Commit | Line | Data |
---|---|---|
ddbcc7e8 | 1 | /* |
ddbcc7e8 PM |
2 | * Generic process-grouping system. |
3 | * | |
4 | * Based originally on the cpuset system, extracted by Paul Menage | |
5 | * Copyright (C) 2006 Google, Inc | |
6 | * | |
0dea1168 KS |
7 | * Notifications support |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
ddbcc7e8 PM |
11 | * Copyright notices from the original cpuset code: |
12 | * -------------------------------------------------- | |
13 | * Copyright (C) 2003 BULL SA. | |
14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
15 | * | |
16 | * Portions derived from Patrick Mochel's sysfs code. | |
17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
18 | * | |
19 | * 2003-10-10 Written by Simon Derr. | |
20 | * 2003-10-22 Updates by Stephen Hemminger. | |
21 | * 2004 May-July Rework by Paul Jackson. | |
22 | * --------------------------------------------------- | |
23 | * | |
24 | * This file is subject to the terms and conditions of the GNU General Public | |
25 | * License. See the file COPYING in the main directory of the Linux | |
26 | * distribution for more details. | |
27 | */ | |
28 | ||
29 | #include <linux/cgroup.h> | |
2ce9738b | 30 | #include <linux/cred.h> |
c6d57f33 | 31 | #include <linux/ctype.h> |
ddbcc7e8 | 32 | #include <linux/errno.h> |
2ce9738b | 33 | #include <linux/init_task.h> |
ddbcc7e8 PM |
34 | #include <linux/kernel.h> |
35 | #include <linux/list.h> | |
36 | #include <linux/mm.h> | |
37 | #include <linux/mutex.h> | |
38 | #include <linux/mount.h> | |
39 | #include <linux/pagemap.h> | |
a424316c | 40 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
41 | #include <linux/rcupdate.h> |
42 | #include <linux/sched.h> | |
ddbcc7e8 | 43 | #include <linux/slab.h> |
ddbcc7e8 | 44 | #include <linux/spinlock.h> |
96d365e0 | 45 | #include <linux/rwsem.h> |
ddbcc7e8 | 46 | #include <linux/string.h> |
bbcb81d0 | 47 | #include <linux/sort.h> |
81a6a5cd | 48 | #include <linux/kmod.h> |
846c7bb0 BS |
49 | #include <linux/delayacct.h> |
50 | #include <linux/cgroupstats.h> | |
0ac801fe | 51 | #include <linux/hashtable.h> |
096b7fe0 | 52 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 53 | #include <linux/idr.h> |
d1d9fd33 | 54 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 55 | #include <linux/kthread.h> |
776f02fa | 56 | #include <linux/delay.h> |
846c7bb0 | 57 | |
60063497 | 58 | #include <linux/atomic.h> |
ddbcc7e8 | 59 | |
b1a21367 TH |
60 | /* |
61 | * pidlists linger the following amount before being destroyed. The goal | |
62 | * is avoiding frequent destruction in the middle of consecutive read calls | |
63 | * Expiring in the middle is a performance problem not a correctness one. | |
64 | * 1 sec should be enough. | |
65 | */ | |
66 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
67 | ||
8d7e6fb0 TH |
68 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
69 | MAX_CFTYPE_NAME + 2) | |
70 | ||
ace2bee8 TH |
71 | /* |
72 | * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file | |
73 | * creation/removal and hierarchy changing operations including cgroup | |
74 | * creation, removal, css association and controller rebinding. This outer | |
75 | * lock is needed mainly to resolve the circular dependency between kernfs | |
76 | * active ref and cgroup_mutex. cgroup_tree_mutex nests above both. | |
77 | */ | |
78 | static DEFINE_MUTEX(cgroup_tree_mutex); | |
79 | ||
e25e2cbb TH |
80 | /* |
81 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
82 | * hierarchy must be performed while holding it. | |
e25e2cbb | 83 | */ |
2219449a TH |
84 | #ifdef CONFIG_PROVE_RCU |
85 | DEFINE_MUTEX(cgroup_mutex); | |
8af01f56 | 86 | EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for lockdep */ |
2219449a | 87 | #else |
81a6a5cd | 88 | static DEFINE_MUTEX(cgroup_mutex); |
2219449a TH |
89 | #endif |
90 | ||
69e943b7 TH |
91 | /* |
92 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
93 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
94 | */ | |
95 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
96 | ||
ace2bee8 | 97 | #define cgroup_assert_mutexes_or_rcu_locked() \ |
87fb54f1 | 98 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
ace2bee8 | 99 | lockdep_is_held(&cgroup_tree_mutex) || \ |
87fb54f1 | 100 | lockdep_is_held(&cgroup_mutex), \ |
ace2bee8 | 101 | "cgroup_[tree_]mutex or RCU read lock required"); |
87fb54f1 | 102 | |
e5fca243 TH |
103 | /* |
104 | * cgroup destruction makes heavy use of work items and there can be a lot | |
105 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
106 | * destruction work items don't end up filling up max_active of system_wq | |
107 | * which may lead to deadlock. | |
108 | */ | |
109 | static struct workqueue_struct *cgroup_destroy_wq; | |
110 | ||
b1a21367 TH |
111 | /* |
112 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
113 | * separate workqueue as flush domain. | |
114 | */ | |
115 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
116 | ||
3ed80a62 | 117 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 118 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 119 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
120 | #include <linux/cgroup_subsys.h> |
121 | }; | |
073219e9 TH |
122 | #undef SUBSYS |
123 | ||
124 | /* array of cgroup subsystem names */ | |
125 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
126 | static const char *cgroup_subsys_name[] = { | |
127 | #include <linux/cgroup_subsys.h> | |
128 | }; | |
129 | #undef SUBSYS | |
ddbcc7e8 | 130 | |
ddbcc7e8 | 131 | /* |
9871bf95 TH |
132 | * The dummy hierarchy, reserved for the subsystems that are otherwise |
133 | * unattached - it never has more than a single cgroup, and all tasks are | |
134 | * part of that cgroup. | |
ddbcc7e8 | 135 | */ |
9871bf95 TH |
136 | static struct cgroupfs_root cgroup_dummy_root; |
137 | ||
138 | /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */ | |
139 | static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup; | |
ddbcc7e8 PM |
140 | |
141 | /* The list of hierarchy roots */ | |
142 | ||
9871bf95 TH |
143 | static LIST_HEAD(cgroup_roots); |
144 | static int cgroup_root_count; | |
ddbcc7e8 | 145 | |
3417ae1f | 146 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 147 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 148 | |
794611a1 LZ |
149 | /* |
150 | * Assign a monotonically increasing serial number to cgroups. It | |
151 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
152 | * numbers. Also, as cgroups are always appended to the parent's | |
153 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
154 | * the ascending serial number order on the list. Protected by |
155 | * cgroup_mutex. | |
794611a1 | 156 | */ |
00356bd5 | 157 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 158 | |
ddbcc7e8 | 159 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
160 | * check for fork/exit handlers to call. This avoids us having to do |
161 | * extra work in the fork/exit path if none of the subsystems need to | |
162 | * be called. | |
ddbcc7e8 | 163 | */ |
8947f9d5 | 164 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 165 | |
628f7cd4 TH |
166 | static struct cftype cgroup_base_files[]; |
167 | ||
59f5296b | 168 | static void cgroup_put(struct cgroup *cgrp); |
f2e85d57 TH |
169 | static int rebind_subsystems(struct cgroupfs_root *root, |
170 | unsigned long added_mask, unsigned removed_mask); | |
f20104de | 171 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 172 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
2bb566cb TH |
173 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
174 | bool is_add); | |
b1a21367 | 175 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 176 | |
95109b62 TH |
177 | /** |
178 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
179 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 180 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 181 | * |
ca8bdcaf TH |
182 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
183 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
184 | * the caller is responsible for pinning the returned css if it wants to | |
185 | * keep accessing it outside the said locks. This function may return | |
186 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
187 | */ |
188 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 189 | struct cgroup_subsys *ss) |
95109b62 | 190 | { |
ca8bdcaf | 191 | if (ss) |
aec25020 | 192 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 TH |
193 | lockdep_is_held(&cgroup_tree_mutex) || |
194 | lockdep_is_held(&cgroup_mutex)); | |
ca8bdcaf TH |
195 | else |
196 | return &cgrp->dummy_css; | |
95109b62 | 197 | } |
42809dd4 | 198 | |
ddbcc7e8 | 199 | /* convenient tests for these bits */ |
54766d4a | 200 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 201 | { |
54766d4a | 202 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
203 | } |
204 | ||
59f5296b TH |
205 | struct cgroup_subsys_state *seq_css(struct seq_file *seq) |
206 | { | |
2bd59d48 TH |
207 | struct kernfs_open_file *of = seq->private; |
208 | struct cgroup *cgrp = of->kn->parent->priv; | |
209 | struct cftype *cft = seq_cft(seq); | |
210 | ||
211 | /* | |
212 | * This is open and unprotected implementation of cgroup_css(). | |
213 | * seq_css() is only called from a kernfs file operation which has | |
214 | * an active reference on the file. Because all the subsystem | |
215 | * files are drained before a css is disassociated with a cgroup, | |
216 | * the matching css from the cgroup's subsys table is guaranteed to | |
217 | * be and stay valid until the enclosing operation is complete. | |
218 | */ | |
219 | if (cft->ss) | |
220 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
221 | else | |
222 | return &cgrp->dummy_css; | |
59f5296b TH |
223 | } |
224 | EXPORT_SYMBOL_GPL(seq_css); | |
225 | ||
78574cf9 LZ |
226 | /** |
227 | * cgroup_is_descendant - test ancestry | |
228 | * @cgrp: the cgroup to be tested | |
229 | * @ancestor: possible ancestor of @cgrp | |
230 | * | |
231 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
232 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
233 | * and @ancestor are accessible. | |
234 | */ | |
235 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
236 | { | |
237 | while (cgrp) { | |
238 | if (cgrp == ancestor) | |
239 | return true; | |
240 | cgrp = cgrp->parent; | |
241 | } | |
242 | return false; | |
243 | } | |
ddbcc7e8 | 244 | |
e9685a03 | 245 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
246 | { |
247 | const int bits = | |
bd89aabc PM |
248 | (1 << CGRP_RELEASABLE) | |
249 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
250 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
251 | } |
252 | ||
e9685a03 | 253 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 254 | { |
bd89aabc | 255 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
256 | } |
257 | ||
1c6727af TH |
258 | /** |
259 | * for_each_css - iterate all css's of a cgroup | |
260 | * @css: the iteration cursor | |
261 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
262 | * @cgrp: the target cgroup to iterate css's of | |
263 | * | |
264 | * Should be called under cgroup_mutex. | |
265 | */ | |
266 | #define for_each_css(css, ssid, cgrp) \ | |
267 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
268 | if (!((css) = rcu_dereference_check( \ | |
269 | (cgrp)->subsys[(ssid)], \ | |
ace2bee8 | 270 | lockdep_is_held(&cgroup_tree_mutex) || \ |
1c6727af TH |
271 | lockdep_is_held(&cgroup_mutex)))) { } \ |
272 | else | |
273 | ||
30159ec7 | 274 | /** |
3ed80a62 | 275 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 276 | * @ss: the iteration cursor |
780cd8b3 | 277 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 278 | */ |
780cd8b3 | 279 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
280 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
281 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 282 | |
5549c497 TH |
283 | /* iterate across the active hierarchies */ |
284 | #define for_each_active_root(root) \ | |
285 | list_for_each_entry((root), &cgroup_roots, root_list) | |
ddbcc7e8 | 286 | |
7ae1bad9 TH |
287 | /** |
288 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
289 | * @cgrp: the cgroup to be checked for liveness | |
290 | * | |
47cfcd09 TH |
291 | * On success, returns true; the mutex should be later unlocked. On |
292 | * failure returns false with no lock held. | |
7ae1bad9 | 293 | */ |
b9777cf8 | 294 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
295 | { |
296 | mutex_lock(&cgroup_mutex); | |
54766d4a | 297 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
298 | mutex_unlock(&cgroup_mutex); |
299 | return false; | |
300 | } | |
301 | return true; | |
302 | } | |
7ae1bad9 | 303 | |
81a6a5cd PM |
304 | /* the list of cgroups eligible for automatic release. Protected by |
305 | * release_list_lock */ | |
306 | static LIST_HEAD(release_list); | |
cdcc136f | 307 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
308 | static void cgroup_release_agent(struct work_struct *work); |
309 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 310 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 311 | |
69d0206c TH |
312 | /* |
313 | * A cgroup can be associated with multiple css_sets as different tasks may | |
314 | * belong to different cgroups on different hierarchies. In the other | |
315 | * direction, a css_set is naturally associated with multiple cgroups. | |
316 | * This M:N relationship is represented by the following link structure | |
317 | * which exists for each association and allows traversing the associations | |
318 | * from both sides. | |
319 | */ | |
320 | struct cgrp_cset_link { | |
321 | /* the cgroup and css_set this link associates */ | |
322 | struct cgroup *cgrp; | |
323 | struct css_set *cset; | |
324 | ||
325 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
326 | struct list_head cset_link; | |
327 | ||
328 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
329 | struct list_head cgrp_link; | |
817929ec PM |
330 | }; |
331 | ||
332 | /* The default css_set - used by init and its children prior to any | |
333 | * hierarchies being mounted. It contains a pointer to the root state | |
334 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
335 | * reference-counted, to improve performance when child cgroups | |
336 | * haven't been created. | |
337 | */ | |
338 | ||
339 | static struct css_set init_css_set; | |
69d0206c | 340 | static struct cgrp_cset_link init_cgrp_cset_link; |
817929ec | 341 | |
0942eeee | 342 | /* |
96d365e0 TH |
343 | * css_set_rwsem protects the list of css_set objects, and the chain of |
344 | * tasks off each css_set. | |
0942eeee | 345 | */ |
96d365e0 | 346 | static DECLARE_RWSEM(css_set_rwsem); |
817929ec PM |
347 | static int css_set_count; |
348 | ||
7717f7ba PM |
349 | /* |
350 | * hash table for cgroup groups. This improves the performance to find | |
351 | * an existing css_set. This hash doesn't (currently) take into | |
352 | * account cgroups in empty hierarchies. | |
353 | */ | |
472b1053 | 354 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 355 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 356 | |
0ac801fe | 357 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 358 | { |
0ac801fe | 359 | unsigned long key = 0UL; |
30159ec7 TH |
360 | struct cgroup_subsys *ss; |
361 | int i; | |
472b1053 | 362 | |
30159ec7 | 363 | for_each_subsys(ss, i) |
0ac801fe LZ |
364 | key += (unsigned long)css[i]; |
365 | key = (key >> 16) ^ key; | |
472b1053 | 366 | |
0ac801fe | 367 | return key; |
472b1053 LZ |
368 | } |
369 | ||
89c5509b | 370 | static void put_css_set_locked(struct css_set *cset, bool taskexit) |
b4f48b63 | 371 | { |
69d0206c | 372 | struct cgrp_cset_link *link, *tmp_link; |
5abb8855 | 373 | |
89c5509b TH |
374 | lockdep_assert_held(&css_set_rwsem); |
375 | ||
376 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 377 | return; |
81a6a5cd | 378 | |
2c6ab6d2 | 379 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
5abb8855 | 380 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
381 | css_set_count--; |
382 | ||
69d0206c | 383 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 384 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 385 | |
69d0206c TH |
386 | list_del(&link->cset_link); |
387 | list_del(&link->cgrp_link); | |
71b5707e | 388 | |
96d365e0 | 389 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
6f3d828f | 390 | if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) { |
81a6a5cd | 391 | if (taskexit) |
bd89aabc PM |
392 | set_bit(CGRP_RELEASABLE, &cgrp->flags); |
393 | check_for_release(cgrp); | |
81a6a5cd | 394 | } |
2c6ab6d2 PM |
395 | |
396 | kfree(link); | |
81a6a5cd | 397 | } |
2c6ab6d2 | 398 | |
5abb8855 | 399 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
400 | } |
401 | ||
89c5509b TH |
402 | static void put_css_set(struct css_set *cset, bool taskexit) |
403 | { | |
404 | /* | |
405 | * Ensure that the refcount doesn't hit zero while any readers | |
406 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
407 | * rwlock | |
408 | */ | |
409 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
410 | return; | |
411 | ||
412 | down_write(&css_set_rwsem); | |
413 | put_css_set_locked(cset, taskexit); | |
414 | up_write(&css_set_rwsem); | |
415 | } | |
416 | ||
817929ec PM |
417 | /* |
418 | * refcounted get/put for css_set objects | |
419 | */ | |
5abb8855 | 420 | static inline void get_css_set(struct css_set *cset) |
817929ec | 421 | { |
5abb8855 | 422 | atomic_inc(&cset->refcount); |
817929ec PM |
423 | } |
424 | ||
b326f9d0 | 425 | /** |
7717f7ba | 426 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
427 | * @cset: candidate css_set being tested |
428 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
429 | * @new_cgrp: cgroup that's being entered by the task |
430 | * @template: desired set of css pointers in css_set (pre-calculated) | |
431 | * | |
6f4b7e63 | 432 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
433 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
434 | */ | |
5abb8855 TH |
435 | static bool compare_css_sets(struct css_set *cset, |
436 | struct css_set *old_cset, | |
7717f7ba PM |
437 | struct cgroup *new_cgrp, |
438 | struct cgroup_subsys_state *template[]) | |
439 | { | |
440 | struct list_head *l1, *l2; | |
441 | ||
5abb8855 | 442 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) { |
7717f7ba PM |
443 | /* Not all subsystems matched */ |
444 | return false; | |
445 | } | |
446 | ||
447 | /* | |
448 | * Compare cgroup pointers in order to distinguish between | |
449 | * different cgroups in heirarchies with no subsystems. We | |
450 | * could get by with just this check alone (and skip the | |
451 | * memcmp above) but on most setups the memcmp check will | |
452 | * avoid the need for this more expensive check on almost all | |
453 | * candidates. | |
454 | */ | |
455 | ||
69d0206c TH |
456 | l1 = &cset->cgrp_links; |
457 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 458 | while (1) { |
69d0206c | 459 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 460 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
461 | |
462 | l1 = l1->next; | |
463 | l2 = l2->next; | |
464 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
465 | if (l1 == &cset->cgrp_links) { |
466 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
467 | break; |
468 | } else { | |
69d0206c | 469 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
470 | } |
471 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
472 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
473 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
474 | cgrp1 = link1->cgrp; | |
475 | cgrp2 = link2->cgrp; | |
7717f7ba | 476 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 477 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
478 | |
479 | /* | |
480 | * If this hierarchy is the hierarchy of the cgroup | |
481 | * that's changing, then we need to check that this | |
482 | * css_set points to the new cgroup; if it's any other | |
483 | * hierarchy, then this css_set should point to the | |
484 | * same cgroup as the old css_set. | |
485 | */ | |
5abb8855 TH |
486 | if (cgrp1->root == new_cgrp->root) { |
487 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
488 | return false; |
489 | } else { | |
5abb8855 | 490 | if (cgrp1 != cgrp2) |
7717f7ba PM |
491 | return false; |
492 | } | |
493 | } | |
494 | return true; | |
495 | } | |
496 | ||
b326f9d0 TH |
497 | /** |
498 | * find_existing_css_set - init css array and find the matching css_set | |
499 | * @old_cset: the css_set that we're using before the cgroup transition | |
500 | * @cgrp: the cgroup that we're moving into | |
501 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 502 | */ |
5abb8855 TH |
503 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
504 | struct cgroup *cgrp, | |
505 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 506 | { |
bd89aabc | 507 | struct cgroupfs_root *root = cgrp->root; |
30159ec7 | 508 | struct cgroup_subsys *ss; |
5abb8855 | 509 | struct css_set *cset; |
0ac801fe | 510 | unsigned long key; |
b326f9d0 | 511 | int i; |
817929ec | 512 | |
aae8aab4 BB |
513 | /* |
514 | * Build the set of subsystem state objects that we want to see in the | |
515 | * new css_set. while subsystems can change globally, the entries here | |
516 | * won't change, so no need for locking. | |
517 | */ | |
30159ec7 | 518 | for_each_subsys(ss, i) { |
a1a71b45 | 519 | if (root->subsys_mask & (1UL << i)) { |
817929ec PM |
520 | /* Subsystem is in this hierarchy. So we want |
521 | * the subsystem state from the new | |
522 | * cgroup */ | |
ca8bdcaf | 523 | template[i] = cgroup_css(cgrp, ss); |
817929ec PM |
524 | } else { |
525 | /* Subsystem is not in this hierarchy, so we | |
526 | * don't want to change the subsystem state */ | |
5abb8855 | 527 | template[i] = old_cset->subsys[i]; |
817929ec PM |
528 | } |
529 | } | |
530 | ||
0ac801fe | 531 | key = css_set_hash(template); |
5abb8855 TH |
532 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
533 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
534 | continue; |
535 | ||
536 | /* This css_set matches what we need */ | |
5abb8855 | 537 | return cset; |
472b1053 | 538 | } |
817929ec PM |
539 | |
540 | /* No existing cgroup group matched */ | |
541 | return NULL; | |
542 | } | |
543 | ||
69d0206c | 544 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 545 | { |
69d0206c | 546 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 547 | |
69d0206c TH |
548 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
549 | list_del(&link->cset_link); | |
36553434 LZ |
550 | kfree(link); |
551 | } | |
552 | } | |
553 | ||
69d0206c TH |
554 | /** |
555 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
556 | * @count: the number of links to allocate | |
557 | * @tmp_links: list_head the allocated links are put on | |
558 | * | |
559 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
560 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 561 | */ |
69d0206c | 562 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 563 | { |
69d0206c | 564 | struct cgrp_cset_link *link; |
817929ec | 565 | int i; |
69d0206c TH |
566 | |
567 | INIT_LIST_HEAD(tmp_links); | |
568 | ||
817929ec | 569 | for (i = 0; i < count; i++) { |
f4f4be2b | 570 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 571 | if (!link) { |
69d0206c | 572 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
573 | return -ENOMEM; |
574 | } | |
69d0206c | 575 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
576 | } |
577 | return 0; | |
578 | } | |
579 | ||
c12f65d4 LZ |
580 | /** |
581 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 582 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 583 | * @cset: the css_set to be linked |
c12f65d4 LZ |
584 | * @cgrp: the destination cgroup |
585 | */ | |
69d0206c TH |
586 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
587 | struct cgroup *cgrp) | |
c12f65d4 | 588 | { |
69d0206c | 589 | struct cgrp_cset_link *link; |
c12f65d4 | 590 | |
69d0206c TH |
591 | BUG_ON(list_empty(tmp_links)); |
592 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); | |
593 | link->cset = cset; | |
7717f7ba | 594 | link->cgrp = cgrp; |
69d0206c | 595 | list_move(&link->cset_link, &cgrp->cset_links); |
7717f7ba PM |
596 | /* |
597 | * Always add links to the tail of the list so that the list | |
598 | * is sorted by order of hierarchy creation | |
599 | */ | |
69d0206c | 600 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
601 | } |
602 | ||
b326f9d0 TH |
603 | /** |
604 | * find_css_set - return a new css_set with one cgroup updated | |
605 | * @old_cset: the baseline css_set | |
606 | * @cgrp: the cgroup to be updated | |
607 | * | |
608 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
609 | * substituted into the appropriate hierarchy. | |
817929ec | 610 | */ |
5abb8855 TH |
611 | static struct css_set *find_css_set(struct css_set *old_cset, |
612 | struct cgroup *cgrp) | |
817929ec | 613 | { |
b326f9d0 | 614 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 615 | struct css_set *cset; |
69d0206c TH |
616 | struct list_head tmp_links; |
617 | struct cgrp_cset_link *link; | |
0ac801fe | 618 | unsigned long key; |
472b1053 | 619 | |
b326f9d0 TH |
620 | lockdep_assert_held(&cgroup_mutex); |
621 | ||
817929ec PM |
622 | /* First see if we already have a cgroup group that matches |
623 | * the desired set */ | |
96d365e0 | 624 | down_read(&css_set_rwsem); |
5abb8855 TH |
625 | cset = find_existing_css_set(old_cset, cgrp, template); |
626 | if (cset) | |
627 | get_css_set(cset); | |
96d365e0 | 628 | up_read(&css_set_rwsem); |
817929ec | 629 | |
5abb8855 TH |
630 | if (cset) |
631 | return cset; | |
817929ec | 632 | |
f4f4be2b | 633 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 634 | if (!cset) |
817929ec PM |
635 | return NULL; |
636 | ||
69d0206c | 637 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 638 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 639 | kfree(cset); |
817929ec PM |
640 | return NULL; |
641 | } | |
642 | ||
5abb8855 | 643 | atomic_set(&cset->refcount, 1); |
69d0206c | 644 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 645 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 646 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 647 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 648 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 649 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
650 | |
651 | /* Copy the set of subsystem state objects generated in | |
652 | * find_existing_css_set() */ | |
5abb8855 | 653 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 654 | |
96d365e0 | 655 | down_write(&css_set_rwsem); |
817929ec | 656 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 657 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 658 | struct cgroup *c = link->cgrp; |
69d0206c | 659 | |
7717f7ba PM |
660 | if (c->root == cgrp->root) |
661 | c = cgrp; | |
69d0206c | 662 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 663 | } |
817929ec | 664 | |
69d0206c | 665 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 666 | |
817929ec | 667 | css_set_count++; |
472b1053 LZ |
668 | |
669 | /* Add this cgroup group to the hash table */ | |
5abb8855 TH |
670 | key = css_set_hash(cset->subsys); |
671 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 672 | |
96d365e0 | 673 | up_write(&css_set_rwsem); |
817929ec | 674 | |
5abb8855 | 675 | return cset; |
b4f48b63 PM |
676 | } |
677 | ||
2bd59d48 TH |
678 | static struct cgroupfs_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
679 | { | |
680 | struct cgroup *top_cgrp = kf_root->kn->priv; | |
681 | ||
682 | return top_cgrp->root; | |
683 | } | |
684 | ||
f2e85d57 TH |
685 | static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end) |
686 | { | |
687 | int id; | |
688 | ||
689 | lockdep_assert_held(&cgroup_mutex); | |
690 | ||
691 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end, | |
692 | GFP_KERNEL); | |
693 | if (id < 0) | |
694 | return id; | |
695 | ||
696 | root->hierarchy_id = id; | |
697 | return 0; | |
698 | } | |
699 | ||
700 | static void cgroup_exit_root_id(struct cgroupfs_root *root) | |
701 | { | |
702 | lockdep_assert_held(&cgroup_mutex); | |
703 | ||
704 | if (root->hierarchy_id) { | |
705 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
706 | root->hierarchy_id = 0; | |
707 | } | |
708 | } | |
709 | ||
710 | static void cgroup_free_root(struct cgroupfs_root *root) | |
711 | { | |
712 | if (root) { | |
713 | /* hierarhcy ID shoulid already have been released */ | |
714 | WARN_ON_ONCE(root->hierarchy_id); | |
715 | ||
716 | idr_destroy(&root->cgroup_idr); | |
717 | kfree(root); | |
718 | } | |
719 | } | |
720 | ||
776f02fa | 721 | static void cgroup_destroy_root(struct cgroupfs_root *root) |
59f5296b | 722 | { |
f2e85d57 TH |
723 | struct cgroup *cgrp = &root->top_cgroup; |
724 | struct cgrp_cset_link *link, *tmp_link; | |
f2e85d57 | 725 | |
2bd59d48 | 726 | mutex_lock(&cgroup_tree_mutex); |
2bd59d48 | 727 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 728 | |
776f02fa | 729 | BUG_ON(atomic_read(&root->nr_cgrps)); |
f2e85d57 TH |
730 | BUG_ON(!list_empty(&cgrp->children)); |
731 | ||
f2e85d57 | 732 | /* Rebind all subsystems back to the default hierarchy */ |
35585573 | 733 | WARN_ON(rebind_subsystems(root, 0, root->subsys_mask)); |
f2e85d57 TH |
734 | |
735 | /* | |
736 | * Release all the links from cset_links to this hierarchy's | |
737 | * root cgroup | |
738 | */ | |
96d365e0 | 739 | down_write(&css_set_rwsem); |
f2e85d57 TH |
740 | |
741 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
742 | list_del(&link->cset_link); | |
743 | list_del(&link->cgrp_link); | |
744 | kfree(link); | |
745 | } | |
96d365e0 | 746 | up_write(&css_set_rwsem); |
f2e85d57 TH |
747 | |
748 | if (!list_empty(&root->root_list)) { | |
749 | list_del(&root->root_list); | |
750 | cgroup_root_count--; | |
751 | } | |
752 | ||
753 | cgroup_exit_root_id(root); | |
754 | ||
755 | mutex_unlock(&cgroup_mutex); | |
756 | mutex_unlock(&cgroup_tree_mutex); | |
f2e85d57 | 757 | |
2bd59d48 | 758 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
759 | cgroup_free_root(root); |
760 | } | |
761 | ||
ceb6a081 TH |
762 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
763 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
7717f7ba PM |
764 | struct cgroupfs_root *root) |
765 | { | |
7717f7ba PM |
766 | struct cgroup *res = NULL; |
767 | ||
96d365e0 TH |
768 | lockdep_assert_held(&cgroup_mutex); |
769 | lockdep_assert_held(&css_set_rwsem); | |
770 | ||
5abb8855 | 771 | if (cset == &init_css_set) { |
7717f7ba PM |
772 | res = &root->top_cgroup; |
773 | } else { | |
69d0206c TH |
774 | struct cgrp_cset_link *link; |
775 | ||
776 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 777 | struct cgroup *c = link->cgrp; |
69d0206c | 778 | |
7717f7ba PM |
779 | if (c->root == root) { |
780 | res = c; | |
781 | break; | |
782 | } | |
783 | } | |
784 | } | |
96d365e0 | 785 | |
7717f7ba PM |
786 | BUG_ON(!res); |
787 | return res; | |
788 | } | |
789 | ||
ceb6a081 TH |
790 | /* |
791 | * Return the cgroup for "task" from the given hierarchy. Must be | |
792 | * called with cgroup_mutex and css_set_rwsem held. | |
793 | */ | |
794 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
795 | struct cgroupfs_root *root) | |
796 | { | |
797 | /* | |
798 | * No need to lock the task - since we hold cgroup_mutex the | |
799 | * task can't change groups, so the only thing that can happen | |
800 | * is that it exits and its css is set back to init_css_set. | |
801 | */ | |
802 | return cset_cgroup_from_root(task_css_set(task), root); | |
803 | } | |
804 | ||
ddbcc7e8 PM |
805 | /* |
806 | * There is one global cgroup mutex. We also require taking | |
807 | * task_lock() when dereferencing a task's cgroup subsys pointers. | |
808 | * See "The task_lock() exception", at the end of this comment. | |
809 | * | |
810 | * A task must hold cgroup_mutex to modify cgroups. | |
811 | * | |
812 | * Any task can increment and decrement the count field without lock. | |
813 | * So in general, code holding cgroup_mutex can't rely on the count | |
814 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 815 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
816 | * means that no tasks are currently attached, therefore there is no |
817 | * way a task attached to that cgroup can fork (the other way to | |
818 | * increment the count). So code holding cgroup_mutex can safely | |
819 | * assume that if the count is zero, it will stay zero. Similarly, if | |
820 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
821 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
822 | * needs that mutex. | |
823 | * | |
ddbcc7e8 PM |
824 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
825 | * (usually) take cgroup_mutex. These are the two most performance | |
826 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
827 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
828 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
829 | * to the release agent with the name of the cgroup (path relative to |
830 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
831 | * |
832 | * A cgroup can only be deleted if both its 'count' of using tasks | |
833 | * is zero, and its list of 'children' cgroups is empty. Since all | |
834 | * tasks in the system use _some_ cgroup, and since there is always at | |
835 | * least one task in the system (init, pid == 1), therefore, top_cgroup | |
836 | * always has either children cgroups and/or using tasks. So we don't | |
837 | * need a special hack to ensure that top_cgroup cannot be deleted. | |
838 | * | |
839 | * The task_lock() exception | |
840 | * | |
841 | * The need for this exception arises from the action of | |
d0b2fdd2 | 842 | * cgroup_attach_task(), which overwrites one task's cgroup pointer with |
a043e3b2 | 843 | * another. It does so using cgroup_mutex, however there are |
ddbcc7e8 PM |
844 | * several performance critical places that need to reference |
845 | * task->cgroup without the expense of grabbing a system global | |
846 | * mutex. Therefore except as noted below, when dereferencing or, as | |
d0b2fdd2 | 847 | * in cgroup_attach_task(), modifying a task's cgroup pointer we use |
ddbcc7e8 PM |
848 | * task_lock(), which acts on a spinlock (task->alloc_lock) already in |
849 | * the task_struct routinely used for such matters. | |
850 | * | |
851 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 852 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
853 | */ |
854 | ||
628f7cd4 | 855 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask); |
2bd59d48 | 856 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 857 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 858 | |
8d7e6fb0 TH |
859 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
860 | char *buf) | |
861 | { | |
862 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && | |
863 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
864 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
865 | cft->ss->name, cft->name); | |
866 | else | |
867 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
868 | return buf; | |
869 | } | |
870 | ||
f2e85d57 TH |
871 | /** |
872 | * cgroup_file_mode - deduce file mode of a control file | |
873 | * @cft: the control file in question | |
874 | * | |
875 | * returns cft->mode if ->mode is not 0 | |
876 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
877 | * returns S_IRUGO if it has only a read handler | |
878 | * returns S_IWUSR if it has only a write hander | |
879 | */ | |
880 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
881 | { | |
882 | umode_t mode = 0; | |
883 | ||
884 | if (cft->mode) | |
885 | return cft->mode; | |
886 | ||
887 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
888 | mode |= S_IRUGO; | |
889 | ||
890 | if (cft->write_u64 || cft->write_s64 || cft->write_string || | |
891 | cft->trigger) | |
892 | mode |= S_IWUSR; | |
893 | ||
894 | return mode; | |
895 | } | |
896 | ||
be445626 LZ |
897 | static void cgroup_free_fn(struct work_struct *work) |
898 | { | |
ea15f8cc | 899 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 | 900 | |
3c9c825b | 901 | atomic_dec(&cgrp->root->nr_cgrps); |
b1a21367 | 902 | cgroup_pidlist_destroy_all(cgrp); |
be445626 | 903 | |
776f02fa TH |
904 | if (cgrp->parent) { |
905 | /* | |
906 | * We get a ref to the parent, and put the ref when this | |
907 | * cgroup is being freed, so it's guaranteed that the | |
908 | * parent won't be destroyed before its children. | |
909 | */ | |
910 | cgroup_put(cgrp->parent); | |
911 | kernfs_put(cgrp->kn); | |
912 | kfree(cgrp); | |
913 | } else { | |
914 | /* | |
915 | * This is top cgroup's refcnt reaching zero, which | |
916 | * indicates that the root should be released. | |
917 | */ | |
918 | cgroup_destroy_root(cgrp->root); | |
919 | } | |
be445626 LZ |
920 | } |
921 | ||
922 | static void cgroup_free_rcu(struct rcu_head *head) | |
923 | { | |
924 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
925 | ||
ea15f8cc | 926 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 927 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
928 | } |
929 | ||
59f5296b TH |
930 | static void cgroup_get(struct cgroup *cgrp) |
931 | { | |
2bd59d48 TH |
932 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
933 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
934 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
935 | } |
936 | ||
59f5296b TH |
937 | static void cgroup_put(struct cgroup *cgrp) |
938 | { | |
2bd59d48 TH |
939 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
940 | return; | |
776f02fa | 941 | if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp))) |
2bd59d48 | 942 | return; |
59f5296b | 943 | |
2bd59d48 TH |
944 | /* |
945 | * XXX: cgrp->id is only used to look up css's. As cgroup and | |
946 | * css's lifetimes will be decoupled, it should be made | |
947 | * per-subsystem and moved to css->id so that lookups are | |
948 | * successful until the target css is released. | |
949 | */ | |
950 | mutex_lock(&cgroup_mutex); | |
951 | idr_remove(&cgrp->root->cgroup_idr, cgrp->id); | |
952 | mutex_unlock(&cgroup_mutex); | |
953 | cgrp->id = -1; | |
ddbcc7e8 | 954 | |
2bd59d48 | 955 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 PM |
956 | } |
957 | ||
2739d3cc | 958 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 959 | { |
2bd59d48 | 960 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 961 | |
ace2bee8 | 962 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 963 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
964 | } |
965 | ||
13af07df | 966 | /** |
628f7cd4 | 967 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 968 | * @cgrp: target cgroup |
13af07df AR |
969 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
970 | */ | |
628f7cd4 | 971 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
05ef1d7c | 972 | { |
13af07df | 973 | struct cgroup_subsys *ss; |
b420ba7d | 974 | int i; |
05ef1d7c | 975 | |
b420ba7d | 976 | for_each_subsys(ss, i) { |
0adb0704 | 977 | struct cftype *cfts; |
b420ba7d TH |
978 | |
979 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 980 | continue; |
0adb0704 TH |
981 | list_for_each_entry(cfts, &ss->cfts, node) |
982 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 983 | } |
ddbcc7e8 PM |
984 | } |
985 | ||
ddbcc7e8 | 986 | static int rebind_subsystems(struct cgroupfs_root *root, |
a8a648c4 | 987 | unsigned long added_mask, unsigned removed_mask) |
ddbcc7e8 | 988 | { |
bd89aabc | 989 | struct cgroup *cgrp = &root->top_cgroup; |
30159ec7 | 990 | struct cgroup_subsys *ss; |
3126121f | 991 | int i, ret; |
ddbcc7e8 | 992 | |
ace2bee8 TH |
993 | lockdep_assert_held(&cgroup_tree_mutex); |
994 | lockdep_assert_held(&cgroup_mutex); | |
aae8aab4 | 995 | |
ddbcc7e8 | 996 | /* Check that any added subsystems are currently free */ |
3ed80a62 TH |
997 | for_each_subsys(ss, i) |
998 | if ((added_mask & (1 << i)) && ss->root != &cgroup_dummy_root) | |
999 | return -EBUSY; | |
ddbcc7e8 | 1000 | |
3126121f TH |
1001 | ret = cgroup_populate_dir(cgrp, added_mask); |
1002 | if (ret) | |
3ed80a62 | 1003 | return ret; |
3126121f TH |
1004 | |
1005 | /* | |
1006 | * Nothing can fail from this point on. Remove files for the | |
1007 | * removed subsystems and rebind each subsystem. | |
1008 | */ | |
4ac06017 | 1009 | mutex_unlock(&cgroup_mutex); |
3126121f | 1010 | cgroup_clear_dir(cgrp, removed_mask); |
4ac06017 | 1011 | mutex_lock(&cgroup_mutex); |
ddbcc7e8 | 1012 | |
30159ec7 | 1013 | for_each_subsys(ss, i) { |
ddbcc7e8 | 1014 | unsigned long bit = 1UL << i; |
30159ec7 | 1015 | |
a1a71b45 | 1016 | if (bit & added_mask) { |
ddbcc7e8 | 1017 | /* We're binding this subsystem to this hierarchy */ |
ca8bdcaf TH |
1018 | BUG_ON(cgroup_css(cgrp, ss)); |
1019 | BUG_ON(!cgroup_css(cgroup_dummy_top, ss)); | |
1020 | BUG_ON(cgroup_css(cgroup_dummy_top, ss)->cgroup != cgroup_dummy_top); | |
a8a648c4 | 1021 | |
73e80ed8 | 1022 | rcu_assign_pointer(cgrp->subsys[i], |
ca8bdcaf TH |
1023 | cgroup_css(cgroup_dummy_top, ss)); |
1024 | cgroup_css(cgrp, ss)->cgroup = cgrp; | |
a8a648c4 | 1025 | |
b2aa30f7 | 1026 | ss->root = root; |
ddbcc7e8 | 1027 | if (ss->bind) |
ca8bdcaf | 1028 | ss->bind(cgroup_css(cgrp, ss)); |
a8a648c4 | 1029 | |
cf5d5941 | 1030 | /* refcount was already taken, and we're keeping it */ |
a8a648c4 | 1031 | root->subsys_mask |= bit; |
a1a71b45 | 1032 | } else if (bit & removed_mask) { |
ddbcc7e8 | 1033 | /* We're removing this subsystem */ |
ca8bdcaf TH |
1034 | BUG_ON(cgroup_css(cgrp, ss) != cgroup_css(cgroup_dummy_top, ss)); |
1035 | BUG_ON(cgroup_css(cgrp, ss)->cgroup != cgrp); | |
a8a648c4 | 1036 | |
ddbcc7e8 | 1037 | if (ss->bind) |
ca8bdcaf | 1038 | ss->bind(cgroup_css(cgroup_dummy_top, ss)); |
73e80ed8 | 1039 | |
ca8bdcaf | 1040 | cgroup_css(cgroup_dummy_top, ss)->cgroup = cgroup_dummy_top; |
73e80ed8 TH |
1041 | RCU_INIT_POINTER(cgrp->subsys[i], NULL); |
1042 | ||
9871bf95 | 1043 | cgroup_subsys[i]->root = &cgroup_dummy_root; |
a8a648c4 | 1044 | root->subsys_mask &= ~bit; |
ddbcc7e8 PM |
1045 | } |
1046 | } | |
ddbcc7e8 | 1047 | |
2bd59d48 | 1048 | kernfs_activate(cgrp->kn); |
ddbcc7e8 PM |
1049 | return 0; |
1050 | } | |
1051 | ||
2bd59d48 TH |
1052 | static int cgroup_show_options(struct seq_file *seq, |
1053 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1054 | { |
2bd59d48 | 1055 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1056 | struct cgroup_subsys *ss; |
b85d2040 | 1057 | int ssid; |
ddbcc7e8 | 1058 | |
b85d2040 TH |
1059 | for_each_subsys(ss, ssid) |
1060 | if (root->subsys_mask & (1 << ssid)) | |
1061 | seq_printf(seq, ",%s", ss->name); | |
873fe09e TH |
1062 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1063 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1064 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1065 | seq_puts(seq, ",noprefix"); |
93438629 | 1066 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1067 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1068 | |
1069 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1070 | if (strlen(root->release_agent_path)) |
1071 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1072 | spin_unlock(&release_agent_path_lock); |
1073 | ||
2260e7fc | 1074 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags)) |
97978e6d | 1075 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1076 | if (strlen(root->name)) |
1077 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1078 | return 0; |
1079 | } | |
1080 | ||
1081 | struct cgroup_sb_opts { | |
a1a71b45 | 1082 | unsigned long subsys_mask; |
ddbcc7e8 | 1083 | unsigned long flags; |
81a6a5cd | 1084 | char *release_agent; |
2260e7fc | 1085 | bool cpuset_clone_children; |
c6d57f33 | 1086 | char *name; |
2c6ab6d2 PM |
1087 | /* User explicitly requested empty subsystem */ |
1088 | bool none; | |
ddbcc7e8 PM |
1089 | }; |
1090 | ||
aae8aab4 | 1091 | /* |
9871bf95 TH |
1092 | * Convert a hierarchy specifier into a bitmask of subsystems and |
1093 | * flags. Call with cgroup_mutex held to protect the cgroup_subsys[] | |
1094 | * array. This function takes refcounts on subsystems to be used, unless it | |
1095 | * returns error, in which case no refcounts are taken. | |
aae8aab4 | 1096 | */ |
cf5d5941 | 1097 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1098 | { |
32a8cf23 DL |
1099 | char *token, *o = data; |
1100 | bool all_ss = false, one_ss = false; | |
f9ab5b5b | 1101 | unsigned long mask = (unsigned long)-1; |
30159ec7 TH |
1102 | struct cgroup_subsys *ss; |
1103 | int i; | |
f9ab5b5b | 1104 | |
aae8aab4 BB |
1105 | BUG_ON(!mutex_is_locked(&cgroup_mutex)); |
1106 | ||
f9ab5b5b | 1107 | #ifdef CONFIG_CPUSETS |
073219e9 | 1108 | mask = ~(1UL << cpuset_cgrp_id); |
f9ab5b5b | 1109 | #endif |
ddbcc7e8 | 1110 | |
c6d57f33 | 1111 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1112 | |
1113 | while ((token = strsep(&o, ",")) != NULL) { | |
1114 | if (!*token) | |
1115 | return -EINVAL; | |
32a8cf23 | 1116 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1117 | /* Explicitly have no subsystems */ |
1118 | opts->none = true; | |
32a8cf23 DL |
1119 | continue; |
1120 | } | |
1121 | if (!strcmp(token, "all")) { | |
1122 | /* Mutually exclusive option 'all' + subsystem name */ | |
1123 | if (one_ss) | |
1124 | return -EINVAL; | |
1125 | all_ss = true; | |
1126 | continue; | |
1127 | } | |
873fe09e TH |
1128 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1129 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1130 | continue; | |
1131 | } | |
32a8cf23 | 1132 | if (!strcmp(token, "noprefix")) { |
93438629 | 1133 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1134 | continue; |
1135 | } | |
1136 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1137 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1138 | continue; |
1139 | } | |
03b1cde6 | 1140 | if (!strcmp(token, "xattr")) { |
93438629 | 1141 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1142 | continue; |
1143 | } | |
32a8cf23 | 1144 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1145 | /* Specifying two release agents is forbidden */ |
1146 | if (opts->release_agent) | |
1147 | return -EINVAL; | |
c6d57f33 | 1148 | opts->release_agent = |
e400c285 | 1149 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1150 | if (!opts->release_agent) |
1151 | return -ENOMEM; | |
32a8cf23 DL |
1152 | continue; |
1153 | } | |
1154 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1155 | const char *name = token + 5; |
1156 | /* Can't specify an empty name */ | |
1157 | if (!strlen(name)) | |
1158 | return -EINVAL; | |
1159 | /* Must match [\w.-]+ */ | |
1160 | for (i = 0; i < strlen(name); i++) { | |
1161 | char c = name[i]; | |
1162 | if (isalnum(c)) | |
1163 | continue; | |
1164 | if ((c == '.') || (c == '-') || (c == '_')) | |
1165 | continue; | |
1166 | return -EINVAL; | |
1167 | } | |
1168 | /* Specifying two names is forbidden */ | |
1169 | if (opts->name) | |
1170 | return -EINVAL; | |
1171 | opts->name = kstrndup(name, | |
e400c285 | 1172 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1173 | GFP_KERNEL); |
1174 | if (!opts->name) | |
1175 | return -ENOMEM; | |
32a8cf23 DL |
1176 | |
1177 | continue; | |
1178 | } | |
1179 | ||
30159ec7 | 1180 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1181 | if (strcmp(token, ss->name)) |
1182 | continue; | |
1183 | if (ss->disabled) | |
1184 | continue; | |
1185 | ||
1186 | /* Mutually exclusive option 'all' + subsystem name */ | |
1187 | if (all_ss) | |
1188 | return -EINVAL; | |
a1a71b45 | 1189 | set_bit(i, &opts->subsys_mask); |
32a8cf23 DL |
1190 | one_ss = true; |
1191 | ||
1192 | break; | |
1193 | } | |
1194 | if (i == CGROUP_SUBSYS_COUNT) | |
1195 | return -ENOENT; | |
1196 | } | |
1197 | ||
1198 | /* | |
1199 | * If the 'all' option was specified select all the subsystems, | |
0d19ea86 LZ |
1200 | * otherwise if 'none', 'name=' and a subsystem name options |
1201 | * were not specified, let's default to 'all' | |
32a8cf23 | 1202 | */ |
30159ec7 TH |
1203 | if (all_ss || (!one_ss && !opts->none && !opts->name)) |
1204 | for_each_subsys(ss, i) | |
1205 | if (!ss->disabled) | |
1206 | set_bit(i, &opts->subsys_mask); | |
ddbcc7e8 | 1207 | |
2c6ab6d2 PM |
1208 | /* Consistency checks */ |
1209 | ||
873fe09e TH |
1210 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1211 | pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); | |
1212 | ||
d3ba07c3 TH |
1213 | if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) || |
1214 | opts->cpuset_clone_children || opts->release_agent || | |
1215 | opts->name) { | |
1216 | pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n"); | |
873fe09e TH |
1217 | return -EINVAL; |
1218 | } | |
873fe09e TH |
1219 | } |
1220 | ||
f9ab5b5b LZ |
1221 | /* |
1222 | * Option noprefix was introduced just for backward compatibility | |
1223 | * with the old cpuset, so we allow noprefix only if mounting just | |
1224 | * the cpuset subsystem. | |
1225 | */ | |
93438629 | 1226 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1227 | return -EINVAL; |
1228 | ||
2c6ab6d2 PM |
1229 | |
1230 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1231 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1232 | return -EINVAL; |
1233 | ||
1234 | /* | |
1235 | * We either have to specify by name or by subsystems. (So all | |
1236 | * empty hierarchies must have a name). | |
1237 | */ | |
a1a71b45 | 1238 | if (!opts->subsys_mask && !opts->name) |
ddbcc7e8 PM |
1239 | return -EINVAL; |
1240 | ||
1241 | return 0; | |
1242 | } | |
1243 | ||
2bd59d48 | 1244 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1245 | { |
1246 | int ret = 0; | |
2bd59d48 | 1247 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1248 | struct cgroup_sb_opts opts; |
a1a71b45 | 1249 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1250 | |
873fe09e TH |
1251 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
1252 | pr_err("cgroup: sane_behavior: remount is not allowed\n"); | |
1253 | return -EINVAL; | |
1254 | } | |
1255 | ||
ace2bee8 | 1256 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1257 | mutex_lock(&cgroup_mutex); |
1258 | ||
1259 | /* See what subsystems are wanted */ | |
1260 | ret = parse_cgroupfs_options(data, &opts); | |
1261 | if (ret) | |
1262 | goto out_unlock; | |
1263 | ||
a8a648c4 | 1264 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
8b5a5a9d TH |
1265 | pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", |
1266 | task_tgid_nr(current), current->comm); | |
1267 | ||
a1a71b45 AR |
1268 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1269 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1270 | |
cf5d5941 | 1271 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1272 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1273 | (opts.name && strcmp(opts.name, root->name))) { |
0ce6cba3 TH |
1274 | pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n", |
1275 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", | |
1276 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1277 | ret = -EINVAL; |
1278 | goto out_unlock; | |
1279 | } | |
1280 | ||
f172e67c | 1281 | /* remounting is not allowed for populated hierarchies */ |
3c9c825b | 1282 | if (!list_empty(&root->top_cgroup.children)) { |
f172e67c | 1283 | ret = -EBUSY; |
0670e08b | 1284 | goto out_unlock; |
cf5d5941 | 1285 | } |
ddbcc7e8 | 1286 | |
a8a648c4 | 1287 | ret = rebind_subsystems(root, added_mask, removed_mask); |
3126121f | 1288 | if (ret) |
0670e08b | 1289 | goto out_unlock; |
ddbcc7e8 | 1290 | |
69e943b7 TH |
1291 | if (opts.release_agent) { |
1292 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1293 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1294 | spin_unlock(&release_agent_path_lock); |
1295 | } | |
ddbcc7e8 | 1296 | out_unlock: |
66bdc9cf | 1297 | kfree(opts.release_agent); |
c6d57f33 | 1298 | kfree(opts.name); |
ddbcc7e8 | 1299 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1300 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1301 | return ret; |
1302 | } | |
1303 | ||
afeb0f9f TH |
1304 | /* |
1305 | * To reduce the fork() overhead for systems that are not actually using | |
1306 | * their cgroups capability, we don't maintain the lists running through | |
1307 | * each css_set to its tasks until we see the list actually used - in other | |
1308 | * words after the first mount. | |
1309 | */ | |
1310 | static bool use_task_css_set_links __read_mostly; | |
1311 | ||
1312 | static void cgroup_enable_task_cg_lists(void) | |
1313 | { | |
1314 | struct task_struct *p, *g; | |
1315 | ||
96d365e0 | 1316 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1317 | |
1318 | if (use_task_css_set_links) | |
1319 | goto out_unlock; | |
1320 | ||
1321 | use_task_css_set_links = true; | |
1322 | ||
1323 | /* | |
1324 | * We need tasklist_lock because RCU is not safe against | |
1325 | * while_each_thread(). Besides, a forking task that has passed | |
1326 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1327 | * is not guaranteed to have its child immediately visible in the | |
1328 | * tasklist if we walk through it with RCU. | |
1329 | */ | |
1330 | read_lock(&tasklist_lock); | |
1331 | do_each_thread(g, p) { | |
1332 | task_lock(p); | |
1333 | ||
1334 | WARN_ON_ONCE(!list_empty(&p->cg_list) || | |
1335 | task_css_set(p) != &init_css_set); | |
1336 | ||
1337 | /* | |
1338 | * We should check if the process is exiting, otherwise | |
1339 | * it will race with cgroup_exit() in that the list | |
1340 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1341 | * Do it while holding siglock so that we don't end up |
1342 | * racing against cgroup_exit(). | |
afeb0f9f | 1343 | */ |
f153ad11 | 1344 | spin_lock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1345 | if (!(p->flags & PF_EXITING)) |
1346 | list_add(&p->cg_list, &task_css_set(p)->tasks); | |
f153ad11 | 1347 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1348 | |
1349 | task_unlock(p); | |
1350 | } while_each_thread(g, p); | |
1351 | read_unlock(&tasklist_lock); | |
1352 | out_unlock: | |
96d365e0 | 1353 | up_write(&css_set_rwsem); |
afeb0f9f TH |
1354 | } |
1355 | ||
cc31edce PM |
1356 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1357 | { | |
2bd59d48 | 1358 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1359 | INIT_LIST_HEAD(&cgrp->sibling); |
1360 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1361 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1362 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1363 | INIT_LIST_HEAD(&cgrp->pidlists); |
1364 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1365 | cgrp->dummy_css.cgroup = cgrp; |
cc31edce | 1366 | } |
c6d57f33 | 1367 | |
ddbcc7e8 PM |
1368 | static void init_cgroup_root(struct cgroupfs_root *root) |
1369 | { | |
bd89aabc | 1370 | struct cgroup *cgrp = &root->top_cgroup; |
b0ca5a84 | 1371 | |
ddbcc7e8 | 1372 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1373 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1374 | cgrp->root = root; |
cc31edce | 1375 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1376 | idr_init(&root->cgroup_idr); |
ddbcc7e8 PM |
1377 | } |
1378 | ||
c6d57f33 PM |
1379 | static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts) |
1380 | { | |
1381 | struct cgroupfs_root *root; | |
1382 | ||
a1a71b45 | 1383 | if (!opts->subsys_mask && !opts->none) |
2bd59d48 | 1384 | return ERR_PTR(-EINVAL); |
c6d57f33 PM |
1385 | |
1386 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
1387 | if (!root) | |
1388 | return ERR_PTR(-ENOMEM); | |
1389 | ||
1390 | init_cgroup_root(root); | |
2c6ab6d2 | 1391 | |
c6d57f33 PM |
1392 | root->flags = opts->flags; |
1393 | if (opts->release_agent) | |
1394 | strcpy(root->release_agent_path, opts->release_agent); | |
1395 | if (opts->name) | |
1396 | strcpy(root->name, opts->name); | |
2260e7fc TH |
1397 | if (opts->cpuset_clone_children) |
1398 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags); | |
c6d57f33 PM |
1399 | return root; |
1400 | } | |
1401 | ||
35585573 | 1402 | static int cgroup_setup_root(struct cgroupfs_root *root, unsigned long ss_mask) |
d427dfeb TH |
1403 | { |
1404 | LIST_HEAD(tmp_links); | |
d427dfeb | 1405 | struct cgroup *root_cgrp = &root->top_cgroup; |
d427dfeb | 1406 | struct css_set *cset; |
d427dfeb TH |
1407 | int i, ret; |
1408 | ||
1409 | lockdep_assert_held(&cgroup_tree_mutex); | |
1410 | lockdep_assert_held(&cgroup_mutex); | |
d427dfeb TH |
1411 | |
1412 | ret = idr_alloc(&root->cgroup_idr, root_cgrp, 0, 1, GFP_KERNEL); | |
1413 | if (ret < 0) | |
2bd59d48 | 1414 | goto out; |
d427dfeb TH |
1415 | root_cgrp->id = ret; |
1416 | ||
d427dfeb | 1417 | /* |
96d365e0 | 1418 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1419 | * but that's OK - it can only be increased by someone holding |
1420 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1421 | * have some link structures left over | |
1422 | */ | |
1423 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1424 | if (ret) | |
2bd59d48 | 1425 | goto out; |
d427dfeb TH |
1426 | |
1427 | /* ID 0 is reserved for dummy root, 1 for unified hierarchy */ | |
1428 | ret = cgroup_init_root_id(root, 2, 0); | |
1429 | if (ret) | |
2bd59d48 | 1430 | goto out; |
d427dfeb | 1431 | |
2bd59d48 TH |
1432 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1433 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1434 | root_cgrp); | |
1435 | if (IS_ERR(root->kf_root)) { | |
1436 | ret = PTR_ERR(root->kf_root); | |
1437 | goto exit_root_id; | |
1438 | } | |
1439 | root_cgrp->kn = root->kf_root->kn; | |
d427dfeb TH |
1440 | |
1441 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); | |
1442 | if (ret) | |
2bd59d48 | 1443 | goto destroy_root; |
d427dfeb | 1444 | |
35585573 | 1445 | ret = rebind_subsystems(root, ss_mask, 0); |
d427dfeb | 1446 | if (ret) |
2bd59d48 | 1447 | goto destroy_root; |
d427dfeb TH |
1448 | |
1449 | /* | |
1450 | * There must be no failure case after here, since rebinding takes | |
1451 | * care of subsystems' refcounts, which are explicitly dropped in | |
1452 | * the failure exit path. | |
1453 | */ | |
1454 | list_add(&root->root_list, &cgroup_roots); | |
1455 | cgroup_root_count++; | |
1456 | ||
1457 | /* | |
1458 | * Link the top cgroup in this hierarchy into all the css_set | |
1459 | * objects. | |
1460 | */ | |
96d365e0 | 1461 | down_write(&css_set_rwsem); |
d427dfeb TH |
1462 | hash_for_each(css_set_table, i, cset, hlist) |
1463 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1464 | up_write(&css_set_rwsem); |
d427dfeb TH |
1465 | |
1466 | BUG_ON(!list_empty(&root_cgrp->children)); | |
3c9c825b | 1467 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
d427dfeb | 1468 | |
2bd59d48 | 1469 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1470 | ret = 0; |
2bd59d48 | 1471 | goto out; |
d427dfeb | 1472 | |
2bd59d48 TH |
1473 | destroy_root: |
1474 | kernfs_destroy_root(root->kf_root); | |
1475 | root->kf_root = NULL; | |
1476 | exit_root_id: | |
d427dfeb | 1477 | cgroup_exit_root_id(root); |
2bd59d48 | 1478 | out: |
d427dfeb TH |
1479 | free_cgrp_cset_links(&tmp_links); |
1480 | return ret; | |
1481 | } | |
1482 | ||
f7e83571 | 1483 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1484 | int flags, const char *unused_dev_name, |
f7e83571 | 1485 | void *data) |
ddbcc7e8 | 1486 | { |
2bd59d48 | 1487 | struct cgroupfs_root *root; |
ddbcc7e8 | 1488 | struct cgroup_sb_opts opts; |
2bd59d48 | 1489 | struct dentry *dentry; |
8e30e2b8 | 1490 | int ret; |
56fde9e0 TH |
1491 | |
1492 | /* | |
1493 | * The first time anyone tries to mount a cgroup, enable the list | |
1494 | * linking each css_set to its tasks and fix up all existing tasks. | |
1495 | */ | |
1496 | if (!use_task_css_set_links) | |
1497 | cgroup_enable_task_cg_lists(); | |
776f02fa | 1498 | retry: |
8e30e2b8 | 1499 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1500 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1501 | |
1502 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1503 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1504 | if (ret) |
8e30e2b8 | 1505 | goto out_unlock; |
ddbcc7e8 | 1506 | |
2bd59d48 TH |
1507 | /* look for a matching existing root */ |
1508 | for_each_active_root(root) { | |
1509 | bool name_match = false; | |
ddbcc7e8 | 1510 | |
2bd59d48 TH |
1511 | /* |
1512 | * If we asked for a name then it must match. Also, if | |
1513 | * name matches but sybsys_mask doesn't, we should fail. | |
1514 | * Remember whether name matched. | |
1515 | */ | |
1516 | if (opts.name) { | |
1517 | if (strcmp(opts.name, root->name)) | |
1518 | continue; | |
1519 | name_match = true; | |
1520 | } | |
ddbcc7e8 | 1521 | |
c6d57f33 | 1522 | /* |
2bd59d48 TH |
1523 | * If we asked for subsystems (or explicitly for no |
1524 | * subsystems) then they must match. | |
c6d57f33 | 1525 | */ |
2bd59d48 TH |
1526 | if ((opts.subsys_mask || opts.none) && |
1527 | (opts.subsys_mask != root->subsys_mask)) { | |
1528 | if (!name_match) | |
1529 | continue; | |
1530 | ret = -EBUSY; | |
1531 | goto out_unlock; | |
1532 | } | |
873fe09e | 1533 | |
c7ba8287 | 1534 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb JL |
1535 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
1536 | pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n"); | |
1537 | ret = -EINVAL; | |
8e30e2b8 | 1538 | goto out_unlock; |
2a0ff3fb JL |
1539 | } else { |
1540 | pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n"); | |
1541 | } | |
873fe09e | 1542 | } |
2bd59d48 | 1543 | |
776f02fa TH |
1544 | /* |
1545 | * A root's lifetime is governed by its top cgroup. Zero | |
1546 | * ref indicate that the root is being destroyed. Wait for | |
1547 | * destruction to complete so that the subsystems are free. | |
1548 | * We can use wait_queue for the wait but this path is | |
1549 | * super cold. Let's just sleep for a bit and retry. | |
1550 | */ | |
1551 | if (!atomic_inc_not_zero(&root->top_cgroup.refcnt)) { | |
1552 | mutex_unlock(&cgroup_mutex); | |
1553 | mutex_unlock(&cgroup_tree_mutex); | |
6534fd6c LZ |
1554 | kfree(opts.release_agent); |
1555 | kfree(opts.name); | |
776f02fa TH |
1556 | msleep(10); |
1557 | goto retry; | |
1558 | } | |
1559 | ||
1560 | ret = 0; | |
2bd59d48 | 1561 | goto out_unlock; |
ddbcc7e8 PM |
1562 | } |
1563 | ||
2bd59d48 TH |
1564 | /* no such thing, create a new one */ |
1565 | root = cgroup_root_from_opts(&opts); | |
1566 | if (IS_ERR(root)) { | |
1567 | ret = PTR_ERR(root); | |
1568 | goto out_unlock; | |
1569 | } | |
1570 | ||
35585573 | 1571 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1572 | if (ret) |
1573 | cgroup_free_root(root); | |
1574 | ||
8e30e2b8 | 1575 | out_unlock: |
e25e2cbb | 1576 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1577 | mutex_unlock(&cgroup_tree_mutex); |
8e30e2b8 | 1578 | |
c6d57f33 PM |
1579 | kfree(opts.release_agent); |
1580 | kfree(opts.name); | |
8e30e2b8 | 1581 | |
2bd59d48 | 1582 | if (ret) |
8e30e2b8 | 1583 | return ERR_PTR(ret); |
2bd59d48 TH |
1584 | |
1585 | dentry = kernfs_mount(fs_type, flags, root->kf_root); | |
1586 | if (IS_ERR(dentry)) | |
776f02fa | 1587 | cgroup_put(&root->top_cgroup); |
2bd59d48 TH |
1588 | return dentry; |
1589 | } | |
1590 | ||
1591 | static void cgroup_kill_sb(struct super_block *sb) | |
1592 | { | |
1593 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
1594 | struct cgroupfs_root *root = cgroup_root_from_kf(kf_root); | |
1595 | ||
776f02fa | 1596 | cgroup_put(&root->top_cgroup); |
2bd59d48 | 1597 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1598 | } |
1599 | ||
ddbcc7e8 PM |
1600 | static struct file_system_type cgroup_fs_type = { |
1601 | .name = "cgroup", | |
f7e83571 | 1602 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1603 | .kill_sb = cgroup_kill_sb, |
1604 | }; | |
1605 | ||
676db4af GK |
1606 | static struct kobject *cgroup_kobj; |
1607 | ||
857a2beb | 1608 | /** |
913ffdb5 | 1609 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1610 | * @task: target task |
857a2beb TH |
1611 | * @buf: the buffer to write the path into |
1612 | * @buflen: the length of the buffer | |
1613 | * | |
913ffdb5 TH |
1614 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1615 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1616 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1617 | * cgroup controller callbacks. | |
1618 | * | |
e61734c5 | 1619 | * Return value is the same as kernfs_path(). |
857a2beb | 1620 | */ |
e61734c5 | 1621 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb TH |
1622 | { |
1623 | struct cgroupfs_root *root; | |
913ffdb5 | 1624 | struct cgroup *cgrp; |
e61734c5 TH |
1625 | int hierarchy_id = 1; |
1626 | char *path = NULL; | |
857a2beb TH |
1627 | |
1628 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1629 | down_read(&css_set_rwsem); |
857a2beb | 1630 | |
913ffdb5 TH |
1631 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1632 | ||
857a2beb TH |
1633 | if (root) { |
1634 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1635 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1636 | } else { |
1637 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1638 | if (strlcpy(buf, "/", buflen) < buflen) |
1639 | path = buf; | |
857a2beb TH |
1640 | } |
1641 | ||
96d365e0 | 1642 | up_read(&css_set_rwsem); |
857a2beb | 1643 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1644 | return path; |
857a2beb | 1645 | } |
913ffdb5 | 1646 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1647 | |
b3dc094e | 1648 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 1649 | struct cgroup_taskset { |
b3dc094e TH |
1650 | /* the src and dst cset list running through cset->mg_node */ |
1651 | struct list_head src_csets; | |
1652 | struct list_head dst_csets; | |
1653 | ||
1654 | /* | |
1655 | * Fields for cgroup_taskset_*() iteration. | |
1656 | * | |
1657 | * Before migration is committed, the target migration tasks are on | |
1658 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
1659 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
1660 | * or ->dst_csets depending on whether migration is committed. | |
1661 | * | |
1662 | * ->cur_csets and ->cur_task point to the current task position | |
1663 | * during iteration. | |
1664 | */ | |
1665 | struct list_head *csets; | |
1666 | struct css_set *cur_cset; | |
1667 | struct task_struct *cur_task; | |
2f7ee569 TH |
1668 | }; |
1669 | ||
1670 | /** | |
1671 | * cgroup_taskset_first - reset taskset and return the first task | |
1672 | * @tset: taskset of interest | |
1673 | * | |
1674 | * @tset iteration is initialized and the first task is returned. | |
1675 | */ | |
1676 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1677 | { | |
b3dc094e TH |
1678 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
1679 | tset->cur_task = NULL; | |
1680 | ||
1681 | return cgroup_taskset_next(tset); | |
2f7ee569 | 1682 | } |
2f7ee569 TH |
1683 | |
1684 | /** | |
1685 | * cgroup_taskset_next - iterate to the next task in taskset | |
1686 | * @tset: taskset of interest | |
1687 | * | |
1688 | * Return the next task in @tset. Iteration must have been initialized | |
1689 | * with cgroup_taskset_first(). | |
1690 | */ | |
1691 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1692 | { | |
b3dc094e TH |
1693 | struct css_set *cset = tset->cur_cset; |
1694 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 1695 | |
b3dc094e TH |
1696 | while (&cset->mg_node != tset->csets) { |
1697 | if (!task) | |
1698 | task = list_first_entry(&cset->mg_tasks, | |
1699 | struct task_struct, cg_list); | |
1700 | else | |
1701 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 1702 | |
b3dc094e TH |
1703 | if (&task->cg_list != &cset->mg_tasks) { |
1704 | tset->cur_cset = cset; | |
1705 | tset->cur_task = task; | |
1706 | return task; | |
1707 | } | |
1708 | ||
1709 | cset = list_next_entry(cset, mg_node); | |
1710 | task = NULL; | |
1711 | } | |
1712 | ||
1713 | return NULL; | |
2f7ee569 | 1714 | } |
2f7ee569 | 1715 | |
cb0f1fe9 | 1716 | /** |
74a1166d | 1717 | * cgroup_task_migrate - move a task from one cgroup to another. |
cb0f1fe9 TH |
1718 | * @old_cgrp; the cgroup @tsk is being migrated from |
1719 | * @tsk: the task being migrated | |
1720 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 1721 | * |
cb0f1fe9 | 1722 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 1723 | */ |
5abb8855 TH |
1724 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1725 | struct task_struct *tsk, | |
1726 | struct css_set *new_cset) | |
74a1166d | 1727 | { |
5abb8855 | 1728 | struct css_set *old_cset; |
74a1166d | 1729 | |
cb0f1fe9 TH |
1730 | lockdep_assert_held(&cgroup_mutex); |
1731 | lockdep_assert_held(&css_set_rwsem); | |
1732 | ||
74a1166d | 1733 | /* |
026085ef MSB |
1734 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1735 | * setting such that we can't race against cgroup_exit() changing the | |
1736 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1737 | */ |
c84cdf75 | 1738 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1739 | old_cset = task_css_set(tsk); |
74a1166d | 1740 | |
b3dc094e TH |
1741 | get_css_set(new_cset); |
1742 | ||
74a1166d | 1743 | task_lock(tsk); |
5abb8855 | 1744 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d BB |
1745 | task_unlock(tsk); |
1746 | ||
b3dc094e | 1747 | list_move(&tsk->cg_list, &new_cset->mg_tasks); |
74a1166d BB |
1748 | |
1749 | /* | |
5abb8855 TH |
1750 | * We just gained a reference on old_cset by taking it from the |
1751 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1752 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1753 | */ |
5abb8855 | 1754 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
cb0f1fe9 | 1755 | put_css_set_locked(old_cset, false); |
74a1166d BB |
1756 | } |
1757 | ||
a043e3b2 | 1758 | /** |
1958d2d5 TH |
1759 | * cgroup_migrate_finish - cleanup after attach |
1760 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 1761 | * |
1958d2d5 TH |
1762 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
1763 | * those functions for details. | |
74a1166d | 1764 | */ |
1958d2d5 TH |
1765 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
1766 | { | |
1767 | struct css_set *cset, *tmp_cset; | |
1768 | ||
1769 | lockdep_assert_held(&cgroup_mutex); | |
1770 | ||
1771 | down_write(&css_set_rwsem); | |
1772 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
1773 | cset->mg_src_cgrp = NULL; | |
1774 | cset->mg_dst_cset = NULL; | |
1775 | list_del_init(&cset->mg_preload_node); | |
1776 | put_css_set_locked(cset, false); | |
1777 | } | |
1778 | up_write(&css_set_rwsem); | |
1779 | } | |
1780 | ||
1781 | /** | |
1782 | * cgroup_migrate_add_src - add a migration source css_set | |
1783 | * @src_cset: the source css_set to add | |
1784 | * @dst_cgrp: the destination cgroup | |
1785 | * @preloaded_csets: list of preloaded css_sets | |
1786 | * | |
1787 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
1788 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
1789 | * up by cgroup_migrate_finish(). | |
1790 | * | |
1791 | * This function may be called without holding threadgroup_lock even if the | |
1792 | * target is a process. Threads may be created and destroyed but as long | |
1793 | * as cgroup_mutex is not dropped, no new css_set can be put into play and | |
1794 | * the preloaded css_sets are guaranteed to cover all migrations. | |
1795 | */ | |
1796 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
1797 | struct cgroup *dst_cgrp, | |
1798 | struct list_head *preloaded_csets) | |
1799 | { | |
1800 | struct cgroup *src_cgrp; | |
1801 | ||
1802 | lockdep_assert_held(&cgroup_mutex); | |
1803 | lockdep_assert_held(&css_set_rwsem); | |
1804 | ||
1805 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
1806 | ||
1807 | /* nothing to do if this cset already belongs to the cgroup */ | |
1808 | if (src_cgrp == dst_cgrp) | |
1809 | return; | |
1810 | ||
1811 | if (!list_empty(&src_cset->mg_preload_node)) | |
1812 | return; | |
1813 | ||
1814 | WARN_ON(src_cset->mg_src_cgrp); | |
1815 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
1816 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
1817 | ||
1818 | src_cset->mg_src_cgrp = src_cgrp; | |
1819 | get_css_set(src_cset); | |
1820 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
1821 | } | |
1822 | ||
1823 | /** | |
1824 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
1825 | * @dst_cgrp: the destination cgroup | |
1826 | * @preloaded_csets: list of preloaded source css_sets | |
1827 | * | |
1828 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
1829 | * have been preloaded to @preloaded_csets. This function looks up and | |
1830 | * pins all destination css_sets, links each to its source, and put them on | |
1831 | * @preloaded_csets. | |
1832 | * | |
1833 | * This function must be called after cgroup_migrate_add_src() has been | |
1834 | * called on each migration source css_set. After migration is performed | |
1835 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
1836 | * @preloaded_csets. | |
1837 | */ | |
1838 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
1839 | struct list_head *preloaded_csets) | |
1840 | { | |
1841 | LIST_HEAD(csets); | |
1842 | struct css_set *src_cset; | |
1843 | ||
1844 | lockdep_assert_held(&cgroup_mutex); | |
1845 | ||
1846 | /* look up the dst cset for each src cset and link it to src */ | |
1847 | list_for_each_entry(src_cset, preloaded_csets, mg_preload_node) { | |
1848 | struct css_set *dst_cset; | |
1849 | ||
1850 | dst_cset = find_css_set(src_cset, dst_cgrp); | |
1851 | if (!dst_cset) | |
1852 | goto err; | |
1853 | ||
1854 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
1855 | src_cset->mg_dst_cset = dst_cset; | |
1856 | ||
1857 | if (list_empty(&dst_cset->mg_preload_node)) | |
1858 | list_add(&dst_cset->mg_preload_node, &csets); | |
1859 | else | |
1860 | put_css_set(dst_cset, false); | |
1861 | } | |
1862 | ||
1863 | list_splice(&csets, preloaded_csets); | |
1864 | return 0; | |
1865 | err: | |
1866 | cgroup_migrate_finish(&csets); | |
1867 | return -ENOMEM; | |
1868 | } | |
1869 | ||
1870 | /** | |
1871 | * cgroup_migrate - migrate a process or task to a cgroup | |
1872 | * @cgrp: the destination cgroup | |
1873 | * @leader: the leader of the process or the task to migrate | |
1874 | * @threadgroup: whether @leader points to the whole process or a single task | |
1875 | * | |
1876 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
1877 | * process, the caller must be holding threadgroup_lock of @leader. The | |
1878 | * caller is also responsible for invoking cgroup_migrate_add_src() and | |
1879 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
1880 | * function and following up with cgroup_migrate_finish(). | |
1881 | * | |
1882 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
1883 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
1884 | * failure, when migrating multiple targets, the success or failure can be | |
1885 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
1886 | * actually starting migrating. | |
1887 | */ | |
1888 | static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader, | |
1889 | bool threadgroup) | |
74a1166d | 1890 | { |
b3dc094e TH |
1891 | struct cgroup_taskset tset = { |
1892 | .src_csets = LIST_HEAD_INIT(tset.src_csets), | |
1893 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), | |
1894 | .csets = &tset.src_csets, | |
1895 | }; | |
1c6727af | 1896 | struct cgroup_subsys_state *css, *failed_css = NULL; |
b3dc094e TH |
1897 | struct css_set *cset, *tmp_cset; |
1898 | struct task_struct *task, *tmp_task; | |
1899 | int i, ret; | |
74a1166d | 1900 | |
fb5d2b4c MSB |
1901 | /* |
1902 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
1903 | * already PF_EXITING could be freed from underneath us unless we | |
1904 | * take an rcu_read_lock. | |
1905 | */ | |
b3dc094e | 1906 | down_write(&css_set_rwsem); |
fb5d2b4c | 1907 | rcu_read_lock(); |
9db8de37 | 1908 | task = leader; |
74a1166d | 1909 | do { |
9db8de37 TH |
1910 | /* @task either already exited or can't exit until the end */ |
1911 | if (task->flags & PF_EXITING) | |
ea84753c | 1912 | goto next; |
cd3d0952 | 1913 | |
b3dc094e | 1914 | cset = task_css_set(task); |
1958d2d5 | 1915 | if (!cset->mg_src_cgrp) |
ea84753c | 1916 | goto next; |
b3dc094e | 1917 | |
b3dc094e | 1918 | list_move(&task->cg_list, &cset->mg_tasks); |
1958d2d5 TH |
1919 | list_move(&cset->mg_node, &tset.src_csets); |
1920 | list_move(&cset->mg_dst_cset->mg_node, &tset.dst_csets); | |
ea84753c | 1921 | next: |
081aa458 LZ |
1922 | if (!threadgroup) |
1923 | break; | |
9db8de37 | 1924 | } while_each_thread(leader, task); |
fb5d2b4c | 1925 | rcu_read_unlock(); |
b3dc094e | 1926 | up_write(&css_set_rwsem); |
74a1166d | 1927 | |
134d3373 | 1928 | /* methods shouldn't be called if no task is actually migrating */ |
b3dc094e TH |
1929 | if (list_empty(&tset.src_csets)) |
1930 | return 0; | |
134d3373 | 1931 | |
1958d2d5 | 1932 | /* check that we can legitimately attach to the cgroup */ |
1c6727af TH |
1933 | for_each_css(css, i, cgrp) { |
1934 | if (css->ss->can_attach) { | |
9db8de37 TH |
1935 | ret = css->ss->can_attach(css, &tset); |
1936 | if (ret) { | |
1c6727af | 1937 | failed_css = css; |
74a1166d BB |
1938 | goto out_cancel_attach; |
1939 | } | |
1940 | } | |
74a1166d BB |
1941 | } |
1942 | ||
1943 | /* | |
1958d2d5 TH |
1944 | * Now that we're guaranteed success, proceed to move all tasks to |
1945 | * the new cgroup. There are no failure cases after here, so this | |
1946 | * is the commit point. | |
74a1166d | 1947 | */ |
cb0f1fe9 | 1948 | down_write(&css_set_rwsem); |
b3dc094e TH |
1949 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
1950 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
1951 | cgroup_task_migrate(cset->mg_src_cgrp, task, | |
1952 | cset->mg_dst_cset); | |
74a1166d | 1953 | } |
cb0f1fe9 | 1954 | up_write(&css_set_rwsem); |
b3dc094e | 1955 | |
74a1166d | 1956 | /* |
1958d2d5 TH |
1957 | * Migration is committed, all target tasks are now on dst_csets. |
1958 | * Nothing is sensitive to fork() after this point. Notify | |
1959 | * controllers that migration is complete. | |
74a1166d | 1960 | */ |
1958d2d5 TH |
1961 | tset.csets = &tset.dst_csets; |
1962 | ||
1c6727af TH |
1963 | for_each_css(css, i, cgrp) |
1964 | if (css->ss->attach) | |
1965 | css->ss->attach(css, &tset); | |
74a1166d | 1966 | |
9db8de37 | 1967 | ret = 0; |
b3dc094e TH |
1968 | goto out_release_tset; |
1969 | ||
74a1166d | 1970 | out_cancel_attach: |
b3dc094e TH |
1971 | for_each_css(css, i, cgrp) { |
1972 | if (css == failed_css) | |
1973 | break; | |
1974 | if (css->ss->cancel_attach) | |
1975 | css->ss->cancel_attach(css, &tset); | |
74a1166d | 1976 | } |
b3dc094e TH |
1977 | out_release_tset: |
1978 | down_write(&css_set_rwsem); | |
1979 | list_splice_init(&tset.dst_csets, &tset.src_csets); | |
1980 | list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { | |
1981 | list_splice_init(&cset->mg_tasks, &cset->tasks); | |
b3dc094e | 1982 | list_del_init(&cset->mg_node); |
b3dc094e TH |
1983 | } |
1984 | up_write(&css_set_rwsem); | |
9db8de37 | 1985 | return ret; |
74a1166d BB |
1986 | } |
1987 | ||
1958d2d5 TH |
1988 | /** |
1989 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
1990 | * @dst_cgrp: the cgroup to attach to | |
1991 | * @leader: the task or the leader of the threadgroup to be attached | |
1992 | * @threadgroup: attach the whole threadgroup? | |
1993 | * | |
1994 | * Call holding cgroup_mutex and the group_rwsem of the leader. Will take | |
1995 | * task_lock of @tsk or each thread in the threadgroup individually in turn. | |
1996 | */ | |
1997 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
1998 | struct task_struct *leader, bool threadgroup) | |
1999 | { | |
2000 | LIST_HEAD(preloaded_csets); | |
2001 | struct task_struct *task; | |
2002 | int ret; | |
2003 | ||
2004 | /* look up all src csets */ | |
2005 | down_read(&css_set_rwsem); | |
2006 | rcu_read_lock(); | |
2007 | task = leader; | |
2008 | do { | |
2009 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2010 | &preloaded_csets); | |
2011 | if (!threadgroup) | |
2012 | break; | |
2013 | } while_each_thread(leader, task); | |
2014 | rcu_read_unlock(); | |
2015 | up_read(&css_set_rwsem); | |
2016 | ||
2017 | /* prepare dst csets and commit */ | |
2018 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2019 | if (!ret) | |
2020 | ret = cgroup_migrate(dst_cgrp, leader, threadgroup); | |
2021 | ||
2022 | cgroup_migrate_finish(&preloaded_csets); | |
2023 | return ret; | |
2024 | } | |
2025 | ||
74a1166d BB |
2026 | /* |
2027 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 TH |
2028 | * function to attach either it or all tasks in its threadgroup. Will lock |
2029 | * cgroup_mutex and threadgroup; may take task_lock of task. | |
bbcb81d0 | 2030 | */ |
74a1166d | 2031 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2032 | { |
bbcb81d0 | 2033 | struct task_struct *tsk; |
c69e8d9c | 2034 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2035 | int ret; |
2036 | ||
74a1166d BB |
2037 | if (!cgroup_lock_live_group(cgrp)) |
2038 | return -ENODEV; | |
2039 | ||
b78949eb MSB |
2040 | retry_find_task: |
2041 | rcu_read_lock(); | |
bbcb81d0 | 2042 | if (pid) { |
73507f33 | 2043 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2044 | if (!tsk) { |
2045 | rcu_read_unlock(); | |
dd4b0a46 | 2046 | ret = -ESRCH; |
b78949eb | 2047 | goto out_unlock_cgroup; |
bbcb81d0 | 2048 | } |
74a1166d BB |
2049 | /* |
2050 | * even if we're attaching all tasks in the thread group, we | |
2051 | * only need to check permissions on one of them. | |
2052 | */ | |
c69e8d9c | 2053 | tcred = __task_cred(tsk); |
14a590c3 EB |
2054 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2055 | !uid_eq(cred->euid, tcred->uid) && | |
2056 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2057 | rcu_read_unlock(); |
b78949eb MSB |
2058 | ret = -EACCES; |
2059 | goto out_unlock_cgroup; | |
bbcb81d0 | 2060 | } |
b78949eb MSB |
2061 | } else |
2062 | tsk = current; | |
cd3d0952 TH |
2063 | |
2064 | if (threadgroup) | |
b78949eb | 2065 | tsk = tsk->group_leader; |
c4c27fbd MG |
2066 | |
2067 | /* | |
14a40ffc | 2068 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2069 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2070 | * with no rt_runtime allocated. Just say no. | |
2071 | */ | |
14a40ffc | 2072 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2073 | ret = -EINVAL; |
2074 | rcu_read_unlock(); | |
2075 | goto out_unlock_cgroup; | |
2076 | } | |
2077 | ||
b78949eb MSB |
2078 | get_task_struct(tsk); |
2079 | rcu_read_unlock(); | |
2080 | ||
2081 | threadgroup_lock(tsk); | |
2082 | if (threadgroup) { | |
2083 | if (!thread_group_leader(tsk)) { | |
2084 | /* | |
2085 | * a race with de_thread from another thread's exec() | |
2086 | * may strip us of our leadership, if this happens, | |
2087 | * there is no choice but to throw this task away and | |
2088 | * try again; this is | |
2089 | * "double-double-toil-and-trouble-check locking". | |
2090 | */ | |
2091 | threadgroup_unlock(tsk); | |
2092 | put_task_struct(tsk); | |
2093 | goto retry_find_task; | |
2094 | } | |
081aa458 LZ |
2095 | } |
2096 | ||
2097 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2098 | ||
cd3d0952 TH |
2099 | threadgroup_unlock(tsk); |
2100 | ||
bbcb81d0 | 2101 | put_task_struct(tsk); |
b78949eb | 2102 | out_unlock_cgroup: |
47cfcd09 | 2103 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
2104 | return ret; |
2105 | } | |
2106 | ||
7ae1bad9 TH |
2107 | /** |
2108 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2109 | * @from: attach to all cgroups of a given task | |
2110 | * @tsk: the task to be attached | |
2111 | */ | |
2112 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2113 | { | |
2114 | struct cgroupfs_root *root; | |
2115 | int retval = 0; | |
2116 | ||
47cfcd09 | 2117 | mutex_lock(&cgroup_mutex); |
7ae1bad9 | 2118 | for_each_active_root(root) { |
96d365e0 TH |
2119 | struct cgroup *from_cgrp; |
2120 | ||
2121 | down_read(&css_set_rwsem); | |
2122 | from_cgrp = task_cgroup_from_root(from, root); | |
2123 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2124 | |
6f4b7e63 | 2125 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2126 | if (retval) |
2127 | break; | |
2128 | } | |
47cfcd09 | 2129 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2130 | |
2131 | return retval; | |
2132 | } | |
2133 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2134 | ||
182446d0 TH |
2135 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2136 | struct cftype *cft, u64 pid) | |
74a1166d | 2137 | { |
182446d0 | 2138 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2139 | } |
2140 | ||
182446d0 TH |
2141 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2142 | struct cftype *cft, u64 tgid) | |
af351026 | 2143 | { |
182446d0 | 2144 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2145 | } |
2146 | ||
182446d0 TH |
2147 | static int cgroup_release_agent_write(struct cgroup_subsys_state *css, |
2148 | struct cftype *cft, const char *buffer) | |
e788e066 | 2149 | { |
5f469907 TH |
2150 | struct cgroupfs_root *root = css->cgroup->root; |
2151 | ||
2152 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
182446d0 | 2153 | if (!cgroup_lock_live_group(css->cgroup)) |
e788e066 | 2154 | return -ENODEV; |
69e943b7 | 2155 | spin_lock(&release_agent_path_lock); |
5f469907 TH |
2156 | strlcpy(root->release_agent_path, buffer, |
2157 | sizeof(root->release_agent_path)); | |
69e943b7 | 2158 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2159 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2160 | return 0; |
2161 | } | |
2162 | ||
2da8ca82 | 2163 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2164 | { |
2da8ca82 | 2165 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2166 | |
e788e066 PM |
2167 | if (!cgroup_lock_live_group(cgrp)) |
2168 | return -ENODEV; | |
2169 | seq_puts(seq, cgrp->root->release_agent_path); | |
2170 | seq_putc(seq, '\n'); | |
47cfcd09 | 2171 | mutex_unlock(&cgroup_mutex); |
e788e066 PM |
2172 | return 0; |
2173 | } | |
2174 | ||
2da8ca82 | 2175 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2176 | { |
2da8ca82 TH |
2177 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2178 | ||
2179 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2180 | return 0; |
2181 | } | |
2182 | ||
2bd59d48 TH |
2183 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2184 | size_t nbytes, loff_t off) | |
355e0c48 | 2185 | { |
2bd59d48 TH |
2186 | struct cgroup *cgrp = of->kn->parent->priv; |
2187 | struct cftype *cft = of->kn->priv; | |
2188 | struct cgroup_subsys_state *css; | |
a742c59d | 2189 | int ret; |
355e0c48 | 2190 | |
2bd59d48 TH |
2191 | /* |
2192 | * kernfs guarantees that a file isn't deleted with operations in | |
2193 | * flight, which means that the matching css is and stays alive and | |
2194 | * doesn't need to be pinned. The RCU locking is not necessary | |
2195 | * either. It's just for the convenience of using cgroup_css(). | |
2196 | */ | |
2197 | rcu_read_lock(); | |
2198 | css = cgroup_css(cgrp, cft->ss); | |
2199 | rcu_read_unlock(); | |
a742c59d TH |
2200 | |
2201 | if (cft->write_string) { | |
2202 | ret = cft->write_string(css, cft, strstrip(buf)); | |
2203 | } else if (cft->write_u64) { | |
2204 | unsigned long long v; | |
2205 | ret = kstrtoull(buf, 0, &v); | |
2206 | if (!ret) | |
2207 | ret = cft->write_u64(css, cft, v); | |
2208 | } else if (cft->write_s64) { | |
2209 | long long v; | |
2210 | ret = kstrtoll(buf, 0, &v); | |
2211 | if (!ret) | |
2212 | ret = cft->write_s64(css, cft, v); | |
2213 | } else if (cft->trigger) { | |
2214 | ret = cft->trigger(css, (unsigned int)cft->private); | |
e73d2c61 | 2215 | } else { |
a742c59d | 2216 | ret = -EINVAL; |
e73d2c61 | 2217 | } |
2bd59d48 | 2218 | |
a742c59d | 2219 | return ret ?: nbytes; |
355e0c48 PM |
2220 | } |
2221 | ||
6612f05b | 2222 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2223 | { |
2bd59d48 | 2224 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2225 | } |
2226 | ||
6612f05b | 2227 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2228 | { |
2bd59d48 | 2229 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2230 | } |
2231 | ||
6612f05b | 2232 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2233 | { |
2bd59d48 | 2234 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2235 | } |
2236 | ||
91796569 | 2237 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2238 | { |
7da11279 TH |
2239 | struct cftype *cft = seq_cft(m); |
2240 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2241 | |
2da8ca82 TH |
2242 | if (cft->seq_show) |
2243 | return cft->seq_show(m, arg); | |
e73d2c61 | 2244 | |
f4c753b7 | 2245 | if (cft->read_u64) |
896f5199 TH |
2246 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2247 | else if (cft->read_s64) | |
2248 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2249 | else | |
2250 | return -EINVAL; | |
2251 | return 0; | |
91796569 PM |
2252 | } |
2253 | ||
2bd59d48 TH |
2254 | static struct kernfs_ops cgroup_kf_single_ops = { |
2255 | .atomic_write_len = PAGE_SIZE, | |
2256 | .write = cgroup_file_write, | |
2257 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2258 | }; |
2259 | ||
2bd59d48 TH |
2260 | static struct kernfs_ops cgroup_kf_ops = { |
2261 | .atomic_write_len = PAGE_SIZE, | |
2262 | .write = cgroup_file_write, | |
2263 | .seq_start = cgroup_seqfile_start, | |
2264 | .seq_next = cgroup_seqfile_next, | |
2265 | .seq_stop = cgroup_seqfile_stop, | |
2266 | .seq_show = cgroup_seqfile_show, | |
2267 | }; | |
ddbcc7e8 PM |
2268 | |
2269 | /* | |
2270 | * cgroup_rename - Only allow simple rename of directories in place. | |
2271 | */ | |
2bd59d48 TH |
2272 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2273 | const char *new_name_str) | |
ddbcc7e8 | 2274 | { |
2bd59d48 | 2275 | struct cgroup *cgrp = kn->priv; |
2bd59d48 | 2276 | int ret; |
65dff759 | 2277 | |
2bd59d48 | 2278 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2279 | return -ENOTDIR; |
2bd59d48 | 2280 | if (kn->parent != new_parent) |
ddbcc7e8 | 2281 | return -EIO; |
65dff759 | 2282 | |
6db8e85c TH |
2283 | /* |
2284 | * This isn't a proper migration and its usefulness is very | |
2285 | * limited. Disallow if sane_behavior. | |
2286 | */ | |
2287 | if (cgroup_sane_behavior(cgrp)) | |
2288 | return -EPERM; | |
2289 | ||
2bd59d48 TH |
2290 | mutex_lock(&cgroup_tree_mutex); |
2291 | mutex_lock(&cgroup_mutex); | |
2292 | ||
2293 | ret = kernfs_rename(kn, new_parent, new_name_str); | |
65dff759 | 2294 | |
2bd59d48 TH |
2295 | mutex_unlock(&cgroup_mutex); |
2296 | mutex_unlock(&cgroup_tree_mutex); | |
2bd59d48 | 2297 | return ret; |
ddbcc7e8 PM |
2298 | } |
2299 | ||
2bb566cb | 2300 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2301 | { |
8d7e6fb0 | 2302 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2303 | struct kernfs_node *kn; |
2304 | struct lock_class_key *key = NULL; | |
05ef1d7c | 2305 | |
2bd59d48 TH |
2306 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2307 | key = &cft->lockdep_key; | |
2308 | #endif | |
2309 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2310 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2311 | NULL, false, key); | |
430af8ad | 2312 | return PTR_ERR_OR_ZERO(kn); |
ddbcc7e8 PM |
2313 | } |
2314 | ||
b1f28d31 TH |
2315 | /** |
2316 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2317 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2318 | * @cfts: array of cftypes to be added |
2319 | * @is_add: whether to add or remove | |
2320 | * | |
2321 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2322 | * For removals, this function never fails. If addition fails, this |
2323 | * function doesn't remove files already added. The caller is responsible | |
2324 | * for cleaning up. | |
b1f28d31 | 2325 | */ |
2bb566cb TH |
2326 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2327 | bool is_add) | |
ddbcc7e8 | 2328 | { |
03b1cde6 | 2329 | struct cftype *cft; |
b1f28d31 TH |
2330 | int ret; |
2331 | ||
ace2bee8 | 2332 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2333 | |
2334 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2335 | /* does cft->flags tell us to skip this file on @cgrp? */ |
873fe09e TH |
2336 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2337 | continue; | |
f33fddc2 G |
2338 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2339 | continue; | |
2340 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2341 | continue; | |
2342 | ||
2739d3cc | 2343 | if (is_add) { |
2bb566cb | 2344 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2345 | if (ret) { |
2739d3cc | 2346 | pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n", |
b1f28d31 TH |
2347 | cft->name, ret); |
2348 | return ret; | |
2349 | } | |
2739d3cc LZ |
2350 | } else { |
2351 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2352 | } |
ddbcc7e8 | 2353 | } |
b1f28d31 | 2354 | return 0; |
ddbcc7e8 PM |
2355 | } |
2356 | ||
21a2d343 | 2357 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2358 | { |
2359 | LIST_HEAD(pending); | |
2bb566cb | 2360 | struct cgroup_subsys *ss = cfts[0].ss; |
492eb21b | 2361 | struct cgroup *root = &ss->root->top_cgroup; |
492eb21b | 2362 | struct cgroup_subsys_state *css; |
9ccece80 | 2363 | int ret = 0; |
8e3f6541 | 2364 | |
21a2d343 | 2365 | lockdep_assert_held(&cgroup_tree_mutex); |
4ac06017 | 2366 | |
21a2d343 TH |
2367 | /* don't bother if @ss isn't attached */ |
2368 | if (ss->root == &cgroup_dummy_root) | |
9ccece80 | 2369 | return 0; |
e8c82d20 | 2370 | |
e8c82d20 | 2371 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2372 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2373 | struct cgroup *cgrp = css->cgroup; |
2374 | ||
e8c82d20 LZ |
2375 | if (cgroup_is_dead(cgrp)) |
2376 | continue; | |
2377 | ||
21a2d343 | 2378 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2379 | if (ret) |
2380 | break; | |
8e3f6541 | 2381 | } |
21a2d343 TH |
2382 | |
2383 | if (is_add && !ret) | |
2384 | kernfs_activate(root->kn); | |
9ccece80 | 2385 | return ret; |
8e3f6541 TH |
2386 | } |
2387 | ||
2da440a2 TH |
2388 | static void cgroup_exit_cftypes(struct cftype *cfts) |
2389 | { | |
2390 | struct cftype *cft; | |
2391 | ||
2bd59d48 TH |
2392 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2393 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2394 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2395 | kfree(cft->kf_ops); | |
2396 | cft->kf_ops = NULL; | |
2da440a2 | 2397 | cft->ss = NULL; |
2bd59d48 | 2398 | } |
2da440a2 TH |
2399 | } |
2400 | ||
2bd59d48 | 2401 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2402 | { |
2403 | struct cftype *cft; | |
2404 | ||
2bd59d48 TH |
2405 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2406 | struct kernfs_ops *kf_ops; | |
2407 | ||
0adb0704 TH |
2408 | WARN_ON(cft->ss || cft->kf_ops); |
2409 | ||
2bd59d48 TH |
2410 | if (cft->seq_start) |
2411 | kf_ops = &cgroup_kf_ops; | |
2412 | else | |
2413 | kf_ops = &cgroup_kf_single_ops; | |
2414 | ||
2415 | /* | |
2416 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2417 | * make a copy of kf_ops to set its atomic_write_len. | |
2418 | */ | |
2419 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2420 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2421 | if (!kf_ops) { | |
2422 | cgroup_exit_cftypes(cfts); | |
2423 | return -ENOMEM; | |
2424 | } | |
2425 | kf_ops->atomic_write_len = cft->max_write_len; | |
2426 | } | |
2427 | ||
2428 | cft->kf_ops = kf_ops; | |
2da440a2 | 2429 | cft->ss = ss; |
2bd59d48 TH |
2430 | } |
2431 | ||
2432 | return 0; | |
2da440a2 TH |
2433 | } |
2434 | ||
21a2d343 TH |
2435 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
2436 | { | |
2437 | lockdep_assert_held(&cgroup_tree_mutex); | |
2438 | ||
2439 | if (!cfts || !cfts[0].ss) | |
2440 | return -ENOENT; | |
2441 | ||
2442 | list_del(&cfts->node); | |
2443 | cgroup_apply_cftypes(cfts, false); | |
2444 | cgroup_exit_cftypes(cfts); | |
2445 | return 0; | |
2446 | } | |
2447 | ||
80b13586 TH |
2448 | /** |
2449 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
2450 | * @cfts: zero-length name terminated array of cftypes | |
2451 | * | |
2452 | * Unregister @cfts. Files described by @cfts are removed from all | |
2453 | * existing cgroups and all future cgroups won't have them either. This | |
2454 | * function can be called anytime whether @cfts' subsys is attached or not. | |
2455 | * | |
2456 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2457 | * registered. | |
2458 | */ | |
2459 | int cgroup_rm_cftypes(struct cftype *cfts) | |
2460 | { | |
21a2d343 | 2461 | int ret; |
80b13586 | 2462 | |
21a2d343 TH |
2463 | mutex_lock(&cgroup_tree_mutex); |
2464 | ret = cgroup_rm_cftypes_locked(cfts); | |
2465 | mutex_unlock(&cgroup_tree_mutex); | |
2466 | return ret; | |
80b13586 TH |
2467 | } |
2468 | ||
8e3f6541 TH |
2469 | /** |
2470 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
2471 | * @ss: target cgroup subsystem | |
2472 | * @cfts: zero-length name terminated array of cftypes | |
2473 | * | |
2474 | * Register @cfts to @ss. Files described by @cfts are created for all | |
2475 | * existing cgroups to which @ss is attached and all future cgroups will | |
2476 | * have them too. This function can be called anytime whether @ss is | |
2477 | * attached or not. | |
2478 | * | |
2479 | * Returns 0 on successful registration, -errno on failure. Note that this | |
2480 | * function currently returns 0 as long as @cfts registration is successful | |
2481 | * even if some file creation attempts on existing cgroups fail. | |
2482 | */ | |
03b1cde6 | 2483 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 2484 | { |
9ccece80 | 2485 | int ret; |
8e3f6541 | 2486 | |
dc5736ed LZ |
2487 | if (!cfts || cfts[0].name[0] == '\0') |
2488 | return 0; | |
2489 | ||
2bd59d48 TH |
2490 | ret = cgroup_init_cftypes(ss, cfts); |
2491 | if (ret) | |
2492 | return ret; | |
2bb566cb | 2493 | |
21a2d343 TH |
2494 | mutex_lock(&cgroup_tree_mutex); |
2495 | ||
0adb0704 | 2496 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 2497 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 2498 | if (ret) |
21a2d343 TH |
2499 | cgroup_rm_cftypes_locked(cfts); |
2500 | ||
2501 | mutex_unlock(&cgroup_tree_mutex); | |
9ccece80 | 2502 | return ret; |
8e3f6541 | 2503 | } |
8e3f6541 | 2504 | |
a043e3b2 LZ |
2505 | /** |
2506 | * cgroup_task_count - count the number of tasks in a cgroup. | |
2507 | * @cgrp: the cgroup in question | |
2508 | * | |
2509 | * Return the number of tasks in the cgroup. | |
2510 | */ | |
07bc356e | 2511 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
2512 | { |
2513 | int count = 0; | |
69d0206c | 2514 | struct cgrp_cset_link *link; |
817929ec | 2515 | |
96d365e0 | 2516 | down_read(&css_set_rwsem); |
69d0206c TH |
2517 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
2518 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 2519 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
2520 | return count; |
2521 | } | |
2522 | ||
53fa5261 | 2523 | /** |
492eb21b TH |
2524 | * css_next_child - find the next child of a given css |
2525 | * @pos_css: the current position (%NULL to initiate traversal) | |
2526 | * @parent_css: css whose children to walk | |
53fa5261 | 2527 | * |
492eb21b | 2528 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
2529 | * under either cgroup_mutex or RCU read lock. The only requirement is |
2530 | * that @parent_css and @pos_css are accessible. The next sibling is | |
2531 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 2532 | */ |
492eb21b TH |
2533 | struct cgroup_subsys_state * |
2534 | css_next_child(struct cgroup_subsys_state *pos_css, | |
2535 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 2536 | { |
492eb21b TH |
2537 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
2538 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
2539 | struct cgroup *next; |
2540 | ||
ace2bee8 | 2541 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
2542 | |
2543 | /* | |
2544 | * @pos could already have been removed. Once a cgroup is removed, | |
2545 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
2546 | * changes. As CGRP_DEAD assertion is serialized and happens |
2547 | * before the cgroup is taken off the ->sibling list, if we see it | |
2548 | * unasserted, it's guaranteed that the next sibling hasn't | |
2549 | * finished its grace period even if it's already removed, and thus | |
2550 | * safe to dereference from this RCU critical section. If | |
2551 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
2552 | * to be visible as %true here. | |
3b287a50 TH |
2553 | * |
2554 | * If @pos is dead, its next pointer can't be dereferenced; | |
2555 | * however, as each cgroup is given a monotonically increasing | |
2556 | * unique serial number and always appended to the sibling list, | |
2557 | * the next one can be found by walking the parent's children until | |
2558 | * we see a cgroup with higher serial number than @pos's. While | |
2559 | * this path can be slower, it's taken only when either the current | |
2560 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 2561 | */ |
3b287a50 TH |
2562 | if (!pos) { |
2563 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
2564 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 2565 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
2566 | } else { |
2567 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
2568 | if (next->serial_nr > pos->serial_nr) | |
2569 | break; | |
53fa5261 TH |
2570 | } |
2571 | ||
492eb21b TH |
2572 | if (&next->sibling == &cgrp->children) |
2573 | return NULL; | |
2574 | ||
ca8bdcaf | 2575 | return cgroup_css(next, parent_css->ss); |
53fa5261 | 2576 | } |
53fa5261 | 2577 | |
574bd9f7 | 2578 | /** |
492eb21b | 2579 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 2580 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2581 | * @root: css whose descendants to walk |
574bd9f7 | 2582 | * |
492eb21b | 2583 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
2584 | * to visit for pre-order traversal of @root's descendants. @root is |
2585 | * included in the iteration and the first node to be visited. | |
75501a6d | 2586 | * |
87fb54f1 TH |
2587 | * While this function requires cgroup_mutex or RCU read locking, it |
2588 | * doesn't require the whole traversal to be contained in a single critical | |
2589 | * section. This function will return the correct next descendant as long | |
2590 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 2591 | */ |
492eb21b TH |
2592 | struct cgroup_subsys_state * |
2593 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
2594 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2595 | { |
492eb21b | 2596 | struct cgroup_subsys_state *next; |
574bd9f7 | 2597 | |
ace2bee8 | 2598 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2599 | |
bd8815a6 | 2600 | /* if first iteration, visit @root */ |
7805d000 | 2601 | if (!pos) |
bd8815a6 | 2602 | return root; |
574bd9f7 TH |
2603 | |
2604 | /* visit the first child if exists */ | |
492eb21b | 2605 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
2606 | if (next) |
2607 | return next; | |
2608 | ||
2609 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
2610 | while (pos != root) { |
2611 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 2612 | if (next) |
574bd9f7 | 2613 | return next; |
492eb21b | 2614 | pos = css_parent(pos); |
7805d000 | 2615 | } |
574bd9f7 TH |
2616 | |
2617 | return NULL; | |
2618 | } | |
574bd9f7 | 2619 | |
12a9d2fe | 2620 | /** |
492eb21b TH |
2621 | * css_rightmost_descendant - return the rightmost descendant of a css |
2622 | * @pos: css of interest | |
12a9d2fe | 2623 | * |
492eb21b TH |
2624 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
2625 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 2626 | * subtree of @pos. |
75501a6d | 2627 | * |
87fb54f1 TH |
2628 | * While this function requires cgroup_mutex or RCU read locking, it |
2629 | * doesn't require the whole traversal to be contained in a single critical | |
2630 | * section. This function will return the correct rightmost descendant as | |
2631 | * long as @pos is accessible. | |
12a9d2fe | 2632 | */ |
492eb21b TH |
2633 | struct cgroup_subsys_state * |
2634 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 2635 | { |
492eb21b | 2636 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 2637 | |
ace2bee8 | 2638 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
2639 | |
2640 | do { | |
2641 | last = pos; | |
2642 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
2643 | pos = NULL; | |
492eb21b | 2644 | css_for_each_child(tmp, last) |
12a9d2fe TH |
2645 | pos = tmp; |
2646 | } while (pos); | |
2647 | ||
2648 | return last; | |
2649 | } | |
12a9d2fe | 2650 | |
492eb21b TH |
2651 | static struct cgroup_subsys_state * |
2652 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 2653 | { |
492eb21b | 2654 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
2655 | |
2656 | do { | |
2657 | last = pos; | |
492eb21b | 2658 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
2659 | } while (pos); |
2660 | ||
2661 | return last; | |
2662 | } | |
2663 | ||
2664 | /** | |
492eb21b | 2665 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 2666 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 2667 | * @root: css whose descendants to walk |
574bd9f7 | 2668 | * |
492eb21b | 2669 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
2670 | * to visit for post-order traversal of @root's descendants. @root is |
2671 | * included in the iteration and the last node to be visited. | |
75501a6d | 2672 | * |
87fb54f1 TH |
2673 | * While this function requires cgroup_mutex or RCU read locking, it |
2674 | * doesn't require the whole traversal to be contained in a single critical | |
2675 | * section. This function will return the correct next descendant as long | |
2676 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
2677 | * @cgroup. | |
574bd9f7 | 2678 | */ |
492eb21b TH |
2679 | struct cgroup_subsys_state * |
2680 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
2681 | struct cgroup_subsys_state *root) | |
574bd9f7 | 2682 | { |
492eb21b | 2683 | struct cgroup_subsys_state *next; |
574bd9f7 | 2684 | |
ace2bee8 | 2685 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 2686 | |
58b79a91 TH |
2687 | /* if first iteration, visit leftmost descendant which may be @root */ |
2688 | if (!pos) | |
2689 | return css_leftmost_descendant(root); | |
574bd9f7 | 2690 | |
bd8815a6 TH |
2691 | /* if we visited @root, we're done */ |
2692 | if (pos == root) | |
2693 | return NULL; | |
2694 | ||
574bd9f7 | 2695 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 2696 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 2697 | if (next) |
492eb21b | 2698 | return css_leftmost_descendant(next); |
574bd9f7 TH |
2699 | |
2700 | /* no sibling left, visit parent */ | |
bd8815a6 | 2701 | return css_parent(pos); |
574bd9f7 | 2702 | } |
574bd9f7 | 2703 | |
0942eeee | 2704 | /** |
72ec7029 | 2705 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
2706 | * @it: the iterator to advance |
2707 | * | |
2708 | * Advance @it to the next css_set to walk. | |
d515876e | 2709 | */ |
72ec7029 | 2710 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e TH |
2711 | { |
2712 | struct list_head *l = it->cset_link; | |
2713 | struct cgrp_cset_link *link; | |
2714 | struct css_set *cset; | |
2715 | ||
2716 | /* Advance to the next non-empty css_set */ | |
2717 | do { | |
2718 | l = l->next; | |
72ec7029 | 2719 | if (l == &it->origin_css->cgroup->cset_links) { |
d515876e TH |
2720 | it->cset_link = NULL; |
2721 | return; | |
2722 | } | |
2723 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
2724 | cset = link->cset; | |
c7561128 TH |
2725 | } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); |
2726 | ||
d515876e | 2727 | it->cset_link = l; |
c7561128 TH |
2728 | |
2729 | if (!list_empty(&cset->tasks)) | |
2730 | it->task = cset->tasks.next; | |
2731 | else | |
2732 | it->task = cset->mg_tasks.next; | |
d515876e TH |
2733 | } |
2734 | ||
0942eeee | 2735 | /** |
72ec7029 TH |
2736 | * css_task_iter_start - initiate task iteration |
2737 | * @css: the css to walk tasks of | |
0942eeee TH |
2738 | * @it: the task iterator to use |
2739 | * | |
72ec7029 TH |
2740 | * Initiate iteration through the tasks of @css. The caller can call |
2741 | * css_task_iter_next() to walk through the tasks until the function | |
2742 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
2743 | * called. | |
0942eeee TH |
2744 | * |
2745 | * Note that this function acquires a lock which is released when the | |
2746 | * iteration finishes. The caller can't sleep while iteration is in | |
2747 | * progress. | |
2748 | */ | |
72ec7029 TH |
2749 | void css_task_iter_start(struct cgroup_subsys_state *css, |
2750 | struct css_task_iter *it) | |
96d365e0 | 2751 | __acquires(css_set_rwsem) |
817929ec | 2752 | { |
56fde9e0 TH |
2753 | /* no one should try to iterate before mounting cgroups */ |
2754 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 2755 | |
96d365e0 | 2756 | down_read(&css_set_rwsem); |
c59cd3d8 | 2757 | |
72ec7029 TH |
2758 | it->origin_css = css; |
2759 | it->cset_link = &css->cgroup->cset_links; | |
c59cd3d8 | 2760 | |
72ec7029 | 2761 | css_advance_task_iter(it); |
817929ec PM |
2762 | } |
2763 | ||
0942eeee | 2764 | /** |
72ec7029 | 2765 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
2766 | * @it: the task iterator being iterated |
2767 | * | |
2768 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
2769 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
2770 | * reaches the end. | |
0942eeee | 2771 | */ |
72ec7029 | 2772 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
2773 | { |
2774 | struct task_struct *res; | |
2775 | struct list_head *l = it->task; | |
c7561128 TH |
2776 | struct cgrp_cset_link *link = list_entry(it->cset_link, |
2777 | struct cgrp_cset_link, cset_link); | |
817929ec PM |
2778 | |
2779 | /* If the iterator cg is NULL, we have no tasks */ | |
69d0206c | 2780 | if (!it->cset_link) |
817929ec PM |
2781 | return NULL; |
2782 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
2783 | |
2784 | /* | |
2785 | * Advance iterator to find next entry. cset->tasks is consumed | |
2786 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
2787 | * next cset. | |
2788 | */ | |
817929ec | 2789 | l = l->next; |
c7561128 TH |
2790 | |
2791 | if (l == &link->cset->tasks) | |
2792 | l = link->cset->mg_tasks.next; | |
2793 | ||
2794 | if (l == &link->cset->mg_tasks) | |
72ec7029 | 2795 | css_advance_task_iter(it); |
c7561128 | 2796 | else |
817929ec | 2797 | it->task = l; |
c7561128 | 2798 | |
817929ec PM |
2799 | return res; |
2800 | } | |
2801 | ||
0942eeee | 2802 | /** |
72ec7029 | 2803 | * css_task_iter_end - finish task iteration |
0942eeee TH |
2804 | * @it: the task iterator to finish |
2805 | * | |
72ec7029 | 2806 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 2807 | */ |
72ec7029 | 2808 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 2809 | __releases(css_set_rwsem) |
817929ec | 2810 | { |
96d365e0 | 2811 | up_read(&css_set_rwsem); |
817929ec PM |
2812 | } |
2813 | ||
8cc99345 TH |
2814 | /** |
2815 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another | |
2816 | * @to: cgroup to which the tasks will be moved | |
2817 | * @from: cgroup in which the tasks currently reside | |
2818 | */ | |
2819 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) | |
2820 | { | |
e406d1cf TH |
2821 | struct css_task_iter it; |
2822 | struct task_struct *task; | |
2823 | int ret = 0; | |
2824 | ||
2825 | do { | |
2826 | css_task_iter_start(&from->dummy_css, &it); | |
2827 | task = css_task_iter_next(&it); | |
2828 | if (task) | |
2829 | get_task_struct(task); | |
2830 | css_task_iter_end(&it); | |
2831 | ||
2832 | if (task) { | |
2833 | mutex_lock(&cgroup_mutex); | |
2834 | ret = cgroup_attach_task(to, task, false); | |
2835 | mutex_unlock(&cgroup_mutex); | |
2836 | put_task_struct(task); | |
2837 | } | |
2838 | } while (task && !ret); | |
2839 | ||
2840 | return ret; | |
8cc99345 TH |
2841 | } |
2842 | ||
bbcb81d0 | 2843 | /* |
102a775e | 2844 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
2845 | * |
2846 | * Reading this file can return large amounts of data if a cgroup has | |
2847 | * *lots* of attached tasks. So it may need several calls to read(), | |
2848 | * but we cannot guarantee that the information we produce is correct | |
2849 | * unless we produce it entirely atomically. | |
2850 | * | |
bbcb81d0 | 2851 | */ |
bbcb81d0 | 2852 | |
24528255 LZ |
2853 | /* which pidlist file are we talking about? */ |
2854 | enum cgroup_filetype { | |
2855 | CGROUP_FILE_PROCS, | |
2856 | CGROUP_FILE_TASKS, | |
2857 | }; | |
2858 | ||
2859 | /* | |
2860 | * A pidlist is a list of pids that virtually represents the contents of one | |
2861 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
2862 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
2863 | * to the cgroup. | |
2864 | */ | |
2865 | struct cgroup_pidlist { | |
2866 | /* | |
2867 | * used to find which pidlist is wanted. doesn't change as long as | |
2868 | * this particular list stays in the list. | |
2869 | */ | |
2870 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
2871 | /* array of xids */ | |
2872 | pid_t *list; | |
2873 | /* how many elements the above list has */ | |
2874 | int length; | |
24528255 LZ |
2875 | /* each of these stored in a list by its cgroup */ |
2876 | struct list_head links; | |
2877 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
2878 | struct cgroup *owner; | |
b1a21367 TH |
2879 | /* for delayed destruction */ |
2880 | struct delayed_work destroy_dwork; | |
24528255 LZ |
2881 | }; |
2882 | ||
d1d9fd33 BB |
2883 | /* |
2884 | * The following two functions "fix" the issue where there are more pids | |
2885 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
2886 | * TODO: replace with a kernel-wide solution to this problem | |
2887 | */ | |
2888 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
2889 | static void *pidlist_allocate(int count) | |
2890 | { | |
2891 | if (PIDLIST_TOO_LARGE(count)) | |
2892 | return vmalloc(count * sizeof(pid_t)); | |
2893 | else | |
2894 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
2895 | } | |
b1a21367 | 2896 | |
d1d9fd33 BB |
2897 | static void pidlist_free(void *p) |
2898 | { | |
2899 | if (is_vmalloc_addr(p)) | |
2900 | vfree(p); | |
2901 | else | |
2902 | kfree(p); | |
2903 | } | |
d1d9fd33 | 2904 | |
b1a21367 TH |
2905 | /* |
2906 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
2907 | * should be left afterwards. | |
2908 | */ | |
2909 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
2910 | { | |
2911 | struct cgroup_pidlist *l, *tmp_l; | |
2912 | ||
2913 | mutex_lock(&cgrp->pidlist_mutex); | |
2914 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
2915 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
2916 | mutex_unlock(&cgrp->pidlist_mutex); | |
2917 | ||
2918 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
2919 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
2920 | } | |
2921 | ||
2922 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
2923 | { | |
2924 | struct delayed_work *dwork = to_delayed_work(work); | |
2925 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
2926 | destroy_dwork); | |
2927 | struct cgroup_pidlist *tofree = NULL; | |
2928 | ||
2929 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
2930 | |
2931 | /* | |
04502365 TH |
2932 | * Destroy iff we didn't get queued again. The state won't change |
2933 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 2934 | */ |
04502365 | 2935 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
2936 | list_del(&l->links); |
2937 | pidlist_free(l->list); | |
2938 | put_pid_ns(l->key.ns); | |
2939 | tofree = l; | |
2940 | } | |
2941 | ||
b1a21367 TH |
2942 | mutex_unlock(&l->owner->pidlist_mutex); |
2943 | kfree(tofree); | |
2944 | } | |
2945 | ||
bbcb81d0 | 2946 | /* |
102a775e | 2947 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 2948 | * Returns the number of unique elements. |
bbcb81d0 | 2949 | */ |
6ee211ad | 2950 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 2951 | { |
102a775e | 2952 | int src, dest = 1; |
102a775e BB |
2953 | |
2954 | /* | |
2955 | * we presume the 0th element is unique, so i starts at 1. trivial | |
2956 | * edge cases first; no work needs to be done for either | |
2957 | */ | |
2958 | if (length == 0 || length == 1) | |
2959 | return length; | |
2960 | /* src and dest walk down the list; dest counts unique elements */ | |
2961 | for (src = 1; src < length; src++) { | |
2962 | /* find next unique element */ | |
2963 | while (list[src] == list[src-1]) { | |
2964 | src++; | |
2965 | if (src == length) | |
2966 | goto after; | |
2967 | } | |
2968 | /* dest always points to where the next unique element goes */ | |
2969 | list[dest] = list[src]; | |
2970 | dest++; | |
2971 | } | |
2972 | after: | |
102a775e BB |
2973 | return dest; |
2974 | } | |
2975 | ||
afb2bc14 TH |
2976 | /* |
2977 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
2978 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
2979 | * different per namespace, each namespace needs differently sorted list, | |
2980 | * making it impossible to use, for example, single rbtree of member tasks | |
2981 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
2982 | * per open file is dangerous, so cgroup had to implement shared pool of | |
2983 | * pidlists keyed by cgroup and namespace. | |
2984 | * | |
2985 | * All this extra complexity was caused by the original implementation | |
2986 | * committing to an entirely unnecessary property. In the long term, we | |
2987 | * want to do away with it. Explicitly scramble sort order if | |
2988 | * sane_behavior so that no such expectation exists in the new interface. | |
2989 | * | |
2990 | * Scrambling is done by swapping every two consecutive bits, which is | |
2991 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
2992 | */ | |
2993 | static pid_t pid_fry(pid_t pid) | |
2994 | { | |
2995 | unsigned a = pid & 0x55555555; | |
2996 | unsigned b = pid & 0xAAAAAAAA; | |
2997 | ||
2998 | return (a << 1) | (b >> 1); | |
2999 | } | |
3000 | ||
3001 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3002 | { | |
3003 | if (cgroup_sane_behavior(cgrp)) | |
3004 | return pid_fry(pid); | |
3005 | else | |
3006 | return pid; | |
3007 | } | |
3008 | ||
102a775e BB |
3009 | static int cmppid(const void *a, const void *b) |
3010 | { | |
3011 | return *(pid_t *)a - *(pid_t *)b; | |
3012 | } | |
3013 | ||
afb2bc14 TH |
3014 | static int fried_cmppid(const void *a, const void *b) |
3015 | { | |
3016 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3017 | } | |
3018 | ||
e6b81710 TH |
3019 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3020 | enum cgroup_filetype type) | |
3021 | { | |
3022 | struct cgroup_pidlist *l; | |
3023 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3024 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3025 | ||
3026 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3027 | ||
3028 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3029 | if (l->key.type == type && l->key.ns == ns) | |
3030 | return l; | |
3031 | return NULL; | |
3032 | } | |
3033 | ||
72a8cb30 BB |
3034 | /* |
3035 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3036 | * returns with the lock on that pidlist already held, and takes care | |
3037 | * of the use count, or returns NULL with no locks held if we're out of | |
3038 | * memory. | |
3039 | */ | |
e6b81710 TH |
3040 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3041 | enum cgroup_filetype type) | |
72a8cb30 BB |
3042 | { |
3043 | struct cgroup_pidlist *l; | |
b70cc5fd | 3044 | |
e6b81710 TH |
3045 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3046 | ||
3047 | l = cgroup_pidlist_find(cgrp, type); | |
3048 | if (l) | |
3049 | return l; | |
3050 | ||
72a8cb30 | 3051 | /* entry not found; create a new one */ |
f4f4be2b | 3052 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3053 | if (!l) |
72a8cb30 | 3054 | return l; |
e6b81710 | 3055 | |
b1a21367 | 3056 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3057 | l->key.type = type; |
e6b81710 TH |
3058 | /* don't need task_nsproxy() if we're looking at ourself */ |
3059 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3060 | l->owner = cgrp; |
3061 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3062 | return l; |
3063 | } | |
3064 | ||
102a775e BB |
3065 | /* |
3066 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3067 | */ | |
72a8cb30 BB |
3068 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3069 | struct cgroup_pidlist **lp) | |
102a775e BB |
3070 | { |
3071 | pid_t *array; | |
3072 | int length; | |
3073 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3074 | struct css_task_iter it; |
817929ec | 3075 | struct task_struct *tsk; |
102a775e BB |
3076 | struct cgroup_pidlist *l; |
3077 | ||
4bac00d1 TH |
3078 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3079 | ||
102a775e BB |
3080 | /* |
3081 | * If cgroup gets more users after we read count, we won't have | |
3082 | * enough space - tough. This race is indistinguishable to the | |
3083 | * caller from the case that the additional cgroup users didn't | |
3084 | * show up until sometime later on. | |
3085 | */ | |
3086 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3087 | array = pidlist_allocate(length); |
102a775e BB |
3088 | if (!array) |
3089 | return -ENOMEM; | |
3090 | /* now, populate the array */ | |
72ec7029 TH |
3091 | css_task_iter_start(&cgrp->dummy_css, &it); |
3092 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3093 | if (unlikely(n == length)) |
817929ec | 3094 | break; |
102a775e | 3095 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3096 | if (type == CGROUP_FILE_PROCS) |
3097 | pid = task_tgid_vnr(tsk); | |
3098 | else | |
3099 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3100 | if (pid > 0) /* make sure to only use valid results */ |
3101 | array[n++] = pid; | |
817929ec | 3102 | } |
72ec7029 | 3103 | css_task_iter_end(&it); |
102a775e BB |
3104 | length = n; |
3105 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3106 | if (cgroup_sane_behavior(cgrp)) |
3107 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3108 | else | |
3109 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3110 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3111 | length = pidlist_uniq(array, length); |
e6b81710 | 3112 | |
e6b81710 | 3113 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3114 | if (!l) { |
e6b81710 | 3115 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3116 | pidlist_free(array); |
72a8cb30 | 3117 | return -ENOMEM; |
102a775e | 3118 | } |
e6b81710 TH |
3119 | |
3120 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3121 | pidlist_free(l->list); |
102a775e BB |
3122 | l->list = array; |
3123 | l->length = length; | |
72a8cb30 | 3124 | *lp = l; |
102a775e | 3125 | return 0; |
bbcb81d0 PM |
3126 | } |
3127 | ||
846c7bb0 | 3128 | /** |
a043e3b2 | 3129 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3130 | * @stats: cgroupstats to fill information into |
3131 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3132 | * been requested. | |
a043e3b2 LZ |
3133 | * |
3134 | * Build and fill cgroupstats so that taskstats can export it to user | |
3135 | * space. | |
846c7bb0 BS |
3136 | */ |
3137 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3138 | { | |
2bd59d48 | 3139 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3140 | struct cgroup *cgrp; |
72ec7029 | 3141 | struct css_task_iter it; |
846c7bb0 | 3142 | struct task_struct *tsk; |
33d283be | 3143 | |
2bd59d48 TH |
3144 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3145 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3146 | kernfs_type(kn) != KERNFS_DIR) | |
3147 | return -EINVAL; | |
3148 | ||
bad34660 LZ |
3149 | mutex_lock(&cgroup_mutex); |
3150 | ||
846c7bb0 | 3151 | /* |
2bd59d48 TH |
3152 | * We aren't being called from kernfs and there's no guarantee on |
3153 | * @kn->priv's validity. For this and css_tryget_from_dir(), | |
3154 | * @kn->priv is RCU safe. Let's do the RCU dancing. | |
846c7bb0 | 3155 | */ |
2bd59d48 TH |
3156 | rcu_read_lock(); |
3157 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 3158 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 3159 | rcu_read_unlock(); |
bad34660 | 3160 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
3161 | return -ENOENT; |
3162 | } | |
bad34660 | 3163 | rcu_read_unlock(); |
846c7bb0 | 3164 | |
72ec7029 TH |
3165 | css_task_iter_start(&cgrp->dummy_css, &it); |
3166 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3167 | switch (tsk->state) { |
3168 | case TASK_RUNNING: | |
3169 | stats->nr_running++; | |
3170 | break; | |
3171 | case TASK_INTERRUPTIBLE: | |
3172 | stats->nr_sleeping++; | |
3173 | break; | |
3174 | case TASK_UNINTERRUPTIBLE: | |
3175 | stats->nr_uninterruptible++; | |
3176 | break; | |
3177 | case TASK_STOPPED: | |
3178 | stats->nr_stopped++; | |
3179 | break; | |
3180 | default: | |
3181 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3182 | stats->nr_io_wait++; | |
3183 | break; | |
3184 | } | |
3185 | } | |
72ec7029 | 3186 | css_task_iter_end(&it); |
846c7bb0 | 3187 | |
bad34660 | 3188 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 3189 | return 0; |
846c7bb0 BS |
3190 | } |
3191 | ||
8f3ff208 | 3192 | |
bbcb81d0 | 3193 | /* |
102a775e | 3194 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3195 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3196 | * in the cgroup->l->list array. |
bbcb81d0 | 3197 | */ |
cc31edce | 3198 | |
102a775e | 3199 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3200 | { |
cc31edce PM |
3201 | /* |
3202 | * Initially we receive a position value that corresponds to | |
3203 | * one more than the last pid shown (or 0 on the first call or | |
3204 | * after a seek to the start). Use a binary-search to find the | |
3205 | * next pid to display, if any | |
3206 | */ | |
2bd59d48 | 3207 | struct kernfs_open_file *of = s->private; |
7da11279 | 3208 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3209 | struct cgroup_pidlist *l; |
7da11279 | 3210 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3211 | int index = 0, pid = *pos; |
4bac00d1 TH |
3212 | int *iter, ret; |
3213 | ||
3214 | mutex_lock(&cgrp->pidlist_mutex); | |
3215 | ||
3216 | /* | |
5d22444f | 3217 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3218 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3219 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3220 | * could already have been destroyed. |
3221 | */ | |
5d22444f TH |
3222 | if (of->priv) |
3223 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3224 | |
3225 | /* | |
3226 | * Either this is the first start() after open or the matching | |
3227 | * pidlist has been destroyed inbetween. Create a new one. | |
3228 | */ | |
5d22444f TH |
3229 | if (!of->priv) { |
3230 | ret = pidlist_array_load(cgrp, type, | |
3231 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3232 | if (ret) |
3233 | return ERR_PTR(ret); | |
3234 | } | |
5d22444f | 3235 | l = of->priv; |
cc31edce | 3236 | |
cc31edce | 3237 | if (pid) { |
102a775e | 3238 | int end = l->length; |
20777766 | 3239 | |
cc31edce PM |
3240 | while (index < end) { |
3241 | int mid = (index + end) / 2; | |
afb2bc14 | 3242 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3243 | index = mid; |
3244 | break; | |
afb2bc14 | 3245 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3246 | index = mid + 1; |
3247 | else | |
3248 | end = mid; | |
3249 | } | |
3250 | } | |
3251 | /* If we're off the end of the array, we're done */ | |
102a775e | 3252 | if (index >= l->length) |
cc31edce PM |
3253 | return NULL; |
3254 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3255 | iter = l->list + index; |
afb2bc14 | 3256 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3257 | return iter; |
3258 | } | |
3259 | ||
102a775e | 3260 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3261 | { |
2bd59d48 | 3262 | struct kernfs_open_file *of = s->private; |
5d22444f | 3263 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3264 | |
5d22444f TH |
3265 | if (l) |
3266 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3267 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3268 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3269 | } |
3270 | ||
102a775e | 3271 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3272 | { |
2bd59d48 | 3273 | struct kernfs_open_file *of = s->private; |
5d22444f | 3274 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3275 | pid_t *p = v; |
3276 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3277 | /* |
3278 | * Advance to the next pid in the array. If this goes off the | |
3279 | * end, we're done | |
3280 | */ | |
3281 | p++; | |
3282 | if (p >= end) { | |
3283 | return NULL; | |
3284 | } else { | |
7da11279 | 3285 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3286 | return p; |
3287 | } | |
3288 | } | |
3289 | ||
102a775e | 3290 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3291 | { |
3292 | return seq_printf(s, "%d\n", *(int *)v); | |
3293 | } | |
bbcb81d0 | 3294 | |
102a775e BB |
3295 | /* |
3296 | * seq_operations functions for iterating on pidlists through seq_file - | |
3297 | * independent of whether it's tasks or procs | |
3298 | */ | |
3299 | static const struct seq_operations cgroup_pidlist_seq_operations = { | |
3300 | .start = cgroup_pidlist_start, | |
3301 | .stop = cgroup_pidlist_stop, | |
3302 | .next = cgroup_pidlist_next, | |
3303 | .show = cgroup_pidlist_show, | |
cc31edce PM |
3304 | }; |
3305 | ||
182446d0 TH |
3306 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3307 | struct cftype *cft) | |
81a6a5cd | 3308 | { |
182446d0 | 3309 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3310 | } |
3311 | ||
182446d0 TH |
3312 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3313 | struct cftype *cft, u64 val) | |
6379c106 | 3314 | { |
182446d0 | 3315 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3316 | if (val) |
182446d0 | 3317 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3318 | else |
182446d0 | 3319 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3320 | return 0; |
3321 | } | |
3322 | ||
182446d0 TH |
3323 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3324 | struct cftype *cft) | |
97978e6d | 3325 | { |
182446d0 | 3326 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3327 | } |
3328 | ||
182446d0 TH |
3329 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3330 | struct cftype *cft, u64 val) | |
97978e6d DL |
3331 | { |
3332 | if (val) | |
182446d0 | 3333 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3334 | else |
182446d0 | 3335 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3336 | return 0; |
3337 | } | |
3338 | ||
d5c56ced | 3339 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3340 | { |
d5c56ced | 3341 | .name = "cgroup.procs", |
6612f05b TH |
3342 | .seq_start = cgroup_pidlist_start, |
3343 | .seq_next = cgroup_pidlist_next, | |
3344 | .seq_stop = cgroup_pidlist_stop, | |
3345 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3346 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3347 | .write_u64 = cgroup_procs_write, |
74a1166d | 3348 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3349 | }, |
97978e6d DL |
3350 | { |
3351 | .name = "cgroup.clone_children", | |
873fe09e | 3352 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3353 | .read_u64 = cgroup_clone_children_read, |
3354 | .write_u64 = cgroup_clone_children_write, | |
3355 | }, | |
873fe09e TH |
3356 | { |
3357 | .name = "cgroup.sane_behavior", | |
3358 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3359 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3360 | }, |
d5c56ced TH |
3361 | |
3362 | /* | |
3363 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3364 | * don't exist if sane_behavior. If you're depending on these, be | |
3365 | * prepared to be burned. | |
3366 | */ | |
3367 | { | |
3368 | .name = "tasks", | |
3369 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
3370 | .seq_start = cgroup_pidlist_start, |
3371 | .seq_next = cgroup_pidlist_next, | |
3372 | .seq_stop = cgroup_pidlist_stop, | |
3373 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3374 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 3375 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
3376 | .mode = S_IRUGO | S_IWUSR, |
3377 | }, | |
3378 | { | |
3379 | .name = "notify_on_release", | |
3380 | .flags = CFTYPE_INSANE, | |
3381 | .read_u64 = cgroup_read_notify_on_release, | |
3382 | .write_u64 = cgroup_write_notify_on_release, | |
3383 | }, | |
6e6ff25b TH |
3384 | { |
3385 | .name = "release_agent", | |
cc5943a7 | 3386 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 3387 | .seq_show = cgroup_release_agent_show, |
6e6ff25b | 3388 | .write_string = cgroup_release_agent_write, |
5f469907 | 3389 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 3390 | }, |
db0416b6 | 3391 | { } /* terminate */ |
bbcb81d0 PM |
3392 | }; |
3393 | ||
13af07df | 3394 | /** |
628f7cd4 | 3395 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 3396 | * @cgrp: target cgroup |
13af07df | 3397 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
3398 | * |
3399 | * On failure, no file is added. | |
13af07df | 3400 | */ |
628f7cd4 | 3401 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask) |
ddbcc7e8 | 3402 | { |
ddbcc7e8 | 3403 | struct cgroup_subsys *ss; |
b420ba7d | 3404 | int i, ret = 0; |
bbcb81d0 | 3405 | |
8e3f6541 | 3406 | /* process cftsets of each subsystem */ |
b420ba7d | 3407 | for_each_subsys(ss, i) { |
0adb0704 | 3408 | struct cftype *cfts; |
b420ba7d TH |
3409 | |
3410 | if (!test_bit(i, &subsys_mask)) | |
13af07df | 3411 | continue; |
8e3f6541 | 3412 | |
0adb0704 TH |
3413 | list_for_each_entry(cfts, &ss->cfts, node) { |
3414 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
3415 | if (ret < 0) |
3416 | goto err; | |
3417 | } | |
ddbcc7e8 | 3418 | } |
ddbcc7e8 | 3419 | return 0; |
bee55099 TH |
3420 | err: |
3421 | cgroup_clear_dir(cgrp, subsys_mask); | |
3422 | return ret; | |
ddbcc7e8 PM |
3423 | } |
3424 | ||
0c21ead1 TH |
3425 | /* |
3426 | * css destruction is four-stage process. | |
3427 | * | |
3428 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3429 | * Implemented in kill_css(). | |
3430 | * | |
3431 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
3432 | * and thus css_tryget() is guaranteed to fail, the css can be offlined | |
3433 | * by invoking offline_css(). After offlining, the base ref is put. | |
3434 | * Implemented in css_killed_work_fn(). | |
3435 | * | |
3436 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3437 | * accessors are inside RCU read sections. css_release() schedules the | |
3438 | * RCU callback. | |
3439 | * | |
3440 | * 4. After the grace period, the css can be freed. Implemented in | |
3441 | * css_free_work_fn(). | |
3442 | * | |
3443 | * It is actually hairier because both step 2 and 4 require process context | |
3444 | * and thus involve punting to css->destroy_work adding two additional | |
3445 | * steps to the already complex sequence. | |
3446 | */ | |
35ef10da | 3447 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3448 | { |
3449 | struct cgroup_subsys_state *css = | |
35ef10da | 3450 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 3451 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3452 | |
0ae78e0b TH |
3453 | if (css->parent) |
3454 | css_put(css->parent); | |
3455 | ||
0c21ead1 | 3456 | css->ss->css_free(css); |
2bd59d48 | 3457 | cgroup_put(cgrp); |
48ddbe19 TH |
3458 | } |
3459 | ||
0c21ead1 | 3460 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3461 | { |
3462 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3463 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3464 | |
35ef10da | 3465 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 3466 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
3467 | } |
3468 | ||
d3daf28d TH |
3469 | static void css_release(struct percpu_ref *ref) |
3470 | { | |
3471 | struct cgroup_subsys_state *css = | |
3472 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3473 | ||
aec25020 | 3474 | rcu_assign_pointer(css->cgroup->subsys[css->ss->id], NULL); |
0c21ead1 | 3475 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
3476 | } |
3477 | ||
623f926b TH |
3478 | static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss, |
3479 | struct cgroup *cgrp) | |
ddbcc7e8 | 3480 | { |
bd89aabc | 3481 | css->cgroup = cgrp; |
72c97e54 | 3482 | css->ss = ss; |
ddbcc7e8 | 3483 | css->flags = 0; |
0ae78e0b TH |
3484 | |
3485 | if (cgrp->parent) | |
ca8bdcaf | 3486 | css->parent = cgroup_css(cgrp->parent, ss); |
0ae78e0b | 3487 | else |
38b53aba | 3488 | css->flags |= CSS_ROOT; |
48ddbe19 | 3489 | |
ca8bdcaf | 3490 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
3491 | } |
3492 | ||
2a4ac633 | 3493 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 3494 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3495 | { |
623f926b | 3496 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
3497 | int ret = 0; |
3498 | ||
ace2bee8 | 3499 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3500 | lockdep_assert_held(&cgroup_mutex); |
3501 | ||
92fb9748 | 3502 | if (ss->css_online) |
eb95419b | 3503 | ret = ss->css_online(css); |
ae7f164a | 3504 | if (!ret) { |
eb95419b | 3505 | css->flags |= CSS_ONLINE; |
f20104de | 3506 | css->cgroup->nr_css++; |
aec25020 | 3507 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 3508 | } |
b1929db4 | 3509 | return ret; |
a31f2d3f TH |
3510 | } |
3511 | ||
2a4ac633 | 3512 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 3513 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3514 | { |
623f926b | 3515 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 3516 | |
ace2bee8 | 3517 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
3518 | lockdep_assert_held(&cgroup_mutex); |
3519 | ||
3520 | if (!(css->flags & CSS_ONLINE)) | |
3521 | return; | |
3522 | ||
d7eeac19 | 3523 | if (ss->css_offline) |
eb95419b | 3524 | ss->css_offline(css); |
a31f2d3f | 3525 | |
eb95419b | 3526 | css->flags &= ~CSS_ONLINE; |
09a503ea | 3527 | css->cgroup->nr_css--; |
aec25020 | 3528 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css); |
a31f2d3f TH |
3529 | } |
3530 | ||
c81c925a TH |
3531 | /** |
3532 | * create_css - create a cgroup_subsys_state | |
3533 | * @cgrp: the cgroup new css will be associated with | |
3534 | * @ss: the subsys of new css | |
3535 | * | |
3536 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
3537 | * css is online and installed in @cgrp with all interface files created. | |
3538 | * Returns 0 on success, -errno on failure. | |
3539 | */ | |
3540 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
3541 | { | |
3542 | struct cgroup *parent = cgrp->parent; | |
3543 | struct cgroup_subsys_state *css; | |
3544 | int err; | |
3545 | ||
c81c925a TH |
3546 | lockdep_assert_held(&cgroup_mutex); |
3547 | ||
3548 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
3549 | if (IS_ERR(css)) | |
3550 | return PTR_ERR(css); | |
3551 | ||
3552 | err = percpu_ref_init(&css->refcnt, css_release); | |
3553 | if (err) | |
3554 | goto err_free; | |
3555 | ||
3556 | init_css(css, ss, cgrp); | |
3557 | ||
aec25020 | 3558 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a TH |
3559 | if (err) |
3560 | goto err_free; | |
3561 | ||
3562 | err = online_css(css); | |
3563 | if (err) | |
3564 | goto err_free; | |
3565 | ||
59f5296b | 3566 | cgroup_get(cgrp); |
c81c925a TH |
3567 | css_get(css->parent); |
3568 | ||
3569 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && | |
3570 | parent->parent) { | |
3571 | 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", | |
3572 | current->comm, current->pid, ss->name); | |
3573 | if (!strcmp(ss->name, "memory")) | |
3574 | pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n"); | |
3575 | ss->warned_broken_hierarchy = true; | |
3576 | } | |
3577 | ||
3578 | return 0; | |
3579 | ||
3580 | err_free: | |
3581 | percpu_ref_cancel_init(&css->refcnt); | |
3582 | ss->css_free(css); | |
3583 | return err; | |
3584 | } | |
3585 | ||
2bd59d48 | 3586 | /** |
a043e3b2 LZ |
3587 | * cgroup_create - create a cgroup |
3588 | * @parent: cgroup that will be parent of the new cgroup | |
e61734c5 | 3589 | * @name: name of the new cgroup |
2bd59d48 | 3590 | * @mode: mode to set on new cgroup |
ddbcc7e8 | 3591 | */ |
e61734c5 | 3592 | static long cgroup_create(struct cgroup *parent, const char *name, |
2bd59d48 | 3593 | umode_t mode) |
ddbcc7e8 | 3594 | { |
bd89aabc | 3595 | struct cgroup *cgrp; |
ddbcc7e8 | 3596 | struct cgroupfs_root *root = parent->root; |
b58c8998 | 3597 | int ssid, err; |
ddbcc7e8 | 3598 | struct cgroup_subsys *ss; |
2bd59d48 | 3599 | struct kernfs_node *kn; |
ddbcc7e8 | 3600 | |
0a950f65 | 3601 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
3602 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
3603 | if (!cgrp) | |
ddbcc7e8 PM |
3604 | return -ENOMEM; |
3605 | ||
ace2bee8 TH |
3606 | mutex_lock(&cgroup_tree_mutex); |
3607 | ||
976c06bc TH |
3608 | /* |
3609 | * Only live parents can have children. Note that the liveliness | |
3610 | * check isn't strictly necessary because cgroup_mkdir() and | |
3611 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
3612 | * anyway so that locking is contained inside cgroup proper and we | |
3613 | * don't get nasty surprises if we ever grow another caller. | |
3614 | */ | |
3615 | if (!cgroup_lock_live_group(parent)) { | |
3616 | err = -ENODEV; | |
ace2bee8 | 3617 | goto err_unlock_tree; |
0ab02ca8 LZ |
3618 | } |
3619 | ||
3620 | /* | |
3621 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
3622 | * a half-baked cgroup. | |
3623 | */ | |
3624 | cgrp->id = idr_alloc(&root->cgroup_idr, NULL, 1, 0, GFP_KERNEL); | |
3625 | if (cgrp->id < 0) { | |
3626 | err = -ENOMEM; | |
3627 | goto err_unlock; | |
976c06bc TH |
3628 | } |
3629 | ||
cc31edce | 3630 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 3631 | |
bd89aabc | 3632 | cgrp->parent = parent; |
0ae78e0b | 3633 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 3634 | cgrp->root = parent->root; |
ddbcc7e8 | 3635 | |
b6abdb0e LZ |
3636 | if (notify_on_release(parent)) |
3637 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
3638 | ||
2260e7fc TH |
3639 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
3640 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 3641 | |
2bd59d48 | 3642 | /* create the directory */ |
e61734c5 | 3643 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 TH |
3644 | if (IS_ERR(kn)) { |
3645 | err = PTR_ERR(kn); | |
0ab02ca8 | 3646 | goto err_free_id; |
2bd59d48 TH |
3647 | } |
3648 | cgrp->kn = kn; | |
ddbcc7e8 | 3649 | |
6f30558f TH |
3650 | /* |
3651 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
3652 | * that @cgrp->kn is always accessible. | |
3653 | */ | |
3654 | kernfs_get(kn); | |
3655 | ||
00356bd5 | 3656 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 3657 | |
4e139afc | 3658 | /* allocation complete, commit to creation */ |
4e139afc | 3659 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3c9c825b | 3660 | atomic_inc(&root->nr_cgrps); |
59f5296b | 3661 | cgroup_get(parent); |
415cf07a | 3662 | |
0d80255e TH |
3663 | /* |
3664 | * @cgrp is now fully operational. If something fails after this | |
3665 | * point, it'll be released via the normal destruction path. | |
3666 | */ | |
4e96ee8e LZ |
3667 | idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
3668 | ||
2bb566cb | 3669 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
3670 | if (err) |
3671 | goto err_destroy; | |
3672 | ||
9d403e99 | 3673 | /* let's create and online css's */ |
b85d2040 TH |
3674 | for_each_subsys(ss, ssid) { |
3675 | if (root->subsys_mask & (1 << ssid)) { | |
3676 | err = create_css(cgrp, ss); | |
3677 | if (err) | |
3678 | goto err_destroy; | |
3679 | } | |
a8638030 | 3680 | } |
ddbcc7e8 | 3681 | |
2bd59d48 TH |
3682 | kernfs_activate(kn); |
3683 | ||
ddbcc7e8 | 3684 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3685 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
3686 | |
3687 | return 0; | |
3688 | ||
0a950f65 | 3689 | err_free_id: |
4e96ee8e | 3690 | idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
3691 | err_unlock: |
3692 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
3693 | err_unlock_tree: |
3694 | mutex_unlock(&cgroup_tree_mutex); | |
bd89aabc | 3695 | kfree(cgrp); |
ddbcc7e8 | 3696 | return err; |
4b8b47eb TH |
3697 | |
3698 | err_destroy: | |
3699 | cgroup_destroy_locked(cgrp); | |
3700 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3701 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 3702 | return err; |
ddbcc7e8 PM |
3703 | } |
3704 | ||
2bd59d48 TH |
3705 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
3706 | umode_t mode) | |
ddbcc7e8 | 3707 | { |
2bd59d48 | 3708 | struct cgroup *parent = parent_kn->priv; |
ddbcc7e8 | 3709 | |
2bd59d48 | 3710 | return cgroup_create(parent, name, mode); |
ddbcc7e8 PM |
3711 | } |
3712 | ||
223dbc38 TH |
3713 | /* |
3714 | * This is called when the refcnt of a css is confirmed to be killed. | |
3715 | * css_tryget() is now guaranteed to fail. | |
3716 | */ | |
3717 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 3718 | { |
223dbc38 TH |
3719 | struct cgroup_subsys_state *css = |
3720 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
3721 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 3722 | |
ace2bee8 | 3723 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
3724 | mutex_lock(&cgroup_mutex); |
3725 | ||
09a503ea TH |
3726 | /* |
3727 | * css_tryget() is guaranteed to fail now. Tell subsystems to | |
3728 | * initate destruction. | |
3729 | */ | |
3730 | offline_css(css); | |
3731 | ||
f20104de TH |
3732 | /* |
3733 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
3734 | * be disabled before proceeding to the second phase of cgroup | |
3735 | * destruction. If we are the last one, kick it off. | |
3736 | */ | |
09a503ea | 3737 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
3738 | cgroup_destroy_css_killed(cgrp); |
3739 | ||
3740 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 3741 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
3742 | |
3743 | /* | |
3744 | * Put the css refs from kill_css(). Each css holds an extra | |
3745 | * reference to the cgroup's dentry and cgroup removal proceeds | |
3746 | * regardless of css refs. On the last put of each css, whenever | |
3747 | * that may be, the extra dentry ref is put so that dentry | |
3748 | * destruction happens only after all css's are released. | |
3749 | */ | |
3750 | css_put(css); | |
d3daf28d TH |
3751 | } |
3752 | ||
223dbc38 TH |
3753 | /* css kill confirmation processing requires process context, bounce */ |
3754 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
3755 | { |
3756 | struct cgroup_subsys_state *css = | |
3757 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3758 | ||
223dbc38 | 3759 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 3760 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
3761 | } |
3762 | ||
edae0c33 TH |
3763 | /** |
3764 | * kill_css - destroy a css | |
3765 | * @css: css to destroy | |
3766 | * | |
3c14f8b4 TH |
3767 | * This function initiates destruction of @css by removing cgroup interface |
3768 | * files and putting its base reference. ->css_offline() will be invoked | |
3769 | * asynchronously once css_tryget() is guaranteed to fail and when the | |
3770 | * reference count reaches zero, @css will be released. | |
edae0c33 TH |
3771 | */ |
3772 | static void kill_css(struct cgroup_subsys_state *css) | |
3773 | { | |
2bd59d48 TH |
3774 | /* |
3775 | * This must happen before css is disassociated with its cgroup. | |
3776 | * See seq_css() for details. | |
3777 | */ | |
aec25020 | 3778 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 3779 | |
edae0c33 TH |
3780 | /* |
3781 | * Killing would put the base ref, but we need to keep it alive | |
3782 | * until after ->css_offline(). | |
3783 | */ | |
3784 | css_get(css); | |
3785 | ||
3786 | /* | |
3787 | * cgroup core guarantees that, by the time ->css_offline() is | |
3788 | * invoked, no new css reference will be given out via | |
3789 | * css_tryget(). We can't simply call percpu_ref_kill() and | |
3790 | * proceed to offlining css's because percpu_ref_kill() doesn't | |
3791 | * guarantee that the ref is seen as killed on all CPUs on return. | |
3792 | * | |
3793 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
3794 | * css is confirmed to be seen as killed on all CPUs. | |
3795 | */ | |
3796 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
3797 | } |
3798 | ||
3799 | /** | |
3800 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
3801 | * @cgrp: cgroup to be destroyed | |
3802 | * | |
3803 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
3804 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
3805 | * guarantee that css_tryget() won't succeed by the time ->css_offline() is | |
3806 | * invoked. To satisfy all the requirements, destruction is implemented in | |
3807 | * the following two steps. | |
3808 | * | |
3809 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
3810 | * userland visible parts and start killing the percpu refcnts of | |
3811 | * css's. Set up so that the next stage will be kicked off once all | |
3812 | * the percpu refcnts are confirmed to be killed. | |
3813 | * | |
3814 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
3815 | * rest of destruction. Once all cgroup references are gone, the | |
3816 | * cgroup is RCU-freed. | |
3817 | * | |
3818 | * This function implements s1. After this step, @cgrp is gone as far as | |
3819 | * the userland is concerned and a new cgroup with the same name may be | |
3820 | * created. As cgroup doesn't care about the names internally, this | |
3821 | * doesn't cause any problem. | |
3822 | */ | |
42809dd4 TH |
3823 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
3824 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 3825 | { |
bb78a92f | 3826 | struct cgroup *child; |
2bd59d48 | 3827 | struct cgroup_subsys_state *css; |
ddd69148 | 3828 | bool empty; |
1c6727af | 3829 | int ssid; |
ddbcc7e8 | 3830 | |
ace2bee8 | 3831 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
3832 | lockdep_assert_held(&cgroup_mutex); |
3833 | ||
ddd69148 | 3834 | /* |
96d365e0 | 3835 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 3836 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 3837 | */ |
96d365e0 | 3838 | down_read(&css_set_rwsem); |
bb78a92f | 3839 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 3840 | up_read(&css_set_rwsem); |
ddd69148 | 3841 | if (!empty) |
ddbcc7e8 | 3842 | return -EBUSY; |
a043e3b2 | 3843 | |
bb78a92f HD |
3844 | /* |
3845 | * Make sure there's no live children. We can't test ->children | |
3846 | * emptiness as dead children linger on it while being destroyed; | |
3847 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
3848 | */ | |
3849 | empty = true; | |
3850 | rcu_read_lock(); | |
3851 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
3852 | empty = cgroup_is_dead(child); | |
3853 | if (!empty) | |
3854 | break; | |
3855 | } | |
3856 | rcu_read_unlock(); | |
3857 | if (!empty) | |
3858 | return -EBUSY; | |
3859 | ||
88703267 | 3860 | /* |
edae0c33 TH |
3861 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
3862 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
3863 | * percpu refs of all css's are confirmed to be killed. This |
3864 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 3865 | */ |
4ac06017 | 3866 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
3867 | for_each_css(css, ssid, cgrp) |
3868 | kill_css(css); | |
4ac06017 | 3869 | mutex_lock(&cgroup_mutex); |
455050d2 TH |
3870 | |
3871 | /* | |
3872 | * Mark @cgrp dead. This prevents further task migration and child | |
3873 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 3874 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 3875 | * resume iteration after dropping RCU read lock. See |
492eb21b | 3876 | * css_next_child() for details. |
455050d2 | 3877 | */ |
54766d4a | 3878 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 3879 | |
455050d2 TH |
3880 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
3881 | raw_spin_lock(&release_list_lock); | |
3882 | if (!list_empty(&cgrp->release_list)) | |
3883 | list_del_init(&cgrp->release_list); | |
3884 | raw_spin_unlock(&release_list_lock); | |
3885 | ||
3886 | /* | |
f20104de TH |
3887 | * If @cgrp has css's attached, the second stage of cgroup |
3888 | * destruction is kicked off from css_killed_work_fn() after the | |
3889 | * refs of all attached css's are killed. If @cgrp doesn't have | |
3890 | * any css, we kick it off here. | |
3891 | */ | |
3892 | if (!cgrp->nr_css) | |
3893 | cgroup_destroy_css_killed(cgrp); | |
3894 | ||
2bd59d48 TH |
3895 | /* remove @cgrp directory along with the base files */ |
3896 | mutex_unlock(&cgroup_mutex); | |
3897 | ||
455050d2 | 3898 | /* |
2bd59d48 TH |
3899 | * There are two control paths which try to determine cgroup from |
3900 | * dentry without going through kernfs - cgroupstats_build() and | |
3901 | * css_tryget_from_dir(). Those are supported by RCU protecting | |
3902 | * clearing of cgrp->kn->priv backpointer, which should happen | |
3903 | * after all files under it have been removed. | |
455050d2 | 3904 | */ |
6f30558f | 3905 | kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ |
2bd59d48 | 3906 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); |
2bd59d48 | 3907 | |
4ac06017 | 3908 | mutex_lock(&cgroup_mutex); |
455050d2 | 3909 | |
ea15f8cc TH |
3910 | return 0; |
3911 | }; | |
3912 | ||
d3daf28d | 3913 | /** |
f20104de | 3914 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
d3daf28d TH |
3915 | * @work: cgroup->destroy_free_work |
3916 | * | |
3917 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
3918 | * destroyed after all css's are offlined and performs the rest of |
3919 | * destruction. This is the second step of destruction described in the | |
3920 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 3921 | */ |
f20104de | 3922 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 3923 | { |
ea15f8cc | 3924 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 3925 | |
ace2bee8 | 3926 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 3927 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 3928 | |
999cd8a4 | 3929 | /* delete this cgroup from parent->children */ |
eb6fd504 | 3930 | list_del_rcu(&cgrp->sibling); |
ed957793 | 3931 | |
59f5296b | 3932 | cgroup_put(cgrp); |
ddbcc7e8 | 3933 | |
bd89aabc | 3934 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 3935 | check_for_release(parent); |
ddbcc7e8 PM |
3936 | } |
3937 | ||
2bd59d48 | 3938 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 3939 | { |
2bd59d48 TH |
3940 | struct cgroup *cgrp = kn->priv; |
3941 | int ret = 0; | |
3942 | ||
3943 | /* | |
3944 | * This is self-destruction but @kn can't be removed while this | |
3945 | * callback is in progress. Let's break active protection. Once | |
3946 | * the protection is broken, @cgrp can be destroyed at any point. | |
3947 | * Pin it so that it stays accessible. | |
3948 | */ | |
3949 | cgroup_get(cgrp); | |
3950 | kernfs_break_active_protection(kn); | |
42809dd4 | 3951 | |
ace2bee8 | 3952 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 3953 | mutex_lock(&cgroup_mutex); |
2bd59d48 TH |
3954 | |
3955 | /* | |
3956 | * @cgrp might already have been destroyed while we're trying to | |
3957 | * grab the mutexes. | |
3958 | */ | |
3959 | if (!cgroup_is_dead(cgrp)) | |
3960 | ret = cgroup_destroy_locked(cgrp); | |
3961 | ||
42809dd4 | 3962 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 3963 | mutex_unlock(&cgroup_tree_mutex); |
42809dd4 | 3964 | |
2bd59d48 TH |
3965 | kernfs_unbreak_active_protection(kn); |
3966 | cgroup_put(cgrp); | |
42809dd4 TH |
3967 | return ret; |
3968 | } | |
3969 | ||
2bd59d48 TH |
3970 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
3971 | .remount_fs = cgroup_remount, | |
3972 | .show_options = cgroup_show_options, | |
3973 | .mkdir = cgroup_mkdir, | |
3974 | .rmdir = cgroup_rmdir, | |
3975 | .rename = cgroup_rename, | |
3976 | }; | |
3977 | ||
06a11920 | 3978 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss) |
ddbcc7e8 | 3979 | { |
ddbcc7e8 | 3980 | struct cgroup_subsys_state *css; |
cfe36bde DC |
3981 | |
3982 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 3983 | |
ace2bee8 | 3984 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
3985 | mutex_lock(&cgroup_mutex); |
3986 | ||
0adb0704 | 3987 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 3988 | |
ddbcc7e8 | 3989 | /* Create the top cgroup state for this subsystem */ |
9871bf95 | 3990 | ss->root = &cgroup_dummy_root; |
ca8bdcaf | 3991 | css = ss->css_alloc(cgroup_css(cgroup_dummy_top, ss)); |
ddbcc7e8 PM |
3992 | /* We don't handle early failures gracefully */ |
3993 | BUG_ON(IS_ERR(css)); | |
623f926b | 3994 | init_css(css, ss, cgroup_dummy_top); |
ddbcc7e8 | 3995 | |
e8d55fde | 3996 | /* Update the init_css_set to contain a subsys |
817929ec | 3997 | * pointer to this state - since the subsystem is |
e8d55fde LZ |
3998 | * newly registered, all tasks and hence the |
3999 | * init_css_set is in the subsystem's top cgroup. */ | |
aec25020 | 4000 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4001 | |
4002 | need_forkexit_callback |= ss->fork || ss->exit; | |
4003 | ||
e8d55fde LZ |
4004 | /* At system boot, before all subsystems have been |
4005 | * registered, no tasks have been forked, so we don't | |
4006 | * need to invoke fork callbacks here. */ | |
4007 | BUG_ON(!list_empty(&init_task.tasks)); | |
4008 | ||
ae7f164a | 4009 | BUG_ON(online_css(css)); |
a8638030 | 4010 | |
648bb56d | 4011 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4012 | mutex_unlock(&cgroup_tree_mutex); |
e6a1105b BB |
4013 | } |
4014 | ||
ddbcc7e8 | 4015 | /** |
a043e3b2 LZ |
4016 | * cgroup_init_early - cgroup initialization at system boot |
4017 | * | |
4018 | * Initialize cgroups at system boot, and initialize any | |
4019 | * subsystems that request early init. | |
ddbcc7e8 PM |
4020 | */ |
4021 | int __init cgroup_init_early(void) | |
4022 | { | |
30159ec7 | 4023 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4024 | int i; |
30159ec7 | 4025 | |
146aa1bd | 4026 | atomic_set(&init_css_set.refcount, 1); |
69d0206c | 4027 | INIT_LIST_HEAD(&init_css_set.cgrp_links); |
817929ec | 4028 | INIT_LIST_HEAD(&init_css_set.tasks); |
b3dc094e | 4029 | INIT_LIST_HEAD(&init_css_set.mg_tasks); |
1958d2d5 | 4030 | INIT_LIST_HEAD(&init_css_set.mg_preload_node); |
b3dc094e | 4031 | INIT_LIST_HEAD(&init_css_set.mg_node); |
472b1053 | 4032 | INIT_HLIST_NODE(&init_css_set.hlist); |
817929ec | 4033 | css_set_count = 1; |
9871bf95 TH |
4034 | init_cgroup_root(&cgroup_dummy_root); |
4035 | cgroup_root_count = 1; | |
a4ea1cc9 | 4036 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4037 | |
69d0206c | 4038 | init_cgrp_cset_link.cset = &init_css_set; |
9871bf95 TH |
4039 | init_cgrp_cset_link.cgrp = cgroup_dummy_top; |
4040 | list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links); | |
69d0206c | 4041 | list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links); |
ddbcc7e8 | 4042 | |
3ed80a62 | 4043 | for_each_subsys(ss, i) { |
aec25020 | 4044 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4045 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4046 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4047 | ss->id, ss->name); |
073219e9 TH |
4048 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4049 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4050 | ||
aec25020 | 4051 | ss->id = i; |
073219e9 | 4052 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4053 | |
4054 | if (ss->early_init) | |
4055 | cgroup_init_subsys(ss); | |
4056 | } | |
4057 | return 0; | |
4058 | } | |
4059 | ||
4060 | /** | |
a043e3b2 LZ |
4061 | * cgroup_init - cgroup initialization |
4062 | * | |
4063 | * Register cgroup filesystem and /proc file, and initialize | |
4064 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4065 | */ |
4066 | int __init cgroup_init(void) | |
4067 | { | |
30159ec7 | 4068 | struct cgroup_subsys *ss; |
0ac801fe | 4069 | unsigned long key; |
30159ec7 | 4070 | int i, err; |
a424316c | 4071 | |
2bd59d48 | 4072 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
2da440a2 | 4073 | |
3ed80a62 | 4074 | for_each_subsys(ss, i) { |
ddbcc7e8 PM |
4075 | if (!ss->early_init) |
4076 | cgroup_init_subsys(ss); | |
de00ffa5 TH |
4077 | |
4078 | /* | |
4079 | * cftype registration needs kmalloc and can't be done | |
4080 | * during early_init. Register base cftypes separately. | |
4081 | */ | |
4082 | if (ss->base_cftypes) | |
4083 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
ddbcc7e8 PM |
4084 | } |
4085 | ||
fa3ca07e | 4086 | /* allocate id for the dummy hierarchy */ |
54e7b4eb | 4087 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4088 | |
82fe9b0d TH |
4089 | /* Add init_css_set to the hash table */ |
4090 | key = css_set_hash(init_css_set.subsys); | |
4091 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4092 | ||
fc76df70 | 4093 | BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1)); |
676db4af | 4094 | |
4e96ee8e LZ |
4095 | err = idr_alloc(&cgroup_dummy_root.cgroup_idr, cgroup_dummy_top, |
4096 | 0, 1, GFP_KERNEL); | |
4097 | BUG_ON(err < 0); | |
4098 | ||
54e7b4eb TH |
4099 | mutex_unlock(&cgroup_mutex); |
4100 | ||
676db4af | 4101 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4102 | if (!cgroup_kobj) |
4103 | return -ENOMEM; | |
676db4af | 4104 | |
ddbcc7e8 | 4105 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
4106 | if (err < 0) { |
4107 | kobject_put(cgroup_kobj); | |
2bd59d48 | 4108 | return err; |
676db4af | 4109 | } |
ddbcc7e8 | 4110 | |
46ae220b | 4111 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 4112 | return 0; |
ddbcc7e8 | 4113 | } |
b4f48b63 | 4114 | |
e5fca243 TH |
4115 | static int __init cgroup_wq_init(void) |
4116 | { | |
4117 | /* | |
4118 | * There isn't much point in executing destruction path in | |
4119 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 4120 | * Use 1 for @max_active. |
e5fca243 TH |
4121 | * |
4122 | * We would prefer to do this in cgroup_init() above, but that | |
4123 | * is called before init_workqueues(): so leave this until after. | |
4124 | */ | |
1a11533f | 4125 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 4126 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4127 | |
4128 | /* | |
4129 | * Used to destroy pidlists and separate to serve as flush domain. | |
4130 | * Cap @max_active to 1 too. | |
4131 | */ | |
4132 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4133 | 0, 1); | |
4134 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4135 | ||
e5fca243 TH |
4136 | return 0; |
4137 | } | |
4138 | core_initcall(cgroup_wq_init); | |
4139 | ||
a424316c PM |
4140 | /* |
4141 | * proc_cgroup_show() | |
4142 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4143 | * - Used for /proc/<pid>/cgroup. | |
4144 | * - No need to task_lock(tsk) on this tsk->cgroup reference, as it | |
4145 | * doesn't really matter if tsk->cgroup changes after we read it, | |
956db3ca | 4146 | * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it |
a424316c PM |
4147 | * anyway. No need to check that tsk->cgroup != NULL, thanks to |
4148 | * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks | |
4149 | * cgroup to top_cgroup. | |
4150 | */ | |
4151 | ||
4152 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4153 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4154 | { |
4155 | struct pid *pid; | |
4156 | struct task_struct *tsk; | |
e61734c5 | 4157 | char *buf, *path; |
a424316c PM |
4158 | int retval; |
4159 | struct cgroupfs_root *root; | |
4160 | ||
4161 | retval = -ENOMEM; | |
e61734c5 | 4162 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4163 | if (!buf) |
4164 | goto out; | |
4165 | ||
4166 | retval = -ESRCH; | |
4167 | pid = m->private; | |
4168 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4169 | if (!tsk) | |
4170 | goto out_free; | |
4171 | ||
4172 | retval = 0; | |
4173 | ||
4174 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 4175 | down_read(&css_set_rwsem); |
a424316c | 4176 | |
e5f6a860 | 4177 | for_each_active_root(root) { |
a424316c | 4178 | struct cgroup_subsys *ss; |
bd89aabc | 4179 | struct cgroup *cgrp; |
b85d2040 | 4180 | int ssid, count = 0; |
a424316c | 4181 | |
2c6ab6d2 | 4182 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 TH |
4183 | for_each_subsys(ss, ssid) |
4184 | if (root->subsys_mask & (1 << ssid)) | |
4185 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); | |
c6d57f33 PM |
4186 | if (strlen(root->name)) |
4187 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4188 | root->name); | |
a424316c | 4189 | seq_putc(m, ':'); |
7717f7ba | 4190 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
4191 | path = cgroup_path(cgrp, buf, PATH_MAX); |
4192 | if (!path) { | |
4193 | retval = -ENAMETOOLONG; | |
a424316c | 4194 | goto out_unlock; |
e61734c5 TH |
4195 | } |
4196 | seq_puts(m, path); | |
a424316c PM |
4197 | seq_putc(m, '\n'); |
4198 | } | |
4199 | ||
4200 | out_unlock: | |
96d365e0 | 4201 | up_read(&css_set_rwsem); |
a424316c PM |
4202 | mutex_unlock(&cgroup_mutex); |
4203 | put_task_struct(tsk); | |
4204 | out_free: | |
4205 | kfree(buf); | |
4206 | out: | |
4207 | return retval; | |
4208 | } | |
4209 | ||
a424316c PM |
4210 | /* Display information about each subsystem and each hierarchy */ |
4211 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4212 | { | |
30159ec7 | 4213 | struct cgroup_subsys *ss; |
a424316c | 4214 | int i; |
a424316c | 4215 | |
8bab8dde | 4216 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4217 | /* |
4218 | * ideally we don't want subsystems moving around while we do this. | |
4219 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4220 | * subsys/hierarchy state. | |
4221 | */ | |
a424316c | 4222 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4223 | |
4224 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4225 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4226 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 4227 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 4228 | |
a424316c PM |
4229 | mutex_unlock(&cgroup_mutex); |
4230 | return 0; | |
4231 | } | |
4232 | ||
4233 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4234 | { | |
9dce07f1 | 4235 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4236 | } |
4237 | ||
828c0950 | 4238 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4239 | .open = cgroupstats_open, |
4240 | .read = seq_read, | |
4241 | .llseek = seq_lseek, | |
4242 | .release = single_release, | |
4243 | }; | |
4244 | ||
b4f48b63 PM |
4245 | /** |
4246 | * cgroup_fork - attach newly forked task to its parents cgroup. | |
a043e3b2 | 4247 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 PM |
4248 | * |
4249 | * Description: A task inherits its parent's cgroup at fork(). | |
4250 | * | |
4251 | * A pointer to the shared css_set was automatically copied in | |
4252 | * fork.c by dup_task_struct(). However, we ignore that copy, since | |
9bb71308 TH |
4253 | * it was not made under the protection of RCU or cgroup_mutex, so |
4254 | * might no longer be a valid cgroup pointer. cgroup_attach_task() might | |
4255 | * have already changed current->cgroups, allowing the previously | |
4256 | * referenced cgroup group to be removed and freed. | |
b4f48b63 PM |
4257 | * |
4258 | * At the point that cgroup_fork() is called, 'current' is the parent | |
4259 | * task, and the passed argument 'child' points to the child task. | |
4260 | */ | |
4261 | void cgroup_fork(struct task_struct *child) | |
4262 | { | |
9bb71308 | 4263 | task_lock(current); |
a8ad805c | 4264 | get_css_set(task_css_set(current)); |
817929ec | 4265 | child->cgroups = current->cgroups; |
9bb71308 | 4266 | task_unlock(current); |
817929ec | 4267 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4268 | } |
4269 | ||
817929ec | 4270 | /** |
a043e3b2 LZ |
4271 | * cgroup_post_fork - called on a new task after adding it to the task list |
4272 | * @child: the task in question | |
4273 | * | |
5edee61e TH |
4274 | * Adds the task to the list running through its css_set if necessary and |
4275 | * call the subsystem fork() callbacks. Has to be after the task is | |
4276 | * visible on the task list in case we race with the first call to | |
0942eeee | 4277 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4278 | * list. |
a043e3b2 | 4279 | */ |
817929ec PM |
4280 | void cgroup_post_fork(struct task_struct *child) |
4281 | { | |
30159ec7 | 4282 | struct cgroup_subsys *ss; |
5edee61e TH |
4283 | int i; |
4284 | ||
3ce3230a FW |
4285 | /* |
4286 | * use_task_css_set_links is set to 1 before we walk the tasklist | |
4287 | * under the tasklist_lock and we read it here after we added the child | |
4288 | * to the tasklist under the tasklist_lock as well. If the child wasn't | |
4289 | * yet in the tasklist when we walked through it from | |
4290 | * cgroup_enable_task_cg_lists(), then use_task_css_set_links value | |
4291 | * should be visible now due to the paired locking and barriers implied | |
4292 | * by LOCK/UNLOCK: it is written before the tasklist_lock unlock | |
4293 | * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock | |
4294 | * lock on fork. | |
4295 | */ | |
817929ec | 4296 | if (use_task_css_set_links) { |
96d365e0 | 4297 | down_write(&css_set_rwsem); |
d8783832 TH |
4298 | task_lock(child); |
4299 | if (list_empty(&child->cg_list)) | |
a8ad805c | 4300 | list_add(&child->cg_list, &task_css_set(child)->tasks); |
d8783832 | 4301 | task_unlock(child); |
96d365e0 | 4302 | up_write(&css_set_rwsem); |
817929ec | 4303 | } |
5edee61e TH |
4304 | |
4305 | /* | |
4306 | * Call ss->fork(). This must happen after @child is linked on | |
4307 | * css_set; otherwise, @child might change state between ->fork() | |
4308 | * and addition to css_set. | |
4309 | */ | |
4310 | if (need_forkexit_callback) { | |
3ed80a62 | 4311 | for_each_subsys(ss, i) |
5edee61e TH |
4312 | if (ss->fork) |
4313 | ss->fork(child); | |
5edee61e | 4314 | } |
817929ec | 4315 | } |
5edee61e | 4316 | |
b4f48b63 PM |
4317 | /** |
4318 | * cgroup_exit - detach cgroup from exiting task | |
4319 | * @tsk: pointer to task_struct of exiting process | |
a043e3b2 | 4320 | * @run_callback: run exit callbacks? |
b4f48b63 PM |
4321 | * |
4322 | * Description: Detach cgroup from @tsk and release it. | |
4323 | * | |
4324 | * Note that cgroups marked notify_on_release force every task in | |
4325 | * them to take the global cgroup_mutex mutex when exiting. | |
4326 | * This could impact scaling on very large systems. Be reluctant to | |
4327 | * use notify_on_release cgroups where very high task exit scaling | |
4328 | * is required on large systems. | |
4329 | * | |
4330 | * the_top_cgroup_hack: | |
4331 | * | |
4332 | * Set the exiting tasks cgroup to the root cgroup (top_cgroup). | |
4333 | * | |
4334 | * We call cgroup_exit() while the task is still competent to | |
4335 | * handle notify_on_release(), then leave the task attached to the | |
4336 | * root cgroup in each hierarchy for the remainder of its exit. | |
4337 | * | |
4338 | * To do this properly, we would increment the reference count on | |
4339 | * top_cgroup, and near the very end of the kernel/exit.c do_exit() | |
4340 | * code we would add a second cgroup function call, to drop that | |
4341 | * reference. This would just create an unnecessary hot spot on | |
4342 | * the top_cgroup reference count, to no avail. | |
4343 | * | |
4344 | * Normally, holding a reference to a cgroup without bumping its | |
4345 | * count is unsafe. The cgroup could go away, or someone could | |
4346 | * attach us to a different cgroup, decrementing the count on | |
4347 | * the first cgroup that we never incremented. But in this case, | |
4348 | * top_cgroup isn't going away, and either task has PF_EXITING set, | |
956db3ca CW |
4349 | * which wards off any cgroup_attach_task() attempts, or task is a failed |
4350 | * fork, never visible to cgroup_attach_task. | |
b4f48b63 PM |
4351 | */ |
4352 | void cgroup_exit(struct task_struct *tsk, int run_callbacks) | |
4353 | { | |
30159ec7 | 4354 | struct cgroup_subsys *ss; |
5abb8855 | 4355 | struct css_set *cset; |
d41d5a01 | 4356 | int i; |
817929ec PM |
4357 | |
4358 | /* | |
96d365e0 TH |
4359 | * Unlink from the css_set task list if necessary. Optimistically |
4360 | * check cg_list before taking css_set_rwsem. | |
817929ec PM |
4361 | */ |
4362 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 4363 | down_write(&css_set_rwsem); |
817929ec | 4364 | if (!list_empty(&tsk->cg_list)) |
8d258797 | 4365 | list_del_init(&tsk->cg_list); |
96d365e0 | 4366 | up_write(&css_set_rwsem); |
817929ec PM |
4367 | } |
4368 | ||
b4f48b63 PM |
4369 | /* Reassign the task to the init_css_set. */ |
4370 | task_lock(tsk); | |
a8ad805c TH |
4371 | cset = task_css_set(tsk); |
4372 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 PZ |
4373 | |
4374 | if (run_callbacks && need_forkexit_callback) { | |
3ed80a62 TH |
4375 | /* see cgroup_post_fork() for details */ |
4376 | for_each_subsys(ss, i) { | |
d41d5a01 | 4377 | if (ss->exit) { |
eb95419b TH |
4378 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
4379 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 4380 | |
eb95419b | 4381 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
4382 | } |
4383 | } | |
4384 | } | |
b4f48b63 | 4385 | task_unlock(tsk); |
d41d5a01 | 4386 | |
89c5509b | 4387 | put_css_set(cset, true); |
b4f48b63 | 4388 | } |
697f4161 | 4389 | |
bd89aabc | 4390 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 4391 | { |
f50daa70 | 4392 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 4393 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
4394 | /* |
4395 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 4396 | * already queued for a userspace notification, queue |
f50daa70 LZ |
4397 | * it now |
4398 | */ | |
81a6a5cd | 4399 | int need_schedule_work = 0; |
f50daa70 | 4400 | |
cdcc136f | 4401 | raw_spin_lock(&release_list_lock); |
54766d4a | 4402 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
4403 | list_empty(&cgrp->release_list)) { |
4404 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
4405 | need_schedule_work = 1; |
4406 | } | |
cdcc136f | 4407 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4408 | if (need_schedule_work) |
4409 | schedule_work(&release_agent_work); | |
4410 | } | |
4411 | } | |
4412 | ||
81a6a5cd PM |
4413 | /* |
4414 | * Notify userspace when a cgroup is released, by running the | |
4415 | * configured release agent with the name of the cgroup (path | |
4416 | * relative to the root of cgroup file system) as the argument. | |
4417 | * | |
4418 | * Most likely, this user command will try to rmdir this cgroup. | |
4419 | * | |
4420 | * This races with the possibility that some other task will be | |
4421 | * attached to this cgroup before it is removed, or that some other | |
4422 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
4423 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
4424 | * unused, and this cgroup will be reprieved from its death sentence, | |
4425 | * to continue to serve a useful existence. Next time it's released, | |
4426 | * we will get notified again, if it still has 'notify_on_release' set. | |
4427 | * | |
4428 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
4429 | * means only wait until the task is successfully execve()'d. The | |
4430 | * separate release agent task is forked by call_usermodehelper(), | |
4431 | * then control in this thread returns here, without waiting for the | |
4432 | * release agent task. We don't bother to wait because the caller of | |
4433 | * this routine has no use for the exit status of the release agent | |
4434 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 4435 | */ |
81a6a5cd PM |
4436 | static void cgroup_release_agent(struct work_struct *work) |
4437 | { | |
4438 | BUG_ON(work != &release_agent_work); | |
4439 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 4440 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
4441 | while (!list_empty(&release_list)) { |
4442 | char *argv[3], *envp[3]; | |
4443 | int i; | |
e61734c5 | 4444 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 4445 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
4446 | struct cgroup, |
4447 | release_list); | |
bd89aabc | 4448 | list_del_init(&cgrp->release_list); |
cdcc136f | 4449 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 4450 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
4451 | if (!pathbuf) |
4452 | goto continue_free; | |
e61734c5 TH |
4453 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
4454 | if (!path) | |
e788e066 PM |
4455 | goto continue_free; |
4456 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
4457 | if (!agentbuf) | |
4458 | goto continue_free; | |
81a6a5cd PM |
4459 | |
4460 | i = 0; | |
e788e066 | 4461 | argv[i++] = agentbuf; |
e61734c5 | 4462 | argv[i++] = path; |
81a6a5cd PM |
4463 | argv[i] = NULL; |
4464 | ||
4465 | i = 0; | |
4466 | /* minimal command environment */ | |
4467 | envp[i++] = "HOME=/"; | |
4468 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
4469 | envp[i] = NULL; | |
4470 | ||
4471 | /* Drop the lock while we invoke the usermode helper, | |
4472 | * since the exec could involve hitting disk and hence | |
4473 | * be a slow process */ | |
4474 | mutex_unlock(&cgroup_mutex); | |
4475 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 4476 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
4477 | continue_free: |
4478 | kfree(pathbuf); | |
4479 | kfree(agentbuf); | |
cdcc136f | 4480 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 4481 | } |
cdcc136f | 4482 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
4483 | mutex_unlock(&cgroup_mutex); |
4484 | } | |
8bab8dde PM |
4485 | |
4486 | static int __init cgroup_disable(char *str) | |
4487 | { | |
30159ec7 | 4488 | struct cgroup_subsys *ss; |
8bab8dde | 4489 | char *token; |
30159ec7 | 4490 | int i; |
8bab8dde PM |
4491 | |
4492 | while ((token = strsep(&str, ",")) != NULL) { | |
4493 | if (!*token) | |
4494 | continue; | |
be45c900 | 4495 | |
3ed80a62 | 4496 | for_each_subsys(ss, i) { |
8bab8dde PM |
4497 | if (!strcmp(token, ss->name)) { |
4498 | ss->disabled = 1; | |
4499 | printk(KERN_INFO "Disabling %s control group" | |
4500 | " subsystem\n", ss->name); | |
4501 | break; | |
4502 | } | |
4503 | } | |
4504 | } | |
4505 | return 1; | |
4506 | } | |
4507 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 4508 | |
b77d7b60 | 4509 | /** |
5a17f543 | 4510 | * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir |
35cf0836 TH |
4511 | * @dentry: directory dentry of interest |
4512 | * @ss: subsystem of interest | |
b77d7b60 | 4513 | * |
5a17f543 TH |
4514 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
4515 | * to get the corresponding css and return it. If such css doesn't exist | |
4516 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 4517 | */ |
5a17f543 TH |
4518 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
4519 | struct cgroup_subsys *ss) | |
e5d1367f | 4520 | { |
2bd59d48 TH |
4521 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
4522 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 4523 | struct cgroup *cgrp; |
b77d7b60 | 4524 | |
35cf0836 | 4525 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
4526 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
4527 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
4528 | return ERR_PTR(-EBADF); |
4529 | ||
5a17f543 TH |
4530 | rcu_read_lock(); |
4531 | ||
2bd59d48 TH |
4532 | /* |
4533 | * This path doesn't originate from kernfs and @kn could already | |
4534 | * have been or be removed at any point. @kn->priv is RCU | |
4535 | * protected for this access. See destroy_locked() for details. | |
4536 | */ | |
4537 | cgrp = rcu_dereference(kn->priv); | |
4538 | if (cgrp) | |
4539 | css = cgroup_css(cgrp, ss); | |
5a17f543 TH |
4540 | |
4541 | if (!css || !css_tryget(css)) | |
4542 | css = ERR_PTR(-ENOENT); | |
4543 | ||
4544 | rcu_read_unlock(); | |
4545 | return css; | |
e5d1367f | 4546 | } |
e5d1367f | 4547 | |
1cb650b9 LZ |
4548 | /** |
4549 | * css_from_id - lookup css by id | |
4550 | * @id: the cgroup id | |
4551 | * @ss: cgroup subsys to be looked into | |
4552 | * | |
4553 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
4554 | * Should be called under rcu_read_lock(). | |
4555 | */ | |
4556 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
4557 | { | |
4558 | struct cgroup *cgrp; | |
4559 | ||
ace2bee8 | 4560 | cgroup_assert_mutexes_or_rcu_locked(); |
1cb650b9 LZ |
4561 | |
4562 | cgrp = idr_find(&ss->root->cgroup_idr, id); | |
4563 | if (cgrp) | |
d1625964 | 4564 | return cgroup_css(cgrp, ss); |
1cb650b9 | 4565 | return NULL; |
e5d1367f SE |
4566 | } |
4567 | ||
fe693435 | 4568 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
4569 | static struct cgroup_subsys_state * |
4570 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
4571 | { |
4572 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
4573 | ||
4574 | if (!css) | |
4575 | return ERR_PTR(-ENOMEM); | |
4576 | ||
4577 | return css; | |
4578 | } | |
4579 | ||
eb95419b | 4580 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 4581 | { |
eb95419b | 4582 | kfree(css); |
fe693435 PM |
4583 | } |
4584 | ||
182446d0 TH |
4585 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
4586 | struct cftype *cft) | |
fe693435 | 4587 | { |
182446d0 | 4588 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
4589 | } |
4590 | ||
182446d0 TH |
4591 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
4592 | struct cftype *cft) | |
fe693435 PM |
4593 | { |
4594 | return (u64)(unsigned long)current->cgroups; | |
4595 | } | |
4596 | ||
182446d0 | 4597 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 4598 | struct cftype *cft) |
fe693435 PM |
4599 | { |
4600 | u64 count; | |
4601 | ||
4602 | rcu_read_lock(); | |
a8ad805c | 4603 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
4604 | rcu_read_unlock(); |
4605 | return count; | |
4606 | } | |
4607 | ||
2da8ca82 | 4608 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4609 | { |
69d0206c | 4610 | struct cgrp_cset_link *link; |
5abb8855 | 4611 | struct css_set *cset; |
e61734c5 TH |
4612 | char *name_buf; |
4613 | ||
4614 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
4615 | if (!name_buf) | |
4616 | return -ENOMEM; | |
7717f7ba | 4617 | |
96d365e0 | 4618 | down_read(&css_set_rwsem); |
7717f7ba | 4619 | rcu_read_lock(); |
5abb8855 | 4620 | cset = rcu_dereference(current->cgroups); |
69d0206c | 4621 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 4622 | struct cgroup *c = link->cgrp; |
59f5296b TH |
4623 | const char *name = "?"; |
4624 | ||
e61734c5 TH |
4625 | if (c != cgroup_dummy_top) { |
4626 | cgroup_name(c, name_buf, NAME_MAX + 1); | |
4627 | name = name_buf; | |
4628 | } | |
7717f7ba | 4629 | |
2c6ab6d2 PM |
4630 | seq_printf(seq, "Root %d group %s\n", |
4631 | c->root->hierarchy_id, name); | |
7717f7ba PM |
4632 | } |
4633 | rcu_read_unlock(); | |
96d365e0 | 4634 | up_read(&css_set_rwsem); |
e61734c5 | 4635 | kfree(name_buf); |
7717f7ba PM |
4636 | return 0; |
4637 | } | |
4638 | ||
4639 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 4640 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 4641 | { |
2da8ca82 | 4642 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 4643 | struct cgrp_cset_link *link; |
7717f7ba | 4644 | |
96d365e0 | 4645 | down_read(&css_set_rwsem); |
182446d0 | 4646 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 4647 | struct css_set *cset = link->cset; |
7717f7ba PM |
4648 | struct task_struct *task; |
4649 | int count = 0; | |
c7561128 | 4650 | |
5abb8855 | 4651 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 4652 | |
5abb8855 | 4653 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
4654 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
4655 | goto overflow; | |
4656 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
4657 | } | |
4658 | ||
4659 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
4660 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
4661 | goto overflow; | |
4662 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 4663 | } |
c7561128 TH |
4664 | continue; |
4665 | overflow: | |
4666 | seq_puts(seq, " ...\n"); | |
7717f7ba | 4667 | } |
96d365e0 | 4668 | up_read(&css_set_rwsem); |
7717f7ba PM |
4669 | return 0; |
4670 | } | |
4671 | ||
182446d0 | 4672 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 4673 | { |
182446d0 | 4674 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
4675 | } |
4676 | ||
4677 | static struct cftype debug_files[] = { | |
fe693435 PM |
4678 | { |
4679 | .name = "taskcount", | |
4680 | .read_u64 = debug_taskcount_read, | |
4681 | }, | |
4682 | ||
4683 | { | |
4684 | .name = "current_css_set", | |
4685 | .read_u64 = current_css_set_read, | |
4686 | }, | |
4687 | ||
4688 | { | |
4689 | .name = "current_css_set_refcount", | |
4690 | .read_u64 = current_css_set_refcount_read, | |
4691 | }, | |
4692 | ||
7717f7ba PM |
4693 | { |
4694 | .name = "current_css_set_cg_links", | |
2da8ca82 | 4695 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
4696 | }, |
4697 | ||
4698 | { | |
4699 | .name = "cgroup_css_links", | |
2da8ca82 | 4700 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
4701 | }, |
4702 | ||
fe693435 PM |
4703 | { |
4704 | .name = "releasable", | |
4705 | .read_u64 = releasable_read, | |
4706 | }, | |
fe693435 | 4707 | |
4baf6e33 TH |
4708 | { } /* terminate */ |
4709 | }; | |
fe693435 | 4710 | |
073219e9 | 4711 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
4712 | .css_alloc = debug_css_alloc, |
4713 | .css_free = debug_css_free, | |
4baf6e33 | 4714 | .base_cftypes = debug_files, |
fe693435 PM |
4715 | }; |
4716 | #endif /* CONFIG_CGROUP_DEBUG */ |