | |
3502 | if (length == 0 || length == 1) | |
3503 | return length; | |
3504 | /* src and dest walk down the list; dest counts unique elements */ | |
3505 | for (src = 1; src < length; src++) { | |
3506 | /* find next unique element */ | |
3507 | while (list[src] == list[src-1]) { | |
3508 | src++; | |
3509 | if (src == length) | |
3510 | goto after; | |
3511 | } | |
3512 | /* dest always points to where the next unique element goes */ | |
3513 | list[dest] = list[src]; | |
3514 | dest++; | |
3515 | } | |
3516 | after: | |
102a775e BB |
3517 | return dest; |
3518 | } | |
3519 | ||
afb2bc14 TH |
3520 | /* |
3521 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3522 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3523 | * different per namespace, each namespace needs differently sorted list, | |
3524 | * making it impossible to use, for example, single rbtree of member tasks | |
3525 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3526 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3527 | * pidlists keyed by cgroup and namespace. | |
3528 | * | |
3529 | * All this extra complexity was caused by the original implementation | |
3530 | * committing to an entirely unnecessary property. In the long term, we | |
3531 | * want to do away with it. Explicitly scramble sort order if | |
3532 | * sane_behavior so that no such expectation exists in the new interface. | |
3533 | * | |
3534 | * Scrambling is done by swapping every two consecutive bits, which is | |
3535 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3536 | */ | |
3537 | static pid_t pid_fry(pid_t pid) | |
3538 | { | |
3539 | unsigned a = pid & 0x55555555; | |
3540 | unsigned b = pid & 0xAAAAAAAA; | |
3541 | ||
3542 | return (a << 1) | (b >> 1); | |
3543 | } | |
3544 | ||
3545 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3546 | { | |
3547 | if (cgroup_sane_behavior(cgrp)) | |
3548 | return pid_fry(pid); | |
3549 | else | |
3550 | return pid; | |
3551 | } | |
3552 | ||
102a775e BB |
3553 | static int cmppid(const void *a, const void *b) |
3554 | { | |
3555 | return *(pid_t *)a - *(pid_t *)b; | |
3556 | } | |
3557 | ||
afb2bc14 TH |
3558 | static int fried_cmppid(const void *a, const void *b) |
3559 | { | |
3560 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3561 | } | |
3562 | ||
e6b81710 TH |
3563 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3564 | enum cgroup_filetype type) | |
3565 | { | |
3566 | struct cgroup_pidlist *l; | |
3567 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3568 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3569 | ||
3570 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3571 | ||
3572 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3573 | if (l->key.type == type && l->key.ns == ns) | |
3574 | return l; | |
3575 | return NULL; | |
3576 | } | |
3577 | ||
72a8cb30 BB |
3578 | /* |
3579 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3580 | * returns with the lock on that pidlist already held, and takes care | |
3581 | * of the use count, or returns NULL with no locks held if we're out of | |
3582 | * memory. | |
3583 | */ | |
e6b81710 TH |
3584 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3585 | enum cgroup_filetype type) | |
72a8cb30 BB |
3586 | { |
3587 | struct cgroup_pidlist *l; | |
b70cc5fd | 3588 | |
e6b81710 TH |
3589 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3590 | ||
3591 | l = cgroup_pidlist_find(cgrp, type); | |
3592 | if (l) | |
3593 | return l; | |
3594 | ||
72a8cb30 | 3595 | /* entry not found; create a new one */ |
f4f4be2b | 3596 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3597 | if (!l) |
72a8cb30 | 3598 | return l; |
e6b81710 | 3599 | |
b1a21367 | 3600 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3601 | l->key.type = type; |
e6b81710 TH |
3602 | /* don't need task_nsproxy() if we're looking at ourself */ |
3603 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3604 | l->owner = cgrp; |
3605 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3606 | return l; |
3607 | } | |
3608 | ||
102a775e BB |
3609 | /* |
3610 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3611 | */ | |
72a8cb30 BB |
3612 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3613 | struct cgroup_pidlist **lp) | |
102a775e BB |
3614 | { |
3615 | pid_t *array; | |
3616 | int length; | |
3617 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3618 | struct css_task_iter it; |
817929ec | 3619 | struct task_struct *tsk; |
102a775e BB |
3620 | struct cgroup_pidlist *l; |
3621 | ||
4bac00d1 TH |
3622 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3623 | ||
102a775e BB |
3624 | /* |
3625 | * If cgroup gets more users after we read count, we won't have | |
3626 | * enough space - tough. This race is indistinguishable to the | |
3627 | * caller from the case that the additional cgroup users didn't | |
3628 | * show up until sometime later on. | |
3629 | */ | |
3630 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3631 | array = pidlist_allocate(length); |
102a775e BB |
3632 | if (!array) |
3633 | return -ENOMEM; | |
3634 | /* now, populate the array */ | |
72ec7029 TH |
3635 | css_task_iter_start(&cgrp->dummy_css, &it); |
3636 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3637 | if (unlikely(n == length)) |
817929ec | 3638 | break; |
102a775e | 3639 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3640 | if (type == CGROUP_FILE_PROCS) |
3641 | pid = task_tgid_vnr(tsk); | |
3642 | else | |
3643 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3644 | if (pid > 0) /* make sure to only use valid results */ |
3645 | array[n++] = pid; | |
817929ec | 3646 | } |
72ec7029 | 3647 | css_task_iter_end(&it); |
102a775e BB |
3648 | length = n; |
3649 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3650 | if (cgroup_sane_behavior(cgrp)) |
3651 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3652 | else | |
3653 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3654 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3655 | length = pidlist_uniq(array, length); |
e6b81710 | 3656 | |
e6b81710 | 3657 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3658 | if (!l) { |
e6b81710 | 3659 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3660 | pidlist_free(array); |
72a8cb30 | 3661 | return -ENOMEM; |
102a775e | 3662 | } |
e6b81710 TH |
3663 | |
3664 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3665 | pidlist_free(l->list); |
102a775e BB |
3666 | l->list = array; |
3667 | l->length = length; | |
72a8cb30 | 3668 | *lp = l; |
102a775e | 3669 | return 0; |
bbcb81d0 PM |
3670 | } |
3671 | ||
846c7bb0 | 3672 | /** |
a043e3b2 | 3673 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3674 | * @stats: cgroupstats to fill information into |
3675 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3676 | * been requested. | |
a043e3b2 LZ |
3677 | * |
3678 | * Build and fill cgroupstats so that taskstats can export it to user | |
3679 | * space. | |
846c7bb0 BS |
3680 | */ |
3681 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3682 | { | |
2bd59d48 | 3683 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3684 | struct cgroup *cgrp; |
72ec7029 | 3685 | struct css_task_iter it; |
846c7bb0 | 3686 | struct task_struct *tsk; |
33d283be | 3687 | |
2bd59d48 TH |
3688 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3689 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3690 | kernfs_type(kn) != KERNFS_DIR) | |
3691 | return -EINVAL; | |
3692 | ||
bad34660 LZ |
3693 | mutex_lock(&cgroup_mutex); |
3694 | ||
846c7bb0 | 3695 | /* |
2bd59d48 TH |
3696 | * We aren't being called from kernfs and there's no guarantee on |
3697 | * @kn->priv's validity. For this and css_tryget_from_dir(), | |
3698 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
846c7bb0 | 3699 | */ |
2bd59d48 TH |
3700 | rcu_read_lock(); |
3701 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 3702 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 3703 | rcu_read_unlock(); |
bad34660 | 3704 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
3705 | return -ENOENT; |
3706 | } | |
bad34660 | 3707 | rcu_read_unlock(); |
846c7bb0 | 3708 | |
72ec7029 TH |
3709 | css_task_iter_start(&cgrp->dummy_css, &it); |
3710 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3711 | switch (tsk->state) { |
3712 | case TASK_RUNNING: | |
3713 | stats->nr_running++; | |
3714 | break; | |
3715 | case TASK_INTERRUPTIBLE: | |
3716 | stats->nr_sleeping++; | |
3717 | break; | |
3718 | case TASK_UNINTERRUPTIBLE: | |
3719 | stats->nr_uninterruptible++; | |
3720 | break; | |
3721 | case TASK_STOPPED: | |
3722 | stats->nr_stopped++; | |
3723 | break; | |
3724 | default: | |
3725 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3726 | stats->nr_io_wait++; | |
3727 | break; | |
3728 | } | |
3729 | } | |
72ec7029 | 3730 | css_task_iter_end(&it); |
846c7bb0 | 3731 | |
bad34660 | 3732 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 3733 | return 0; |
846c7bb0 BS |
3734 | } |
3735 | ||
8f3ff208 | 3736 | |
bbcb81d0 | 3737 | /* |
102a775e | 3738 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3739 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3740 | * in the cgroup->l->list array. |
bbcb81d0 | 3741 | */ |
cc31edce | 3742 | |
102a775e | 3743 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3744 | { |
cc31edce PM |
3745 | /* |
3746 | * Initially we receive a position value that corresponds to | |
3747 | * one more than the last pid shown (or 0 on the first call or | |
3748 | * after a seek to the start). Use a binary-search to find the | |
3749 | * next pid to display, if any | |
3750 | */ | |
2bd59d48 | 3751 | struct kernfs_open_file *of = s->private; |
7da11279 | 3752 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3753 | struct cgroup_pidlist *l; |
7da11279 | 3754 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3755 | int index = 0, pid = *pos; |
4bac00d1 TH |
3756 | int *iter, ret; |
3757 | ||
3758 | mutex_lock(&cgrp->pidlist_mutex); | |
3759 | ||
3760 | /* | |
5d22444f | 3761 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3762 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3763 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3764 | * could already have been destroyed. |
3765 | */ | |
5d22444f TH |
3766 | if (of->priv) |
3767 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3768 | |
3769 | /* | |
3770 | * Either this is the first start() after open or the matching | |
3771 | * pidlist has been destroyed inbetween. Create a new one. | |
3772 | */ | |
5d22444f TH |
3773 | if (!of->priv) { |
3774 | ret = pidlist_array_load(cgrp, type, | |
3775 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3776 | if (ret) |
3777 | return ERR_PTR(ret); | |
3778 | } | |
5d22444f | 3779 | l = of->priv; |
cc31edce | 3780 | |
cc31edce | 3781 | if (pid) { |
102a775e | 3782 | int end = l->length; |
20777766 | 3783 | |
cc31edce PM |
3784 | while (index < end) { |
3785 | int mid = (index + end) / 2; | |
afb2bc14 | 3786 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3787 | index = mid; |
3788 | break; | |
afb2bc14 | 3789 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3790 | index = mid + 1; |
3791 | else | |
3792 | end = mid; | |
3793 | } | |
3794 | } | |
3795 | /* If we're off the end of the array, we're done */ | |
102a775e | 3796 | if (index >= l->length) |
cc31edce PM |
3797 | return NULL; |
3798 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3799 | iter = l->list + index; |
afb2bc14 | 3800 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3801 | return iter; |
3802 | } | |
3803 | ||
102a775e | 3804 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3805 | { |
2bd59d48 | 3806 | struct kernfs_open_file *of = s->private; |
5d22444f | 3807 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3808 | |
5d22444f TH |
3809 | if (l) |
3810 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3811 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3812 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3813 | } |
3814 | ||
102a775e | 3815 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3816 | { |
2bd59d48 | 3817 | struct kernfs_open_file *of = s->private; |
5d22444f | 3818 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3819 | pid_t *p = v; |
3820 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3821 | /* |
3822 | * Advance to the next pid in the array. If this goes off the | |
3823 | * end, we're done | |
3824 | */ | |
3825 | p++; | |
3826 | if (p >= end) { | |
3827 | return NULL; | |
3828 | } else { | |
7da11279 | 3829 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3830 | return p; |
3831 | } | |
3832 | } | |
3833 | ||
102a775e | 3834 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3835 | { |
3836 | return seq_printf(s, "%d\n", *(int *)v); | |
3837 | } | |
bbcb81d0 | 3838 | |
102a775e BB |
3839 | /* |
3840 | * seq_operations functions for iterating on pidlists through seq_file - | |
3841 | * independent of whether it's tasks or procs | |
3842 | */ | |
3843 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3844 | .start = cgroup_pidlist_start, | |
3845 | .stop = cgroup_pidlist_stop, | |
3846 | .next = cgroup_pidlist_next, | |
3847 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3848 | }; |
3849 | ||
182446d0 TH |
3850 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3851 | struct cftype *cft) | |
81a6a5cd | 3852 | { |
182446d0 | 3853 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3854 | } |
3855 | ||
182446d0 TH |
3856 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3857 | struct cftype *cft, u64 val) | |
6379c106 | 3858 | { |
182446d0 | 3859 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3860 | if (val) |
182446d0 | 3861 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3862 | else |
182446d0 | 3863 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3864 | return 0; |
3865 | } | |
3866 | ||
182446d0 TH |
3867 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3868 | struct cftype *cft) | |
97978e6d | 3869 | { |
182446d0 | 3870 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3871 | } |
3872 | ||
182446d0 TH |
3873 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3874 | struct cftype *cft, u64 val) | |
97978e6d DL |
3875 | { |
3876 | if (val) | |
182446d0 | 3877 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3878 | else |
182446d0 | 3879 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3880 | return 0; |
3881 | } | |
3882 | ||
d5c56ced | 3883 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3884 | { |
d5c56ced | 3885 | .name = "cgroup.procs", |
6612f05b TH |
3886 | .seq_start = cgroup_pidlist_start, |
3887 | .seq_next = cgroup_pidlist_next, | |
3888 | .seq_stop = cgroup_pidlist_stop, | |
3889 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3890 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3891 | .write_u64 = cgroup_procs_write, |
74a1166d | 3892 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3893 | }, |
97978e6d DL |
3894 | { |
3895 | .name = "cgroup.clone_children", | |
873fe09e | 3896 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3897 | .read_u64 = cgroup_clone_children_read, |
3898 | .write_u64 = cgroup_clone_children_write, | |
3899 | }, | |
873fe09e TH |
3900 | { |
3901 | .name = "cgroup.sane_behavior", | |
3902 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3903 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3904 | }, |
f8f22e53 TH |
3905 | { |
3906 | .name = "cgroup.controllers", | |
3907 | .flags = CFTYPE_ONLY_ON_DFL | CFTYPE_ONLY_ON_ROOT, | |
3908 | .seq_show = cgroup_root_controllers_show, | |
3909 | }, | |
3910 | { | |
3911 | .name = "cgroup.controllers", | |
3912 | .flags = CFTYPE_ONLY_ON_DFL | CFTYPE_NOT_ON_ROOT, | |
3913 | .seq_show = cgroup_controllers_show, | |
3914 | }, | |
3915 | { | |
3916 | .name = "cgroup.subtree_control", | |
3917 | .flags = CFTYPE_ONLY_ON_DFL, | |
3918 | .seq_show = cgroup_subtree_control_show, | |
3919 | .write_string = cgroup_subtree_control_write, | |
3920 | }, | |
d5c56ced TH |
3921 | |
3922 | /* | |
3923 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3924 | * don't exist if sane_behavior. If you're depending on these, be | |
3925 | * prepared to be burned. | |
3926 | */ | |
3927 | { | |
3928 | .name = "tasks", | |
3929 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3930 | .seq_start = cgroup_pidlist_start, |
3931 | .seq_next = cgroup_pidlist_next, | |
3932 | .seq_stop = cgroup_pidlist_stop, | |
3933 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3934 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3935 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3936 | .mode = S_IRUGO | S_IWUSR, |
3937 | }, | |
3938 | { | |
3939 | .name = "notify_on_release", | |
3940 | .flags = CFTYPE_INSANE, | |
3941 | .read_u64 = cgroup_read_notify_on_release, | |
3942 | .write_u64 = cgroup_write_notify_on_release, | |
3943 | }, | |
6e6ff25b TH |
3944 | { |
3945 | .name = "release_agent", | |
cc5943a7 | 3946 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3947 | .seq_show = cgroup_release_agent_show, |
6e6ff25b | 3948 | .write_string = cgroup_release_agent_write, |
5f469907 | 3949 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 3950 | }, |
db0416b6 | 3951 | { } /* terminate */ |
bbcb81d0 PM |
3952 | }; |
3953 | ||
13af07df | 3954 | /** |
628f7cd4 | 3955 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3956 | * @cgrp: target cgroup |
13af07df | 3957 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3958 | * |
3959 | * On failure, no file is added. | |
13af07df | 3960 | */ |
628f7cd4 | 3961 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3962 | { |
ddbcc7e8 | 3963 | struct cgroup_subsys *ss; |
b420ba7d | 3964 | int i, ret = 0; |
bbcb81d0 | 3965 | |
8e3f6541 | 3966 | /* process cftsets of each subsystem */ |
b420ba7d | 3967 | for_each_subsys(ss, i) { |
0adb0704 | 3968 | struct cftype *cfts; |
b420ba7d TH |
3969 | |
3970 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3971 | continue; |
8e3f6541 | 3972 | |
0adb0704 TH |
3973 | list_for_each_entry(cfts, &ss->cfts, node) { |
3974 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
3975 | if (ret < 0) |
3976 | goto err; | |
3977 | } | |
ddbcc7e8 | 3978 | } |
ddbcc7e8 | 3979 | return 0; |
bee55099 TH |
3980 | err: |
3981 | cgroup_clear_dir(cgrp, subsys_mask); | |
3982 | return ret; | |
ddbcc7e8 PM |
3983 | } |
3984 | ||
0c21ead1 TH |
3985 | /* |
3986 | * css destruction is four-stage process. | |
3987 | * | |
3988 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3989 | * Implemented in kill_css(). | |
3990 | * | |
3991 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3992 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3993 | * by invoking offline_css(). After offlining, the base ref is put. | |
3994 | * Implemented in css_killed_work_fn(). | |
3995 | * | |
3996 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3997 | * accessors are inside RCU read sections. css_release() schedules the | |
3998 | * RCU callback. | |
3999 | * | |
4000 | * 4. After the grace period, the css can be freed. Implemented in | |
4001 | * css_free_work_fn(). | |
4002 | * | |
4003 | * It is actually hairier because both step 2 and 4 require process context | |
4004 | * and thus involve punting to css->destroy_work adding two additional | |
4005 | * steps to the already complex sequence. | |
4006 | */ | |
35ef10da | 4007 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4008 | { |
4009 | struct cgroup_subsys_state *css = | |
35ef10da | 4010 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 4011 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4012 | |
0ae78e0b TH |
4013 | if (css->parent) |
4014 | css_put(css->parent); | |
4015 | ||
0c21ead1 | 4016 | css->ss->css_free(css); |
2bd59d48 | 4017 | cgroup_put(cgrp); |
48ddbe19 TH |
4018 | } |
4019 | ||
0c21ead1 | 4020 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4021 | { |
4022 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4023 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4024 | |
35ef10da | 4025 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4026 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4027 | } |
4028 | ||
d3daf28d TH |
4029 | static void css_release(struct percpu_ref *ref) |
4030 | { | |
4031 | struct cgroup_subsys_state *css = | |
4032 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4033 | ||
01a97140 | 4034 | RCU_INIT_POINTER(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 4035 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4036 | } |
4037 | ||
623f926b TH |
4038 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
4039 | struct cgroup *cgrp) | |
ddbcc7e8 | 4040 | { |
bd89aabc | 4041 | css->cgroup = cgrp; |
72c97e54 | 4042 | css->ss = ss; |
ddbcc7e8 | 4043 | css->flags = 0; |
0ae78e0b TH |
4044 | |
4045 | if (cgrp->parent) | |
ca8bdcaf | 4046 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 4047 | else |
38b53aba | 4048 | css->flags |= CSS_ROOT; |
48ddbe19 | 4049 | |
ca8bdcaf | 4050 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4051 | } |
4052 | ||
2a4ac633 | 4053 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4054 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4055 | { |
623f926b | 4056 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4057 | int ret = 0; |
4058 | ||
ace2bee8 | 4059 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
4060 | lockdep_assert_held(&cgroup_mutex); |
4061 | ||
92fb9748 | 4062 | if (ss->css_online) |
eb95419b | 4063 | ret = ss->css_online(css); |
ae7f164a | 4064 | if (!ret) { |
eb95419b | 4065 | css->flags |= CSS_ONLINE; |
f20104de | 4066 | css->cgroup->nr_css++; |
aec25020 | 4067 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 4068 | } |
b1929db4 | 4069 | return ret; |
a31f2d3f TH |
4070 | } |
4071 | ||
2a4ac633 | 4072 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4073 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4074 | { |
623f926b | 4075 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 4076 | |
ace2bee8 | 4077 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
4078 | lockdep_assert_held(&cgroup_mutex); |
4079 | ||
4080 | if (!(css->flags & CSS_ONLINE)) | |
4081 | return; | |
4082 | ||
d7eeac19 | 4083 | if (ss->css_offline) |
eb95419b | 4084 | ss->css_offline(css); |
a31f2d3f | 4085 | |
eb95419b | 4086 | css->flags &= ~CSS_ONLINE; |
09a503ea | 4087 | css->cgroup->nr_css--; |
e3297803 | 4088 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4089 | |
4090 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4091 | } |
4092 | ||
c81c925a TH |
4093 | /** |
4094 | * create_css - create a cgroup_subsys_state | |
4095 | * @cgrp: the cgroup new css will be associated with | |
4096 | * @ss: the subsys of new css | |
4097 | * | |
4098 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
4099 | * css is online and installed in @cgrp with all interface files created. | |
4100 | * Returns 0 on success, -errno on failure. | |
4101 | */ | |
4102 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
4103 | { | |
4104 | struct cgroup *parent = cgrp->parent; | |
4105 | struct cgroup_subsys_state *css; | |
4106 | int err; | |
4107 | ||
c81c925a TH |
4108 | lockdep_assert_held(&cgroup_mutex); |
4109 | ||
4110 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
4111 | if (IS_ERR(css)) | |
4112 | return PTR_ERR(css); | |
4113 | ||
4114 | err = percpu_ref_init(&css->refcnt, css_release); | |
4115 | if (err) | |
3eb59ec6 | 4116 | goto err_free_css; |
c81c925a TH |
4117 | |
4118 | init_css(css, ss, cgrp); | |
4119 | ||
aec25020 | 4120 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a | 4121 | if (err) |
3eb59ec6 | 4122 | goto err_free_percpu_ref; |
c81c925a TH |
4123 | |
4124 | err = online_css(css); | |
4125 | if (err) | |
3eb59ec6 | 4126 | goto err_clear_dir; |
c81c925a | 4127 | |
59f5296b | 4128 | cgroup_get(cgrp); |
c81c925a TH |
4129 | css_get(css->parent); |
4130 | ||
4131 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
4132 | parent->parent) { | |
4133 | 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", | |
4134 | current->comm, current->pid, ss->name); | |
4135 | if (!strcmp(ss->name, "memory")) | |
4136 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
4137 | ss->warned_broken_hierarchy = true; | |
4138 | } | |
4139 | ||
4140 | return 0; | |
4141 | ||
3eb59ec6 | 4142 | err_clear_dir: |
32d01dc7 | 4143 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3eb59ec6 | 4144 | err_free_percpu_ref: |
c81c925a | 4145 | percpu_ref_cancel_init(&css->refcnt); |
3eb59ec6 | 4146 | err_free_css: |
c81c925a TH |
4147 | ss->css_free(css); |
4148 | return err; | |
4149 | } | |
4150 | ||
2bd59d48 | 4151 | /** |
a043e3b2 LZ |
4152 | * cgroup_create - create a cgroup |
4153 | * @parent: cgroup that will be parent of the new cgroup | |
e61734c5 | 4154 | * @name: name of the new cgroup |
2bd59d48 | 4155 | * @mode: mode to set on new cgroup |
ddbcc7e8 | 4156 | */ |
e61734c5 | 4157 | static long cgroup_create(struct cgroup *parent, const char *name, |
2bd59d48 | 4158 | umode_t mode) |
ddbcc7e8 | 4159 | { |
bd89aabc | 4160 | struct cgroup *cgrp; |
3dd06ffa | 4161 | struct cgroup_root *root = parent->root; |
b58c8998 | 4162 | int ssid, err; |
ddbcc7e8 | 4163 | struct cgroup_subsys *ss; |
2bd59d48 | 4164 | struct kernfs_node *kn; |
ddbcc7e8 | 4165 | |
0a950f65 | 4166 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
4167 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
4168 | if (!cgrp) | |
ddbcc7e8 PM |
4169 | return -ENOMEM; |
4170 | ||
ace2bee8 | 4171 | mutex_lock(&cgroup_tree_mutex); |
65dff759 | 4172 | |
976c06bc TH |
4173 | /* |
4174 | * Only live parents can have children. Note that the liveliness | |
4175 | * check isn't strictly necessary because cgroup_mkdir() and | |
4176 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
4177 | * anyway so that locking is contained inside cgroup proper and we | |
4178 | * don't get nasty surprises if we ever grow another caller. | |
4179 | */ | |
4180 | if (!cgroup_lock_live_group(parent)) { | |
4181 | err = -ENODEV; | |
ace2bee8 | 4182 | goto err_unlock_tree; |
0ab02ca8 LZ |
4183 | } |
4184 | ||
4185 | /* | |
4186 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4187 | * a half-baked cgroup. | |
4188 | */ | |
4189 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
4190 | if (cgrp->id < 0) { | |
4191 | err = -ENOMEM; | |
4192 | goto err_unlock; | |
976c06bc TH |
4193 | } |
4194 | ||
cc31edce | 4195 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4196 | |
bd89aabc | 4197 | cgrp->parent = parent; |
0ae78e0b | 4198 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 4199 | cgrp->root = parent->root; |
ddbcc7e8 | 4200 | |
b6abdb0e LZ |
4201 | if (notify_on_release(parent)) |
4202 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4203 | ||
2260e7fc TH |
4204 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4205 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4206 | |
2bd59d48 | 4207 | /* create the directory */ |
e61734c5 | 4208 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 TH |
4209 | if (IS_ERR(kn)) { |
4210 | err = PTR_ERR(kn); | |
0ab02ca8 | 4211 | goto err_free_id; |
2bd59d48 TH |
4212 | } |
4213 | cgrp->kn = kn; | |
ddbcc7e8 | 4214 | |
4e139afc | 4215 | /* |
6f30558f TH |
4216 | * This extra ref will be put in cgroup_free_fn() and guarantees |
4217 | * that @cgrp->kn is always accessible. | |
4e139afc | 4218 | */ |
6f30558f | 4219 | kernfs_get(kn); |
ddbcc7e8 | 4220 | |
00356bd5 | 4221 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 4222 | |
4e139afc | 4223 | /* allocation complete, commit to creation */ |
4e139afc | 4224 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3c9c825b | 4225 | atomic_inc(&root->nr_cgrps); |
59f5296b | 4226 | cgroup_get(parent); |
415cf07a | 4227 | |
0d80255e TH |
4228 | /* |
4229 | * @cgrp is now fully operational. If something fails after this | |
4230 | * point, it'll be released via the normal destruction path. | |
4231 | */ | |
4e96ee8e LZ |
4232 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4233 | ||
49957f8e TH |
4234 | err = cgroup_kn_set_ugid(kn); |
4235 | if (err) | |
4236 | goto err_destroy; | |
4237 | ||
2bb566cb | 4238 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
4239 | if (err) |
4240 | goto err_destroy; | |
4241 | ||
9d403e99 | 4242 | /* let's create and online css's */ |
b85d2040 | 4243 | for_each_subsys(ss, ssid) { |
f392e51c | 4244 | if (parent->child_subsys_mask & (1 << ssid)) { |
b85d2040 TH |
4245 | err = create_css(cgrp, ss); |
4246 | if (err) | |
4247 | goto err_destroy; | |
4248 | } | |
a8638030 | 4249 | } |
ddbcc7e8 | 4250 | |
bd53d617 TH |
4251 | /* |
4252 | * On the default hierarchy, a child doesn't automatically inherit | |
4253 | * child_subsys_mask from the parent. Each is configured manually. | |
4254 | */ | |
4255 | if (!cgroup_on_dfl(cgrp)) | |
4256 | cgrp->child_subsys_mask = parent->child_subsys_mask; | |
f392e51c | 4257 | |
2bd59d48 TH |
4258 | kernfs_activate(kn); |
4259 | ||
ddbcc7e8 | 4260 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4261 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
4262 | |
4263 | return 0; | |
4264 | ||
0a950f65 | 4265 | err_free_id: |
4e96ee8e | 4266 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
4267 | err_unlock: |
4268 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
4269 | err_unlock_tree: |
4270 | mutex_unlock(&cgroup_tree_mutex); | |
bd89aabc | 4271 | kfree(cgrp); |
ddbcc7e8 | 4272 | return err; |
4b8b47eb TH |
4273 | |
4274 | err_destroy: | |
4275 | cgroup_destroy_locked(cgrp); | |
4276 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4277 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 4278 | return err; |
ddbcc7e8 PM |
4279 | } |
4280 | ||
2bd59d48 TH |
4281 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
4282 | umode_t mode) | |
ddbcc7e8 | 4283 | { |
2bd59d48 | 4284 | struct cgroup *parent = parent_kn->priv; |
e1b2dc17 | 4285 | int ret; |
ddbcc7e8 | 4286 | |
e1b2dc17 TH |
4287 | /* |
4288 | * cgroup_create() grabs cgroup_tree_mutex which nests outside | |
4289 | * kernfs active_ref and cgroup_create() already synchronizes | |
4290 | * properly against removal through cgroup_lock_live_group(). | |
4291 | * Break it before calling cgroup_create(). | |
4292 | */ | |
4293 | cgroup_get(parent); | |
4294 | kernfs_break_active_protection(parent_kn); | |
ddbcc7e8 | 4295 | |
e1b2dc17 TH |
4296 | ret = cgroup_create(parent, name, mode); |
4297 | ||
4298 | kernfs_unbreak_active_protection(parent_kn); | |
4299 | cgroup_put(parent); | |
4300 | return ret; | |
ddbcc7e8 PM |
4301 | } |
4302 | ||
223dbc38 TH |
4303 | /* |
4304 | * This is called when the refcnt of a css is confirmed to be killed. | |
4305 | * css_tryget() is now guaranteed to fail. | |
4306 | */ | |
4307 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4308 | { |
223dbc38 TH |
4309 | struct cgroup_subsys_state *css = |
4310 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
4311 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 4312 | |
ace2bee8 | 4313 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
4314 | mutex_lock(&cgroup_mutex); |
4315 | ||
09a503ea TH |
4316 | /* |
4317 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
4318 | * initate destruction. | |
4319 | */ | |
4320 | offline_css(css); | |
4321 | ||
f20104de TH |
4322 | /* |
4323 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
4324 | * be disabled before proceeding to the second phase of cgroup | |
4325 | * destruction. If we are the last one, kick it off. | |
4326 | */ | |
09a503ea | 4327 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
4328 | cgroup_destroy_css_killed(cgrp); |
4329 | ||
4330 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4331 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
4332 | |
4333 | /* | |
4334 | * Put the css refs from kill_css(). Each css holds an extra | |
4335 | * reference to the cgroup's dentry and cgroup removal proceeds | |
4336 | * regardless of css refs. On the last put of each css, whenever | |
4337 | * that may be, the extra dentry ref is put so that dentry | |
4338 | * destruction happens only after all css's are released. | |
4339 | */ | |
4340 | css_put(css); | |
d3daf28d TH |
4341 | } |
4342 | ||
223dbc38 TH |
4343 | /* css kill confirmation processing requires process context, bounce */ |
4344 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4345 | { |
4346 | struct cgroup_subsys_state *css = | |
4347 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4348 | ||
223dbc38 | 4349 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 4350 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
4351 | } |
4352 | ||
f392e51c TH |
4353 | /** |
4354 | * kill_css - destroy a css | |
4355 | * @css: css to destroy | |
4356 | * | |
4357 | * This function initiates destruction of @css by removing cgroup interface | |
4358 | * files and putting its base reference. ->css_offline() will be invoked | |
4359 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
4360 | * reference count reaches zero, @css will be released. | |
4361 | */ | |
4362 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 4363 | { |
94419627 TH |
4364 | lockdep_assert_held(&cgroup_tree_mutex); |
4365 | ||
2bd59d48 TH |
4366 | /* |
4367 | * This must happen before css is disassociated with its cgroup. | |
4368 | * See seq_css() for details. | |
4369 | */ | |
aec25020 | 4370 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 4371 | |
edae0c33 TH |
4372 | /* |
4373 | * Killing would put the base ref, but we need to keep it alive | |
4374 | * until after ->css_offline(). | |
4375 | */ | |
4376 | css_get(css); | |
4377 | ||
4378 | /* | |
4379 | * cgroup core guarantees that, by the time ->css_offline() is | |
4380 | * invoked, no new css reference will be given out via | |
4381 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
4382 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
4383 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4384 | * | |
4385 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4386 | * css is confirmed to be seen as killed on all CPUs. | |
4387 | */ | |
4388 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4389 | } |
4390 | ||
4391 | /** | |
4392 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4393 | * @cgrp: cgroup to be destroyed | |
4394 | * | |
4395 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4396 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
4397 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
4398 | * invoked. To satisfy all the requirements, destruction is implemented in | |
4399 | * the following two steps. | |
4400 | * | |
4401 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4402 | * userland visible parts and start killing the percpu refcnts of | |
4403 | * css's. Set up so that the next stage will be kicked off once all | |
4404 | * the percpu refcnts are confirmed to be killed. | |
4405 | * | |
4406 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4407 | * rest of destruction. Once all cgroup references are gone, the | |
4408 | * cgroup is RCU-freed. | |
4409 | * | |
4410 | * This function implements s1. After this step, @cgrp is gone as far as | |
4411 | * the userland is concerned and a new cgroup with the same name may be | |
4412 | * created. As cgroup doesn't care about the names internally, this | |
4413 | * doesn't cause any problem. | |
4414 | */ | |
42809dd4 TH |
4415 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4416 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4417 | { |
bb78a92f | 4418 | struct cgroup *child; |
2bd59d48 | 4419 | struct cgroup_subsys_state *css; |
ddd69148 | 4420 | bool empty; |
1c6727af | 4421 | int ssid; |
ddbcc7e8 | 4422 | |
ace2bee8 | 4423 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
4424 | lockdep_assert_held(&cgroup_mutex); |
4425 | ||
ddd69148 | 4426 | /* |
96d365e0 | 4427 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 4428 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 4429 | */ |
96d365e0 | 4430 | down_read(&css_set_rwsem); |
bb78a92f | 4431 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 4432 | up_read(&css_set_rwsem); |
ddd69148 | 4433 | if (!empty) |
ddbcc7e8 | 4434 | return -EBUSY; |
a043e3b2 | 4435 | |
bb78a92f HD |
4436 | /* |
4437 | * Make sure there's no live children. We can't test ->children | |
4438 | * emptiness as dead children linger on it while being destroyed; | |
4439 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
4440 | */ | |
4441 | empty = true; | |
4442 | rcu_read_lock(); | |
4443 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
4444 | empty = cgroup_is_dead(child); | |
4445 | if (!empty) | |
4446 | break; | |
4447 | } | |
4448 | rcu_read_unlock(); | |
4449 | if (!empty) | |
4450 | return -EBUSY; | |
4451 | ||
455050d2 TH |
4452 | /* |
4453 | * Mark @cgrp dead. This prevents further task migration and child | |
4454 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 4455 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 4456 | * resume iteration after dropping RCU read lock. See |
492eb21b | 4457 | * css_next_child() for details. |
455050d2 | 4458 | */ |
54766d4a | 4459 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 4460 | |
88703267 | 4461 | /* |
edae0c33 TH |
4462 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
4463 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
4464 | * percpu refs of all css's are confirmed to be killed. This |
4465 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 4466 | */ |
4ac06017 | 4467 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
4468 | for_each_css(css, ssid, cgrp) |
4469 | kill_css(css); | |
4ac06017 | 4470 | mutex_lock(&cgroup_mutex); |
455050d2 | 4471 | |
455050d2 TH |
4472 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
4473 | raw_spin_lock(&release_list_lock); | |
4474 | if (!list_empty(&cgrp->release_list)) | |
4475 | list_del_init(&cgrp->release_list); | |
4476 | raw_spin_unlock(&release_list_lock); | |
4477 | ||
4478 | /* | |
f20104de TH |
4479 | * If @cgrp has css's attached, the second stage of cgroup |
4480 | * destruction is kicked off from css_killed_work_fn() after the | |
4481 | * refs of all attached css's are killed. If @cgrp doesn't have | |
4482 | * any css, we kick it off here. | |
4483 | */ | |
4484 | if (!cgrp->nr_css) | |
4485 | cgroup_destroy_css_killed(cgrp); | |
4486 | ||
2bd59d48 TH |
4487 | /* remove @cgrp directory along with the base files */ |
4488 | mutex_unlock(&cgroup_mutex); | |
4489 | ||
455050d2 | 4490 | /* |
2bd59d48 TH |
4491 | * There are two control paths which try to determine cgroup from |
4492 | * dentry without going through kernfs - cgroupstats_build() and | |
4493 | * css_tryget_from_dir(). Those are supported by RCU protecting | |
4494 | * clearing of cgrp->kn->priv backpointer, which should happen | |
4495 | * after all files under it have been removed. | |
455050d2 | 4496 | */ |
6f30558f | 4497 | kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ |
2bd59d48 | 4498 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); |
2bd59d48 | 4499 | |
4ac06017 | 4500 | mutex_lock(&cgroup_mutex); |
455050d2 | 4501 | |
ea15f8cc TH |
4502 | return 0; |
4503 | }; | |
4504 | ||
d3daf28d | 4505 | /** |
f20104de | 4506 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
4507 | * @work: cgroup->destroy_free_work |
4508 | * | |
4509 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
4510 | * destroyed after all css's are offlined and performs the rest of |
4511 | * destruction. This is the second step of destruction described in the | |
4512 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 4513 | */ |
f20104de | 4514 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 4515 | { |
ea15f8cc | 4516 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 4517 | |
ace2bee8 | 4518 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 4519 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 4520 | |
999cd8a4 | 4521 | /* delete this cgroup from parent->children */ |
eb6fd504 | 4522 | list_del_rcu(&cgrp->sibling); |
ed957793 | 4523 | |
59f5296b | 4524 | cgroup_put(cgrp); |
ddbcc7e8 | 4525 | |
bd89aabc | 4526 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 4527 | check_for_release(parent); |
ddbcc7e8 PM |
4528 | } |
4529 | ||
2bd59d48 | 4530 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4531 | { |
2bd59d48 TH |
4532 | struct cgroup *cgrp = kn->priv; |
4533 | int ret = 0; | |
42809dd4 | 4534 | |
2bd59d48 TH |
4535 | /* |
4536 | * This is self-destruction but @kn can't be removed while this | |
4537 | * callback is in progress. Let's break active protection. Once | |
4538 | * the protection is broken, @cgrp can be destroyed at any point. | |
4539 | * Pin it so that it stays accessible. | |
4540 | */ | |
4541 | cgroup_get(cgrp); | |
4542 | kernfs_break_active_protection(kn); | |
42809dd4 | 4543 | |
ace2bee8 | 4544 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 4545 | mutex_lock(&cgroup_mutex); |
8e3f6541 TH |
4546 | |
4547 | /* | |
2bd59d48 TH |
4548 | * @cgrp might already have been destroyed while we're trying to |
4549 | * grab the mutexes. | |
8e3f6541 | 4550 | */ |
2bd59d48 TH |
4551 | if (!cgroup_is_dead(cgrp)) |
4552 | ret = cgroup_destroy_locked(cgrp); | |
2bb566cb | 4553 | |
42809dd4 | 4554 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4555 | mutex_unlock(&cgroup_tree_mutex); |
2bb566cb | 4556 | |
2bd59d48 TH |
4557 | kernfs_unbreak_active_protection(kn); |
4558 | cgroup_put(cgrp); | |
42809dd4 | 4559 | return ret; |
8e3f6541 TH |
4560 | } |
4561 | ||
2bd59d48 TH |
4562 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4563 | .remount_fs = cgroup_remount, | |
4564 | .show_options = cgroup_show_options, | |
4565 | .mkdir = cgroup_mkdir, | |
4566 | .rmdir = cgroup_rmdir, | |
4567 | .rename = cgroup_rename, | |
4568 | }; | |
4569 | ||
06a11920 | 4570 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 4571 | { |
ddbcc7e8 | 4572 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4573 | |
4574 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4575 | |
ace2bee8 | 4576 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
4577 | mutex_lock(&cgroup_mutex); |
4578 | ||
0adb0704 | 4579 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4580 | |
3dd06ffa TH |
4581 | /* Create the root cgroup state for this subsystem */ |
4582 | ss->root = &cgrp_dfl_root; | |
4583 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
4584 | /* We don't handle early failures gracefully */ |
4585 | BUG_ON(IS_ERR(css)); | |
3dd06ffa | 4586 | init_css(css, ss, &cgrp_dfl_root.cgrp); |
ddbcc7e8 | 4587 | |
e8d55fde | 4588 | /* Update the init_css_set to contain a subsys |
817929ec | 4589 | * pointer to this state - since the subsystem is |
e8d55fde | 4590 | * newly registered, all tasks and hence the |
3dd06ffa | 4591 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 4592 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4593 | |
4594 | need_forkexit_callback |= ss->fork || ss->exit; | |
4595 | ||
e8d55fde LZ |
4596 | /* At system boot, before all subsystems have been |
4597 | * registered, no tasks have been forked, so we don't | |
4598 | * need to invoke fork callbacks here. */ | |
4599 | BUG_ON(!list_empty(&init_task.tasks)); | |
4600 | ||
ae7f164a | 4601 | BUG_ON(online_css(css)); |
a8638030 | 4602 | |
f392e51c | 4603 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
cf5d5941 BB |
4604 | |
4605 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4606 | mutex_unlock(&cgroup_tree_mutex); |
cf5d5941 | 4607 | } |
cf5d5941 | 4608 | |
ddbcc7e8 | 4609 | /** |
a043e3b2 LZ |
4610 | * cgroup_init_early - cgroup initialization at system boot |
4611 | * | |
4612 | * Initialize cgroups at system boot, and initialize any | |
4613 | * subsystems that request early init. | |
ddbcc7e8 PM |
4614 | */ |
4615 | int __init cgroup_init_early(void) | |
4616 | { | |
a2dd4247 TH |
4617 | static struct cgroup_sb_opts __initdata opts = |
4618 | { .flags = CGRP_ROOT_SANE_BEHAVIOR }; | |
30159ec7 | 4619 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4620 | int i; |
30159ec7 | 4621 | |
3dd06ffa | 4622 | init_cgroup_root(&cgrp_dfl_root, &opts); |
a4ea1cc9 | 4623 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4624 | |
3ed80a62 | 4625 | for_each_subsys(ss, i) { |
aec25020 | 4626 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4627 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4628 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4629 | ss->id, ss->name); |
073219e9 TH |
4630 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4631 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4632 | ||
aec25020 | 4633 | ss->id = i; |
073219e9 | 4634 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4635 | |
4636 | if (ss->early_init) | |
4637 | cgroup_init_subsys(ss); | |
4638 | } | |
4639 | return 0; | |
4640 | } | |
4641 | ||
4642 | /** | |
a043e3b2 LZ |
4643 | * cgroup_init - cgroup initialization |
4644 | * | |
4645 | * Register cgroup filesystem and /proc file, and initialize | |
4646 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4647 | */ |
4648 | int __init cgroup_init(void) | |
4649 | { | |
30159ec7 | 4650 | struct cgroup_subsys *ss; |
0ac801fe | 4651 | unsigned long key; |
172a2c06 | 4652 | int ssid, err; |
ddbcc7e8 | 4653 | |
2bd59d48 | 4654 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
ddbcc7e8 | 4655 | |
985ed670 | 4656 | mutex_lock(&cgroup_tree_mutex); |
54e7b4eb | 4657 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4658 | |
82fe9b0d TH |
4659 | /* Add init_css_set to the hash table */ |
4660 | key = css_set_hash(init_css_set.subsys); | |
4661 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4662 | ||
3dd06ffa | 4663 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 4664 | |
54e7b4eb | 4665 | mutex_unlock(&cgroup_mutex); |
985ed670 | 4666 | mutex_unlock(&cgroup_tree_mutex); |
54e7b4eb | 4667 | |
172a2c06 TH |
4668 | for_each_subsys(ss, ssid) { |
4669 | if (!ss->early_init) | |
4670 | cgroup_init_subsys(ss); | |
4671 | ||
2d8f243a TH |
4672 | list_add_tail(&init_css_set.e_cset_node[ssid], |
4673 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
4674 | ||
172a2c06 TH |
4675 | /* |
4676 | * cftype registration needs kmalloc and can't be done | |
4677 | * during early_init. Register base cftypes separately. | |
4678 | */ | |
4679 | if (ss->base_cftypes) | |
4680 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
676db4af GK |
4681 | } |
4682 | ||
676db4af | 4683 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4684 | if (!cgroup_kobj) |
4685 | return -ENOMEM; | |
676db4af | 4686 | |
ddbcc7e8 | 4687 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
4688 | if (err < 0) { |
4689 | kobject_put(cgroup_kobj); | |
2bd59d48 | 4690 | return err; |
676db4af | 4691 | } |
ddbcc7e8 | 4692 | |
46ae220b | 4693 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 4694 | return 0; |
ddbcc7e8 | 4695 | } |
b4f48b63 | 4696 | |
e5fca243 TH |
4697 | static int __init cgroup_wq_init(void) |
4698 | { | |
4699 | /* | |
4700 | * There isn't much point in executing destruction path in | |
4701 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 4702 | * Use 1 for @max_active. |
e5fca243 TH |
4703 | * |
4704 | * We would prefer to do this in cgroup_init() above, but that | |
4705 | * is called before init_workqueues(): so leave this until after. | |
4706 | */ | |
1a11533f | 4707 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 4708 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4709 | |
4710 | /* | |
4711 | * Used to destroy pidlists and separate to serve as flush domain. | |
4712 | * Cap @max_active to 1 too. | |
4713 | */ | |
4714 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4715 | 0, 1); | |
4716 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4717 | ||
e5fca243 TH |
4718 | return 0; |
4719 | } | |
4720 | core_initcall(cgroup_wq_init); | |
4721 | ||
a424316c PM |
4722 | /* |
4723 | * proc_cgroup_show() | |
4724 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4725 | * - Used for /proc/<pid>/cgroup. | |
a424316c PM |
4726 | */ |
4727 | ||
4728 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4729 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4730 | { |
4731 | struct pid *pid; | |
4732 | struct task_struct *tsk; | |
e61734c5 | 4733 | char *buf, *path; |
a424316c | 4734 | int retval; |
3dd06ffa | 4735 | struct cgroup_root *root; |
a424316c PM |
4736 | |
4737 | retval = -ENOMEM; | |
e61734c5 | 4738 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4739 | if (!buf) |
4740 | goto out; | |
4741 | ||
4742 | retval = -ESRCH; | |
4743 | pid = m->private; | |
4744 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4745 | if (!tsk) | |
4746 | goto out_free; | |
4747 | ||
4748 | retval = 0; | |
4749 | ||
4750 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 4751 | down_read(&css_set_rwsem); |
a424316c | 4752 | |
985ed670 | 4753 | for_each_root(root) { |
a424316c | 4754 | struct cgroup_subsys *ss; |
bd89aabc | 4755 | struct cgroup *cgrp; |
b85d2040 | 4756 | int ssid, count = 0; |
a424316c | 4757 | |
a2dd4247 | 4758 | if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) |
985ed670 TH |
4759 | continue; |
4760 | ||
2c6ab6d2 | 4761 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 | 4762 | for_each_subsys(ss, ssid) |
f392e51c | 4763 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 4764 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); |
c6d57f33 PM |
4765 | if (strlen(root->name)) |
4766 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4767 | root->name); | |
a424316c | 4768 | seq_putc(m, ':'); |
7717f7ba | 4769 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
4770 | path = cgroup_path(cgrp, buf, PATH_MAX); |
4771 | if (!path) { | |
4772 | retval = -ENAMETOOLONG; | |
a424316c | 4773 | goto out_unlock; |
e61734c5 TH |
4774 | } |
4775 | seq_puts(m, path); | |
a424316c PM |
4776 | seq_putc(m, '\n'); |
4777 | } | |
4778 | ||
4779 | out_unlock: | |
96d365e0 | 4780 | up_read(&css_set_rwsem); |
a424316c PM |
4781 | mutex_unlock(&cgroup_mutex); |
4782 | put_task_struct(tsk); | |
4783 | out_free: | |
4784 | kfree(buf); | |
4785 | out: | |
4786 | return retval; | |
4787 | } | |
4788 | ||
a424316c PM |
4789 | /* Display information about each subsystem and each hierarchy */ |
4790 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4791 | { | |
30159ec7 | 4792 | struct cgroup_subsys *ss; |
a424316c | 4793 | int i; |
a424316c | 4794 | |
8bab8dde | 4795 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4796 | /* |
4797 | * ideally we don't want subsystems moving around while we do this. | |
4798 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4799 | * subsys/hierarchy state. | |
4800 | */ | |
a424316c | 4801 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4802 | |
4803 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4804 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4805 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 4806 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 4807 | |
a424316c PM |
4808 | mutex_unlock(&cgroup_mutex); |
4809 | return 0; | |
4810 | } | |
4811 | ||
4812 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4813 | { | |
9dce07f1 | 4814 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4815 | } |
4816 | ||
828c0950 | 4817 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4818 | .open = cgroupstats_open, |
4819 | .read = seq_read, | |
4820 | .llseek = seq_lseek, | |
4821 | .release = single_release, | |
4822 | }; | |
4823 | ||
b4f48b63 | 4824 | /** |
eaf797ab | 4825 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 4826 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 4827 | * |
eaf797ab TH |
4828 | * A task is associated with the init_css_set until cgroup_post_fork() |
4829 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
4830 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
4831 | */ |
4832 | void cgroup_fork(struct task_struct *child) | |
4833 | { | |
eaf797ab | 4834 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 4835 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4836 | } |
4837 | ||
817929ec | 4838 | /** |
a043e3b2 LZ |
4839 | * cgroup_post_fork - called on a new task after adding it to the task list |
4840 | * @child: the task in question | |
4841 | * | |
5edee61e TH |
4842 | * Adds the task to the list running through its css_set if necessary and |
4843 | * call the subsystem fork() callbacks. Has to be after the task is | |
4844 | * visible on the task list in case we race with the first call to | |
0942eeee | 4845 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4846 | * list. |
a043e3b2 | 4847 | */ |
817929ec PM |
4848 | void cgroup_post_fork(struct task_struct *child) |
4849 | { | |
30159ec7 | 4850 | struct cgroup_subsys *ss; |
5edee61e TH |
4851 | int i; |
4852 | ||
3ce3230a | 4853 | /* |
eaf797ab TH |
4854 | * This may race against cgroup_enable_task_cg_links(). As that |
4855 | * function sets use_task_css_set_links before grabbing | |
4856 | * tasklist_lock and we just went through tasklist_lock to add | |
4857 | * @child, it's guaranteed that either we see the set | |
4858 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
4859 | * @child during its iteration. | |
4860 | * | |
4861 | * If we won the race, @child is associated with %current's | |
4862 | * css_set. Grabbing css_set_rwsem guarantees both that the | |
4863 | * association is stable, and, on completion of the parent's | |
4864 | * migration, @child is visible in the source of migration or | |
4865 | * already in the destination cgroup. This guarantee is necessary | |
4866 | * when implementing operations which need to migrate all tasks of | |
4867 | * a cgroup to another. | |
4868 | * | |
4869 | * Note that if we lose to cgroup_enable_task_cg_links(), @child | |
4870 | * will remain in init_css_set. This is safe because all tasks are | |
4871 | * in the init_css_set before cg_links is enabled and there's no | |
4872 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 4873 | */ |
817929ec | 4874 | if (use_task_css_set_links) { |
eaf797ab TH |
4875 | struct css_set *cset; |
4876 | ||
96d365e0 | 4877 | down_write(&css_set_rwsem); |
0e1d768f | 4878 | cset = task_css_set(current); |
eaf797ab TH |
4879 | if (list_empty(&child->cg_list)) { |
4880 | rcu_assign_pointer(child->cgroups, cset); | |
4881 | list_add(&child->cg_list, &cset->tasks); | |
4882 | get_css_set(cset); | |
4883 | } | |
96d365e0 | 4884 | up_write(&css_set_rwsem); |
817929ec | 4885 | } |
5edee61e TH |
4886 | |
4887 | /* | |
4888 | * Call ss->fork(). This must happen after @child is linked on | |
4889 | * css_set; otherwise, @child might change state between ->fork() | |
4890 | * and addition to css_set. | |
4891 | */ | |
4892 | if (need_forkexit_callback) { | |
3ed80a62 | 4893 | for_each_subsys(ss, i) |
5edee61e TH |
4894 | if (ss->fork) |
4895 | ss->fork(child); | |
5edee61e | 4896 | } |
817929ec | 4897 | } |
5edee61e | 4898 | |
b4f48b63 PM |
4899 | /** |
4900 | * cgroup_exit - detach cgroup from exiting task | |
4901 | * @tsk: pointer to task_struct of exiting process | |
4902 | * | |
4903 | * Description: Detach cgroup from @tsk and release it. | |
4904 | * | |
4905 | * Note that cgroups marked notify_on_release force every task in | |
4906 | * them to take the global cgroup_mutex mutex when exiting. | |
4907 | * This could impact scaling on very large systems. Be reluctant to | |
4908 | * use notify_on_release cgroups where very high task exit scaling | |
4909 | * is required on large systems. | |
4910 | * | |
0e1d768f TH |
4911 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
4912 | * call cgroup_exit() while the task is still competent to handle | |
4913 | * notify_on_release(), then leave the task attached to the root cgroup in | |
4914 | * each hierarchy for the remainder of its exit. No need to bother with | |
4915 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 4916 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 4917 | */ |
1ec41830 | 4918 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 4919 | { |
30159ec7 | 4920 | struct cgroup_subsys *ss; |
5abb8855 | 4921 | struct css_set *cset; |
eaf797ab | 4922 | bool put_cset = false; |
d41d5a01 | 4923 | int i; |
817929ec PM |
4924 | |
4925 | /* | |
0e1d768f TH |
4926 | * Unlink from @tsk from its css_set. As migration path can't race |
4927 | * with us, we can check cg_list without grabbing css_set_rwsem. | |
817929ec PM |
4928 | */ |
4929 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 4930 | down_write(&css_set_rwsem); |
0e1d768f | 4931 | list_del_init(&tsk->cg_list); |
96d365e0 | 4932 | up_write(&css_set_rwsem); |
0e1d768f | 4933 | put_cset = true; |
817929ec PM |
4934 | } |
4935 | ||
b4f48b63 | 4936 | /* Reassign the task to the init_css_set. */ |
a8ad805c TH |
4937 | cset = task_css_set(tsk); |
4938 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 | 4939 | |
1ec41830 | 4940 | if (need_forkexit_callback) { |
3ed80a62 TH |
4941 | /* see cgroup_post_fork() for details */ |
4942 | for_each_subsys(ss, i) { | |
d41d5a01 | 4943 | if (ss->exit) { |
eb95419b TH |
4944 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4945 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4946 | |
eb95419b | 4947 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4948 | } |
4949 | } | |
4950 | } | |
d41d5a01 | 4951 | |
eaf797ab TH |
4952 | if (put_cset) |
4953 | put_css_set(cset, true); | |
b4f48b63 | 4954 | } |
697f4161 | 4955 | |
bd89aabc | 4956 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4957 | { |
f50daa70 | 4958 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4959 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4960 | /* |
4961 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4962 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4963 | * it now |
4964 | */ | |
81a6a5cd | 4965 | int need_schedule_work = 0; |
f50daa70 | 4966 | |
cdcc136f | 4967 | raw_spin_lock(&release_list_lock); |
54766d4a | 4968 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4969 | list_empty(&cgrp->release_list)) { |
4970 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4971 | need_schedule_work = 1; |
4972 | } | |
cdcc136f | 4973 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4974 | if (need_schedule_work) |
4975 | schedule_work(&release_agent_work); | |
4976 | } | |
4977 | } | |
4978 | ||
81a6a5cd PM |
4979 | /* |
4980 | * Notify userspace when a cgroup is released, by running the | |
4981 | * configured release agent with the name of the cgroup (path | |
4982 | * relative to the root of cgroup file system) as the argument. | |
4983 | * | |
4984 | * Most likely, this user command will try to rmdir this cgroup. | |
4985 | * | |
4986 | * This races with the possibility that some other task will be | |
4987 | * attached to this cgroup before it is removed, or that some other | |
4988 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4989 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4990 | * unused, and this cgroup will be reprieved from its death sentence, | |
4991 | * to continue to serve a useful existence. Next time it's released, | |
4992 | * we will get notified again, if it still has 'notify_on_release' set. | |
4993 | * | |
4994 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4995 | * means only wait until the task is successfully execve()'d. The | |
4996 | * separate release agent task is forked by call_usermodehelper(), | |
4997 | * then control in this thread returns here, without waiting for the | |
4998 | * release agent task. We don't bother to wait because the caller of | |
4999 | * this routine has no use for the exit status of the release agent | |
5000 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5001 | */ |
81a6a5cd PM |
5002 | static void cgroup_release_agent(struct work_struct *work) |
5003 | { | |
5004 | BUG_ON(work != &release_agent_work); | |
5005 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 5006 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
5007 | while (!list_empty(&release_list)) { |
5008 | char *argv[3], *envp[3]; | |
5009 | int i; | |
e61734c5 | 5010 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 5011 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
5012 | struct cgroup, |
5013 | release_list); | |
bd89aabc | 5014 | list_del_init(&cgrp->release_list); |
cdcc136f | 5015 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 5016 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
5017 | if (!pathbuf) |
5018 | goto continue_free; | |
e61734c5 TH |
5019 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
5020 | if (!path) | |
e788e066 PM |
5021 | goto continue_free; |
5022 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5023 | if (!agentbuf) | |
5024 | goto continue_free; | |
81a6a5cd PM |
5025 | |
5026 | i = 0; | |
e788e066 | 5027 | argv[i++] = agentbuf; |
e61734c5 | 5028 | argv[i++] = path; |
81a6a5cd PM |
5029 | argv[i] = NULL; |
5030 | ||
5031 | i = 0; | |
5032 | /* minimal command environment */ | |
5033 | envp[i++] = "HOME=/"; | |
5034 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5035 | envp[i] = NULL; | |
5036 | ||
5037 | /* Drop the lock while we invoke the usermode helper, | |
5038 | * since the exec could involve hitting disk and hence | |
5039 | * be a slow process */ | |
5040 | mutex_unlock(&cgroup_mutex); | |
5041 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 5042 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
5043 | continue_free: |
5044 | kfree(pathbuf); | |
5045 | kfree(agentbuf); | |
cdcc136f | 5046 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 5047 | } |
cdcc136f | 5048 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5049 | mutex_unlock(&cgroup_mutex); |
5050 | } | |
8bab8dde PM |
5051 | |
5052 | static int __init cgroup_disable(char *str) | |
5053 | { | |
30159ec7 | 5054 | struct cgroup_subsys *ss; |
8bab8dde | 5055 | char *token; |
30159ec7 | 5056 | int i; |
8bab8dde PM |
5057 | |
5058 | while ((token = strsep(&str, ",")) != NULL) { | |
5059 | if (!*token) | |
5060 | continue; | |
be45c900 | 5061 | |
3ed80a62 | 5062 | for_each_subsys(ss, i) { |
8bab8dde PM |
5063 | if (!strcmp(token, ss->name)) { |
5064 | ss->disabled = 1; | |
5065 | printk(KERN_INFO "Disabling %s control group" | |
5066 | " subsystem\n", ss->name); | |
5067 | break; | |
5068 | } | |
5069 | } | |
5070 | } | |
5071 | return 1; | |
5072 | } | |
5073 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5074 | |
b77d7b60 | 5075 | /** |
5a17f543 | 5076 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
5077 | * @dentry: directory dentry of interest |
5078 | * @ss: subsystem of interest | |
b77d7b60 | 5079 | * |
5a17f543 TH |
5080 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5081 | * to get the corresponding css and return it. If such css doesn't exist | |
5082 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5083 | */ |
5a17f543 TH |
5084 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
5085 | struct cgroup_subsys *ss) | |
e5d1367f | 5086 | { |
2bd59d48 TH |
5087 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
5088 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 5089 | struct cgroup *cgrp; |
e5d1367f | 5090 | |
35cf0836 | 5091 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
5092 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
5093 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5094 | return ERR_PTR(-EBADF); |
5095 | ||
5a17f543 TH |
5096 | rcu_read_lock(); |
5097 | ||
2bd59d48 TH |
5098 | /* |
5099 | * This path doesn't originate from kernfs and @kn could already | |
5100 | * have been or be removed at any point. @kn->priv is RCU | |
5101 | * protected for this access. See destroy_locked() for details. | |
5102 | */ | |
5103 | cgrp = rcu_dereference(kn->priv); | |
5104 | if (cgrp) | |
5105 | css = cgroup_css(cgrp, ss); | |
5a17f543 TH |
5106 | |
5107 | if (!css || !css_tryget(css)) | |
5108 | css = ERR_PTR(-ENOENT); | |
5109 | ||
5110 | rcu_read_unlock(); | |
5111 | return css; | |
e5d1367f | 5112 | } |
e5d1367f | 5113 | |
1cb650b9 LZ |
5114 | /** |
5115 | * css_from_id - lookup css by id | |
5116 | * @id: the cgroup id | |
5117 | * @ss: cgroup subsys to be looked into | |
5118 | * | |
5119 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5120 | * Should be called under rcu_read_lock(). | |
5121 | */ | |
5122 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5123 | { | |
5124 | struct cgroup *cgrp; | |
5125 | ||
ace2bee8 | 5126 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
5127 | |
5128 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
5129 | if (cgrp) | |
d1625964 | 5130 | return cgroup_css(cgrp, ss); |
1cb650b9 | 5131 | return NULL; |
e5d1367f SE |
5132 | } |
5133 | ||
fe693435 | 5134 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5135 | static struct cgroup_subsys_state * |
5136 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5137 | { |
5138 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5139 | ||
5140 | if (!css) | |
5141 | return ERR_PTR(-ENOMEM); | |
5142 | ||
5143 | return css; | |
5144 | } | |
5145 | ||
eb95419b | 5146 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5147 | { |
eb95419b | 5148 | kfree(css); |
fe693435 PM |
5149 | } |
5150 | ||
182446d0 TH |
5151 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5152 | struct cftype *cft) | |
fe693435 | 5153 | { |
182446d0 | 5154 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5155 | } |
5156 | ||
182446d0 TH |
5157 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5158 | struct cftype *cft) | |
fe693435 PM |
5159 | { |
5160 | return (u64)(unsigned long)current->cgroups; | |
5161 | } | |
5162 | ||
182446d0 | 5163 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5164 | struct cftype *cft) |
fe693435 PM |
5165 | { |
5166 | u64 count; | |
5167 | ||
5168 | rcu_read_lock(); | |
a8ad805c | 5169 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5170 | rcu_read_unlock(); |
5171 | return count; | |
5172 | } | |
5173 | ||
2da8ca82 | 5174 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5175 | { |
69d0206c | 5176 | struct cgrp_cset_link *link; |
5abb8855 | 5177 | struct css_set *cset; |
e61734c5 TH |
5178 | char *name_buf; |
5179 | ||
5180 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
5181 | if (!name_buf) | |
5182 | return -ENOMEM; | |
7717f7ba | 5183 | |
96d365e0 | 5184 | down_read(&css_set_rwsem); |
7717f7ba | 5185 | rcu_read_lock(); |
5abb8855 | 5186 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5187 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 5188 | struct cgroup *c = link->cgrp; |
7717f7ba | 5189 | |
a2dd4247 | 5190 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 5191 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 5192 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
5193 | } |
5194 | rcu_read_unlock(); | |
96d365e0 | 5195 | up_read(&css_set_rwsem); |
e61734c5 | 5196 | kfree(name_buf); |
7717f7ba PM |
5197 | return 0; |
5198 | } | |
5199 | ||
5200 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 5201 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5202 | { |
2da8ca82 | 5203 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 5204 | struct cgrp_cset_link *link; |
7717f7ba | 5205 | |
96d365e0 | 5206 | down_read(&css_set_rwsem); |
182446d0 | 5207 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5208 | struct css_set *cset = link->cset; |
7717f7ba PM |
5209 | struct task_struct *task; |
5210 | int count = 0; | |
c7561128 | 5211 | |
5abb8855 | 5212 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 5213 | |
5abb8855 | 5214 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
5215 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
5216 | goto overflow; | |
5217 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
5218 | } | |
5219 | ||
5220 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
5221 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
5222 | goto overflow; | |
5223 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 5224 | } |
c7561128 TH |
5225 | continue; |
5226 | overflow: | |
5227 | seq_puts(seq, " ...\n"); | |
7717f7ba | 5228 | } |
96d365e0 | 5229 | up_read(&css_set_rwsem); |
7717f7ba PM |
5230 | return 0; |
5231 | } | |
5232 | ||
182446d0 | 5233 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5234 | { |
182446d0 | 5235 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
5236 | } |
5237 | ||
5238 | static struct cftype debug_files[] = { | |
fe693435 PM |
5239 | { |
5240 | .name = "taskcount", | |
5241 | .read_u64 = debug_taskcount_read, | |
5242 | }, | |
5243 | ||
5244 | { | |
5245 | .name = "current_css_set", | |
5246 | .read_u64 = current_css_set_read, | |
5247 | }, | |
5248 | ||
5249 | { | |
5250 | .name = "current_css_set_refcount", | |
5251 | .read_u64 = current_css_set_refcount_read, | |
5252 | }, | |
5253 | ||
7717f7ba PM |
5254 | { |
5255 | .name = "current_css_set_cg_links", | |
2da8ca82 | 5256 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
5257 | }, |
5258 | ||
5259 | { | |
5260 | .name = "cgroup_css_links", | |
2da8ca82 | 5261 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
5262 | }, |
5263 | ||
fe693435 PM |
5264 | { |
5265 | .name = "releasable", | |
5266 | .read_u64 = releasable_read, | |
5267 | }, | |
fe693435 | 5268 | |
4baf6e33 TH |
5269 | { } /* terminate */ |
5270 | }; | |
fe693435 | 5271 | |
073219e9 | 5272 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
5273 | .css_alloc = debug_css_alloc, |
5274 | .css_free = debug_css_free, | |
4baf6e33 | 5275 | .base_cftypes = debug_files, |
fe693435 PM |
5276 | }; |
5277 | #endif /* CONFIG_CGROUP_DEBUG */ |