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 | ||
ed3d261b JP |
29 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
30 | ||
ddbcc7e8 | 31 | #include <linux/cgroup.h> |
2ce9738b | 32 | #include <linux/cred.h> |
c6d57f33 | 33 | #include <linux/ctype.h> |
ddbcc7e8 | 34 | #include <linux/errno.h> |
2ce9738b | 35 | #include <linux/init_task.h> |
ddbcc7e8 PM |
36 | #include <linux/kernel.h> |
37 | #include <linux/list.h> | |
38 | #include <linux/mm.h> | |
39 | #include <linux/mutex.h> | |
40 | #include <linux/mount.h> | |
41 | #include <linux/pagemap.h> | |
a424316c | 42 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
43 | #include <linux/rcupdate.h> |
44 | #include <linux/sched.h> | |
ddbcc7e8 | 45 | #include <linux/slab.h> |
ddbcc7e8 | 46 | #include <linux/spinlock.h> |
96d365e0 | 47 | #include <linux/rwsem.h> |
ddbcc7e8 | 48 | #include <linux/string.h> |
bbcb81d0 | 49 | #include <linux/sort.h> |
81a6a5cd | 50 | #include <linux/kmod.h> |
846c7bb0 BS |
51 | #include <linux/delayacct.h> |
52 | #include <linux/cgroupstats.h> | |
0ac801fe | 53 | #include <linux/hashtable.h> |
096b7fe0 | 54 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 55 | #include <linux/idr.h> |
d1d9fd33 | 56 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 57 | #include <linux/kthread.h> |
776f02fa | 58 | #include <linux/delay.h> |
846c7bb0 | 59 | |
60063497 | 60 | #include <linux/atomic.h> |
ddbcc7e8 | 61 | |
b1a21367 TH |
62 | /* |
63 | * pidlists linger the following amount before being destroyed. The goal | |
64 | * is avoiding frequent destruction in the middle of consecutive read calls | |
65 | * Expiring in the middle is a performance problem not a correctness one. | |
66 | * 1 sec should be enough. | |
67 | */ | |
68 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
69 | ||
8d7e6fb0 TH |
70 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
71 | MAX_CFTYPE_NAME + 2) | |
72 | ||
ace2bee8 TH |
73 | /* |
74 | * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file | |
75 | * creation/removal and hierarchy changing operations including cgroup | |
76 | * creation, removal, css association and controller rebinding. This outer | |
77 | * lock is needed mainly to resolve the circular dependency between kernfs | |
78 | * active ref and cgroup_mutex. cgroup_tree_mutex nests above both. | |
79 | */ | |
80 | static DEFINE_MUTEX(cgroup_tree_mutex); | |
81 | ||
e25e2cbb TH |
82 | /* |
83 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
84 | * hierarchy must be performed while holding it. | |
85 | * | |
0e1d768f TH |
86 | * css_set_rwsem protects task->cgroups pointer, the list of css_set |
87 | * objects, and the chain of tasks off each css_set. | |
e25e2cbb | 88 | * |
0e1d768f TH |
89 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
90 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 91 | */ |
2219449a TH |
92 | #ifdef CONFIG_PROVE_RCU |
93 | DEFINE_MUTEX(cgroup_mutex); | |
0e1d768f TH |
94 | DECLARE_RWSEM(css_set_rwsem); |
95 | EXPORT_SYMBOL_GPL(cgroup_mutex); | |
96 | EXPORT_SYMBOL_GPL(css_set_rwsem); | |
2219449a | 97 | #else |
81a6a5cd | 98 | static DEFINE_MUTEX(cgroup_mutex); |
0e1d768f | 99 | static DECLARE_RWSEM(css_set_rwsem); |
2219449a TH |
100 | #endif |
101 | ||
6fa4918d | 102 | /* |
15a4c835 TH |
103 | * Protects cgroup_idr and css_idr so that IDs can be released without |
104 | * grabbing cgroup_mutex. | |
6fa4918d TH |
105 | */ |
106 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
107 | ||
69e943b7 TH |
108 | /* |
109 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
110 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
111 | */ | |
112 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 113 | |
ace2bee8 | 114 | #define cgroup_assert_mutexes_or_rcu_locked() \ |
87fb54f1 | 115 | rcu_lockdep_assert(rcu_read_lock_held() || \ |
ace2bee8 | 116 | lockdep_is_held(&cgroup_tree_mutex) || \ |
87fb54f1 | 117 | lockdep_is_held(&cgroup_mutex), \ |
ace2bee8 | 118 | "cgroup_[tree_]mutex or RCU read lock required"); |
780cd8b3 | 119 | |
e5fca243 TH |
120 | /* |
121 | * cgroup destruction makes heavy use of work items and there can be a lot | |
122 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
123 | * destruction work items don't end up filling up max_active of system_wq | |
124 | * which may lead to deadlock. | |
125 | */ | |
126 | static struct workqueue_struct *cgroup_destroy_wq; | |
127 | ||
b1a21367 TH |
128 | /* |
129 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
130 | * separate workqueue as flush domain. | |
131 | */ | |
132 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
133 | ||
3ed80a62 | 134 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 135 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 136 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
137 | #include <linux/cgroup_subsys.h> |
138 | }; | |
073219e9 TH |
139 | #undef SUBSYS |
140 | ||
141 | /* array of cgroup subsystem names */ | |
142 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
143 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
144 | #include <linux/cgroup_subsys.h> |
145 | }; | |
073219e9 | 146 | #undef SUBSYS |
ddbcc7e8 | 147 | |
ddbcc7e8 | 148 | /* |
3dd06ffa | 149 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
150 | * unattached - it never has more than a single cgroup, and all tasks are |
151 | * part of that cgroup. | |
ddbcc7e8 | 152 | */ |
a2dd4247 | 153 | struct cgroup_root cgrp_dfl_root; |
9871bf95 | 154 | |
a2dd4247 TH |
155 | /* |
156 | * The default hierarchy always exists but is hidden until mounted for the | |
157 | * first time. This is for backward compatibility. | |
158 | */ | |
159 | static bool cgrp_dfl_root_visible; | |
ddbcc7e8 PM |
160 | |
161 | /* The list of hierarchy roots */ | |
162 | ||
9871bf95 TH |
163 | static LIST_HEAD(cgroup_roots); |
164 | static int cgroup_root_count; | |
ddbcc7e8 | 165 | |
3417ae1f | 166 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 167 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 168 | |
794611a1 LZ |
169 | /* |
170 | * Assign a monotonically increasing serial number to cgroups. It | |
171 | * guarantees cgroups with bigger numbers are newer than those with smaller | |
172 | * numbers. Also, as cgroups are always appended to the parent's | |
173 | * ->children list, it guarantees that sibling cgroups are always sorted in | |
00356bd5 TH |
174 | * the ascending serial number order on the list. Protected by |
175 | * cgroup_mutex. | |
794611a1 | 176 | */ |
00356bd5 | 177 | static u64 cgroup_serial_nr_next = 1; |
794611a1 | 178 | |
ddbcc7e8 | 179 | /* This flag indicates whether tasks in the fork and exit paths should |
a043e3b2 LZ |
180 | * check for fork/exit handlers to call. This avoids us having to do |
181 | * extra work in the fork/exit path if none of the subsystems need to | |
182 | * be called. | |
ddbcc7e8 | 183 | */ |
8947f9d5 | 184 | static int need_forkexit_callback __read_mostly; |
ddbcc7e8 | 185 | |
628f7cd4 TH |
186 | static struct cftype cgroup_base_files[]; |
187 | ||
59f5296b | 188 | static void cgroup_put(struct cgroup *cgrp); |
3dd06ffa | 189 | static int rebind_subsystems(struct cgroup_root *dst_root, |
69dfa00c | 190 | unsigned int ss_mask); |
f20104de | 191 | static void cgroup_destroy_css_killed(struct cgroup *cgrp); |
42809dd4 | 192 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
f8f22e53 TH |
193 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss); |
194 | static void kill_css(struct cgroup_subsys_state *css); | |
2bb566cb TH |
195 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
196 | bool is_add); | |
b1a21367 | 197 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp); |
42809dd4 | 198 | |
6fa4918d TH |
199 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
200 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
201 | gfp_t gfp_mask) | |
202 | { | |
203 | int ret; | |
204 | ||
205 | idr_preload(gfp_mask); | |
54504e97 | 206 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 207 | ret = idr_alloc(idr, ptr, start, end, gfp_mask); |
54504e97 | 208 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
209 | idr_preload_end(); |
210 | return ret; | |
211 | } | |
212 | ||
213 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
214 | { | |
215 | void *ret; | |
216 | ||
54504e97 | 217 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 218 | ret = idr_replace(idr, ptr, id); |
54504e97 | 219 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
220 | return ret; |
221 | } | |
222 | ||
223 | static void cgroup_idr_remove(struct idr *idr, int id) | |
224 | { | |
54504e97 | 225 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 226 | idr_remove(idr, id); |
54504e97 | 227 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
228 | } |
229 | ||
95109b62 TH |
230 | /** |
231 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
232 | * @cgrp: the cgroup of interest | |
ca8bdcaf | 233 | * @ss: the subsystem of interest (%NULL returns the dummy_css) |
95109b62 | 234 | * |
ca8bdcaf TH |
235 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
236 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
237 | * the caller is responsible for pinning the returned css if it wants to | |
238 | * keep accessing it outside the said locks. This function may return | |
239 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
240 | */ |
241 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 242 | struct cgroup_subsys *ss) |
95109b62 | 243 | { |
ca8bdcaf | 244 | if (ss) |
aec25020 | 245 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 TH |
246 | lockdep_is_held(&cgroup_tree_mutex) || |
247 | lockdep_is_held(&cgroup_mutex)); | |
ca8bdcaf TH |
248 | else |
249 | return &cgrp->dummy_css; | |
95109b62 | 250 | } |
42809dd4 | 251 | |
aec3dfcb TH |
252 | /** |
253 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
254 | * @cgrp: the cgroup of interest | |
255 | * @ss: the subsystem of interest (%NULL returns the dummy_css) | |
256 | * | |
257 | * Similar to cgroup_css() but returns the effctive css, which is defined | |
258 | * as the matching css of the nearest ancestor including self which has @ss | |
259 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
260 | * function is guaranteed to return non-NULL css. | |
261 | */ | |
262 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
263 | struct cgroup_subsys *ss) | |
264 | { | |
265 | lockdep_assert_held(&cgroup_mutex); | |
266 | ||
267 | if (!ss) | |
268 | return &cgrp->dummy_css; | |
269 | ||
270 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
271 | return NULL; | |
272 | ||
273 | while (cgrp->parent && | |
274 | !(cgrp->parent->child_subsys_mask & (1 << ss->id))) | |
275 | cgrp = cgrp->parent; | |
276 | ||
277 | return cgroup_css(cgrp, ss); | |
278 | } | |
279 | ||
ddbcc7e8 | 280 | /* convenient tests for these bits */ |
54766d4a | 281 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 282 | { |
54766d4a | 283 | return test_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 PM |
284 | } |
285 | ||
b4168640 | 286 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 287 | { |
2bd59d48 | 288 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 289 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
290 | |
291 | /* | |
292 | * This is open and unprotected implementation of cgroup_css(). | |
293 | * seq_css() is only called from a kernfs file operation which has | |
294 | * an active reference on the file. Because all the subsystem | |
295 | * files are drained before a css is disassociated with a cgroup, | |
296 | * the matching css from the cgroup's subsys table is guaranteed to | |
297 | * be and stay valid until the enclosing operation is complete. | |
298 | */ | |
299 | if (cft->ss) | |
300 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
301 | else | |
302 | return &cgrp->dummy_css; | |
59f5296b | 303 | } |
b4168640 | 304 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 305 | |
78574cf9 LZ |
306 | /** |
307 | * cgroup_is_descendant - test ancestry | |
308 | * @cgrp: the cgroup to be tested | |
309 | * @ancestor: possible ancestor of @cgrp | |
310 | * | |
311 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
312 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
313 | * and @ancestor are accessible. | |
314 | */ | |
315 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
316 | { | |
317 | while (cgrp) { | |
318 | if (cgrp == ancestor) | |
319 | return true; | |
320 | cgrp = cgrp->parent; | |
321 | } | |
322 | return false; | |
323 | } | |
ddbcc7e8 | 324 | |
e9685a03 | 325 | static int cgroup_is_releasable(const struct cgroup *cgrp) |
81a6a5cd PM |
326 | { |
327 | const int bits = | |
bd89aabc PM |
328 | (1 << CGRP_RELEASABLE) | |
329 | (1 << CGRP_NOTIFY_ON_RELEASE); | |
330 | return (cgrp->flags & bits) == bits; | |
81a6a5cd PM |
331 | } |
332 | ||
e9685a03 | 333 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 334 | { |
bd89aabc | 335 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
336 | } |
337 | ||
1c6727af TH |
338 | /** |
339 | * for_each_css - iterate all css's of a cgroup | |
340 | * @css: the iteration cursor | |
341 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
342 | * @cgrp: the target cgroup to iterate css's of | |
343 | * | |
aec3dfcb | 344 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
345 | */ |
346 | #define for_each_css(css, ssid, cgrp) \ | |
347 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
348 | if (!((css) = rcu_dereference_check( \ | |
349 | (cgrp)->subsys[(ssid)], \ | |
ace2bee8 | 350 | lockdep_is_held(&cgroup_tree_mutex) || \ |
1c6727af TH |
351 | lockdep_is_held(&cgroup_mutex)))) { } \ |
352 | else | |
353 | ||
aec3dfcb TH |
354 | /** |
355 | * for_each_e_css - iterate all effective css's of a cgroup | |
356 | * @css: the iteration cursor | |
357 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
358 | * @cgrp: the target cgroup to iterate css's of | |
359 | * | |
360 | * Should be called under cgroup_[tree_]mutex. | |
361 | */ | |
362 | #define for_each_e_css(css, ssid, cgrp) \ | |
363 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
364 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
365 | ; \ | |
366 | else | |
367 | ||
30159ec7 | 368 | /** |
3ed80a62 | 369 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 370 | * @ss: the iteration cursor |
780cd8b3 | 371 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 372 | */ |
780cd8b3 | 373 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
374 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
375 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 376 | |
985ed670 TH |
377 | /* iterate across the hierarchies */ |
378 | #define for_each_root(root) \ | |
5549c497 | 379 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 380 | |
f8f22e53 TH |
381 | /* iterate over child cgrps, lock should be held throughout iteration */ |
382 | #define cgroup_for_each_live_child(child, cgrp) \ | |
383 | list_for_each_entry((child), &(cgrp)->children, sibling) \ | |
384 | if (({ lockdep_assert_held(&cgroup_tree_mutex); \ | |
385 | cgroup_is_dead(child); })) \ | |
386 | ; \ | |
387 | else | |
388 | ||
7ae1bad9 TH |
389 | /** |
390 | * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive. | |
391 | * @cgrp: the cgroup to be checked for liveness | |
392 | * | |
47cfcd09 TH |
393 | * On success, returns true; the mutex should be later unlocked. On |
394 | * failure returns false with no lock held. | |
7ae1bad9 | 395 | */ |
b9777cf8 | 396 | static bool cgroup_lock_live_group(struct cgroup *cgrp) |
7ae1bad9 TH |
397 | { |
398 | mutex_lock(&cgroup_mutex); | |
54766d4a | 399 | if (cgroup_is_dead(cgrp)) { |
7ae1bad9 TH |
400 | mutex_unlock(&cgroup_mutex); |
401 | return false; | |
402 | } | |
403 | return true; | |
404 | } | |
7ae1bad9 | 405 | |
81a6a5cd PM |
406 | /* the list of cgroups eligible for automatic release. Protected by |
407 | * release_list_lock */ | |
408 | static LIST_HEAD(release_list); | |
cdcc136f | 409 | static DEFINE_RAW_SPINLOCK(release_list_lock); |
81a6a5cd PM |
410 | static void cgroup_release_agent(struct work_struct *work); |
411 | static DECLARE_WORK(release_agent_work, cgroup_release_agent); | |
bd89aabc | 412 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 413 | |
69d0206c TH |
414 | /* |
415 | * A cgroup can be associated with multiple css_sets as different tasks may | |
416 | * belong to different cgroups on different hierarchies. In the other | |
417 | * direction, a css_set is naturally associated with multiple cgroups. | |
418 | * This M:N relationship is represented by the following link structure | |
419 | * which exists for each association and allows traversing the associations | |
420 | * from both sides. | |
421 | */ | |
422 | struct cgrp_cset_link { | |
423 | /* the cgroup and css_set this link associates */ | |
424 | struct cgroup *cgrp; | |
425 | struct css_set *cset; | |
426 | ||
427 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
428 | struct list_head cset_link; | |
429 | ||
430 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
431 | struct list_head cgrp_link; | |
817929ec PM |
432 | }; |
433 | ||
172a2c06 TH |
434 | /* |
435 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
436 | * hierarchies being mounted. It contains a pointer to the root state |
437 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
438 | * reference-counted, to improve performance when child cgroups | |
439 | * haven't been created. | |
440 | */ | |
5024ae29 | 441 | struct css_set init_css_set = { |
172a2c06 TH |
442 | .refcount = ATOMIC_INIT(1), |
443 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
444 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
445 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
446 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
447 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
448 | }; | |
817929ec | 449 | |
172a2c06 | 450 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 451 | |
842b597e TH |
452 | /** |
453 | * cgroup_update_populated - updated populated count of a cgroup | |
454 | * @cgrp: the target cgroup | |
455 | * @populated: inc or dec populated count | |
456 | * | |
457 | * @cgrp is either getting the first task (css_set) or losing the last. | |
458 | * Update @cgrp->populated_cnt accordingly. The count is propagated | |
459 | * towards root so that a given cgroup's populated_cnt is zero iff the | |
460 | * cgroup and all its descendants are empty. | |
461 | * | |
462 | * @cgrp's interface file "cgroup.populated" is zero if | |
463 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
464 | * changes from or to zero, userland is notified that the content of the | |
465 | * interface file has changed. This can be used to detect when @cgrp and | |
466 | * its descendants become populated or empty. | |
467 | */ | |
468 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
469 | { | |
470 | lockdep_assert_held(&css_set_rwsem); | |
471 | ||
472 | do { | |
473 | bool trigger; | |
474 | ||
475 | if (populated) | |
476 | trigger = !cgrp->populated_cnt++; | |
477 | else | |
478 | trigger = !--cgrp->populated_cnt; | |
479 | ||
480 | if (!trigger) | |
481 | break; | |
482 | ||
483 | if (cgrp->populated_kn) | |
484 | kernfs_notify(cgrp->populated_kn); | |
485 | cgrp = cgrp->parent; | |
486 | } while (cgrp); | |
487 | } | |
488 | ||
7717f7ba PM |
489 | /* |
490 | * hash table for cgroup groups. This improves the performance to find | |
491 | * an existing css_set. This hash doesn't (currently) take into | |
492 | * account cgroups in empty hierarchies. | |
493 | */ | |
472b1053 | 494 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 495 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 496 | |
0ac801fe | 497 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 498 | { |
0ac801fe | 499 | unsigned long key = 0UL; |
30159ec7 TH |
500 | struct cgroup_subsys *ss; |
501 | int i; | |
472b1053 | 502 | |
30159ec7 | 503 | for_each_subsys(ss, i) |
0ac801fe LZ |
504 | key += (unsigned long)css[i]; |
505 | key = (key >> 16) ^ key; | |
472b1053 | 506 | |
0ac801fe | 507 | return key; |
472b1053 LZ |
508 | } |
509 | ||
89c5509b | 510 | static void put_css_set_locked(struct css_set *cset, bool taskexit) |
b4f48b63 | 511 | { |
69d0206c | 512 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
513 | struct cgroup_subsys *ss; |
514 | int ssid; | |
5abb8855 | 515 | |
89c5509b TH |
516 | lockdep_assert_held(&css_set_rwsem); |
517 | ||
518 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 519 | return; |
81a6a5cd | 520 | |
2c6ab6d2 | 521 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
2d8f243a TH |
522 | for_each_subsys(ss, ssid) |
523 | list_del(&cset->e_cset_node[ssid]); | |
5abb8855 | 524 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
525 | css_set_count--; |
526 | ||
69d0206c | 527 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 528 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 529 | |
69d0206c TH |
530 | list_del(&link->cset_link); |
531 | list_del(&link->cgrp_link); | |
71b5707e | 532 | |
96d365e0 | 533 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
842b597e TH |
534 | if (list_empty(&cgrp->cset_links)) { |
535 | cgroup_update_populated(cgrp, false); | |
536 | if (notify_on_release(cgrp)) { | |
537 | if (taskexit) | |
538 | set_bit(CGRP_RELEASABLE, &cgrp->flags); | |
539 | check_for_release(cgrp); | |
540 | } | |
81a6a5cd | 541 | } |
2c6ab6d2 PM |
542 | |
543 | kfree(link); | |
81a6a5cd | 544 | } |
2c6ab6d2 | 545 | |
5abb8855 | 546 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
547 | } |
548 | ||
89c5509b TH |
549 | static void put_css_set(struct css_set *cset, bool taskexit) |
550 | { | |
551 | /* | |
552 | * Ensure that the refcount doesn't hit zero while any readers | |
553 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
554 | * rwlock | |
555 | */ | |
556 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
557 | return; | |
558 | ||
559 | down_write(&css_set_rwsem); | |
560 | put_css_set_locked(cset, taskexit); | |
561 | up_write(&css_set_rwsem); | |
562 | } | |
563 | ||
817929ec PM |
564 | /* |
565 | * refcounted get/put for css_set objects | |
566 | */ | |
5abb8855 | 567 | static inline void get_css_set(struct css_set *cset) |
817929ec | 568 | { |
5abb8855 | 569 | atomic_inc(&cset->refcount); |
817929ec PM |
570 | } |
571 | ||
b326f9d0 | 572 | /** |
7717f7ba | 573 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
574 | * @cset: candidate css_set being tested |
575 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
576 | * @new_cgrp: cgroup that's being entered by the task |
577 | * @template: desired set of css pointers in css_set (pre-calculated) | |
578 | * | |
6f4b7e63 | 579 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
580 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
581 | */ | |
5abb8855 TH |
582 | static bool compare_css_sets(struct css_set *cset, |
583 | struct css_set *old_cset, | |
7717f7ba PM |
584 | struct cgroup *new_cgrp, |
585 | struct cgroup_subsys_state *template[]) | |
586 | { | |
587 | struct list_head *l1, *l2; | |
588 | ||
aec3dfcb TH |
589 | /* |
590 | * On the default hierarchy, there can be csets which are | |
591 | * associated with the same set of cgroups but different csses. | |
592 | * Let's first ensure that csses match. | |
593 | */ | |
594 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 595 | return false; |
7717f7ba PM |
596 | |
597 | /* | |
598 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
599 | * different cgroups in hierarchies. As different cgroups may |
600 | * share the same effective css, this comparison is always | |
601 | * necessary. | |
7717f7ba | 602 | */ |
69d0206c TH |
603 | l1 = &cset->cgrp_links; |
604 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 605 | while (1) { |
69d0206c | 606 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 607 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
608 | |
609 | l1 = l1->next; | |
610 | l2 = l2->next; | |
611 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
612 | if (l1 == &cset->cgrp_links) { |
613 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
614 | break; |
615 | } else { | |
69d0206c | 616 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
617 | } |
618 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
619 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
620 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
621 | cgrp1 = link1->cgrp; | |
622 | cgrp2 = link2->cgrp; | |
7717f7ba | 623 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 624 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
625 | |
626 | /* | |
627 | * If this hierarchy is the hierarchy of the cgroup | |
628 | * that's changing, then we need to check that this | |
629 | * css_set points to the new cgroup; if it's any other | |
630 | * hierarchy, then this css_set should point to the | |
631 | * same cgroup as the old css_set. | |
632 | */ | |
5abb8855 TH |
633 | if (cgrp1->root == new_cgrp->root) { |
634 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
635 | return false; |
636 | } else { | |
5abb8855 | 637 | if (cgrp1 != cgrp2) |
7717f7ba PM |
638 | return false; |
639 | } | |
640 | } | |
641 | return true; | |
642 | } | |
643 | ||
b326f9d0 TH |
644 | /** |
645 | * find_existing_css_set - init css array and find the matching css_set | |
646 | * @old_cset: the css_set that we're using before the cgroup transition | |
647 | * @cgrp: the cgroup that we're moving into | |
648 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 649 | */ |
5abb8855 TH |
650 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
651 | struct cgroup *cgrp, | |
652 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 653 | { |
3dd06ffa | 654 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 655 | struct cgroup_subsys *ss; |
5abb8855 | 656 | struct css_set *cset; |
0ac801fe | 657 | unsigned long key; |
b326f9d0 | 658 | int i; |
817929ec | 659 | |
aae8aab4 BB |
660 | /* |
661 | * Build the set of subsystem state objects that we want to see in the | |
662 | * new css_set. while subsystems can change globally, the entries here | |
663 | * won't change, so no need for locking. | |
664 | */ | |
30159ec7 | 665 | for_each_subsys(ss, i) { |
f392e51c | 666 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
667 | /* |
668 | * @ss is in this hierarchy, so we want the | |
669 | * effective css from @cgrp. | |
670 | */ | |
671 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 672 | } else { |
aec3dfcb TH |
673 | /* |
674 | * @ss is not in this hierarchy, so we don't want | |
675 | * to change the css. | |
676 | */ | |
5abb8855 | 677 | template[i] = old_cset->subsys[i]; |
817929ec PM |
678 | } |
679 | } | |
680 | ||
0ac801fe | 681 | key = css_set_hash(template); |
5abb8855 TH |
682 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
683 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
684 | continue; |
685 | ||
686 | /* This css_set matches what we need */ | |
5abb8855 | 687 | return cset; |
472b1053 | 688 | } |
817929ec PM |
689 | |
690 | /* No existing cgroup group matched */ | |
691 | return NULL; | |
692 | } | |
693 | ||
69d0206c | 694 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 695 | { |
69d0206c | 696 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 697 | |
69d0206c TH |
698 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
699 | list_del(&link->cset_link); | |
36553434 LZ |
700 | kfree(link); |
701 | } | |
702 | } | |
703 | ||
69d0206c TH |
704 | /** |
705 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
706 | * @count: the number of links to allocate | |
707 | * @tmp_links: list_head the allocated links are put on | |
708 | * | |
709 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
710 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 711 | */ |
69d0206c | 712 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 713 | { |
69d0206c | 714 | struct cgrp_cset_link *link; |
817929ec | 715 | int i; |
69d0206c TH |
716 | |
717 | INIT_LIST_HEAD(tmp_links); | |
718 | ||
817929ec | 719 | for (i = 0; i < count; i++) { |
f4f4be2b | 720 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 721 | if (!link) { |
69d0206c | 722 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
723 | return -ENOMEM; |
724 | } | |
69d0206c | 725 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
726 | } |
727 | return 0; | |
728 | } | |
729 | ||
c12f65d4 LZ |
730 | /** |
731 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 732 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 733 | * @cset: the css_set to be linked |
c12f65d4 LZ |
734 | * @cgrp: the destination cgroup |
735 | */ | |
69d0206c TH |
736 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
737 | struct cgroup *cgrp) | |
c12f65d4 | 738 | { |
69d0206c | 739 | struct cgrp_cset_link *link; |
c12f65d4 | 740 | |
69d0206c | 741 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
742 | |
743 | if (cgroup_on_dfl(cgrp)) | |
744 | cset->dfl_cgrp = cgrp; | |
745 | ||
69d0206c TH |
746 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
747 | link->cset = cset; | |
7717f7ba | 748 | link->cgrp = cgrp; |
842b597e TH |
749 | |
750 | if (list_empty(&cgrp->cset_links)) | |
751 | cgroup_update_populated(cgrp, true); | |
69d0206c | 752 | list_move(&link->cset_link, &cgrp->cset_links); |
842b597e | 753 | |
7717f7ba PM |
754 | /* |
755 | * Always add links to the tail of the list so that the list | |
756 | * is sorted by order of hierarchy creation | |
757 | */ | |
69d0206c | 758 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
759 | } |
760 | ||
b326f9d0 TH |
761 | /** |
762 | * find_css_set - return a new css_set with one cgroup updated | |
763 | * @old_cset: the baseline css_set | |
764 | * @cgrp: the cgroup to be updated | |
765 | * | |
766 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
767 | * substituted into the appropriate hierarchy. | |
817929ec | 768 | */ |
5abb8855 TH |
769 | static struct css_set *find_css_set(struct css_set *old_cset, |
770 | struct cgroup *cgrp) | |
817929ec | 771 | { |
b326f9d0 | 772 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 773 | struct css_set *cset; |
69d0206c TH |
774 | struct list_head tmp_links; |
775 | struct cgrp_cset_link *link; | |
2d8f243a | 776 | struct cgroup_subsys *ss; |
0ac801fe | 777 | unsigned long key; |
2d8f243a | 778 | int ssid; |
472b1053 | 779 | |
b326f9d0 TH |
780 | lockdep_assert_held(&cgroup_mutex); |
781 | ||
817929ec PM |
782 | /* First see if we already have a cgroup group that matches |
783 | * the desired set */ | |
96d365e0 | 784 | down_read(&css_set_rwsem); |
5abb8855 TH |
785 | cset = find_existing_css_set(old_cset, cgrp, template); |
786 | if (cset) | |
787 | get_css_set(cset); | |
96d365e0 | 788 | up_read(&css_set_rwsem); |
817929ec | 789 | |
5abb8855 TH |
790 | if (cset) |
791 | return cset; | |
817929ec | 792 | |
f4f4be2b | 793 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 794 | if (!cset) |
817929ec PM |
795 | return NULL; |
796 | ||
69d0206c | 797 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 798 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 799 | kfree(cset); |
817929ec PM |
800 | return NULL; |
801 | } | |
802 | ||
5abb8855 | 803 | atomic_set(&cset->refcount, 1); |
69d0206c | 804 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 805 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 806 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 807 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 808 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 809 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
810 | |
811 | /* Copy the set of subsystem state objects generated in | |
812 | * find_existing_css_set() */ | |
5abb8855 | 813 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 814 | |
96d365e0 | 815 | down_write(&css_set_rwsem); |
817929ec | 816 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 817 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 818 | struct cgroup *c = link->cgrp; |
69d0206c | 819 | |
7717f7ba PM |
820 | if (c->root == cgrp->root) |
821 | c = cgrp; | |
69d0206c | 822 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 823 | } |
817929ec | 824 | |
69d0206c | 825 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 826 | |
817929ec | 827 | css_set_count++; |
472b1053 | 828 | |
2d8f243a | 829 | /* Add @cset to the hash table */ |
5abb8855 TH |
830 | key = css_set_hash(cset->subsys); |
831 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 832 | |
2d8f243a TH |
833 | for_each_subsys(ss, ssid) |
834 | list_add_tail(&cset->e_cset_node[ssid], | |
835 | &cset->subsys[ssid]->cgroup->e_csets[ssid]); | |
836 | ||
96d365e0 | 837 | up_write(&css_set_rwsem); |
817929ec | 838 | |
5abb8855 | 839 | return cset; |
b4f48b63 PM |
840 | } |
841 | ||
3dd06ffa | 842 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 843 | { |
3dd06ffa | 844 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 845 | |
3dd06ffa | 846 | return root_cgrp->root; |
2bd59d48 TH |
847 | } |
848 | ||
3dd06ffa | 849 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
850 | { |
851 | int id; | |
852 | ||
853 | lockdep_assert_held(&cgroup_mutex); | |
854 | ||
985ed670 | 855 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
856 | if (id < 0) |
857 | return id; | |
858 | ||
859 | root->hierarchy_id = id; | |
860 | return 0; | |
861 | } | |
862 | ||
3dd06ffa | 863 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
864 | { |
865 | lockdep_assert_held(&cgroup_mutex); | |
866 | ||
867 | if (root->hierarchy_id) { | |
868 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
869 | root->hierarchy_id = 0; | |
870 | } | |
871 | } | |
872 | ||
3dd06ffa | 873 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
874 | { |
875 | if (root) { | |
876 | /* hierarhcy ID shoulid already have been released */ | |
877 | WARN_ON_ONCE(root->hierarchy_id); | |
878 | ||
879 | idr_destroy(&root->cgroup_idr); | |
880 | kfree(root); | |
881 | } | |
882 | } | |
883 | ||
3dd06ffa | 884 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 885 | { |
3dd06ffa | 886 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 887 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 888 | |
2bd59d48 | 889 | mutex_lock(&cgroup_tree_mutex); |
2bd59d48 | 890 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 891 | |
776f02fa | 892 | BUG_ON(atomic_read(&root->nr_cgrps)); |
f2e85d57 TH |
893 | BUG_ON(!list_empty(&cgrp->children)); |
894 | ||
f2e85d57 | 895 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 896 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 897 | |
7717f7ba | 898 | /* |
f2e85d57 TH |
899 | * Release all the links from cset_links to this hierarchy's |
900 | * root cgroup | |
7717f7ba | 901 | */ |
96d365e0 | 902 | down_write(&css_set_rwsem); |
f2e85d57 TH |
903 | |
904 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
905 | list_del(&link->cset_link); | |
906 | list_del(&link->cgrp_link); | |
907 | kfree(link); | |
908 | } | |
96d365e0 | 909 | up_write(&css_set_rwsem); |
f2e85d57 TH |
910 | |
911 | if (!list_empty(&root->root_list)) { | |
912 | list_del(&root->root_list); | |
913 | cgroup_root_count--; | |
914 | } | |
915 | ||
916 | cgroup_exit_root_id(root); | |
917 | ||
918 | mutex_unlock(&cgroup_mutex); | |
919 | mutex_unlock(&cgroup_tree_mutex); | |
f2e85d57 | 920 | |
2bd59d48 | 921 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
922 | cgroup_free_root(root); |
923 | } | |
924 | ||
ceb6a081 TH |
925 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
926 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 927 | struct cgroup_root *root) |
7717f7ba | 928 | { |
7717f7ba PM |
929 | struct cgroup *res = NULL; |
930 | ||
96d365e0 TH |
931 | lockdep_assert_held(&cgroup_mutex); |
932 | lockdep_assert_held(&css_set_rwsem); | |
933 | ||
5abb8855 | 934 | if (cset == &init_css_set) { |
3dd06ffa | 935 | res = &root->cgrp; |
7717f7ba | 936 | } else { |
69d0206c TH |
937 | struct cgrp_cset_link *link; |
938 | ||
939 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 940 | struct cgroup *c = link->cgrp; |
69d0206c | 941 | |
7717f7ba PM |
942 | if (c->root == root) { |
943 | res = c; | |
944 | break; | |
945 | } | |
946 | } | |
947 | } | |
96d365e0 | 948 | |
7717f7ba PM |
949 | BUG_ON(!res); |
950 | return res; | |
951 | } | |
952 | ||
ddbcc7e8 | 953 | /* |
ceb6a081 TH |
954 | * Return the cgroup for "task" from the given hierarchy. Must be |
955 | * called with cgroup_mutex and css_set_rwsem held. | |
956 | */ | |
957 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 958 | struct cgroup_root *root) |
ceb6a081 TH |
959 | { |
960 | /* | |
961 | * No need to lock the task - since we hold cgroup_mutex the | |
962 | * task can't change groups, so the only thing that can happen | |
963 | * is that it exits and its css is set back to init_css_set. | |
964 | */ | |
965 | return cset_cgroup_from_root(task_css_set(task), root); | |
966 | } | |
967 | ||
ddbcc7e8 | 968 | /* |
ddbcc7e8 PM |
969 | * A task must hold cgroup_mutex to modify cgroups. |
970 | * | |
971 | * Any task can increment and decrement the count field without lock. | |
972 | * So in general, code holding cgroup_mutex can't rely on the count | |
973 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 974 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
975 | * means that no tasks are currently attached, therefore there is no |
976 | * way a task attached to that cgroup can fork (the other way to | |
977 | * increment the count). So code holding cgroup_mutex can safely | |
978 | * assume that if the count is zero, it will stay zero. Similarly, if | |
979 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
980 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
981 | * needs that mutex. | |
982 | * | |
ddbcc7e8 PM |
983 | * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't |
984 | * (usually) take cgroup_mutex. These are the two most performance | |
985 | * critical pieces of code here. The exception occurs on cgroup_exit(), | |
986 | * when a task in a notify_on_release cgroup exits. Then cgroup_mutex | |
987 | * is taken, and if the cgroup count is zero, a usermode call made | |
a043e3b2 LZ |
988 | * to the release agent with the name of the cgroup (path relative to |
989 | * the root of cgroup file system) as the argument. | |
ddbcc7e8 PM |
990 | * |
991 | * A cgroup can only be deleted if both its 'count' of using tasks | |
992 | * is zero, and its list of 'children' cgroups is empty. Since all | |
993 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 994 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 995 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 996 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
997 | * |
998 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 999 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1000 | */ |
1001 | ||
69dfa00c | 1002 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned int subsys_mask); |
2bd59d48 | 1003 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1004 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1005 | |
8d7e6fb0 TH |
1006 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1007 | char *buf) | |
ddbcc7e8 | 1008 | { |
8d7e6fb0 TH |
1009 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1010 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1011 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
1012 | cft->ss->name, cft->name); | |
1013 | else | |
1014 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1015 | return buf; | |
ddbcc7e8 PM |
1016 | } |
1017 | ||
f2e85d57 TH |
1018 | /** |
1019 | * cgroup_file_mode - deduce file mode of a control file | |
1020 | * @cft: the control file in question | |
1021 | * | |
1022 | * returns cft->mode if ->mode is not 0 | |
1023 | * returns S_IRUGO|S_IWUSR if it has both a read and a write handler | |
1024 | * returns S_IRUGO if it has only a read handler | |
1025 | * returns S_IWUSR if it has only a write hander | |
1026 | */ | |
1027 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1028 | { |
f2e85d57 | 1029 | umode_t mode = 0; |
65dff759 | 1030 | |
f2e85d57 TH |
1031 | if (cft->mode) |
1032 | return cft->mode; | |
1033 | ||
1034 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) | |
1035 | mode |= S_IRUGO; | |
1036 | ||
6770c64e | 1037 | if (cft->write_u64 || cft->write_s64 || cft->write) |
f2e85d57 TH |
1038 | mode |= S_IWUSR; |
1039 | ||
1040 | return mode; | |
65dff759 LZ |
1041 | } |
1042 | ||
be445626 LZ |
1043 | static void cgroup_free_fn(struct work_struct *work) |
1044 | { | |
ea15f8cc | 1045 | struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work); |
be445626 | 1046 | |
3c9c825b | 1047 | atomic_dec(&cgrp->root->nr_cgrps); |
b1a21367 | 1048 | cgroup_pidlist_destroy_all(cgrp); |
be445626 | 1049 | |
776f02fa TH |
1050 | if (cgrp->parent) { |
1051 | /* | |
1052 | * We get a ref to the parent, and put the ref when this | |
1053 | * cgroup is being freed, so it's guaranteed that the | |
1054 | * parent won't be destroyed before its children. | |
1055 | */ | |
1056 | cgroup_put(cgrp->parent); | |
1057 | kernfs_put(cgrp->kn); | |
1058 | kfree(cgrp); | |
1059 | } else { | |
1060 | /* | |
3dd06ffa | 1061 | * This is root cgroup's refcnt reaching zero, which |
776f02fa TH |
1062 | * indicates that the root should be released. |
1063 | */ | |
1064 | cgroup_destroy_root(cgrp->root); | |
1065 | } | |
be445626 LZ |
1066 | } |
1067 | ||
1068 | static void cgroup_free_rcu(struct rcu_head *head) | |
1069 | { | |
1070 | struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head); | |
1071 | ||
ea15f8cc | 1072 | INIT_WORK(&cgrp->destroy_work, cgroup_free_fn); |
e5fca243 | 1073 | queue_work(cgroup_destroy_wq, &cgrp->destroy_work); |
be445626 LZ |
1074 | } |
1075 | ||
59f5296b | 1076 | static void cgroup_get(struct cgroup *cgrp) |
ddbcc7e8 | 1077 | { |
2bd59d48 TH |
1078 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
1079 | WARN_ON_ONCE(atomic_read(&cgrp->refcnt) <= 0); | |
1080 | atomic_inc(&cgrp->refcnt); | |
ddbcc7e8 PM |
1081 | } |
1082 | ||
59f5296b | 1083 | static void cgroup_put(struct cgroup *cgrp) |
05ef1d7c | 1084 | { |
2bd59d48 TH |
1085 | if (!atomic_dec_and_test(&cgrp->refcnt)) |
1086 | return; | |
776f02fa | 1087 | if (WARN_ON_ONCE(cgrp->parent && !cgroup_is_dead(cgrp))) |
2bd59d48 | 1088 | return; |
05ef1d7c | 1089 | |
6fa4918d | 1090 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
2bd59d48 | 1091 | cgrp->id = -1; |
05ef1d7c | 1092 | |
2bd59d48 | 1093 | call_rcu(&cgrp->rcu_head, cgroup_free_rcu); |
ddbcc7e8 | 1094 | } |
05ef1d7c | 1095 | |
2739d3cc | 1096 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1097 | { |
2bd59d48 | 1098 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1099 | |
ace2bee8 | 1100 | lockdep_assert_held(&cgroup_tree_mutex); |
2bd59d48 | 1101 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1102 | } |
1103 | ||
13af07df | 1104 | /** |
628f7cd4 | 1105 | * cgroup_clear_dir - remove subsys files in a cgroup directory |
8f89140a | 1106 | * @cgrp: target cgroup |
13af07df AR |
1107 | * @subsys_mask: mask of the subsystem ids whose files should be removed |
1108 | */ | |
69dfa00c | 1109 | static void cgroup_clear_dir(struct cgroup *cgrp, unsigned int subsys_mask) |
05ef1d7c | 1110 | { |
13af07df | 1111 | struct cgroup_subsys *ss; |
b420ba7d | 1112 | int i; |
05ef1d7c | 1113 | |
b420ba7d | 1114 | for_each_subsys(ss, i) { |
0adb0704 | 1115 | struct cftype *cfts; |
b420ba7d | 1116 | |
69dfa00c | 1117 | if (!(subsys_mask & (1 << i))) |
13af07df | 1118 | continue; |
0adb0704 TH |
1119 | list_for_each_entry(cfts, &ss->cfts, node) |
1120 | cgroup_addrm_files(cgrp, cfts, false); | |
13af07df | 1121 | } |
ddbcc7e8 PM |
1122 | } |
1123 | ||
69dfa00c | 1124 | static int rebind_subsystems(struct cgroup_root *dst_root, unsigned int ss_mask) |
ddbcc7e8 | 1125 | { |
30159ec7 | 1126 | struct cgroup_subsys *ss; |
2d8f243a | 1127 | int ssid, i, ret; |
ddbcc7e8 | 1128 | |
ace2bee8 TH |
1129 | lockdep_assert_held(&cgroup_tree_mutex); |
1130 | lockdep_assert_held(&cgroup_mutex); | |
ddbcc7e8 | 1131 | |
5df36032 TH |
1132 | for_each_subsys(ss, ssid) { |
1133 | if (!(ss_mask & (1 << ssid))) | |
1134 | continue; | |
aae8aab4 | 1135 | |
7fd8c565 TH |
1136 | /* if @ss has non-root csses attached to it, can't move */ |
1137 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1138 | return -EBUSY; |
1d5be6b2 | 1139 | |
5df36032 | 1140 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1141 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1142 | return -EBUSY; |
ddbcc7e8 PM |
1143 | } |
1144 | ||
a2dd4247 TH |
1145 | ret = cgroup_populate_dir(&dst_root->cgrp, ss_mask); |
1146 | if (ret) { | |
1147 | if (dst_root != &cgrp_dfl_root) | |
5df36032 | 1148 | return ret; |
ddbcc7e8 | 1149 | |
a2dd4247 TH |
1150 | /* |
1151 | * Rebinding back to the default root is not allowed to | |
1152 | * fail. Using both default and non-default roots should | |
1153 | * be rare. Moving subsystems back and forth even more so. | |
1154 | * Just warn about it and continue. | |
1155 | */ | |
1156 | if (cgrp_dfl_root_visible) { | |
69dfa00c | 1157 | pr_warn("failed to create files (%d) while rebinding 0x%x to default root\n", |
a2a1f9ea | 1158 | ret, ss_mask); |
ed3d261b | 1159 | pr_warn("you may retry by moving them to a different hierarchy and unbinding\n"); |
a2dd4247 | 1160 | } |
5df36032 | 1161 | } |
3126121f TH |
1162 | |
1163 | /* | |
1164 | * Nothing can fail from this point on. Remove files for the | |
1165 | * removed subsystems and rebind each subsystem. | |
1166 | */ | |
4ac06017 | 1167 | mutex_unlock(&cgroup_mutex); |
5df36032 | 1168 | for_each_subsys(ss, ssid) |
a2dd4247 | 1169 | if (ss_mask & (1 << ssid)) |
3dd06ffa | 1170 | cgroup_clear_dir(&ss->root->cgrp, 1 << ssid); |
4ac06017 | 1171 | mutex_lock(&cgroup_mutex); |
a8a648c4 | 1172 | |
5df36032 | 1173 | for_each_subsys(ss, ssid) { |
3dd06ffa | 1174 | struct cgroup_root *src_root; |
5df36032 | 1175 | struct cgroup_subsys_state *css; |
2d8f243a | 1176 | struct css_set *cset; |
a8a648c4 | 1177 | |
5df36032 TH |
1178 | if (!(ss_mask & (1 << ssid))) |
1179 | continue; | |
a8a648c4 | 1180 | |
5df36032 | 1181 | src_root = ss->root; |
3dd06ffa | 1182 | css = cgroup_css(&src_root->cgrp, ss); |
a8a648c4 | 1183 | |
3dd06ffa | 1184 | WARN_ON(!css || cgroup_css(&dst_root->cgrp, ss)); |
73e80ed8 | 1185 | |
3dd06ffa TH |
1186 | RCU_INIT_POINTER(src_root->cgrp.subsys[ssid], NULL); |
1187 | rcu_assign_pointer(dst_root->cgrp.subsys[ssid], css); | |
5df36032 | 1188 | ss->root = dst_root; |
3dd06ffa | 1189 | css->cgroup = &dst_root->cgrp; |
73e80ed8 | 1190 | |
2d8f243a TH |
1191 | down_write(&css_set_rwsem); |
1192 | hash_for_each(css_set_table, i, cset, hlist) | |
1193 | list_move_tail(&cset->e_cset_node[ss->id], | |
1194 | &dst_root->cgrp.e_csets[ss->id]); | |
1195 | up_write(&css_set_rwsem); | |
1196 | ||
f392e51c TH |
1197 | src_root->subsys_mask &= ~(1 << ssid); |
1198 | src_root->cgrp.child_subsys_mask &= ~(1 << ssid); | |
1199 | ||
bd53d617 | 1200 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1201 | dst_root->subsys_mask |= 1 << ssid; |
bd53d617 TH |
1202 | if (dst_root != &cgrp_dfl_root) |
1203 | dst_root->cgrp.child_subsys_mask |= 1 << ssid; | |
a8a648c4 | 1204 | |
5df36032 TH |
1205 | if (ss->bind) |
1206 | ss->bind(css); | |
ddbcc7e8 | 1207 | } |
ddbcc7e8 | 1208 | |
a2dd4247 | 1209 | kernfs_activate(dst_root->cgrp.kn); |
ddbcc7e8 PM |
1210 | return 0; |
1211 | } | |
1212 | ||
2bd59d48 TH |
1213 | static int cgroup_show_options(struct seq_file *seq, |
1214 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1215 | { |
3dd06ffa | 1216 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1217 | struct cgroup_subsys *ss; |
b85d2040 | 1218 | int ssid; |
ddbcc7e8 | 1219 | |
b85d2040 | 1220 | for_each_subsys(ss, ssid) |
f392e51c | 1221 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 1222 | seq_printf(seq, ",%s", ss->name); |
873fe09e TH |
1223 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) |
1224 | seq_puts(seq, ",sane_behavior"); | |
93438629 | 1225 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1226 | seq_puts(seq, ",noprefix"); |
93438629 | 1227 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1228 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1229 | |
1230 | spin_lock(&release_agent_path_lock); | |
81a6a5cd PM |
1231 | if (strlen(root->release_agent_path)) |
1232 | seq_printf(seq, ",release_agent=%s", root->release_agent_path); | |
69e943b7 TH |
1233 | spin_unlock(&release_agent_path_lock); |
1234 | ||
3dd06ffa | 1235 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1236 | seq_puts(seq, ",clone_children"); |
c6d57f33 PM |
1237 | if (strlen(root->name)) |
1238 | seq_printf(seq, ",name=%s", root->name); | |
ddbcc7e8 PM |
1239 | return 0; |
1240 | } | |
1241 | ||
1242 | struct cgroup_sb_opts { | |
69dfa00c TH |
1243 | unsigned int subsys_mask; |
1244 | unsigned int flags; | |
81a6a5cd | 1245 | char *release_agent; |
2260e7fc | 1246 | bool cpuset_clone_children; |
c6d57f33 | 1247 | char *name; |
2c6ab6d2 PM |
1248 | /* User explicitly requested empty subsystem */ |
1249 | bool none; | |
ddbcc7e8 PM |
1250 | }; |
1251 | ||
cf5d5941 | 1252 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1253 | { |
32a8cf23 DL |
1254 | char *token, *o = data; |
1255 | bool all_ss = false, one_ss = false; | |
69dfa00c | 1256 | unsigned int mask = -1U; |
30159ec7 TH |
1257 | struct cgroup_subsys *ss; |
1258 | int i; | |
f9ab5b5b LZ |
1259 | |
1260 | #ifdef CONFIG_CPUSETS | |
69dfa00c | 1261 | mask = ~(1U << cpuset_cgrp_id); |
f9ab5b5b | 1262 | #endif |
ddbcc7e8 | 1263 | |
c6d57f33 | 1264 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1265 | |
1266 | while ((token = strsep(&o, ",")) != NULL) { | |
1267 | if (!*token) | |
1268 | return -EINVAL; | |
32a8cf23 | 1269 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1270 | /* Explicitly have no subsystems */ |
1271 | opts->none = true; | |
32a8cf23 DL |
1272 | continue; |
1273 | } | |
1274 | if (!strcmp(token, "all")) { | |
1275 | /* Mutually exclusive option 'all' + subsystem name */ | |
1276 | if (one_ss) | |
1277 | return -EINVAL; | |
1278 | all_ss = true; | |
1279 | continue; | |
1280 | } | |
873fe09e TH |
1281 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1282 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1283 | continue; | |
1284 | } | |
32a8cf23 | 1285 | if (!strcmp(token, "noprefix")) { |
93438629 | 1286 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1287 | continue; |
1288 | } | |
1289 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1290 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1291 | continue; |
1292 | } | |
03b1cde6 | 1293 | if (!strcmp(token, "xattr")) { |
93438629 | 1294 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1295 | continue; |
1296 | } | |
32a8cf23 | 1297 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1298 | /* Specifying two release agents is forbidden */ |
1299 | if (opts->release_agent) | |
1300 | return -EINVAL; | |
c6d57f33 | 1301 | opts->release_agent = |
e400c285 | 1302 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1303 | if (!opts->release_agent) |
1304 | return -ENOMEM; | |
32a8cf23 DL |
1305 | continue; |
1306 | } | |
1307 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1308 | const char *name = token + 5; |
1309 | /* Can't specify an empty name */ | |
1310 | if (!strlen(name)) | |
1311 | return -EINVAL; | |
1312 | /* Must match [\w.-]+ */ | |
1313 | for (i = 0; i < strlen(name); i++) { | |
1314 | char c = name[i]; | |
1315 | if (isalnum(c)) | |
1316 | continue; | |
1317 | if ((c == '.') || (c == '-') || (c == '_')) | |
1318 | continue; | |
1319 | return -EINVAL; | |
1320 | } | |
1321 | /* Specifying two names is forbidden */ | |
1322 | if (opts->name) | |
1323 | return -EINVAL; | |
1324 | opts->name = kstrndup(name, | |
e400c285 | 1325 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1326 | GFP_KERNEL); |
1327 | if (!opts->name) | |
1328 | return -ENOMEM; | |
32a8cf23 DL |
1329 | |
1330 | continue; | |
1331 | } | |
1332 | ||
30159ec7 | 1333 | for_each_subsys(ss, i) { |
32a8cf23 DL |
1334 | if (strcmp(token, ss->name)) |
1335 | continue; | |
1336 | if (ss->disabled) | |
1337 | continue; | |
1338 | ||
1339 | /* Mutually exclusive option 'all' + subsystem name */ | |
1340 | if (all_ss) | |
1341 | return -EINVAL; | |
69dfa00c | 1342 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1343 | one_ss = true; |
1344 | ||
1345 | break; | |
1346 | } | |
1347 | if (i == CGROUP_SUBSYS_COUNT) | |
1348 | return -ENOENT; | |
1349 | } | |
1350 | ||
2c6ab6d2 PM |
1351 | /* Consistency checks */ |
1352 | ||
873fe09e | 1353 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
ed3d261b | 1354 | pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); |
873fe09e | 1355 | |
d3ba07c3 TH |
1356 | if ((opts->flags & (CGRP_ROOT_NOPREFIX | CGRP_ROOT_XATTR)) || |
1357 | opts->cpuset_clone_children || opts->release_agent || | |
1358 | opts->name) { | |
ed3d261b | 1359 | pr_err("sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n"); |
873fe09e TH |
1360 | return -EINVAL; |
1361 | } | |
a2dd4247 TH |
1362 | } else { |
1363 | /* | |
1364 | * If the 'all' option was specified select all the | |
1365 | * subsystems, otherwise if 'none', 'name=' and a subsystem | |
1366 | * name options were not specified, let's default to 'all' | |
1367 | */ | |
1368 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1369 | for_each_subsys(ss, i) | |
1370 | if (!ss->disabled) | |
69dfa00c | 1371 | opts->subsys_mask |= (1 << i); |
873fe09e | 1372 | |
a2dd4247 TH |
1373 | /* |
1374 | * We either have to specify by name or by subsystems. (So | |
1375 | * all empty hierarchies must have a name). | |
1376 | */ | |
1377 | if (!opts->subsys_mask && !opts->name) | |
873fe09e | 1378 | return -EINVAL; |
873fe09e TH |
1379 | } |
1380 | ||
f9ab5b5b LZ |
1381 | /* |
1382 | * Option noprefix was introduced just for backward compatibility | |
1383 | * with the old cpuset, so we allow noprefix only if mounting just | |
1384 | * the cpuset subsystem. | |
1385 | */ | |
93438629 | 1386 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1387 | return -EINVAL; |
1388 | ||
2c6ab6d2 PM |
1389 | |
1390 | /* Can't specify "none" and some subsystems */ | |
a1a71b45 | 1391 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1392 | return -EINVAL; |
1393 | ||
ddbcc7e8 PM |
1394 | return 0; |
1395 | } | |
1396 | ||
2bd59d48 | 1397 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1398 | { |
1399 | int ret = 0; | |
3dd06ffa | 1400 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1401 | struct cgroup_sb_opts opts; |
69dfa00c | 1402 | unsigned int added_mask, removed_mask; |
ddbcc7e8 | 1403 | |
873fe09e | 1404 | if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
ed3d261b | 1405 | pr_err("sane_behavior: remount is not allowed\n"); |
873fe09e TH |
1406 | return -EINVAL; |
1407 | } | |
1408 | ||
ace2bee8 | 1409 | mutex_lock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1410 | mutex_lock(&cgroup_mutex); |
1411 | ||
1412 | /* See what subsystems are wanted */ | |
1413 | ret = parse_cgroupfs_options(data, &opts); | |
1414 | if (ret) | |
1415 | goto out_unlock; | |
1416 | ||
f392e51c | 1417 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1418 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1419 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1420 | |
f392e51c TH |
1421 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1422 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1423 | |
cf5d5941 | 1424 | /* Don't allow flags or name to change at remount */ |
0ce6cba3 | 1425 | if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) || |
cf5d5941 | 1426 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1427 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
0ce6cba3 TH |
1428 | opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "", |
1429 | root->flags & CGRP_ROOT_OPTION_MASK, root->name); | |
c6d57f33 PM |
1430 | ret = -EINVAL; |
1431 | goto out_unlock; | |
1432 | } | |
1433 | ||
f172e67c | 1434 | /* remounting is not allowed for populated hierarchies */ |
3dd06ffa | 1435 | if (!list_empty(&root->cgrp.children)) { |
f172e67c | 1436 | ret = -EBUSY; |
0670e08b | 1437 | goto out_unlock; |
cf5d5941 | 1438 | } |
ddbcc7e8 | 1439 | |
5df36032 | 1440 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1441 | if (ret) |
0670e08b | 1442 | goto out_unlock; |
ddbcc7e8 | 1443 | |
3dd06ffa | 1444 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1445 | |
69e943b7 TH |
1446 | if (opts.release_agent) { |
1447 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1448 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1449 | spin_unlock(&release_agent_path_lock); |
1450 | } | |
ddbcc7e8 | 1451 | out_unlock: |
66bdc9cf | 1452 | kfree(opts.release_agent); |
c6d57f33 | 1453 | kfree(opts.name); |
ddbcc7e8 | 1454 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1455 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
1456 | return ret; |
1457 | } | |
1458 | ||
afeb0f9f TH |
1459 | /* |
1460 | * To reduce the fork() overhead for systems that are not actually using | |
1461 | * their cgroups capability, we don't maintain the lists running through | |
1462 | * each css_set to its tasks until we see the list actually used - in other | |
1463 | * words after the first mount. | |
1464 | */ | |
1465 | static bool use_task_css_set_links __read_mostly; | |
1466 | ||
1467 | static void cgroup_enable_task_cg_lists(void) | |
1468 | { | |
1469 | struct task_struct *p, *g; | |
1470 | ||
96d365e0 | 1471 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1472 | |
1473 | if (use_task_css_set_links) | |
1474 | goto out_unlock; | |
1475 | ||
1476 | use_task_css_set_links = true; | |
1477 | ||
1478 | /* | |
1479 | * We need tasklist_lock because RCU is not safe against | |
1480 | * while_each_thread(). Besides, a forking task that has passed | |
1481 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1482 | * is not guaranteed to have its child immediately visible in the | |
1483 | * tasklist if we walk through it with RCU. | |
1484 | */ | |
1485 | read_lock(&tasklist_lock); | |
1486 | do_each_thread(g, p) { | |
afeb0f9f TH |
1487 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1488 | task_css_set(p) != &init_css_set); | |
1489 | ||
1490 | /* | |
1491 | * We should check if the process is exiting, otherwise | |
1492 | * it will race with cgroup_exit() in that the list | |
1493 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1494 | * Do it while holding siglock so that we don't end up |
1495 | * racing against cgroup_exit(). | |
afeb0f9f | 1496 | */ |
f153ad11 | 1497 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1498 | if (!(p->flags & PF_EXITING)) { |
1499 | struct css_set *cset = task_css_set(p); | |
1500 | ||
1501 | list_add(&p->cg_list, &cset->tasks); | |
1502 | get_css_set(cset); | |
1503 | } | |
f153ad11 | 1504 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1505 | } while_each_thread(g, p); |
1506 | read_unlock(&tasklist_lock); | |
1507 | out_unlock: | |
96d365e0 | 1508 | up_write(&css_set_rwsem); |
afeb0f9f | 1509 | } |
ddbcc7e8 | 1510 | |
cc31edce PM |
1511 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1512 | { | |
2d8f243a TH |
1513 | struct cgroup_subsys *ss; |
1514 | int ssid; | |
1515 | ||
2bd59d48 | 1516 | atomic_set(&cgrp->refcnt, 1); |
cc31edce PM |
1517 | INIT_LIST_HEAD(&cgrp->sibling); |
1518 | INIT_LIST_HEAD(&cgrp->children); | |
69d0206c | 1519 | INIT_LIST_HEAD(&cgrp->cset_links); |
cc31edce | 1520 | INIT_LIST_HEAD(&cgrp->release_list); |
72a8cb30 BB |
1521 | INIT_LIST_HEAD(&cgrp->pidlists); |
1522 | mutex_init(&cgrp->pidlist_mutex); | |
67f4c36f | 1523 | cgrp->dummy_css.cgroup = cgrp; |
2d8f243a TH |
1524 | |
1525 | for_each_subsys(ss, ssid) | |
1526 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1527 | |
1528 | init_waitqueue_head(&cgrp->offline_waitq); | |
cc31edce | 1529 | } |
c6d57f33 | 1530 | |
3dd06ffa | 1531 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1532 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1533 | { |
3dd06ffa | 1534 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1535 | |
ddbcc7e8 | 1536 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1537 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1538 | cgrp->root = root; |
cc31edce | 1539 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1540 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1541 | |
c6d57f33 PM |
1542 | root->flags = opts->flags; |
1543 | if (opts->release_agent) | |
1544 | strcpy(root->release_agent_path, opts->release_agent); | |
1545 | if (opts->name) | |
1546 | strcpy(root->name, opts->name); | |
2260e7fc | 1547 | if (opts->cpuset_clone_children) |
3dd06ffa | 1548 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1549 | } |
1550 | ||
69dfa00c | 1551 | static int cgroup_setup_root(struct cgroup_root *root, unsigned int ss_mask) |
2c6ab6d2 | 1552 | { |
d427dfeb | 1553 | LIST_HEAD(tmp_links); |
3dd06ffa | 1554 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1555 | struct css_set *cset; |
d427dfeb | 1556 | int i, ret; |
2c6ab6d2 | 1557 | |
d427dfeb TH |
1558 | lockdep_assert_held(&cgroup_tree_mutex); |
1559 | lockdep_assert_held(&cgroup_mutex); | |
c6d57f33 | 1560 | |
6fa4918d | 1561 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_NOWAIT); |
d427dfeb | 1562 | if (ret < 0) |
2bd59d48 | 1563 | goto out; |
d427dfeb | 1564 | root_cgrp->id = ret; |
c6d57f33 | 1565 | |
d427dfeb | 1566 | /* |
96d365e0 | 1567 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1568 | * but that's OK - it can only be increased by someone holding |
1569 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1570 | * have some link structures left over | |
1571 | */ | |
1572 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1573 | if (ret) | |
2bd59d48 | 1574 | goto out; |
ddbcc7e8 | 1575 | |
985ed670 | 1576 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1577 | if (ret) |
2bd59d48 | 1578 | goto out; |
ddbcc7e8 | 1579 | |
2bd59d48 TH |
1580 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1581 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1582 | root_cgrp); | |
1583 | if (IS_ERR(root->kf_root)) { | |
1584 | ret = PTR_ERR(root->kf_root); | |
1585 | goto exit_root_id; | |
1586 | } | |
1587 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1588 | |
d427dfeb TH |
1589 | ret = cgroup_addrm_files(root_cgrp, cgroup_base_files, true); |
1590 | if (ret) | |
2bd59d48 | 1591 | goto destroy_root; |
ddbcc7e8 | 1592 | |
5df36032 | 1593 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1594 | if (ret) |
2bd59d48 | 1595 | goto destroy_root; |
ddbcc7e8 | 1596 | |
d427dfeb TH |
1597 | /* |
1598 | * There must be no failure case after here, since rebinding takes | |
1599 | * care of subsystems' refcounts, which are explicitly dropped in | |
1600 | * the failure exit path. | |
1601 | */ | |
1602 | list_add(&root->root_list, &cgroup_roots); | |
1603 | cgroup_root_count++; | |
0df6a63f | 1604 | |
d427dfeb | 1605 | /* |
3dd06ffa | 1606 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1607 | * objects. |
1608 | */ | |
96d365e0 | 1609 | down_write(&css_set_rwsem); |
d427dfeb TH |
1610 | hash_for_each(css_set_table, i, cset, hlist) |
1611 | link_css_set(&tmp_links, cset, root_cgrp); | |
96d365e0 | 1612 | up_write(&css_set_rwsem); |
ddbcc7e8 | 1613 | |
d427dfeb | 1614 | BUG_ON(!list_empty(&root_cgrp->children)); |
3c9c825b | 1615 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1616 | |
2bd59d48 | 1617 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1618 | ret = 0; |
2bd59d48 | 1619 | goto out; |
d427dfeb | 1620 | |
2bd59d48 TH |
1621 | destroy_root: |
1622 | kernfs_destroy_root(root->kf_root); | |
1623 | root->kf_root = NULL; | |
1624 | exit_root_id: | |
d427dfeb | 1625 | cgroup_exit_root_id(root); |
2bd59d48 | 1626 | out: |
d427dfeb TH |
1627 | free_cgrp_cset_links(&tmp_links); |
1628 | return ret; | |
ddbcc7e8 PM |
1629 | } |
1630 | ||
f7e83571 | 1631 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1632 | int flags, const char *unused_dev_name, |
f7e83571 | 1633 | void *data) |
ddbcc7e8 | 1634 | { |
3dd06ffa | 1635 | struct cgroup_root *root; |
ddbcc7e8 | 1636 | struct cgroup_sb_opts opts; |
2bd59d48 | 1637 | struct dentry *dentry; |
8e30e2b8 | 1638 | int ret; |
c6b3d5bc | 1639 | bool new_sb; |
ddbcc7e8 | 1640 | |
56fde9e0 TH |
1641 | /* |
1642 | * The first time anyone tries to mount a cgroup, enable the list | |
1643 | * linking each css_set to its tasks and fix up all existing tasks. | |
1644 | */ | |
1645 | if (!use_task_css_set_links) | |
1646 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 1647 | |
8e30e2b8 | 1648 | mutex_lock(&cgroup_tree_mutex); |
aae8aab4 | 1649 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1650 | |
1651 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1652 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1653 | if (ret) |
8e30e2b8 | 1654 | goto out_unlock; |
e37a06f1 | 1655 | retry: |
2bd59d48 | 1656 | /* look for a matching existing root */ |
a2dd4247 TH |
1657 | if (!opts.subsys_mask && !opts.none && !opts.name) { |
1658 | cgrp_dfl_root_visible = true; | |
1659 | root = &cgrp_dfl_root; | |
1660 | cgroup_get(&root->cgrp); | |
1661 | ret = 0; | |
1662 | goto out_unlock; | |
ddbcc7e8 PM |
1663 | } |
1664 | ||
985ed670 | 1665 | for_each_root(root) { |
2bd59d48 | 1666 | bool name_match = false; |
3126121f | 1667 | |
3dd06ffa | 1668 | if (root == &cgrp_dfl_root) |
985ed670 | 1669 | continue; |
3126121f | 1670 | |
cf5d5941 | 1671 | /* |
2bd59d48 TH |
1672 | * If we asked for a name then it must match. Also, if |
1673 | * name matches but sybsys_mask doesn't, we should fail. | |
1674 | * Remember whether name matched. | |
cf5d5941 | 1675 | */ |
2bd59d48 TH |
1676 | if (opts.name) { |
1677 | if (strcmp(opts.name, root->name)) | |
1678 | continue; | |
1679 | name_match = true; | |
1680 | } | |
ddbcc7e8 | 1681 | |
c6d57f33 | 1682 | /* |
2bd59d48 TH |
1683 | * If we asked for subsystems (or explicitly for no |
1684 | * subsystems) then they must match. | |
c6d57f33 | 1685 | */ |
2bd59d48 | 1686 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 1687 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
1688 | if (!name_match) |
1689 | continue; | |
1690 | ret = -EBUSY; | |
1691 | goto out_unlock; | |
1692 | } | |
873fe09e | 1693 | |
c7ba8287 | 1694 | if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) { |
2a0ff3fb | 1695 | if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) { |
ed3d261b | 1696 | pr_err("sane_behavior: new mount options should match the existing superblock\n"); |
2a0ff3fb | 1697 | ret = -EINVAL; |
8e30e2b8 | 1698 | goto out_unlock; |
2a0ff3fb | 1699 | } else { |
ed3d261b | 1700 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); |
2a0ff3fb | 1701 | } |
873fe09e | 1702 | } |
ddbcc7e8 | 1703 | |
776f02fa | 1704 | /* |
3dd06ffa | 1705 | * A root's lifetime is governed by its root cgroup. Zero |
776f02fa TH |
1706 | * ref indicate that the root is being destroyed. Wait for |
1707 | * destruction to complete so that the subsystems are free. | |
1708 | * We can use wait_queue for the wait but this path is | |
1709 | * super cold. Let's just sleep for a bit and retry. | |
1710 | */ | |
3dd06ffa | 1711 | if (!atomic_inc_not_zero(&root->cgrp.refcnt)) { |
776f02fa TH |
1712 | mutex_unlock(&cgroup_mutex); |
1713 | mutex_unlock(&cgroup_tree_mutex); | |
1714 | msleep(10); | |
e37a06f1 LZ |
1715 | mutex_lock(&cgroup_tree_mutex); |
1716 | mutex_lock(&cgroup_mutex); | |
776f02fa TH |
1717 | goto retry; |
1718 | } | |
ddbcc7e8 | 1719 | |
776f02fa | 1720 | ret = 0; |
2bd59d48 | 1721 | goto out_unlock; |
ddbcc7e8 | 1722 | } |
ddbcc7e8 | 1723 | |
817929ec | 1724 | /* |
172a2c06 TH |
1725 | * No such thing, create a new one. name= matching without subsys |
1726 | * specification is allowed for already existing hierarchies but we | |
1727 | * can't create new one without subsys specification. | |
817929ec | 1728 | */ |
172a2c06 TH |
1729 | if (!opts.subsys_mask && !opts.none) { |
1730 | ret = -EINVAL; | |
1731 | goto out_unlock; | |
817929ec | 1732 | } |
817929ec | 1733 | |
172a2c06 TH |
1734 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
1735 | if (!root) { | |
1736 | ret = -ENOMEM; | |
2bd59d48 | 1737 | goto out_unlock; |
839ec545 | 1738 | } |
e5f6a860 | 1739 | |
172a2c06 TH |
1740 | init_cgroup_root(root, &opts); |
1741 | ||
35585573 | 1742 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
1743 | if (ret) |
1744 | cgroup_free_root(root); | |
fa3ca07e | 1745 | |
8e30e2b8 | 1746 | out_unlock: |
ddbcc7e8 | 1747 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 1748 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 | 1749 | |
c6d57f33 PM |
1750 | kfree(opts.release_agent); |
1751 | kfree(opts.name); | |
03b1cde6 | 1752 | |
2bd59d48 | 1753 | if (ret) |
8e30e2b8 | 1754 | return ERR_PTR(ret); |
2bd59d48 | 1755 | |
c6b3d5bc LZ |
1756 | dentry = kernfs_mount(fs_type, flags, root->kf_root, &new_sb); |
1757 | if (IS_ERR(dentry) || !new_sb) | |
3dd06ffa | 1758 | cgroup_put(&root->cgrp); |
2bd59d48 TH |
1759 | return dentry; |
1760 | } | |
1761 | ||
1762 | static void cgroup_kill_sb(struct super_block *sb) | |
1763 | { | |
1764 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 1765 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 1766 | |
3dd06ffa | 1767 | cgroup_put(&root->cgrp); |
2bd59d48 | 1768 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
1769 | } |
1770 | ||
1771 | static struct file_system_type cgroup_fs_type = { | |
1772 | .name = "cgroup", | |
f7e83571 | 1773 | .mount = cgroup_mount, |
ddbcc7e8 PM |
1774 | .kill_sb = cgroup_kill_sb, |
1775 | }; | |
1776 | ||
676db4af GK |
1777 | static struct kobject *cgroup_kobj; |
1778 | ||
857a2beb | 1779 | /** |
913ffdb5 | 1780 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1781 | * @task: target task |
857a2beb TH |
1782 | * @buf: the buffer to write the path into |
1783 | * @buflen: the length of the buffer | |
1784 | * | |
913ffdb5 TH |
1785 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1786 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1787 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1788 | * cgroup controller callbacks. | |
1789 | * | |
e61734c5 | 1790 | * Return value is the same as kernfs_path(). |
857a2beb | 1791 | */ |
e61734c5 | 1792 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 1793 | { |
3dd06ffa | 1794 | struct cgroup_root *root; |
913ffdb5 | 1795 | struct cgroup *cgrp; |
e61734c5 TH |
1796 | int hierarchy_id = 1; |
1797 | char *path = NULL; | |
857a2beb TH |
1798 | |
1799 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 1800 | down_read(&css_set_rwsem); |
857a2beb | 1801 | |
913ffdb5 TH |
1802 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1803 | ||
857a2beb TH |
1804 | if (root) { |
1805 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 1806 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
1807 | } else { |
1808 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
1809 | if (strlcpy(buf, "/", buflen) < buflen) |
1810 | path = buf; | |
857a2beb TH |
1811 | } |
1812 | ||
96d365e0 | 1813 | up_read(&css_set_rwsem); |
857a2beb | 1814 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 1815 | return path; |
857a2beb | 1816 | } |
913ffdb5 | 1817 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1818 | |
b3dc094e | 1819 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 1820 | struct cgroup_taskset { |
b3dc094e TH |
1821 | /* the src and dst cset list running through cset->mg_node */ |
1822 | struct list_head src_csets; | |
1823 | struct list_head dst_csets; | |
1824 | ||
1825 | /* | |
1826 | * Fields for cgroup_taskset_*() iteration. | |
1827 | * | |
1828 | * Before migration is committed, the target migration tasks are on | |
1829 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
1830 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
1831 | * or ->dst_csets depending on whether migration is committed. | |
1832 | * | |
1833 | * ->cur_csets and ->cur_task point to the current task position | |
1834 | * during iteration. | |
1835 | */ | |
1836 | struct list_head *csets; | |
1837 | struct css_set *cur_cset; | |
1838 | struct task_struct *cur_task; | |
2f7ee569 TH |
1839 | }; |
1840 | ||
1841 | /** | |
1842 | * cgroup_taskset_first - reset taskset and return the first task | |
1843 | * @tset: taskset of interest | |
1844 | * | |
1845 | * @tset iteration is initialized and the first task is returned. | |
1846 | */ | |
1847 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
1848 | { | |
b3dc094e TH |
1849 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
1850 | tset->cur_task = NULL; | |
1851 | ||
1852 | return cgroup_taskset_next(tset); | |
2f7ee569 | 1853 | } |
2f7ee569 TH |
1854 | |
1855 | /** | |
1856 | * cgroup_taskset_next - iterate to the next task in taskset | |
1857 | * @tset: taskset of interest | |
1858 | * | |
1859 | * Return the next task in @tset. Iteration must have been initialized | |
1860 | * with cgroup_taskset_first(). | |
1861 | */ | |
1862 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
1863 | { | |
b3dc094e TH |
1864 | struct css_set *cset = tset->cur_cset; |
1865 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 1866 | |
b3dc094e TH |
1867 | while (&cset->mg_node != tset->csets) { |
1868 | if (!task) | |
1869 | task = list_first_entry(&cset->mg_tasks, | |
1870 | struct task_struct, cg_list); | |
1871 | else | |
1872 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 1873 | |
b3dc094e TH |
1874 | if (&task->cg_list != &cset->mg_tasks) { |
1875 | tset->cur_cset = cset; | |
1876 | tset->cur_task = task; | |
1877 | return task; | |
1878 | } | |
2f7ee569 | 1879 | |
b3dc094e TH |
1880 | cset = list_next_entry(cset, mg_node); |
1881 | task = NULL; | |
1882 | } | |
2f7ee569 | 1883 | |
b3dc094e | 1884 | return NULL; |
2f7ee569 | 1885 | } |
2f7ee569 | 1886 | |
cb0f1fe9 | 1887 | /** |
74a1166d | 1888 | * cgroup_task_migrate - move a task from one cgroup to another. |
60106946 | 1889 | * @old_cgrp: the cgroup @tsk is being migrated from |
cb0f1fe9 TH |
1890 | * @tsk: the task being migrated |
1891 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 1892 | * |
cb0f1fe9 | 1893 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 1894 | */ |
5abb8855 TH |
1895 | static void cgroup_task_migrate(struct cgroup *old_cgrp, |
1896 | struct task_struct *tsk, | |
1897 | struct css_set *new_cset) | |
74a1166d | 1898 | { |
5abb8855 | 1899 | struct css_set *old_cset; |
74a1166d | 1900 | |
cb0f1fe9 TH |
1901 | lockdep_assert_held(&cgroup_mutex); |
1902 | lockdep_assert_held(&css_set_rwsem); | |
1903 | ||
74a1166d | 1904 | /* |
026085ef MSB |
1905 | * We are synchronized through threadgroup_lock() against PF_EXITING |
1906 | * setting such that we can't race against cgroup_exit() changing the | |
1907 | * css_set to init_css_set and dropping the old one. | |
74a1166d | 1908 | */ |
c84cdf75 | 1909 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 1910 | old_cset = task_css_set(tsk); |
74a1166d | 1911 | |
b3dc094e | 1912 | get_css_set(new_cset); |
5abb8855 | 1913 | rcu_assign_pointer(tsk->cgroups, new_cset); |
74a1166d | 1914 | |
1b9aba49 TH |
1915 | /* |
1916 | * Use move_tail so that cgroup_taskset_first() still returns the | |
1917 | * leader after migration. This works because cgroup_migrate() | |
1918 | * ensures that the dst_cset of the leader is the first on the | |
1919 | * tset's dst_csets list. | |
1920 | */ | |
1921 | list_move_tail(&tsk->cg_list, &new_cset->mg_tasks); | |
74a1166d BB |
1922 | |
1923 | /* | |
5abb8855 TH |
1924 | * We just gained a reference on old_cset by taking it from the |
1925 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
1926 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 1927 | */ |
5abb8855 | 1928 | set_bit(CGRP_RELEASABLE, &old_cgrp->flags); |
cb0f1fe9 | 1929 | put_css_set_locked(old_cset, false); |
74a1166d BB |
1930 | } |
1931 | ||
a043e3b2 | 1932 | /** |
1958d2d5 TH |
1933 | * cgroup_migrate_finish - cleanup after attach |
1934 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 1935 | * |
1958d2d5 TH |
1936 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
1937 | * those functions for details. | |
74a1166d | 1938 | */ |
1958d2d5 | 1939 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 1940 | { |
1958d2d5 | 1941 | struct css_set *cset, *tmp_cset; |
74a1166d | 1942 | |
1958d2d5 TH |
1943 | lockdep_assert_held(&cgroup_mutex); |
1944 | ||
1945 | down_write(&css_set_rwsem); | |
1946 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
1947 | cset->mg_src_cgrp = NULL; | |
1948 | cset->mg_dst_cset = NULL; | |
1949 | list_del_init(&cset->mg_preload_node); | |
1950 | put_css_set_locked(cset, false); | |
1951 | } | |
1952 | up_write(&css_set_rwsem); | |
1953 | } | |
1954 | ||
1955 | /** | |
1956 | * cgroup_migrate_add_src - add a migration source css_set | |
1957 | * @src_cset: the source css_set to add | |
1958 | * @dst_cgrp: the destination cgroup | |
1959 | * @preloaded_csets: list of preloaded css_sets | |
1960 | * | |
1961 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
1962 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
1963 | * up by cgroup_migrate_finish(). | |
1964 | * | |
1965 | * This function may be called without holding threadgroup_lock even if the | |
1966 | * target is a process. Threads may be created and destroyed but as long | |
1967 | * as cgroup_mutex is not dropped, no new css_set can be put into play and | |
1968 | * the preloaded css_sets are guaranteed to cover all migrations. | |
1969 | */ | |
1970 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
1971 | struct cgroup *dst_cgrp, | |
1972 | struct list_head *preloaded_csets) | |
1973 | { | |
1974 | struct cgroup *src_cgrp; | |
1975 | ||
1976 | lockdep_assert_held(&cgroup_mutex); | |
1977 | lockdep_assert_held(&css_set_rwsem); | |
1978 | ||
1979 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
1980 | ||
1958d2d5 TH |
1981 | if (!list_empty(&src_cset->mg_preload_node)) |
1982 | return; | |
1983 | ||
1984 | WARN_ON(src_cset->mg_src_cgrp); | |
1985 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
1986 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
1987 | ||
1988 | src_cset->mg_src_cgrp = src_cgrp; | |
1989 | get_css_set(src_cset); | |
1990 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
1991 | } | |
1992 | ||
1993 | /** | |
1994 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 1995 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
1996 | * @preloaded_csets: list of preloaded source css_sets |
1997 | * | |
1998 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
1999 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
2000 | * pins all destination css_sets, links each to its source, and append them |
2001 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
2002 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
2003 | * |
2004 | * This function must be called after cgroup_migrate_add_src() has been | |
2005 | * called on each migration source css_set. After migration is performed | |
2006 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2007 | * @preloaded_csets. | |
2008 | */ | |
2009 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
2010 | struct list_head *preloaded_csets) | |
2011 | { | |
2012 | LIST_HEAD(csets); | |
f817de98 | 2013 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2014 | |
2015 | lockdep_assert_held(&cgroup_mutex); | |
2016 | ||
f8f22e53 TH |
2017 | /* |
2018 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2019 | * with tasks so that child cgroups don't compete against tasks. | |
2020 | */ | |
2021 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && dst_cgrp->parent && | |
2022 | dst_cgrp->child_subsys_mask) | |
2023 | return -EBUSY; | |
2024 | ||
1958d2d5 | 2025 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 2026 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2027 | struct css_set *dst_cset; |
2028 | ||
f817de98 TH |
2029 | dst_cset = find_css_set(src_cset, |
2030 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
2031 | if (!dst_cset) |
2032 | goto err; | |
2033 | ||
2034 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2035 | |
2036 | /* | |
2037 | * If src cset equals dst, it's noop. Drop the src. | |
2038 | * cgroup_migrate() will skip the cset too. Note that we | |
2039 | * can't handle src == dst as some nodes are used by both. | |
2040 | */ | |
2041 | if (src_cset == dst_cset) { | |
2042 | src_cset->mg_src_cgrp = NULL; | |
2043 | list_del_init(&src_cset->mg_preload_node); | |
2044 | put_css_set(src_cset, false); | |
2045 | put_css_set(dst_cset, false); | |
2046 | continue; | |
2047 | } | |
2048 | ||
1958d2d5 TH |
2049 | src_cset->mg_dst_cset = dst_cset; |
2050 | ||
2051 | if (list_empty(&dst_cset->mg_preload_node)) | |
2052 | list_add(&dst_cset->mg_preload_node, &csets); | |
2053 | else | |
2054 | put_css_set(dst_cset, false); | |
2055 | } | |
2056 | ||
f817de98 | 2057 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2058 | return 0; |
2059 | err: | |
2060 | cgroup_migrate_finish(&csets); | |
2061 | return -ENOMEM; | |
2062 | } | |
2063 | ||
2064 | /** | |
2065 | * cgroup_migrate - migrate a process or task to a cgroup | |
2066 | * @cgrp: the destination cgroup | |
2067 | * @leader: the leader of the process or the task to migrate | |
2068 | * @threadgroup: whether @leader points to the whole process or a single task | |
2069 | * | |
2070 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
2071 | * process, the caller must be holding threadgroup_lock of @leader. The | |
2072 | * caller is also responsible for invoking cgroup_migrate_add_src() and | |
2073 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2074 | * function and following up with cgroup_migrate_finish(). | |
2075 | * | |
2076 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2077 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2078 | * failure, when migrating multiple targets, the success or failure can be | |
2079 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2080 | * actually starting migrating. | |
2081 | */ | |
2082 | static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader, | |
2083 | bool threadgroup) | |
74a1166d | 2084 | { |
b3dc094e TH |
2085 | struct cgroup_taskset tset = { |
2086 | .src_csets = LIST_HEAD_INIT(tset.src_csets), | |
2087 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), | |
2088 | .csets = &tset.src_csets, | |
2089 | }; | |
1c6727af | 2090 | struct cgroup_subsys_state *css, *failed_css = NULL; |
b3dc094e TH |
2091 | struct css_set *cset, *tmp_cset; |
2092 | struct task_struct *task, *tmp_task; | |
2093 | int i, ret; | |
74a1166d | 2094 | |
fb5d2b4c MSB |
2095 | /* |
2096 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2097 | * already PF_EXITING could be freed from underneath us unless we | |
2098 | * take an rcu_read_lock. | |
2099 | */ | |
b3dc094e | 2100 | down_write(&css_set_rwsem); |
fb5d2b4c | 2101 | rcu_read_lock(); |
9db8de37 | 2102 | task = leader; |
74a1166d | 2103 | do { |
9db8de37 TH |
2104 | /* @task either already exited or can't exit until the end */ |
2105 | if (task->flags & PF_EXITING) | |
ea84753c | 2106 | goto next; |
134d3373 | 2107 | |
eaf797ab TH |
2108 | /* leave @task alone if post_fork() hasn't linked it yet */ |
2109 | if (list_empty(&task->cg_list)) | |
ea84753c | 2110 | goto next; |
cd3d0952 | 2111 | |
b3dc094e | 2112 | cset = task_css_set(task); |
1958d2d5 | 2113 | if (!cset->mg_src_cgrp) |
ea84753c | 2114 | goto next; |
b3dc094e | 2115 | |
61d1d219 | 2116 | /* |
1b9aba49 TH |
2117 | * cgroup_taskset_first() must always return the leader. |
2118 | * Take care to avoid disturbing the ordering. | |
61d1d219 | 2119 | */ |
1b9aba49 TH |
2120 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2121 | if (list_empty(&cset->mg_node)) | |
2122 | list_add_tail(&cset->mg_node, &tset.src_csets); | |
2123 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2124 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2125 | &tset.dst_csets); | |
ea84753c | 2126 | next: |
081aa458 LZ |
2127 | if (!threadgroup) |
2128 | break; | |
9db8de37 | 2129 | } while_each_thread(leader, task); |
fb5d2b4c | 2130 | rcu_read_unlock(); |
b3dc094e | 2131 | up_write(&css_set_rwsem); |
74a1166d | 2132 | |
134d3373 | 2133 | /* methods shouldn't be called if no task is actually migrating */ |
b3dc094e TH |
2134 | if (list_empty(&tset.src_csets)) |
2135 | return 0; | |
134d3373 | 2136 | |
1958d2d5 | 2137 | /* check that we can legitimately attach to the cgroup */ |
aec3dfcb | 2138 | for_each_e_css(css, i, cgrp) { |
1c6727af | 2139 | if (css->ss->can_attach) { |
9db8de37 TH |
2140 | ret = css->ss->can_attach(css, &tset); |
2141 | if (ret) { | |
1c6727af | 2142 | failed_css = css; |
74a1166d BB |
2143 | goto out_cancel_attach; |
2144 | } | |
2145 | } | |
74a1166d BB |
2146 | } |
2147 | ||
2148 | /* | |
1958d2d5 TH |
2149 | * Now that we're guaranteed success, proceed to move all tasks to |
2150 | * the new cgroup. There are no failure cases after here, so this | |
2151 | * is the commit point. | |
74a1166d | 2152 | */ |
cb0f1fe9 | 2153 | down_write(&css_set_rwsem); |
b3dc094e TH |
2154 | list_for_each_entry(cset, &tset.src_csets, mg_node) { |
2155 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
2156 | cgroup_task_migrate(cset->mg_src_cgrp, task, | |
2157 | cset->mg_dst_cset); | |
74a1166d | 2158 | } |
cb0f1fe9 | 2159 | up_write(&css_set_rwsem); |
74a1166d BB |
2160 | |
2161 | /* | |
1958d2d5 TH |
2162 | * Migration is committed, all target tasks are now on dst_csets. |
2163 | * Nothing is sensitive to fork() after this point. Notify | |
2164 | * controllers that migration is complete. | |
74a1166d | 2165 | */ |
1958d2d5 | 2166 | tset.csets = &tset.dst_csets; |
74a1166d | 2167 | |
aec3dfcb | 2168 | for_each_e_css(css, i, cgrp) |
1c6727af TH |
2169 | if (css->ss->attach) |
2170 | css->ss->attach(css, &tset); | |
74a1166d | 2171 | |
9db8de37 | 2172 | ret = 0; |
b3dc094e TH |
2173 | goto out_release_tset; |
2174 | ||
74a1166d | 2175 | out_cancel_attach: |
aec3dfcb | 2176 | for_each_e_css(css, i, cgrp) { |
b3dc094e TH |
2177 | if (css == failed_css) |
2178 | break; | |
2179 | if (css->ss->cancel_attach) | |
2180 | css->ss->cancel_attach(css, &tset); | |
74a1166d | 2181 | } |
b3dc094e TH |
2182 | out_release_tset: |
2183 | down_write(&css_set_rwsem); | |
2184 | list_splice_init(&tset.dst_csets, &tset.src_csets); | |
2185 | list_for_each_entry_safe(cset, tmp_cset, &tset.src_csets, mg_node) { | |
1b9aba49 | 2186 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); |
b3dc094e | 2187 | list_del_init(&cset->mg_node); |
b3dc094e TH |
2188 | } |
2189 | up_write(&css_set_rwsem); | |
9db8de37 | 2190 | return ret; |
74a1166d BB |
2191 | } |
2192 | ||
1958d2d5 TH |
2193 | /** |
2194 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2195 | * @dst_cgrp: the cgroup to attach to | |
2196 | * @leader: the task or the leader of the threadgroup to be attached | |
2197 | * @threadgroup: attach the whole threadgroup? | |
2198 | * | |
0e1d768f | 2199 | * Call holding cgroup_mutex and threadgroup_lock of @leader. |
1958d2d5 TH |
2200 | */ |
2201 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2202 | struct task_struct *leader, bool threadgroup) | |
2203 | { | |
2204 | LIST_HEAD(preloaded_csets); | |
2205 | struct task_struct *task; | |
2206 | int ret; | |
2207 | ||
2208 | /* look up all src csets */ | |
2209 | down_read(&css_set_rwsem); | |
2210 | rcu_read_lock(); | |
2211 | task = leader; | |
2212 | do { | |
2213 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2214 | &preloaded_csets); | |
2215 | if (!threadgroup) | |
2216 | break; | |
2217 | } while_each_thread(leader, task); | |
2218 | rcu_read_unlock(); | |
2219 | up_read(&css_set_rwsem); | |
2220 | ||
2221 | /* prepare dst csets and commit */ | |
2222 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2223 | if (!ret) | |
2224 | ret = cgroup_migrate(dst_cgrp, leader, threadgroup); | |
2225 | ||
2226 | cgroup_migrate_finish(&preloaded_csets); | |
2227 | return ret; | |
74a1166d BB |
2228 | } |
2229 | ||
2230 | /* | |
2231 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2232 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2233 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2234 | */ |
74a1166d | 2235 | static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup) |
bbcb81d0 | 2236 | { |
bbcb81d0 | 2237 | struct task_struct *tsk; |
c69e8d9c | 2238 | const struct cred *cred = current_cred(), *tcred; |
bbcb81d0 PM |
2239 | int ret; |
2240 | ||
74a1166d BB |
2241 | if (!cgroup_lock_live_group(cgrp)) |
2242 | return -ENODEV; | |
2243 | ||
b78949eb MSB |
2244 | retry_find_task: |
2245 | rcu_read_lock(); | |
bbcb81d0 | 2246 | if (pid) { |
73507f33 | 2247 | tsk = find_task_by_vpid(pid); |
74a1166d BB |
2248 | if (!tsk) { |
2249 | rcu_read_unlock(); | |
dd4b0a46 | 2250 | ret = -ESRCH; |
b78949eb | 2251 | goto out_unlock_cgroup; |
bbcb81d0 | 2252 | } |
74a1166d BB |
2253 | /* |
2254 | * even if we're attaching all tasks in the thread group, we | |
2255 | * only need to check permissions on one of them. | |
2256 | */ | |
c69e8d9c | 2257 | tcred = __task_cred(tsk); |
14a590c3 EB |
2258 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && |
2259 | !uid_eq(cred->euid, tcred->uid) && | |
2260 | !uid_eq(cred->euid, tcred->suid)) { | |
c69e8d9c | 2261 | rcu_read_unlock(); |
b78949eb MSB |
2262 | ret = -EACCES; |
2263 | goto out_unlock_cgroup; | |
bbcb81d0 | 2264 | } |
b78949eb MSB |
2265 | } else |
2266 | tsk = current; | |
cd3d0952 TH |
2267 | |
2268 | if (threadgroup) | |
b78949eb | 2269 | tsk = tsk->group_leader; |
c4c27fbd MG |
2270 | |
2271 | /* | |
14a40ffc | 2272 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2273 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2274 | * with no rt_runtime allocated. Just say no. | |
2275 | */ | |
14a40ffc | 2276 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd MG |
2277 | ret = -EINVAL; |
2278 | rcu_read_unlock(); | |
2279 | goto out_unlock_cgroup; | |
2280 | } | |
2281 | ||
b78949eb MSB |
2282 | get_task_struct(tsk); |
2283 | rcu_read_unlock(); | |
2284 | ||
2285 | threadgroup_lock(tsk); | |
2286 | if (threadgroup) { | |
2287 | if (!thread_group_leader(tsk)) { | |
2288 | /* | |
2289 | * a race with de_thread from another thread's exec() | |
2290 | * may strip us of our leadership, if this happens, | |
2291 | * there is no choice but to throw this task away and | |
2292 | * try again; this is | |
2293 | * "double-double-toil-and-trouble-check locking". | |
2294 | */ | |
2295 | threadgroup_unlock(tsk); | |
2296 | put_task_struct(tsk); | |
2297 | goto retry_find_task; | |
2298 | } | |
081aa458 LZ |
2299 | } |
2300 | ||
2301 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
2302 | ||
cd3d0952 TH |
2303 | threadgroup_unlock(tsk); |
2304 | ||
bbcb81d0 | 2305 | put_task_struct(tsk); |
b78949eb | 2306 | out_unlock_cgroup: |
47cfcd09 | 2307 | mutex_unlock(&cgroup_mutex); |
bbcb81d0 PM |
2308 | return ret; |
2309 | } | |
2310 | ||
7ae1bad9 TH |
2311 | /** |
2312 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2313 | * @from: attach to all cgroups of a given task | |
2314 | * @tsk: the task to be attached | |
2315 | */ | |
2316 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2317 | { | |
3dd06ffa | 2318 | struct cgroup_root *root; |
7ae1bad9 TH |
2319 | int retval = 0; |
2320 | ||
47cfcd09 | 2321 | mutex_lock(&cgroup_mutex); |
985ed670 | 2322 | for_each_root(root) { |
96d365e0 TH |
2323 | struct cgroup *from_cgrp; |
2324 | ||
3dd06ffa | 2325 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2326 | continue; |
2327 | ||
96d365e0 TH |
2328 | down_read(&css_set_rwsem); |
2329 | from_cgrp = task_cgroup_from_root(from, root); | |
2330 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2331 | |
6f4b7e63 | 2332 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2333 | if (retval) |
2334 | break; | |
2335 | } | |
47cfcd09 | 2336 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2337 | |
2338 | return retval; | |
2339 | } | |
2340 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2341 | ||
182446d0 TH |
2342 | static int cgroup_tasks_write(struct cgroup_subsys_state *css, |
2343 | struct cftype *cft, u64 pid) | |
74a1166d | 2344 | { |
182446d0 | 2345 | return attach_task_by_pid(css->cgroup, pid, false); |
74a1166d BB |
2346 | } |
2347 | ||
182446d0 TH |
2348 | static int cgroup_procs_write(struct cgroup_subsys_state *css, |
2349 | struct cftype *cft, u64 tgid) | |
af351026 | 2350 | { |
182446d0 | 2351 | return attach_task_by_pid(css->cgroup, tgid, true); |
af351026 PM |
2352 | } |
2353 | ||
451af504 TH |
2354 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2355 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2356 | { |
451af504 TH |
2357 | struct cgroup *cgrp = of_css(of)->cgroup; |
2358 | struct cgroup_root *root = cgrp->root; | |
5f469907 TH |
2359 | |
2360 | BUILD_BUG_ON(sizeof(root->release_agent_path) < PATH_MAX); | |
451af504 | 2361 | if (!cgroup_lock_live_group(cgrp)) |
e788e066 | 2362 | return -ENODEV; |
69e943b7 | 2363 | spin_lock(&release_agent_path_lock); |
451af504 | 2364 | strlcpy(root->release_agent_path, strstrip(buf), |
5f469907 | 2365 | sizeof(root->release_agent_path)); |
69e943b7 | 2366 | spin_unlock(&release_agent_path_lock); |
47cfcd09 | 2367 | mutex_unlock(&cgroup_mutex); |
451af504 | 2368 | return nbytes; |
e788e066 PM |
2369 | } |
2370 | ||
2da8ca82 | 2371 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2372 | { |
2da8ca82 | 2373 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2374 | |
46cfeb04 | 2375 | spin_lock(&release_agent_path_lock); |
e788e066 | 2376 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 2377 | spin_unlock(&release_agent_path_lock); |
e788e066 | 2378 | seq_putc(seq, '\n'); |
e788e066 PM |
2379 | return 0; |
2380 | } | |
2381 | ||
2da8ca82 | 2382 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2383 | { |
2da8ca82 TH |
2384 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2385 | ||
2386 | seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp)); | |
e788e066 PM |
2387 | return 0; |
2388 | } | |
2389 | ||
f8f22e53 TH |
2390 | static void cgroup_print_ss_mask(struct seq_file *seq, unsigned int ss_mask) |
2391 | { | |
2392 | struct cgroup_subsys *ss; | |
2393 | bool printed = false; | |
2394 | int ssid; | |
2395 | ||
2396 | for_each_subsys(ss, ssid) { | |
2397 | if (ss_mask & (1 << ssid)) { | |
2398 | if (printed) | |
2399 | seq_putc(seq, ' '); | |
2400 | seq_printf(seq, "%s", ss->name); | |
2401 | printed = true; | |
2402 | } | |
2403 | } | |
2404 | if (printed) | |
2405 | seq_putc(seq, '\n'); | |
2406 | } | |
2407 | ||
2408 | /* show controllers which are currently attached to the default hierarchy */ | |
2409 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
2410 | { | |
2411 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2412 | ||
2413 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask); | |
2414 | return 0; | |
2415 | } | |
2416 | ||
2417 | /* show controllers which are enabled from the parent */ | |
2418 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
2419 | { | |
2420 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2421 | ||
2422 | cgroup_print_ss_mask(seq, cgrp->parent->child_subsys_mask); | |
2423 | return 0; | |
2424 | } | |
2425 | ||
2426 | /* show controllers which are enabled for a given cgroup's children */ | |
2427 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
2428 | { | |
2429 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
2430 | ||
2431 | cgroup_print_ss_mask(seq, cgrp->child_subsys_mask); | |
2432 | return 0; | |
2433 | } | |
2434 | ||
2435 | /** | |
2436 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2437 | * @cgrp: root of the subtree to update csses for | |
2438 | * | |
2439 | * @cgrp's child_subsys_mask has changed and its subtree's (self excluded) | |
2440 | * css associations need to be updated accordingly. This function looks up | |
2441 | * all css_sets which are attached to the subtree, creates the matching | |
2442 | * updated css_sets and migrates the tasks to the new ones. | |
2443 | */ | |
2444 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2445 | { | |
2446 | LIST_HEAD(preloaded_csets); | |
2447 | struct cgroup_subsys_state *css; | |
2448 | struct css_set *src_cset; | |
2449 | int ret; | |
2450 | ||
2451 | lockdep_assert_held(&cgroup_tree_mutex); | |
2452 | lockdep_assert_held(&cgroup_mutex); | |
2453 | ||
2454 | /* look up all csses currently attached to @cgrp's subtree */ | |
2455 | down_read(&css_set_rwsem); | |
2456 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { | |
2457 | struct cgrp_cset_link *link; | |
2458 | ||
2459 | /* self is not affected by child_subsys_mask change */ | |
2460 | if (css->cgroup == cgrp) | |
2461 | continue; | |
2462 | ||
2463 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2464 | cgroup_migrate_add_src(link->cset, cgrp, | |
2465 | &preloaded_csets); | |
2466 | } | |
2467 | up_read(&css_set_rwsem); | |
2468 | ||
2469 | /* NULL dst indicates self on default hierarchy */ | |
2470 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2471 | if (ret) | |
2472 | goto out_finish; | |
2473 | ||
2474 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { | |
2475 | struct task_struct *last_task = NULL, *task; | |
2476 | ||
2477 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2478 | if (!src_cset->mg_src_cgrp) | |
2479 | break; | |
2480 | ||
2481 | /* | |
2482 | * All tasks in src_cset need to be migrated to the | |
2483 | * matching dst_cset. Empty it process by process. We | |
2484 | * walk tasks but migrate processes. The leader might even | |
2485 | * belong to a different cset but such src_cset would also | |
2486 | * be among the target src_csets because the default | |
2487 | * hierarchy enforces per-process membership. | |
2488 | */ | |
2489 | while (true) { | |
2490 | down_read(&css_set_rwsem); | |
2491 | task = list_first_entry_or_null(&src_cset->tasks, | |
2492 | struct task_struct, cg_list); | |
2493 | if (task) { | |
2494 | task = task->group_leader; | |
2495 | WARN_ON_ONCE(!task_css_set(task)->mg_src_cgrp); | |
2496 | get_task_struct(task); | |
2497 | } | |
2498 | up_read(&css_set_rwsem); | |
2499 | ||
2500 | if (!task) | |
2501 | break; | |
2502 | ||
2503 | /* guard against possible infinite loop */ | |
2504 | if (WARN(last_task == task, | |
2505 | "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n")) | |
2506 | goto out_finish; | |
2507 | last_task = task; | |
2508 | ||
2509 | threadgroup_lock(task); | |
2510 | /* raced against de_thread() from another thread? */ | |
2511 | if (!thread_group_leader(task)) { | |
2512 | threadgroup_unlock(task); | |
2513 | put_task_struct(task); | |
2514 | continue; | |
2515 | } | |
2516 | ||
2517 | ret = cgroup_migrate(src_cset->dfl_cgrp, task, true); | |
2518 | ||
2519 | threadgroup_unlock(task); | |
2520 | put_task_struct(task); | |
2521 | ||
2522 | if (WARN(ret, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret)) | |
2523 | goto out_finish; | |
2524 | } | |
2525 | } | |
2526 | ||
2527 | out_finish: | |
2528 | cgroup_migrate_finish(&preloaded_csets); | |
2529 | return ret; | |
2530 | } | |
2531 | ||
2532 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
451af504 TH |
2533 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
2534 | char *buf, size_t nbytes, | |
2535 | loff_t off) | |
f8f22e53 | 2536 | { |
7d331fa9 | 2537 | unsigned int enable = 0, disable = 0; |
451af504 | 2538 | struct cgroup *cgrp = of_css(of)->cgroup, *child; |
f8f22e53 | 2539 | struct cgroup_subsys *ss; |
451af504 | 2540 | char *tok; |
f8f22e53 TH |
2541 | int ssid, ret; |
2542 | ||
2543 | /* | |
d37167ab TH |
2544 | * Parse input - space separated list of subsystem names prefixed |
2545 | * with either + or -. | |
f8f22e53 | 2546 | */ |
451af504 TH |
2547 | buf = strstrip(buf); |
2548 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
2549 | if (tok[0] == '\0') |
2550 | continue; | |
f8f22e53 TH |
2551 | for_each_subsys(ss, ssid) { |
2552 | if (ss->disabled || strcmp(tok + 1, ss->name)) | |
2553 | continue; | |
2554 | ||
2555 | if (*tok == '+') { | |
7d331fa9 TH |
2556 | enable |= 1 << ssid; |
2557 | disable &= ~(1 << ssid); | |
f8f22e53 | 2558 | } else if (*tok == '-') { |
7d331fa9 TH |
2559 | disable |= 1 << ssid; |
2560 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
2561 | } else { |
2562 | return -EINVAL; | |
2563 | } | |
2564 | break; | |
2565 | } | |
2566 | if (ssid == CGROUP_SUBSYS_COUNT) | |
2567 | return -EINVAL; | |
2568 | } | |
2569 | ||
2570 | /* | |
2571 | * We're gonna grab cgroup_tree_mutex which nests outside kernfs | |
2572 | * active_ref. cgroup_lock_live_group() already provides enough | |
2573 | * protection. Ensure @cgrp stays accessible and break the | |
2574 | * active_ref protection. | |
2575 | */ | |
2576 | cgroup_get(cgrp); | |
2577 | kernfs_break_active_protection(cgrp->control_kn); | |
f8f22e53 TH |
2578 | |
2579 | mutex_lock(&cgroup_tree_mutex); | |
2580 | ||
2581 | for_each_subsys(ss, ssid) { | |
2582 | if (enable & (1 << ssid)) { | |
2583 | if (cgrp->child_subsys_mask & (1 << ssid)) { | |
2584 | enable &= ~(1 << ssid); | |
2585 | continue; | |
2586 | } | |
2587 | ||
2588 | /* | |
2589 | * Because css offlining is asynchronous, userland | |
2590 | * might try to re-enable the same controller while | |
2591 | * the previous instance is still around. In such | |
2592 | * cases, wait till it's gone using offline_waitq. | |
2593 | */ | |
2594 | cgroup_for_each_live_child(child, cgrp) { | |
0cee8b77 | 2595 | DEFINE_WAIT(wait); |
f8f22e53 TH |
2596 | |
2597 | if (!cgroup_css(child, ss)) | |
2598 | continue; | |
2599 | ||
0cee8b77 | 2600 | cgroup_get(child); |
f8f22e53 TH |
2601 | prepare_to_wait(&child->offline_waitq, &wait, |
2602 | TASK_UNINTERRUPTIBLE); | |
2603 | mutex_unlock(&cgroup_tree_mutex); | |
2604 | schedule(); | |
2605 | finish_wait(&child->offline_waitq, &wait); | |
0cee8b77 | 2606 | cgroup_put(child); |
7d331fa9 TH |
2607 | |
2608 | ret = restart_syscall(); | |
2609 | goto out_unbreak; | |
f8f22e53 TH |
2610 | } |
2611 | ||
2612 | /* unavailable or not enabled on the parent? */ | |
2613 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
2614 | (cgrp->parent && | |
2615 | !(cgrp->parent->child_subsys_mask & (1 << ssid)))) { | |
2616 | ret = -ENOENT; | |
2617 | goto out_unlock_tree; | |
2618 | } | |
2619 | } else if (disable & (1 << ssid)) { | |
2620 | if (!(cgrp->child_subsys_mask & (1 << ssid))) { | |
2621 | disable &= ~(1 << ssid); | |
2622 | continue; | |
2623 | } | |
2624 | ||
2625 | /* a child has it enabled? */ | |
2626 | cgroup_for_each_live_child(child, cgrp) { | |
2627 | if (child->child_subsys_mask & (1 << ssid)) { | |
2628 | ret = -EBUSY; | |
2629 | goto out_unlock_tree; | |
2630 | } | |
2631 | } | |
2632 | } | |
2633 | } | |
2634 | ||
2635 | if (!enable && !disable) { | |
2636 | ret = 0; | |
2637 | goto out_unlock_tree; | |
2638 | } | |
2639 | ||
2640 | if (!cgroup_lock_live_group(cgrp)) { | |
2641 | ret = -ENODEV; | |
2642 | goto out_unlock_tree; | |
2643 | } | |
2644 | ||
2645 | /* | |
2646 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2647 | * with tasks so that child cgroups don't compete against tasks. | |
2648 | */ | |
2649 | if (enable && cgrp->parent && !list_empty(&cgrp->cset_links)) { | |
2650 | ret = -EBUSY; | |
2651 | goto out_unlock; | |
2652 | } | |
2653 | ||
2654 | /* | |
2655 | * Create csses for enables and update child_subsys_mask. This | |
2656 | * changes cgroup_e_css() results which in turn makes the | |
2657 | * subsequent cgroup_update_dfl_csses() associate all tasks in the | |
2658 | * subtree to the updated csses. | |
2659 | */ | |
2660 | for_each_subsys(ss, ssid) { | |
2661 | if (!(enable & (1 << ssid))) | |
2662 | continue; | |
2663 | ||
2664 | cgroup_for_each_live_child(child, cgrp) { | |
2665 | ret = create_css(child, ss); | |
2666 | if (ret) | |
2667 | goto err_undo_css; | |
2668 | } | |
2669 | } | |
2670 | ||
2671 | cgrp->child_subsys_mask |= enable; | |
2672 | cgrp->child_subsys_mask &= ~disable; | |
2673 | ||
2674 | ret = cgroup_update_dfl_csses(cgrp); | |
2675 | if (ret) | |
2676 | goto err_undo_css; | |
2677 | ||
2678 | /* all tasks are now migrated away from the old csses, kill them */ | |
2679 | for_each_subsys(ss, ssid) { | |
2680 | if (!(disable & (1 << ssid))) | |
2681 | continue; | |
2682 | ||
2683 | cgroup_for_each_live_child(child, cgrp) | |
2684 | kill_css(cgroup_css(child, ss)); | |
2685 | } | |
2686 | ||
2687 | kernfs_activate(cgrp->kn); | |
2688 | ret = 0; | |
2689 | out_unlock: | |
2690 | mutex_unlock(&cgroup_mutex); | |
2691 | out_unlock_tree: | |
2692 | mutex_unlock(&cgroup_tree_mutex); | |
7d331fa9 | 2693 | out_unbreak: |
f8f22e53 TH |
2694 | kernfs_unbreak_active_protection(cgrp->control_kn); |
2695 | cgroup_put(cgrp); | |
451af504 | 2696 | return ret ?: nbytes; |
f8f22e53 TH |
2697 | |
2698 | err_undo_css: | |
2699 | cgrp->child_subsys_mask &= ~enable; | |
2700 | cgrp->child_subsys_mask |= disable; | |
2701 | ||
2702 | for_each_subsys(ss, ssid) { | |
2703 | if (!(enable & (1 << ssid))) | |
2704 | continue; | |
2705 | ||
2706 | cgroup_for_each_live_child(child, cgrp) { | |
2707 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
2708 | if (css) | |
2709 | kill_css(css); | |
2710 | } | |
2711 | } | |
2712 | goto out_unlock; | |
2713 | } | |
2714 | ||
842b597e TH |
2715 | static int cgroup_populated_show(struct seq_file *seq, void *v) |
2716 | { | |
2717 | seq_printf(seq, "%d\n", (bool)seq_css(seq)->cgroup->populated_cnt); | |
2718 | return 0; | |
2719 | } | |
2720 | ||
2bd59d48 TH |
2721 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2722 | size_t nbytes, loff_t off) | |
355e0c48 | 2723 | { |
2bd59d48 TH |
2724 | struct cgroup *cgrp = of->kn->parent->priv; |
2725 | struct cftype *cft = of->kn->priv; | |
2726 | struct cgroup_subsys_state *css; | |
a742c59d | 2727 | int ret; |
355e0c48 | 2728 | |
b4168640 TH |
2729 | if (cft->write) |
2730 | return cft->write(of, buf, nbytes, off); | |
2731 | ||
2bd59d48 TH |
2732 | /* |
2733 | * kernfs guarantees that a file isn't deleted with operations in | |
2734 | * flight, which means that the matching css is and stays alive and | |
2735 | * doesn't need to be pinned. The RCU locking is not necessary | |
2736 | * either. It's just for the convenience of using cgroup_css(). | |
2737 | */ | |
2738 | rcu_read_lock(); | |
2739 | css = cgroup_css(cgrp, cft->ss); | |
2740 | rcu_read_unlock(); | |
a742c59d | 2741 | |
451af504 | 2742 | if (cft->write_u64) { |
a742c59d TH |
2743 | unsigned long long v; |
2744 | ret = kstrtoull(buf, 0, &v); | |
2745 | if (!ret) | |
2746 | ret = cft->write_u64(css, cft, v); | |
2747 | } else if (cft->write_s64) { | |
2748 | long long v; | |
2749 | ret = kstrtoll(buf, 0, &v); | |
2750 | if (!ret) | |
2751 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 2752 | } else { |
a742c59d | 2753 | ret = -EINVAL; |
e73d2c61 | 2754 | } |
2bd59d48 | 2755 | |
a742c59d | 2756 | return ret ?: nbytes; |
355e0c48 PM |
2757 | } |
2758 | ||
6612f05b | 2759 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2760 | { |
2bd59d48 | 2761 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2762 | } |
2763 | ||
6612f05b | 2764 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2765 | { |
2bd59d48 | 2766 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
2767 | } |
2768 | ||
6612f05b | 2769 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 2770 | { |
2bd59d48 | 2771 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
2772 | } |
2773 | ||
91796569 | 2774 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 2775 | { |
7da11279 TH |
2776 | struct cftype *cft = seq_cft(m); |
2777 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 2778 | |
2da8ca82 TH |
2779 | if (cft->seq_show) |
2780 | return cft->seq_show(m, arg); | |
e73d2c61 | 2781 | |
f4c753b7 | 2782 | if (cft->read_u64) |
896f5199 TH |
2783 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
2784 | else if (cft->read_s64) | |
2785 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
2786 | else | |
2787 | return -EINVAL; | |
2788 | return 0; | |
91796569 PM |
2789 | } |
2790 | ||
2bd59d48 TH |
2791 | static struct kernfs_ops cgroup_kf_single_ops = { |
2792 | .atomic_write_len = PAGE_SIZE, | |
2793 | .write = cgroup_file_write, | |
2794 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
2795 | }; |
2796 | ||
2bd59d48 TH |
2797 | static struct kernfs_ops cgroup_kf_ops = { |
2798 | .atomic_write_len = PAGE_SIZE, | |
2799 | .write = cgroup_file_write, | |
2800 | .seq_start = cgroup_seqfile_start, | |
2801 | .seq_next = cgroup_seqfile_next, | |
2802 | .seq_stop = cgroup_seqfile_stop, | |
2803 | .seq_show = cgroup_seqfile_show, | |
2804 | }; | |
ddbcc7e8 PM |
2805 | |
2806 | /* | |
2807 | * cgroup_rename - Only allow simple rename of directories in place. | |
2808 | */ | |
2bd59d48 TH |
2809 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
2810 | const char *new_name_str) | |
ddbcc7e8 | 2811 | { |
2bd59d48 | 2812 | struct cgroup *cgrp = kn->priv; |
65dff759 | 2813 | int ret; |
65dff759 | 2814 | |
2bd59d48 | 2815 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 2816 | return -ENOTDIR; |
2bd59d48 | 2817 | if (kn->parent != new_parent) |
ddbcc7e8 | 2818 | return -EIO; |
65dff759 | 2819 | |
6db8e85c TH |
2820 | /* |
2821 | * This isn't a proper migration and its usefulness is very | |
2822 | * limited. Disallow if sane_behavior. | |
2823 | */ | |
2824 | if (cgroup_sane_behavior(cgrp)) | |
2825 | return -EPERM; | |
099fca32 | 2826 | |
e1b2dc17 TH |
2827 | /* |
2828 | * We're gonna grab cgroup_tree_mutex which nests outside kernfs | |
2829 | * active_ref. kernfs_rename() doesn't require active_ref | |
2830 | * protection. Break them before grabbing cgroup_tree_mutex. | |
2831 | */ | |
2832 | kernfs_break_active_protection(new_parent); | |
2833 | kernfs_break_active_protection(kn); | |
099fca32 | 2834 | |
2bd59d48 TH |
2835 | mutex_lock(&cgroup_tree_mutex); |
2836 | mutex_lock(&cgroup_mutex); | |
099fca32 | 2837 | |
2bd59d48 | 2838 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 2839 | |
2bd59d48 TH |
2840 | mutex_unlock(&cgroup_mutex); |
2841 | mutex_unlock(&cgroup_tree_mutex); | |
e1b2dc17 TH |
2842 | |
2843 | kernfs_unbreak_active_protection(kn); | |
2844 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 2845 | return ret; |
099fca32 LZ |
2846 | } |
2847 | ||
49957f8e TH |
2848 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
2849 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
2850 | { | |
2851 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
2852 | .ia_uid = current_fsuid(), | |
2853 | .ia_gid = current_fsgid(), }; | |
2854 | ||
2855 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
2856 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
2857 | return 0; | |
2858 | ||
2859 | return kernfs_setattr(kn, &iattr); | |
2860 | } | |
2861 | ||
2bb566cb | 2862 | static int cgroup_add_file(struct cgroup *cgrp, struct cftype *cft) |
ddbcc7e8 | 2863 | { |
8d7e6fb0 | 2864 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
2865 | struct kernfs_node *kn; |
2866 | struct lock_class_key *key = NULL; | |
49957f8e | 2867 | int ret; |
05ef1d7c | 2868 | |
2bd59d48 TH |
2869 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
2870 | key = &cft->lockdep_key; | |
2871 | #endif | |
2872 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
2873 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
2874 | NULL, false, key); | |
49957f8e TH |
2875 | if (IS_ERR(kn)) |
2876 | return PTR_ERR(kn); | |
2877 | ||
2878 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 2879 | if (ret) { |
49957f8e | 2880 | kernfs_remove(kn); |
f8f22e53 TH |
2881 | return ret; |
2882 | } | |
2883 | ||
2884 | if (cft->seq_show == cgroup_subtree_control_show) | |
2885 | cgrp->control_kn = kn; | |
842b597e TH |
2886 | else if (cft->seq_show == cgroup_populated_show) |
2887 | cgrp->populated_kn = kn; | |
f8f22e53 | 2888 | return 0; |
ddbcc7e8 PM |
2889 | } |
2890 | ||
b1f28d31 TH |
2891 | /** |
2892 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
2893 | * @cgrp: the target cgroup | |
b1f28d31 TH |
2894 | * @cfts: array of cftypes to be added |
2895 | * @is_add: whether to add or remove | |
2896 | * | |
2897 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
2bb566cb TH |
2898 | * For removals, this function never fails. If addition fails, this |
2899 | * function doesn't remove files already added. The caller is responsible | |
2900 | * for cleaning up. | |
b1f28d31 | 2901 | */ |
2bb566cb TH |
2902 | static int cgroup_addrm_files(struct cgroup *cgrp, struct cftype cfts[], |
2903 | bool is_add) | |
ddbcc7e8 | 2904 | { |
03b1cde6 | 2905 | struct cftype *cft; |
b1f28d31 TH |
2906 | int ret; |
2907 | ||
ace2bee8 | 2908 | lockdep_assert_held(&cgroup_tree_mutex); |
db0416b6 TH |
2909 | |
2910 | for (cft = cfts; cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 2911 | /* does cft->flags tell us to skip this file on @cgrp? */ |
8cbbf2c9 TH |
2912 | if ((cft->flags & CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
2913 | continue; | |
873fe09e TH |
2914 | if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp)) |
2915 | continue; | |
f33fddc2 G |
2916 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) |
2917 | continue; | |
2918 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent) | |
2919 | continue; | |
2920 | ||
2739d3cc | 2921 | if (is_add) { |
2bb566cb | 2922 | ret = cgroup_add_file(cgrp, cft); |
b1f28d31 | 2923 | if (ret) { |
ed3d261b JP |
2924 | pr_warn("%s: failed to add %s, err=%d\n", |
2925 | __func__, cft->name, ret); | |
b1f28d31 TH |
2926 | return ret; |
2927 | } | |
2739d3cc LZ |
2928 | } else { |
2929 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 2930 | } |
ddbcc7e8 | 2931 | } |
b1f28d31 | 2932 | return 0; |
ddbcc7e8 PM |
2933 | } |
2934 | ||
21a2d343 | 2935 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
2936 | { |
2937 | LIST_HEAD(pending); | |
2bb566cb | 2938 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 2939 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 2940 | struct cgroup_subsys_state *css; |
9ccece80 | 2941 | int ret = 0; |
8e3f6541 | 2942 | |
21a2d343 | 2943 | lockdep_assert_held(&cgroup_tree_mutex); |
8e3f6541 | 2944 | |
e8c82d20 | 2945 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 2946 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
2947 | struct cgroup *cgrp = css->cgroup; |
2948 | ||
e8c82d20 LZ |
2949 | if (cgroup_is_dead(cgrp)) |
2950 | continue; | |
2951 | ||
21a2d343 | 2952 | ret = cgroup_addrm_files(cgrp, cfts, is_add); |
9ccece80 TH |
2953 | if (ret) |
2954 | break; | |
8e3f6541 | 2955 | } |
21a2d343 TH |
2956 | |
2957 | if (is_add && !ret) | |
2958 | kernfs_activate(root->kn); | |
9ccece80 | 2959 | return ret; |
8e3f6541 TH |
2960 | } |
2961 | ||
2da440a2 | 2962 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 2963 | { |
2bb566cb | 2964 | struct cftype *cft; |
8e3f6541 | 2965 | |
2bd59d48 TH |
2966 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2967 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
2968 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
2969 | kfree(cft->kf_ops); | |
2970 | cft->kf_ops = NULL; | |
2da440a2 | 2971 | cft->ss = NULL; |
2bd59d48 | 2972 | } |
2da440a2 TH |
2973 | } |
2974 | ||
2bd59d48 | 2975 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
2976 | { |
2977 | struct cftype *cft; | |
2978 | ||
2bd59d48 TH |
2979 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
2980 | struct kernfs_ops *kf_ops; | |
2981 | ||
0adb0704 TH |
2982 | WARN_ON(cft->ss || cft->kf_ops); |
2983 | ||
2bd59d48 TH |
2984 | if (cft->seq_start) |
2985 | kf_ops = &cgroup_kf_ops; | |
2986 | else | |
2987 | kf_ops = &cgroup_kf_single_ops; | |
2988 | ||
2989 | /* | |
2990 | * Ugh... if @cft wants a custom max_write_len, we need to | |
2991 | * make a copy of kf_ops to set its atomic_write_len. | |
2992 | */ | |
2993 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
2994 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
2995 | if (!kf_ops) { | |
2996 | cgroup_exit_cftypes(cfts); | |
2997 | return -ENOMEM; | |
2998 | } | |
2999 | kf_ops->atomic_write_len = cft->max_write_len; | |
3000 | } | |
8e3f6541 | 3001 | |
2bd59d48 | 3002 | cft->kf_ops = kf_ops; |
2bb566cb | 3003 | cft->ss = ss; |
2bd59d48 | 3004 | } |
2bb566cb | 3005 | |
2bd59d48 | 3006 | return 0; |
2da440a2 TH |
3007 | } |
3008 | ||
21a2d343 TH |
3009 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3010 | { | |
3011 | lockdep_assert_held(&cgroup_tree_mutex); | |
3012 | ||
3013 | if (!cfts || !cfts[0].ss) | |
3014 | return -ENOENT; | |
3015 | ||
3016 | list_del(&cfts->node); | |
3017 | cgroup_apply_cftypes(cfts, false); | |
3018 | cgroup_exit_cftypes(cfts); | |
3019 | return 0; | |
8e3f6541 | 3020 | } |
8e3f6541 | 3021 | |
79578621 TH |
3022 | /** |
3023 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3024 | * @cfts: zero-length name terminated array of cftypes |
3025 | * | |
2bb566cb TH |
3026 | * Unregister @cfts. Files described by @cfts are removed from all |
3027 | * existing cgroups and all future cgroups won't have them either. This | |
3028 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3029 | * |
3030 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3031 | * registered. |
79578621 | 3032 | */ |
2bb566cb | 3033 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3034 | { |
21a2d343 | 3035 | int ret; |
79578621 | 3036 | |
21a2d343 TH |
3037 | mutex_lock(&cgroup_tree_mutex); |
3038 | ret = cgroup_rm_cftypes_locked(cfts); | |
3039 | mutex_unlock(&cgroup_tree_mutex); | |
3040 | return ret; | |
80b13586 TH |
3041 | } |
3042 | ||
8e3f6541 TH |
3043 | /** |
3044 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3045 | * @ss: target cgroup subsystem | |
3046 | * @cfts: zero-length name terminated array of cftypes | |
3047 | * | |
3048 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3049 | * existing cgroups to which @ss is attached and all future cgroups will | |
3050 | * have them too. This function can be called anytime whether @ss is | |
3051 | * attached or not. | |
3052 | * | |
3053 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3054 | * function currently returns 0 as long as @cfts registration is successful | |
3055 | * even if some file creation attempts on existing cgroups fail. | |
3056 | */ | |
03b1cde6 | 3057 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3058 | { |
9ccece80 | 3059 | int ret; |
8e3f6541 | 3060 | |
dc5736ed LZ |
3061 | if (!cfts || cfts[0].name[0] == '\0') |
3062 | return 0; | |
2bb566cb | 3063 | |
2bd59d48 TH |
3064 | ret = cgroup_init_cftypes(ss, cfts); |
3065 | if (ret) | |
3066 | return ret; | |
79578621 | 3067 | |
21a2d343 TH |
3068 | mutex_lock(&cgroup_tree_mutex); |
3069 | ||
0adb0704 | 3070 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3071 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3072 | if (ret) |
21a2d343 | 3073 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3074 | |
21a2d343 | 3075 | mutex_unlock(&cgroup_tree_mutex); |
9ccece80 | 3076 | return ret; |
79578621 TH |
3077 | } |
3078 | ||
a043e3b2 LZ |
3079 | /** |
3080 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3081 | * @cgrp: the cgroup in question | |
3082 | * | |
3083 | * Return the number of tasks in the cgroup. | |
3084 | */ | |
07bc356e | 3085 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3086 | { |
3087 | int count = 0; | |
69d0206c | 3088 | struct cgrp_cset_link *link; |
817929ec | 3089 | |
96d365e0 | 3090 | down_read(&css_set_rwsem); |
69d0206c TH |
3091 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3092 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 3093 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
3094 | return count; |
3095 | } | |
3096 | ||
53fa5261 | 3097 | /** |
492eb21b TH |
3098 | * css_next_child - find the next child of a given css |
3099 | * @pos_css: the current position (%NULL to initiate traversal) | |
3100 | * @parent_css: css whose children to walk | |
53fa5261 | 3101 | * |
492eb21b | 3102 | * This function returns the next child of @parent_css and should be called |
87fb54f1 TH |
3103 | * under either cgroup_mutex or RCU read lock. The only requirement is |
3104 | * that @parent_css and @pos_css are accessible. The next sibling is | |
3105 | * guaranteed to be returned regardless of their states. | |
53fa5261 | 3106 | */ |
492eb21b TH |
3107 | struct cgroup_subsys_state * |
3108 | css_next_child(struct cgroup_subsys_state *pos_css, | |
3109 | struct cgroup_subsys_state *parent_css) | |
53fa5261 | 3110 | { |
492eb21b TH |
3111 | struct cgroup *pos = pos_css ? pos_css->cgroup : NULL; |
3112 | struct cgroup *cgrp = parent_css->cgroup; | |
53fa5261 TH |
3113 | struct cgroup *next; |
3114 | ||
ace2bee8 | 3115 | cgroup_assert_mutexes_or_rcu_locked(); |
53fa5261 TH |
3116 | |
3117 | /* | |
3118 | * @pos could already have been removed. Once a cgroup is removed, | |
3119 | * its ->sibling.next is no longer updated when its next sibling | |
ea15f8cc TH |
3120 | * changes. As CGRP_DEAD assertion is serialized and happens |
3121 | * before the cgroup is taken off the ->sibling list, if we see it | |
3122 | * unasserted, it's guaranteed that the next sibling hasn't | |
3123 | * finished its grace period even if it's already removed, and thus | |
3124 | * safe to dereference from this RCU critical section. If | |
3125 | * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed | |
3126 | * to be visible as %true here. | |
3b287a50 TH |
3127 | * |
3128 | * If @pos is dead, its next pointer can't be dereferenced; | |
3129 | * however, as each cgroup is given a monotonically increasing | |
3130 | * unique serial number and always appended to the sibling list, | |
3131 | * the next one can be found by walking the parent's children until | |
3132 | * we see a cgroup with higher serial number than @pos's. While | |
3133 | * this path can be slower, it's taken only when either the current | |
3134 | * cgroup is removed or iteration and removal race. | |
53fa5261 | 3135 | */ |
3b287a50 TH |
3136 | if (!pos) { |
3137 | next = list_entry_rcu(cgrp->children.next, struct cgroup, sibling); | |
3138 | } else if (likely(!cgroup_is_dead(pos))) { | |
53fa5261 | 3139 | next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling); |
3b287a50 TH |
3140 | } else { |
3141 | list_for_each_entry_rcu(next, &cgrp->children, sibling) | |
3142 | if (next->serial_nr > pos->serial_nr) | |
3143 | break; | |
53fa5261 TH |
3144 | } |
3145 | ||
3b281afb TH |
3146 | /* |
3147 | * @next, if not pointing to the head, can be dereferenced and is | |
3148 | * the next sibling; however, it might have @ss disabled. If so, | |
3149 | * fast-forward to the next enabled one. | |
3150 | */ | |
3151 | while (&next->sibling != &cgrp->children) { | |
3152 | struct cgroup_subsys_state *next_css = cgroup_css(next, parent_css->ss); | |
492eb21b | 3153 | |
3b281afb TH |
3154 | if (next_css) |
3155 | return next_css; | |
3156 | next = list_entry_rcu(next->sibling.next, struct cgroup, sibling); | |
3157 | } | |
3158 | return NULL; | |
53fa5261 | 3159 | } |
53fa5261 | 3160 | |
574bd9f7 | 3161 | /** |
492eb21b | 3162 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3163 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3164 | * @root: css whose descendants to walk |
574bd9f7 | 3165 | * |
492eb21b | 3166 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3167 | * to visit for pre-order traversal of @root's descendants. @root is |
3168 | * included in the iteration and the first node to be visited. | |
75501a6d | 3169 | * |
87fb54f1 TH |
3170 | * While this function requires cgroup_mutex or RCU read locking, it |
3171 | * doesn't require the whole traversal to be contained in a single critical | |
3172 | * section. This function will return the correct next descendant as long | |
3173 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
574bd9f7 | 3174 | */ |
492eb21b TH |
3175 | struct cgroup_subsys_state * |
3176 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3177 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3178 | { |
492eb21b | 3179 | struct cgroup_subsys_state *next; |
574bd9f7 | 3180 | |
ace2bee8 | 3181 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 3182 | |
bd8815a6 | 3183 | /* if first iteration, visit @root */ |
7805d000 | 3184 | if (!pos) |
bd8815a6 | 3185 | return root; |
574bd9f7 TH |
3186 | |
3187 | /* visit the first child if exists */ | |
492eb21b | 3188 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3189 | if (next) |
3190 | return next; | |
3191 | ||
3192 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b TH |
3193 | while (pos != root) { |
3194 | next = css_next_child(pos, css_parent(pos)); | |
75501a6d | 3195 | if (next) |
574bd9f7 | 3196 | return next; |
492eb21b | 3197 | pos = css_parent(pos); |
7805d000 | 3198 | } |
574bd9f7 TH |
3199 | |
3200 | return NULL; | |
3201 | } | |
574bd9f7 | 3202 | |
12a9d2fe | 3203 | /** |
492eb21b TH |
3204 | * css_rightmost_descendant - return the rightmost descendant of a css |
3205 | * @pos: css of interest | |
12a9d2fe | 3206 | * |
492eb21b TH |
3207 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3208 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3209 | * subtree of @pos. |
75501a6d | 3210 | * |
87fb54f1 TH |
3211 | * While this function requires cgroup_mutex or RCU read locking, it |
3212 | * doesn't require the whole traversal to be contained in a single critical | |
3213 | * section. This function will return the correct rightmost descendant as | |
3214 | * long as @pos is accessible. | |
12a9d2fe | 3215 | */ |
492eb21b TH |
3216 | struct cgroup_subsys_state * |
3217 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3218 | { |
492eb21b | 3219 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3220 | |
ace2bee8 | 3221 | cgroup_assert_mutexes_or_rcu_locked(); |
12a9d2fe TH |
3222 | |
3223 | do { | |
3224 | last = pos; | |
3225 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3226 | pos = NULL; | |
492eb21b | 3227 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3228 | pos = tmp; |
3229 | } while (pos); | |
3230 | ||
3231 | return last; | |
3232 | } | |
12a9d2fe | 3233 | |
492eb21b TH |
3234 | static struct cgroup_subsys_state * |
3235 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3236 | { |
492eb21b | 3237 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3238 | |
3239 | do { | |
3240 | last = pos; | |
492eb21b | 3241 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3242 | } while (pos); |
3243 | ||
3244 | return last; | |
3245 | } | |
3246 | ||
3247 | /** | |
492eb21b | 3248 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3249 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3250 | * @root: css whose descendants to walk |
574bd9f7 | 3251 | * |
492eb21b | 3252 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3253 | * to visit for post-order traversal of @root's descendants. @root is |
3254 | * included in the iteration and the last node to be visited. | |
75501a6d | 3255 | * |
87fb54f1 TH |
3256 | * While this function requires cgroup_mutex or RCU read locking, it |
3257 | * doesn't require the whole traversal to be contained in a single critical | |
3258 | * section. This function will return the correct next descendant as long | |
3259 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3260 | * @cgroup. | |
574bd9f7 | 3261 | */ |
492eb21b TH |
3262 | struct cgroup_subsys_state * |
3263 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3264 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3265 | { |
492eb21b | 3266 | struct cgroup_subsys_state *next; |
574bd9f7 | 3267 | |
ace2bee8 | 3268 | cgroup_assert_mutexes_or_rcu_locked(); |
574bd9f7 | 3269 | |
58b79a91 TH |
3270 | /* if first iteration, visit leftmost descendant which may be @root */ |
3271 | if (!pos) | |
3272 | return css_leftmost_descendant(root); | |
574bd9f7 | 3273 | |
bd8815a6 TH |
3274 | /* if we visited @root, we're done */ |
3275 | if (pos == root) | |
3276 | return NULL; | |
3277 | ||
574bd9f7 | 3278 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
492eb21b | 3279 | next = css_next_child(pos, css_parent(pos)); |
75501a6d | 3280 | if (next) |
492eb21b | 3281 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3282 | |
3283 | /* no sibling left, visit parent */ | |
bd8815a6 | 3284 | return css_parent(pos); |
574bd9f7 | 3285 | } |
574bd9f7 | 3286 | |
0942eeee | 3287 | /** |
72ec7029 | 3288 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
3289 | * @it: the iterator to advance |
3290 | * | |
3291 | * Advance @it to the next css_set to walk. | |
d515876e | 3292 | */ |
72ec7029 | 3293 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e | 3294 | { |
0f0a2b4f | 3295 | struct list_head *l = it->cset_pos; |
d515876e TH |
3296 | struct cgrp_cset_link *link; |
3297 | struct css_set *cset; | |
3298 | ||
3299 | /* Advance to the next non-empty css_set */ | |
3300 | do { | |
3301 | l = l->next; | |
0f0a2b4f TH |
3302 | if (l == it->cset_head) { |
3303 | it->cset_pos = NULL; | |
d515876e TH |
3304 | return; |
3305 | } | |
3ebb2b6e TH |
3306 | |
3307 | if (it->ss) { | |
3308 | cset = container_of(l, struct css_set, | |
3309 | e_cset_node[it->ss->id]); | |
3310 | } else { | |
3311 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3312 | cset = link->cset; | |
3313 | } | |
c7561128 TH |
3314 | } while (list_empty(&cset->tasks) && list_empty(&cset->mg_tasks)); |
3315 | ||
0f0a2b4f | 3316 | it->cset_pos = l; |
c7561128 TH |
3317 | |
3318 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3319 | it->task_pos = cset->tasks.next; |
c7561128 | 3320 | else |
0f0a2b4f TH |
3321 | it->task_pos = cset->mg_tasks.next; |
3322 | ||
3323 | it->tasks_head = &cset->tasks; | |
3324 | it->mg_tasks_head = &cset->mg_tasks; | |
d515876e TH |
3325 | } |
3326 | ||
0942eeee | 3327 | /** |
72ec7029 TH |
3328 | * css_task_iter_start - initiate task iteration |
3329 | * @css: the css to walk tasks of | |
0942eeee TH |
3330 | * @it: the task iterator to use |
3331 | * | |
72ec7029 TH |
3332 | * Initiate iteration through the tasks of @css. The caller can call |
3333 | * css_task_iter_next() to walk through the tasks until the function | |
3334 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3335 | * called. | |
0942eeee TH |
3336 | * |
3337 | * Note that this function acquires a lock which is released when the | |
3338 | * iteration finishes. The caller can't sleep while iteration is in | |
3339 | * progress. | |
3340 | */ | |
72ec7029 TH |
3341 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3342 | struct css_task_iter *it) | |
96d365e0 | 3343 | __acquires(css_set_rwsem) |
817929ec | 3344 | { |
56fde9e0 TH |
3345 | /* no one should try to iterate before mounting cgroups */ |
3346 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3347 | |
96d365e0 | 3348 | down_read(&css_set_rwsem); |
c59cd3d8 | 3349 | |
3ebb2b6e TH |
3350 | it->ss = css->ss; |
3351 | ||
3352 | if (it->ss) | |
3353 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3354 | else | |
3355 | it->cset_pos = &css->cgroup->cset_links; | |
3356 | ||
0f0a2b4f | 3357 | it->cset_head = it->cset_pos; |
c59cd3d8 | 3358 | |
72ec7029 | 3359 | css_advance_task_iter(it); |
817929ec PM |
3360 | } |
3361 | ||
0942eeee | 3362 | /** |
72ec7029 | 3363 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3364 | * @it: the task iterator being iterated |
3365 | * | |
3366 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3367 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3368 | * reaches the end. | |
0942eeee | 3369 | */ |
72ec7029 | 3370 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
3371 | { |
3372 | struct task_struct *res; | |
0f0a2b4f | 3373 | struct list_head *l = it->task_pos; |
817929ec PM |
3374 | |
3375 | /* If the iterator cg is NULL, we have no tasks */ | |
0f0a2b4f | 3376 | if (!it->cset_pos) |
817929ec PM |
3377 | return NULL; |
3378 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
3379 | |
3380 | /* | |
3381 | * Advance iterator to find next entry. cset->tasks is consumed | |
3382 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3383 | * next cset. | |
3384 | */ | |
817929ec | 3385 | l = l->next; |
c7561128 | 3386 | |
0f0a2b4f TH |
3387 | if (l == it->tasks_head) |
3388 | l = it->mg_tasks_head->next; | |
c7561128 | 3389 | |
0f0a2b4f | 3390 | if (l == it->mg_tasks_head) |
72ec7029 | 3391 | css_advance_task_iter(it); |
c7561128 | 3392 | else |
0f0a2b4f | 3393 | it->task_pos = l; |
c7561128 | 3394 | |
817929ec PM |
3395 | return res; |
3396 | } | |
3397 | ||
0942eeee | 3398 | /** |
72ec7029 | 3399 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3400 | * @it: the task iterator to finish |
3401 | * | |
72ec7029 | 3402 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3403 | */ |
72ec7029 | 3404 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 3405 | __releases(css_set_rwsem) |
31a7df01 | 3406 | { |
96d365e0 | 3407 | up_read(&css_set_rwsem); |
31a7df01 CW |
3408 | } |
3409 | ||
3410 | /** | |
8cc99345 TH |
3411 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
3412 | * @to: cgroup to which the tasks will be moved | |
3413 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 3414 | * |
eaf797ab TH |
3415 | * Locking rules between cgroup_post_fork() and the migration path |
3416 | * guarantee that, if a task is forking while being migrated, the new child | |
3417 | * is guaranteed to be either visible in the source cgroup after the | |
3418 | * parent's migration is complete or put into the target cgroup. No task | |
3419 | * can slip out of migration through forking. | |
31a7df01 | 3420 | */ |
8cc99345 | 3421 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 3422 | { |
952aaa12 TH |
3423 | LIST_HEAD(preloaded_csets); |
3424 | struct cgrp_cset_link *link; | |
72ec7029 | 3425 | struct css_task_iter it; |
e406d1cf | 3426 | struct task_struct *task; |
952aaa12 | 3427 | int ret; |
31a7df01 | 3428 | |
952aaa12 | 3429 | mutex_lock(&cgroup_mutex); |
31a7df01 | 3430 | |
952aaa12 TH |
3431 | /* all tasks in @from are being moved, all csets are source */ |
3432 | down_read(&css_set_rwsem); | |
3433 | list_for_each_entry(link, &from->cset_links, cset_link) | |
3434 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
3435 | up_read(&css_set_rwsem); | |
31a7df01 | 3436 | |
952aaa12 TH |
3437 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
3438 | if (ret) | |
3439 | goto out_err; | |
8cc99345 | 3440 | |
952aaa12 TH |
3441 | /* |
3442 | * Migrate tasks one-by-one until @form is empty. This fails iff | |
3443 | * ->can_attach() fails. | |
3444 | */ | |
e406d1cf TH |
3445 | do { |
3446 | css_task_iter_start(&from->dummy_css, &it); | |
3447 | task = css_task_iter_next(&it); | |
3448 | if (task) | |
3449 | get_task_struct(task); | |
3450 | css_task_iter_end(&it); | |
3451 | ||
3452 | if (task) { | |
952aaa12 | 3453 | ret = cgroup_migrate(to, task, false); |
e406d1cf TH |
3454 | put_task_struct(task); |
3455 | } | |
3456 | } while (task && !ret); | |
952aaa12 TH |
3457 | out_err: |
3458 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 3459 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 3460 | return ret; |
8cc99345 TH |
3461 | } |
3462 | ||
bbcb81d0 | 3463 | /* |
102a775e | 3464 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3465 | * |
3466 | * Reading this file can return large amounts of data if a cgroup has | |
3467 | * *lots* of attached tasks. So it may need several calls to read(), | |
3468 | * but we cannot guarantee that the information we produce is correct | |
3469 | * unless we produce it entirely atomically. | |
3470 | * | |
bbcb81d0 | 3471 | */ |
bbcb81d0 | 3472 | |
24528255 LZ |
3473 | /* which pidlist file are we talking about? */ |
3474 | enum cgroup_filetype { | |
3475 | CGROUP_FILE_PROCS, | |
3476 | CGROUP_FILE_TASKS, | |
3477 | }; | |
3478 | ||
3479 | /* | |
3480 | * A pidlist is a list of pids that virtually represents the contents of one | |
3481 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3482 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3483 | * to the cgroup. | |
3484 | */ | |
3485 | struct cgroup_pidlist { | |
3486 | /* | |
3487 | * used to find which pidlist is wanted. doesn't change as long as | |
3488 | * this particular list stays in the list. | |
3489 | */ | |
3490 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3491 | /* array of xids */ | |
3492 | pid_t *list; | |
3493 | /* how many elements the above list has */ | |
3494 | int length; | |
24528255 LZ |
3495 | /* each of these stored in a list by its cgroup */ |
3496 | struct list_head links; | |
3497 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3498 | struct cgroup *owner; | |
b1a21367 TH |
3499 | /* for delayed destruction */ |
3500 | struct delayed_work destroy_dwork; | |
24528255 LZ |
3501 | }; |
3502 | ||
d1d9fd33 BB |
3503 | /* |
3504 | * The following two functions "fix" the issue where there are more pids | |
3505 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
3506 | * TODO: replace with a kernel-wide solution to this problem | |
3507 | */ | |
3508 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
3509 | static void *pidlist_allocate(int count) | |
3510 | { | |
3511 | if (PIDLIST_TOO_LARGE(count)) | |
3512 | return vmalloc(count * sizeof(pid_t)); | |
3513 | else | |
3514 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
3515 | } | |
b1a21367 | 3516 | |
d1d9fd33 BB |
3517 | static void pidlist_free(void *p) |
3518 | { | |
3519 | if (is_vmalloc_addr(p)) | |
3520 | vfree(p); | |
3521 | else | |
3522 | kfree(p); | |
3523 | } | |
d1d9fd33 | 3524 | |
b1a21367 TH |
3525 | /* |
3526 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
3527 | * should be left afterwards. | |
3528 | */ | |
3529 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
3530 | { | |
3531 | struct cgroup_pidlist *l, *tmp_l; | |
3532 | ||
3533 | mutex_lock(&cgrp->pidlist_mutex); | |
3534 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
3535 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
3536 | mutex_unlock(&cgrp->pidlist_mutex); | |
3537 | ||
3538 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
3539 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
3540 | } | |
3541 | ||
3542 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
3543 | { | |
3544 | struct delayed_work *dwork = to_delayed_work(work); | |
3545 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
3546 | destroy_dwork); | |
3547 | struct cgroup_pidlist *tofree = NULL; | |
3548 | ||
3549 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
3550 | |
3551 | /* | |
04502365 TH |
3552 | * Destroy iff we didn't get queued again. The state won't change |
3553 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 3554 | */ |
04502365 | 3555 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
3556 | list_del(&l->links); |
3557 | pidlist_free(l->list); | |
3558 | put_pid_ns(l->key.ns); | |
3559 | tofree = l; | |
3560 | } | |
3561 | ||
b1a21367 TH |
3562 | mutex_unlock(&l->owner->pidlist_mutex); |
3563 | kfree(tofree); | |
3564 | } | |
3565 | ||
bbcb81d0 | 3566 | /* |
102a775e | 3567 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 3568 | * Returns the number of unique elements. |
bbcb81d0 | 3569 | */ |
6ee211ad | 3570 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 3571 | { |
102a775e | 3572 | int src, dest = 1; |
102a775e BB |
3573 | |
3574 | /* | |
3575 | * we presume the 0th element is unique, so i starts at 1. trivial | |
3576 | * edge cases first; no work needs to be done for either | |
3577 | */ | |
3578 | if (length == 0 || length == 1) | |
3579 | return length; | |
3580 | /* src and dest walk down the list; dest counts unique elements */ | |
3581 | for (src = 1; src < length; src++) { | |
3582 | /* find next unique element */ | |
3583 | while (list[src] == list[src-1]) { | |
3584 | src++; | |
3585 | if (src == length) | |
3586 | goto after; | |
3587 | } | |
3588 | /* dest always points to where the next unique element goes */ | |
3589 | list[dest] = list[src]; | |
3590 | dest++; | |
3591 | } | |
3592 | after: | |
102a775e BB |
3593 | return dest; |
3594 | } | |
3595 | ||
afb2bc14 TH |
3596 | /* |
3597 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
3598 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
3599 | * different per namespace, each namespace needs differently sorted list, | |
3600 | * making it impossible to use, for example, single rbtree of member tasks | |
3601 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
3602 | * per open file is dangerous, so cgroup had to implement shared pool of | |
3603 | * pidlists keyed by cgroup and namespace. | |
3604 | * | |
3605 | * All this extra complexity was caused by the original implementation | |
3606 | * committing to an entirely unnecessary property. In the long term, we | |
3607 | * want to do away with it. Explicitly scramble sort order if | |
3608 | * sane_behavior so that no such expectation exists in the new interface. | |
3609 | * | |
3610 | * Scrambling is done by swapping every two consecutive bits, which is | |
3611 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
3612 | */ | |
3613 | static pid_t pid_fry(pid_t pid) | |
3614 | { | |
3615 | unsigned a = pid & 0x55555555; | |
3616 | unsigned b = pid & 0xAAAAAAAA; | |
3617 | ||
3618 | return (a << 1) | (b >> 1); | |
3619 | } | |
3620 | ||
3621 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
3622 | { | |
3623 | if (cgroup_sane_behavior(cgrp)) | |
3624 | return pid_fry(pid); | |
3625 | else | |
3626 | return pid; | |
3627 | } | |
3628 | ||
102a775e BB |
3629 | static int cmppid(const void *a, const void *b) |
3630 | { | |
3631 | return *(pid_t *)a - *(pid_t *)b; | |
3632 | } | |
3633 | ||
afb2bc14 TH |
3634 | static int fried_cmppid(const void *a, const void *b) |
3635 | { | |
3636 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
3637 | } | |
3638 | ||
e6b81710 TH |
3639 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
3640 | enum cgroup_filetype type) | |
3641 | { | |
3642 | struct cgroup_pidlist *l; | |
3643 | /* don't need task_nsproxy() if we're looking at ourself */ | |
3644 | struct pid_namespace *ns = task_active_pid_ns(current); | |
3645 | ||
3646 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
3647 | ||
3648 | list_for_each_entry(l, &cgrp->pidlists, links) | |
3649 | if (l->key.type == type && l->key.ns == ns) | |
3650 | return l; | |
3651 | return NULL; | |
3652 | } | |
3653 | ||
72a8cb30 BB |
3654 | /* |
3655 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
3656 | * returns with the lock on that pidlist already held, and takes care | |
3657 | * of the use count, or returns NULL with no locks held if we're out of | |
3658 | * memory. | |
3659 | */ | |
e6b81710 TH |
3660 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
3661 | enum cgroup_filetype type) | |
72a8cb30 BB |
3662 | { |
3663 | struct cgroup_pidlist *l; | |
b70cc5fd | 3664 | |
e6b81710 TH |
3665 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3666 | ||
3667 | l = cgroup_pidlist_find(cgrp, type); | |
3668 | if (l) | |
3669 | return l; | |
3670 | ||
72a8cb30 | 3671 | /* entry not found; create a new one */ |
f4f4be2b | 3672 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 3673 | if (!l) |
72a8cb30 | 3674 | return l; |
e6b81710 | 3675 | |
b1a21367 | 3676 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 3677 | l->key.type = type; |
e6b81710 TH |
3678 | /* don't need task_nsproxy() if we're looking at ourself */ |
3679 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
3680 | l->owner = cgrp; |
3681 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
3682 | return l; |
3683 | } | |
3684 | ||
102a775e BB |
3685 | /* |
3686 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
3687 | */ | |
72a8cb30 BB |
3688 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
3689 | struct cgroup_pidlist **lp) | |
102a775e BB |
3690 | { |
3691 | pid_t *array; | |
3692 | int length; | |
3693 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 3694 | struct css_task_iter it; |
817929ec | 3695 | struct task_struct *tsk; |
102a775e BB |
3696 | struct cgroup_pidlist *l; |
3697 | ||
4bac00d1 TH |
3698 | lockdep_assert_held(&cgrp->pidlist_mutex); |
3699 | ||
102a775e BB |
3700 | /* |
3701 | * If cgroup gets more users after we read count, we won't have | |
3702 | * enough space - tough. This race is indistinguishable to the | |
3703 | * caller from the case that the additional cgroup users didn't | |
3704 | * show up until sometime later on. | |
3705 | */ | |
3706 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 3707 | array = pidlist_allocate(length); |
102a775e BB |
3708 | if (!array) |
3709 | return -ENOMEM; | |
3710 | /* now, populate the array */ | |
72ec7029 TH |
3711 | css_task_iter_start(&cgrp->dummy_css, &it); |
3712 | while ((tsk = css_task_iter_next(&it))) { | |
102a775e | 3713 | if (unlikely(n == length)) |
817929ec | 3714 | break; |
102a775e | 3715 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
3716 | if (type == CGROUP_FILE_PROCS) |
3717 | pid = task_tgid_vnr(tsk); | |
3718 | else | |
3719 | pid = task_pid_vnr(tsk); | |
102a775e BB |
3720 | if (pid > 0) /* make sure to only use valid results */ |
3721 | array[n++] = pid; | |
817929ec | 3722 | } |
72ec7029 | 3723 | css_task_iter_end(&it); |
102a775e BB |
3724 | length = n; |
3725 | /* now sort & (if procs) strip out duplicates */ | |
afb2bc14 TH |
3726 | if (cgroup_sane_behavior(cgrp)) |
3727 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); | |
3728 | else | |
3729 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 3730 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 3731 | length = pidlist_uniq(array, length); |
e6b81710 | 3732 | |
e6b81710 | 3733 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 3734 | if (!l) { |
e6b81710 | 3735 | mutex_unlock(&cgrp->pidlist_mutex); |
d1d9fd33 | 3736 | pidlist_free(array); |
72a8cb30 | 3737 | return -ENOMEM; |
102a775e | 3738 | } |
e6b81710 TH |
3739 | |
3740 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 3741 | pidlist_free(l->list); |
102a775e BB |
3742 | l->list = array; |
3743 | l->length = length; | |
72a8cb30 | 3744 | *lp = l; |
102a775e | 3745 | return 0; |
bbcb81d0 PM |
3746 | } |
3747 | ||
846c7bb0 | 3748 | /** |
a043e3b2 | 3749 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
3750 | * @stats: cgroupstats to fill information into |
3751 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
3752 | * been requested. | |
a043e3b2 LZ |
3753 | * |
3754 | * Build and fill cgroupstats so that taskstats can export it to user | |
3755 | * space. | |
846c7bb0 BS |
3756 | */ |
3757 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
3758 | { | |
2bd59d48 | 3759 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 3760 | struct cgroup *cgrp; |
72ec7029 | 3761 | struct css_task_iter it; |
846c7bb0 | 3762 | struct task_struct *tsk; |
33d283be | 3763 | |
2bd59d48 TH |
3764 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
3765 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
3766 | kernfs_type(kn) != KERNFS_DIR) | |
3767 | return -EINVAL; | |
3768 | ||
bad34660 LZ |
3769 | mutex_lock(&cgroup_mutex); |
3770 | ||
846c7bb0 | 3771 | /* |
2bd59d48 | 3772 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 3773 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 3774 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 3775 | */ |
2bd59d48 TH |
3776 | rcu_read_lock(); |
3777 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 3778 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 3779 | rcu_read_unlock(); |
bad34660 | 3780 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
3781 | return -ENOENT; |
3782 | } | |
bad34660 | 3783 | rcu_read_unlock(); |
846c7bb0 | 3784 | |
72ec7029 TH |
3785 | css_task_iter_start(&cgrp->dummy_css, &it); |
3786 | while ((tsk = css_task_iter_next(&it))) { | |
846c7bb0 BS |
3787 | switch (tsk->state) { |
3788 | case TASK_RUNNING: | |
3789 | stats->nr_running++; | |
3790 | break; | |
3791 | case TASK_INTERRUPTIBLE: | |
3792 | stats->nr_sleeping++; | |
3793 | break; | |
3794 | case TASK_UNINTERRUPTIBLE: | |
3795 | stats->nr_uninterruptible++; | |
3796 | break; | |
3797 | case TASK_STOPPED: | |
3798 | stats->nr_stopped++; | |
3799 | break; | |
3800 | default: | |
3801 | if (delayacct_is_task_waiting_on_io(tsk)) | |
3802 | stats->nr_io_wait++; | |
3803 | break; | |
3804 | } | |
3805 | } | |
72ec7029 | 3806 | css_task_iter_end(&it); |
846c7bb0 | 3807 | |
bad34660 | 3808 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 3809 | return 0; |
846c7bb0 BS |
3810 | } |
3811 | ||
8f3ff208 | 3812 | |
bbcb81d0 | 3813 | /* |
102a775e | 3814 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 3815 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 3816 | * in the cgroup->l->list array. |
bbcb81d0 | 3817 | */ |
cc31edce | 3818 | |
102a775e | 3819 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 3820 | { |
cc31edce PM |
3821 | /* |
3822 | * Initially we receive a position value that corresponds to | |
3823 | * one more than the last pid shown (or 0 on the first call or | |
3824 | * after a seek to the start). Use a binary-search to find the | |
3825 | * next pid to display, if any | |
3826 | */ | |
2bd59d48 | 3827 | struct kernfs_open_file *of = s->private; |
7da11279 | 3828 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 3829 | struct cgroup_pidlist *l; |
7da11279 | 3830 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 3831 | int index = 0, pid = *pos; |
4bac00d1 TH |
3832 | int *iter, ret; |
3833 | ||
3834 | mutex_lock(&cgrp->pidlist_mutex); | |
3835 | ||
3836 | /* | |
5d22444f | 3837 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 3838 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 3839 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
3840 | * could already have been destroyed. |
3841 | */ | |
5d22444f TH |
3842 | if (of->priv) |
3843 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
3844 | |
3845 | /* | |
3846 | * Either this is the first start() after open or the matching | |
3847 | * pidlist has been destroyed inbetween. Create a new one. | |
3848 | */ | |
5d22444f TH |
3849 | if (!of->priv) { |
3850 | ret = pidlist_array_load(cgrp, type, | |
3851 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
3852 | if (ret) |
3853 | return ERR_PTR(ret); | |
3854 | } | |
5d22444f | 3855 | l = of->priv; |
cc31edce | 3856 | |
cc31edce | 3857 | if (pid) { |
102a775e | 3858 | int end = l->length; |
20777766 | 3859 | |
cc31edce PM |
3860 | while (index < end) { |
3861 | int mid = (index + end) / 2; | |
afb2bc14 | 3862 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
3863 | index = mid; |
3864 | break; | |
afb2bc14 | 3865 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
3866 | index = mid + 1; |
3867 | else | |
3868 | end = mid; | |
3869 | } | |
3870 | } | |
3871 | /* If we're off the end of the array, we're done */ | |
102a775e | 3872 | if (index >= l->length) |
cc31edce PM |
3873 | return NULL; |
3874 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 3875 | iter = l->list + index; |
afb2bc14 | 3876 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
3877 | return iter; |
3878 | } | |
3879 | ||
102a775e | 3880 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 3881 | { |
2bd59d48 | 3882 | struct kernfs_open_file *of = s->private; |
5d22444f | 3883 | struct cgroup_pidlist *l = of->priv; |
62236858 | 3884 | |
5d22444f TH |
3885 | if (l) |
3886 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 3887 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 3888 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
3889 | } |
3890 | ||
102a775e | 3891 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 3892 | { |
2bd59d48 | 3893 | struct kernfs_open_file *of = s->private; |
5d22444f | 3894 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
3895 | pid_t *p = v; |
3896 | pid_t *end = l->list + l->length; | |
cc31edce PM |
3897 | /* |
3898 | * Advance to the next pid in the array. If this goes off the | |
3899 | * end, we're done | |
3900 | */ | |
3901 | p++; | |
3902 | if (p >= end) { | |
3903 | return NULL; | |
3904 | } else { | |
7da11279 | 3905 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
3906 | return p; |
3907 | } | |
3908 | } | |
3909 | ||
102a775e | 3910 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce PM |
3911 | { |
3912 | return seq_printf(s, "%d\n", *(int *)v); | |
3913 | } | |
bbcb81d0 | 3914 | |
182446d0 TH |
3915 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
3916 | struct cftype *cft) | |
81a6a5cd | 3917 | { |
182446d0 | 3918 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
3919 | } |
3920 | ||
182446d0 TH |
3921 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
3922 | struct cftype *cft, u64 val) | |
6379c106 | 3923 | { |
182446d0 | 3924 | clear_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
6379c106 | 3925 | if (val) |
182446d0 | 3926 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 3927 | else |
182446d0 | 3928 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
3929 | return 0; |
3930 | } | |
3931 | ||
182446d0 TH |
3932 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
3933 | struct cftype *cft) | |
97978e6d | 3934 | { |
182446d0 | 3935 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3936 | } |
3937 | ||
182446d0 TH |
3938 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
3939 | struct cftype *cft, u64 val) | |
97978e6d DL |
3940 | { |
3941 | if (val) | |
182446d0 | 3942 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 3943 | else |
182446d0 | 3944 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
3945 | return 0; |
3946 | } | |
3947 | ||
d5c56ced | 3948 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3949 | { |
d5c56ced | 3950 | .name = "cgroup.procs", |
6612f05b TH |
3951 | .seq_start = cgroup_pidlist_start, |
3952 | .seq_next = cgroup_pidlist_next, | |
3953 | .seq_stop = cgroup_pidlist_stop, | |
3954 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 3955 | .private = CGROUP_FILE_PROCS, |
74a1166d | 3956 | .write_u64 = cgroup_procs_write, |
74a1166d | 3957 | .mode = S_IRUGO | S_IWUSR, |
102a775e | 3958 | }, |
97978e6d DL |
3959 | { |
3960 | .name = "cgroup.clone_children", | |
873fe09e | 3961 | .flags = CFTYPE_INSANE, |
97978e6d DL |
3962 | .read_u64 = cgroup_clone_children_read, |
3963 | .write_u64 = cgroup_clone_children_write, | |
3964 | }, | |
873fe09e TH |
3965 | { |
3966 | .name = "cgroup.sane_behavior", | |
3967 | .flags = CFTYPE_ONLY_ON_ROOT, | |
2da8ca82 | 3968 | .seq_show = cgroup_sane_behavior_show, |
873fe09e | 3969 | }, |
f8f22e53 TH |
3970 | { |
3971 | .name = "cgroup.controllers", | |
3972 | .flags = CFTYPE_ONLY_ON_DFL | CFTYPE_ONLY_ON_ROOT, | |
3973 | .seq_show = cgroup_root_controllers_show, | |
3974 | }, | |
3975 | { | |
3976 | .name = "cgroup.controllers", | |
3977 | .flags = CFTYPE_ONLY_ON_DFL | CFTYPE_NOT_ON_ROOT, | |
3978 | .seq_show = cgroup_controllers_show, | |
3979 | }, | |
3980 | { | |
3981 | .name = "cgroup.subtree_control", | |
3982 | .flags = CFTYPE_ONLY_ON_DFL, | |
3983 | .seq_show = cgroup_subtree_control_show, | |
451af504 | 3984 | .write = cgroup_subtree_control_write, |
f8f22e53 | 3985 | }, |
842b597e TH |
3986 | { |
3987 | .name = "cgroup.populated", | |
3988 | .flags = CFTYPE_ONLY_ON_DFL | CFTYPE_NOT_ON_ROOT, | |
3989 | .seq_show = cgroup_populated_show, | |
3990 | }, | |
d5c56ced TH |
3991 | |
3992 | /* | |
3993 | * Historical crazy stuff. These don't have "cgroup." prefix and | |
3994 | * don't exist if sane_behavior. If you're depending on these, be | |
3995 | * prepared to be burned. | |
3996 | */ | |
3997 | { | |
3998 | .name = "tasks", | |
3999 | .flags = CFTYPE_INSANE, /* use "procs" instead */ | |
6612f05b TH |
4000 | .seq_start = cgroup_pidlist_start, |
4001 | .seq_next = cgroup_pidlist_next, | |
4002 | .seq_stop = cgroup_pidlist_stop, | |
4003 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4004 | .private = CGROUP_FILE_TASKS, |
d5c56ced | 4005 | .write_u64 = cgroup_tasks_write, |
d5c56ced TH |
4006 | .mode = S_IRUGO | S_IWUSR, |
4007 | }, | |
4008 | { | |
4009 | .name = "notify_on_release", | |
4010 | .flags = CFTYPE_INSANE, | |
4011 | .read_u64 = cgroup_read_notify_on_release, | |
4012 | .write_u64 = cgroup_write_notify_on_release, | |
4013 | }, | |
6e6ff25b TH |
4014 | { |
4015 | .name = "release_agent", | |
cc5943a7 | 4016 | .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4017 | .seq_show = cgroup_release_agent_show, |
451af504 | 4018 | .write = cgroup_release_agent_write, |
5f469907 | 4019 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4020 | }, |
db0416b6 | 4021 | { } /* terminate */ |
bbcb81d0 PM |
4022 | }; |
4023 | ||
13af07df | 4024 | /** |
628f7cd4 | 4025 | * cgroup_populate_dir - create subsys files in a cgroup directory |
13af07df | 4026 | * @cgrp: target cgroup |
13af07df | 4027 | * @subsys_mask: mask of the subsystem ids whose files should be added |
bee55099 TH |
4028 | * |
4029 | * On failure, no file is added. | |
13af07df | 4030 | */ |
69dfa00c | 4031 | static int cgroup_populate_dir(struct cgroup *cgrp, unsigned int subsys_mask) |
ddbcc7e8 | 4032 | { |
ddbcc7e8 | 4033 | struct cgroup_subsys *ss; |
b420ba7d | 4034 | int i, ret = 0; |
bbcb81d0 | 4035 | |
8e3f6541 | 4036 | /* process cftsets of each subsystem */ |
b420ba7d | 4037 | for_each_subsys(ss, i) { |
0adb0704 | 4038 | struct cftype *cfts; |
b420ba7d | 4039 | |
69dfa00c | 4040 | if (!(subsys_mask & (1 << i))) |
13af07df | 4041 | continue; |
8e3f6541 | 4042 | |
0adb0704 TH |
4043 | list_for_each_entry(cfts, &ss->cfts, node) { |
4044 | ret = cgroup_addrm_files(cgrp, cfts, true); | |
bee55099 TH |
4045 | if (ret < 0) |
4046 | goto err; | |
4047 | } | |
ddbcc7e8 | 4048 | } |
ddbcc7e8 | 4049 | return 0; |
bee55099 TH |
4050 | err: |
4051 | cgroup_clear_dir(cgrp, subsys_mask); | |
4052 | return ret; | |
ddbcc7e8 PM |
4053 | } |
4054 | ||
0c21ead1 TH |
4055 | /* |
4056 | * css destruction is four-stage process. | |
4057 | * | |
4058 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4059 | * Implemented in kill_css(). | |
4060 | * | |
4061 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4062 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4063 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4064 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4065 | * |
4066 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4067 | * accessors are inside RCU read sections. css_release() schedules the | |
4068 | * RCU callback. | |
4069 | * | |
4070 | * 4. After the grace period, the css can be freed. Implemented in | |
4071 | * css_free_work_fn(). | |
4072 | * | |
4073 | * It is actually hairier because both step 2 and 4 require process context | |
4074 | * and thus involve punting to css->destroy_work adding two additional | |
4075 | * steps to the already complex sequence. | |
4076 | */ | |
35ef10da | 4077 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4078 | { |
4079 | struct cgroup_subsys_state *css = | |
35ef10da | 4080 | container_of(work, struct cgroup_subsys_state, destroy_work); |
0c21ead1 | 4081 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4082 | |
0ae78e0b TH |
4083 | if (css->parent) |
4084 | css_put(css->parent); | |
4085 | ||
0c21ead1 | 4086 | css->ss->css_free(css); |
2bd59d48 | 4087 | cgroup_put(cgrp); |
48ddbe19 TH |
4088 | } |
4089 | ||
0c21ead1 | 4090 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4091 | { |
4092 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4093 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4094 | |
35ef10da | 4095 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4096 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4097 | } |
4098 | ||
d3daf28d TH |
4099 | static void css_release(struct percpu_ref *ref) |
4100 | { | |
4101 | struct cgroup_subsys_state *css = | |
4102 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
15a4c835 TH |
4103 | struct cgroup_subsys *ss = css->ss; |
4104 | ||
15a4c835 | 4105 | cgroup_idr_remove(&ss->css_idr, css->id); |
d3daf28d | 4106 | |
0c21ead1 | 4107 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4108 | } |
4109 | ||
ddfcadab TH |
4110 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4111 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4112 | { |
ddfcadab TH |
4113 | cgroup_get(cgrp); |
4114 | ||
bd89aabc | 4115 | css->cgroup = cgrp; |
72c97e54 | 4116 | css->ss = ss; |
ddbcc7e8 | 4117 | css->flags = 0; |
0ae78e0b | 4118 | |
ddfcadab | 4119 | if (cgrp->parent) { |
ca8bdcaf | 4120 | css->parent = cgroup_css(cgrp->parent, ss); |
ddfcadab TH |
4121 | css_get(css->parent); |
4122 | } else { | |
38b53aba | 4123 | css->flags |= CSS_ROOT; |
ddfcadab | 4124 | } |
48ddbe19 | 4125 | |
ca8bdcaf | 4126 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4127 | } |
4128 | ||
2a4ac633 | 4129 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4130 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4131 | { |
623f926b | 4132 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4133 | int ret = 0; |
4134 | ||
ace2bee8 | 4135 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
4136 | lockdep_assert_held(&cgroup_mutex); |
4137 | ||
92fb9748 | 4138 | if (ss->css_online) |
eb95419b | 4139 | ret = ss->css_online(css); |
ae7f164a | 4140 | if (!ret) { |
eb95419b | 4141 | css->flags |= CSS_ONLINE; |
f20104de | 4142 | css->cgroup->nr_css++; |
aec25020 | 4143 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 4144 | } |
b1929db4 | 4145 | return ret; |
a31f2d3f TH |
4146 | } |
4147 | ||
2a4ac633 | 4148 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4149 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4150 | { |
623f926b | 4151 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f | 4152 | |
ace2bee8 | 4153 | lockdep_assert_held(&cgroup_tree_mutex); |
a31f2d3f TH |
4154 | lockdep_assert_held(&cgroup_mutex); |
4155 | ||
4156 | if (!(css->flags & CSS_ONLINE)) | |
4157 | return; | |
4158 | ||
d7eeac19 | 4159 | if (ss->css_offline) |
eb95419b | 4160 | ss->css_offline(css); |
a31f2d3f | 4161 | |
eb95419b | 4162 | css->flags &= ~CSS_ONLINE; |
09a503ea | 4163 | css->cgroup->nr_css--; |
e3297803 | 4164 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4165 | |
4166 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4167 | } |
4168 | ||
c81c925a TH |
4169 | /** |
4170 | * create_css - create a cgroup_subsys_state | |
4171 | * @cgrp: the cgroup new css will be associated with | |
4172 | * @ss: the subsys of new css | |
4173 | * | |
4174 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
4175 | * css is online and installed in @cgrp with all interface files created. | |
4176 | * Returns 0 on success, -errno on failure. | |
4177 | */ | |
4178 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss) | |
4179 | { | |
4180 | struct cgroup *parent = cgrp->parent; | |
4181 | struct cgroup_subsys_state *css; | |
4182 | int err; | |
4183 | ||
c81c925a TH |
4184 | lockdep_assert_held(&cgroup_mutex); |
4185 | ||
4186 | css = ss->css_alloc(cgroup_css(parent, ss)); | |
4187 | if (IS_ERR(css)) | |
4188 | return PTR_ERR(css); | |
4189 | ||
ddfcadab | 4190 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4191 | |
c81c925a TH |
4192 | err = percpu_ref_init(&css->refcnt, css_release); |
4193 | if (err) | |
3eb59ec6 | 4194 | goto err_free_css; |
c81c925a | 4195 | |
15a4c835 TH |
4196 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_NOWAIT); |
4197 | if (err < 0) | |
4198 | goto err_free_percpu_ref; | |
4199 | css->id = err; | |
4200 | ||
aec25020 | 4201 | err = cgroup_populate_dir(cgrp, 1 << ss->id); |
c81c925a | 4202 | if (err) |
15a4c835 TH |
4203 | goto err_free_id; |
4204 | ||
4205 | /* @css is ready to be brought online now, make it visible */ | |
4206 | cgroup_idr_replace(&ss->css_idr, css, css->id); | |
c81c925a TH |
4207 | |
4208 | err = online_css(css); | |
4209 | if (err) | |
3eb59ec6 | 4210 | goto err_clear_dir; |
c81c925a | 4211 | |
c81c925a TH |
4212 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
4213 | parent->parent) { | |
ed3d261b | 4214 | pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n", |
a2a1f9ea | 4215 | current->comm, current->pid, ss->name); |
c81c925a | 4216 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4217 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4218 | ss->warned_broken_hierarchy = true; |
4219 | } | |
4220 | ||
4221 | return 0; | |
4222 | ||
3eb59ec6 | 4223 | err_clear_dir: |
32d01dc7 | 4224 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
15a4c835 TH |
4225 | err_free_id: |
4226 | cgroup_idr_remove(&ss->css_idr, css->id); | |
3eb59ec6 | 4227 | err_free_percpu_ref: |
c81c925a | 4228 | percpu_ref_cancel_init(&css->refcnt); |
3eb59ec6 | 4229 | err_free_css: |
a2bed820 | 4230 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
c81c925a TH |
4231 | return err; |
4232 | } | |
4233 | ||
2bd59d48 | 4234 | /** |
a043e3b2 LZ |
4235 | * cgroup_create - create a cgroup |
4236 | * @parent: cgroup that will be parent of the new cgroup | |
e61734c5 | 4237 | * @name: name of the new cgroup |
2bd59d48 | 4238 | * @mode: mode to set on new cgroup |
ddbcc7e8 | 4239 | */ |
e61734c5 | 4240 | static long cgroup_create(struct cgroup *parent, const char *name, |
2bd59d48 | 4241 | umode_t mode) |
ddbcc7e8 | 4242 | { |
bd89aabc | 4243 | struct cgroup *cgrp; |
3dd06ffa | 4244 | struct cgroup_root *root = parent->root; |
b58c8998 | 4245 | int ssid, err; |
ddbcc7e8 | 4246 | struct cgroup_subsys *ss; |
2bd59d48 | 4247 | struct kernfs_node *kn; |
ddbcc7e8 | 4248 | |
0a950f65 | 4249 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc PM |
4250 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
4251 | if (!cgrp) | |
ddbcc7e8 PM |
4252 | return -ENOMEM; |
4253 | ||
ace2bee8 | 4254 | mutex_lock(&cgroup_tree_mutex); |
65dff759 | 4255 | |
976c06bc TH |
4256 | /* |
4257 | * Only live parents can have children. Note that the liveliness | |
4258 | * check isn't strictly necessary because cgroup_mkdir() and | |
4259 | * cgroup_rmdir() are fully synchronized by i_mutex; however, do it | |
4260 | * anyway so that locking is contained inside cgroup proper and we | |
4261 | * don't get nasty surprises if we ever grow another caller. | |
4262 | */ | |
4263 | if (!cgroup_lock_live_group(parent)) { | |
4264 | err = -ENODEV; | |
ace2bee8 | 4265 | goto err_unlock_tree; |
0ab02ca8 LZ |
4266 | } |
4267 | ||
4268 | /* | |
4269 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4270 | * a half-baked cgroup. | |
4271 | */ | |
6fa4918d | 4272 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_NOWAIT); |
0ab02ca8 LZ |
4273 | if (cgrp->id < 0) { |
4274 | err = -ENOMEM; | |
4275 | goto err_unlock; | |
976c06bc TH |
4276 | } |
4277 | ||
cc31edce | 4278 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4279 | |
bd89aabc | 4280 | cgrp->parent = parent; |
0ae78e0b | 4281 | cgrp->dummy_css.parent = &parent->dummy_css; |
bd89aabc | 4282 | cgrp->root = parent->root; |
ddbcc7e8 | 4283 | |
b6abdb0e LZ |
4284 | if (notify_on_release(parent)) |
4285 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4286 | ||
2260e7fc TH |
4287 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4288 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4289 | |
2bd59d48 | 4290 | /* create the directory */ |
e61734c5 | 4291 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 TH |
4292 | if (IS_ERR(kn)) { |
4293 | err = PTR_ERR(kn); | |
0ab02ca8 | 4294 | goto err_free_id; |
2bd59d48 TH |
4295 | } |
4296 | cgrp->kn = kn; | |
ddbcc7e8 | 4297 | |
4e139afc | 4298 | /* |
6f30558f TH |
4299 | * This extra ref will be put in cgroup_free_fn() and guarantees |
4300 | * that @cgrp->kn is always accessible. | |
4e139afc | 4301 | */ |
6f30558f | 4302 | kernfs_get(kn); |
ddbcc7e8 | 4303 | |
00356bd5 | 4304 | cgrp->serial_nr = cgroup_serial_nr_next++; |
53fa5261 | 4305 | |
4e139afc | 4306 | /* allocation complete, commit to creation */ |
4e139afc | 4307 | list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children); |
3c9c825b | 4308 | atomic_inc(&root->nr_cgrps); |
59f5296b | 4309 | cgroup_get(parent); |
415cf07a | 4310 | |
0d80255e TH |
4311 | /* |
4312 | * @cgrp is now fully operational. If something fails after this | |
4313 | * point, it'll be released via the normal destruction path. | |
4314 | */ | |
6fa4918d | 4315 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 4316 | |
49957f8e TH |
4317 | err = cgroup_kn_set_ugid(kn); |
4318 | if (err) | |
4319 | goto err_destroy; | |
4320 | ||
2bb566cb | 4321 | err = cgroup_addrm_files(cgrp, cgroup_base_files, true); |
628f7cd4 TH |
4322 | if (err) |
4323 | goto err_destroy; | |
4324 | ||
9d403e99 | 4325 | /* let's create and online css's */ |
b85d2040 | 4326 | for_each_subsys(ss, ssid) { |
f392e51c | 4327 | if (parent->child_subsys_mask & (1 << ssid)) { |
b85d2040 TH |
4328 | err = create_css(cgrp, ss); |
4329 | if (err) | |
4330 | goto err_destroy; | |
4331 | } | |
a8638030 | 4332 | } |
ddbcc7e8 | 4333 | |
bd53d617 TH |
4334 | /* |
4335 | * On the default hierarchy, a child doesn't automatically inherit | |
4336 | * child_subsys_mask from the parent. Each is configured manually. | |
4337 | */ | |
4338 | if (!cgroup_on_dfl(cgrp)) | |
4339 | cgrp->child_subsys_mask = parent->child_subsys_mask; | |
f392e51c | 4340 | |
2bd59d48 TH |
4341 | kernfs_activate(kn); |
4342 | ||
ddbcc7e8 | 4343 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4344 | mutex_unlock(&cgroup_tree_mutex); |
ddbcc7e8 PM |
4345 | |
4346 | return 0; | |
4347 | ||
0a950f65 | 4348 | err_free_id: |
6fa4918d | 4349 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); |
0ab02ca8 LZ |
4350 | err_unlock: |
4351 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 TH |
4352 | err_unlock_tree: |
4353 | mutex_unlock(&cgroup_tree_mutex); | |
bd89aabc | 4354 | kfree(cgrp); |
ddbcc7e8 | 4355 | return err; |
4b8b47eb TH |
4356 | |
4357 | err_destroy: | |
4358 | cgroup_destroy_locked(cgrp); | |
4359 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4360 | mutex_unlock(&cgroup_tree_mutex); |
4b8b47eb | 4361 | return err; |
ddbcc7e8 PM |
4362 | } |
4363 | ||
2bd59d48 TH |
4364 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
4365 | umode_t mode) | |
ddbcc7e8 | 4366 | { |
2bd59d48 | 4367 | struct cgroup *parent = parent_kn->priv; |
e1b2dc17 | 4368 | int ret; |
ddbcc7e8 | 4369 | |
e1b2dc17 TH |
4370 | /* |
4371 | * cgroup_create() grabs cgroup_tree_mutex which nests outside | |
4372 | * kernfs active_ref and cgroup_create() already synchronizes | |
4373 | * properly against removal through cgroup_lock_live_group(). | |
4374 | * Break it before calling cgroup_create(). | |
4375 | */ | |
4376 | cgroup_get(parent); | |
4377 | kernfs_break_active_protection(parent_kn); | |
ddbcc7e8 | 4378 | |
e1b2dc17 TH |
4379 | ret = cgroup_create(parent, name, mode); |
4380 | ||
4381 | kernfs_unbreak_active_protection(parent_kn); | |
4382 | cgroup_put(parent); | |
4383 | return ret; | |
ddbcc7e8 PM |
4384 | } |
4385 | ||
223dbc38 TH |
4386 | /* |
4387 | * This is called when the refcnt of a css is confirmed to be killed. | |
ec903c0c | 4388 | * css_tryget_online() is now guaranteed to fail. |
223dbc38 TH |
4389 | */ |
4390 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4391 | { |
223dbc38 TH |
4392 | struct cgroup_subsys_state *css = |
4393 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
4394 | struct cgroup *cgrp = css->cgroup; | |
d3daf28d | 4395 | |
ace2bee8 | 4396 | mutex_lock(&cgroup_tree_mutex); |
f20104de TH |
4397 | mutex_lock(&cgroup_mutex); |
4398 | ||
09a503ea | 4399 | /* |
ec903c0c TH |
4400 | * css_tryget_online() is guaranteed to fail now. Tell subsystems |
4401 | * to initate destruction. | |
09a503ea TH |
4402 | */ |
4403 | offline_css(css); | |
4404 | ||
f20104de TH |
4405 | /* |
4406 | * If @cgrp is marked dead, it's waiting for refs of all css's to | |
4407 | * be disabled before proceeding to the second phase of cgroup | |
4408 | * destruction. If we are the last one, kick it off. | |
4409 | */ | |
09a503ea | 4410 | if (!cgrp->nr_css && cgroup_is_dead(cgrp)) |
f20104de TH |
4411 | cgroup_destroy_css_killed(cgrp); |
4412 | ||
4413 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4414 | mutex_unlock(&cgroup_tree_mutex); |
09a503ea TH |
4415 | |
4416 | /* | |
4417 | * Put the css refs from kill_css(). Each css holds an extra | |
4418 | * reference to the cgroup's dentry and cgroup removal proceeds | |
4419 | * regardless of css refs. On the last put of each css, whenever | |
4420 | * that may be, the extra dentry ref is put so that dentry | |
4421 | * destruction happens only after all css's are released. | |
4422 | */ | |
4423 | css_put(css); | |
d3daf28d TH |
4424 | } |
4425 | ||
223dbc38 TH |
4426 | /* css kill confirmation processing requires process context, bounce */ |
4427 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4428 | { |
4429 | struct cgroup_subsys_state *css = | |
4430 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4431 | ||
223dbc38 | 4432 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 4433 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
4434 | } |
4435 | ||
f392e51c TH |
4436 | /** |
4437 | * kill_css - destroy a css | |
4438 | * @css: css to destroy | |
4439 | * | |
4440 | * This function initiates destruction of @css by removing cgroup interface | |
4441 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
4442 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
4443 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
4444 | */ |
4445 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 4446 | { |
94419627 TH |
4447 | lockdep_assert_held(&cgroup_tree_mutex); |
4448 | ||
2bd59d48 TH |
4449 | /* |
4450 | * This must happen before css is disassociated with its cgroup. | |
4451 | * See seq_css() for details. | |
4452 | */ | |
aec25020 | 4453 | cgroup_clear_dir(css->cgroup, 1 << css->ss->id); |
3c14f8b4 | 4454 | |
edae0c33 TH |
4455 | /* |
4456 | * Killing would put the base ref, but we need to keep it alive | |
4457 | * until after ->css_offline(). | |
4458 | */ | |
4459 | css_get(css); | |
4460 | ||
4461 | /* | |
4462 | * cgroup core guarantees that, by the time ->css_offline() is | |
4463 | * invoked, no new css reference will be given out via | |
ec903c0c | 4464 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
4465 | * proceed to offlining css's because percpu_ref_kill() doesn't |
4466 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4467 | * | |
4468 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4469 | * css is confirmed to be seen as killed on all CPUs. | |
4470 | */ | |
4471 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4472 | } |
4473 | ||
4474 | /** | |
4475 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4476 | * @cgrp: cgroup to be destroyed | |
4477 | * | |
4478 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4479 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
4480 | * guarantee that css_tryget_online() won't succeed by the time |
4481 | * ->css_offline() is invoked. To satisfy all the requirements, | |
4482 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
4483 | * |
4484 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4485 | * userland visible parts and start killing the percpu refcnts of | |
4486 | * css's. Set up so that the next stage will be kicked off once all | |
4487 | * the percpu refcnts are confirmed to be killed. | |
4488 | * | |
4489 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4490 | * rest of destruction. Once all cgroup references are gone, the | |
4491 | * cgroup is RCU-freed. | |
4492 | * | |
4493 | * This function implements s1. After this step, @cgrp is gone as far as | |
4494 | * the userland is concerned and a new cgroup with the same name may be | |
4495 | * created. As cgroup doesn't care about the names internally, this | |
4496 | * doesn't cause any problem. | |
4497 | */ | |
42809dd4 TH |
4498 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4499 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4500 | { |
bb78a92f | 4501 | struct cgroup *child; |
2bd59d48 | 4502 | struct cgroup_subsys_state *css; |
ddd69148 | 4503 | bool empty; |
1c6727af | 4504 | int ssid; |
ddbcc7e8 | 4505 | |
ace2bee8 | 4506 | lockdep_assert_held(&cgroup_tree_mutex); |
42809dd4 TH |
4507 | lockdep_assert_held(&cgroup_mutex); |
4508 | ||
ddd69148 | 4509 | /* |
96d365e0 | 4510 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 4511 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 4512 | */ |
96d365e0 | 4513 | down_read(&css_set_rwsem); |
bb78a92f | 4514 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 4515 | up_read(&css_set_rwsem); |
ddd69148 | 4516 | if (!empty) |
ddbcc7e8 | 4517 | return -EBUSY; |
a043e3b2 | 4518 | |
bb78a92f HD |
4519 | /* |
4520 | * Make sure there's no live children. We can't test ->children | |
4521 | * emptiness as dead children linger on it while being destroyed; | |
4522 | * otherwise, "rmdir parent/child parent" may fail with -EBUSY. | |
4523 | */ | |
4524 | empty = true; | |
4525 | rcu_read_lock(); | |
4526 | list_for_each_entry_rcu(child, &cgrp->children, sibling) { | |
4527 | empty = cgroup_is_dead(child); | |
4528 | if (!empty) | |
4529 | break; | |
4530 | } | |
4531 | rcu_read_unlock(); | |
4532 | if (!empty) | |
4533 | return -EBUSY; | |
4534 | ||
455050d2 TH |
4535 | /* |
4536 | * Mark @cgrp dead. This prevents further task migration and child | |
4537 | * creation by disabling cgroup_lock_live_group(). Note that | |
492eb21b | 4538 | * CGRP_DEAD assertion is depended upon by css_next_child() to |
455050d2 | 4539 | * resume iteration after dropping RCU read lock. See |
492eb21b | 4540 | * css_next_child() for details. |
455050d2 | 4541 | */ |
54766d4a | 4542 | set_bit(CGRP_DEAD, &cgrp->flags); |
ddbcc7e8 | 4543 | |
88703267 | 4544 | /* |
edae0c33 TH |
4545 | * Initiate massacre of all css's. cgroup_destroy_css_killed() |
4546 | * will be invoked to perform the rest of destruction once the | |
4ac06017 TH |
4547 | * percpu refs of all css's are confirmed to be killed. This |
4548 | * involves removing the subsystem's files, drop cgroup_mutex. | |
88703267 | 4549 | */ |
4ac06017 | 4550 | mutex_unlock(&cgroup_mutex); |
1c6727af TH |
4551 | for_each_css(css, ssid, cgrp) |
4552 | kill_css(css); | |
4ac06017 | 4553 | mutex_lock(&cgroup_mutex); |
455050d2 | 4554 | |
455050d2 TH |
4555 | /* CGRP_DEAD is set, remove from ->release_list for the last time */ |
4556 | raw_spin_lock(&release_list_lock); | |
4557 | if (!list_empty(&cgrp->release_list)) | |
4558 | list_del_init(&cgrp->release_list); | |
4559 | raw_spin_unlock(&release_list_lock); | |
4560 | ||
4561 | /* | |
f20104de TH |
4562 | * If @cgrp has css's attached, the second stage of cgroup |
4563 | * destruction is kicked off from css_killed_work_fn() after the | |
4564 | * refs of all attached css's are killed. If @cgrp doesn't have | |
4565 | * any css, we kick it off here. | |
4566 | */ | |
4567 | if (!cgrp->nr_css) | |
4568 | cgroup_destroy_css_killed(cgrp); | |
4569 | ||
2bd59d48 TH |
4570 | /* remove @cgrp directory along with the base files */ |
4571 | mutex_unlock(&cgroup_mutex); | |
4572 | ||
455050d2 | 4573 | /* |
2bd59d48 TH |
4574 | * There are two control paths which try to determine cgroup from |
4575 | * dentry without going through kernfs - cgroupstats_build() and | |
ec903c0c TH |
4576 | * css_tryget_online_from_dir(). Those are supported by RCU |
4577 | * protecting clearing of cgrp->kn->priv backpointer, which should | |
4578 | * happen after all files under it have been removed. | |
455050d2 | 4579 | */ |
6f30558f | 4580 | kernfs_remove(cgrp->kn); /* @cgrp has an extra ref on its kn */ |
2bd59d48 | 4581 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); |
2bd59d48 | 4582 | |
4ac06017 | 4583 | mutex_lock(&cgroup_mutex); |
455050d2 | 4584 | |
ea15f8cc TH |
4585 | return 0; |
4586 | }; | |
4587 | ||
d3daf28d | 4588 | /** |
f20104de | 4589 | * cgroup_destroy_css_killed - the second step of cgroup destruction |
60106946 | 4590 | * @cgrp: the cgroup whose csses have just finished offlining |
d3daf28d TH |
4591 | * |
4592 | * This function is invoked from a work item for a cgroup which is being | |
09a503ea TH |
4593 | * destroyed after all css's are offlined and performs the rest of |
4594 | * destruction. This is the second step of destruction described in the | |
4595 | * comment above cgroup_destroy_locked(). | |
d3daf28d | 4596 | */ |
f20104de | 4597 | static void cgroup_destroy_css_killed(struct cgroup *cgrp) |
ea15f8cc | 4598 | { |
ea15f8cc | 4599 | struct cgroup *parent = cgrp->parent; |
ea15f8cc | 4600 | |
ace2bee8 | 4601 | lockdep_assert_held(&cgroup_tree_mutex); |
f20104de | 4602 | lockdep_assert_held(&cgroup_mutex); |
ea15f8cc | 4603 | |
999cd8a4 | 4604 | /* delete this cgroup from parent->children */ |
eb6fd504 | 4605 | list_del_rcu(&cgrp->sibling); |
ed957793 | 4606 | |
59f5296b | 4607 | cgroup_put(cgrp); |
ddbcc7e8 | 4608 | |
bd89aabc | 4609 | set_bit(CGRP_RELEASABLE, &parent->flags); |
81a6a5cd | 4610 | check_for_release(parent); |
ddbcc7e8 PM |
4611 | } |
4612 | ||
2bd59d48 | 4613 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4614 | { |
2bd59d48 TH |
4615 | struct cgroup *cgrp = kn->priv; |
4616 | int ret = 0; | |
42809dd4 | 4617 | |
2bd59d48 TH |
4618 | /* |
4619 | * This is self-destruction but @kn can't be removed while this | |
4620 | * callback is in progress. Let's break active protection. Once | |
4621 | * the protection is broken, @cgrp can be destroyed at any point. | |
4622 | * Pin it so that it stays accessible. | |
4623 | */ | |
4624 | cgroup_get(cgrp); | |
4625 | kernfs_break_active_protection(kn); | |
42809dd4 | 4626 | |
ace2bee8 | 4627 | mutex_lock(&cgroup_tree_mutex); |
42809dd4 | 4628 | mutex_lock(&cgroup_mutex); |
8e3f6541 TH |
4629 | |
4630 | /* | |
2bd59d48 TH |
4631 | * @cgrp might already have been destroyed while we're trying to |
4632 | * grab the mutexes. | |
8e3f6541 | 4633 | */ |
2bd59d48 TH |
4634 | if (!cgroup_is_dead(cgrp)) |
4635 | ret = cgroup_destroy_locked(cgrp); | |
2bb566cb | 4636 | |
42809dd4 | 4637 | mutex_unlock(&cgroup_mutex); |
ace2bee8 | 4638 | mutex_unlock(&cgroup_tree_mutex); |
2bb566cb | 4639 | |
2bd59d48 TH |
4640 | kernfs_unbreak_active_protection(kn); |
4641 | cgroup_put(cgrp); | |
42809dd4 | 4642 | return ret; |
8e3f6541 TH |
4643 | } |
4644 | ||
2bd59d48 TH |
4645 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4646 | .remount_fs = cgroup_remount, | |
4647 | .show_options = cgroup_show_options, | |
4648 | .mkdir = cgroup_mkdir, | |
4649 | .rmdir = cgroup_rmdir, | |
4650 | .rename = cgroup_rename, | |
4651 | }; | |
4652 | ||
15a4c835 | 4653 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 4654 | { |
ddbcc7e8 | 4655 | struct cgroup_subsys_state *css; |
cfe36bde DC |
4656 | |
4657 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 4658 | |
ace2bee8 | 4659 | mutex_lock(&cgroup_tree_mutex); |
648bb56d TH |
4660 | mutex_lock(&cgroup_mutex); |
4661 | ||
15a4c835 | 4662 | idr_init(&ss->css_idr); |
0adb0704 | 4663 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4664 | |
3dd06ffa TH |
4665 | /* Create the root cgroup state for this subsystem */ |
4666 | ss->root = &cgrp_dfl_root; | |
4667 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
4668 | /* We don't handle early failures gracefully */ |
4669 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 4670 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
15a4c835 TH |
4671 | if (early) { |
4672 | /* idr_alloc() can't be called safely during early init */ | |
4673 | css->id = 1; | |
4674 | } else { | |
4675 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
4676 | BUG_ON(css->id < 0); | |
4677 | } | |
ddbcc7e8 | 4678 | |
e8d55fde | 4679 | /* Update the init_css_set to contain a subsys |
817929ec | 4680 | * pointer to this state - since the subsystem is |
e8d55fde | 4681 | * newly registered, all tasks and hence the |
3dd06ffa | 4682 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 4683 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 PM |
4684 | |
4685 | need_forkexit_callback |= ss->fork || ss->exit; | |
4686 | ||
e8d55fde LZ |
4687 | /* At system boot, before all subsystems have been |
4688 | * registered, no tasks have been forked, so we don't | |
4689 | * need to invoke fork callbacks here. */ | |
4690 | BUG_ON(!list_empty(&init_task.tasks)); | |
4691 | ||
ae7f164a | 4692 | BUG_ON(online_css(css)); |
a8638030 | 4693 | |
f392e51c | 4694 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
cf5d5941 BB |
4695 | |
4696 | mutex_unlock(&cgroup_mutex); | |
ace2bee8 | 4697 | mutex_unlock(&cgroup_tree_mutex); |
cf5d5941 | 4698 | } |
cf5d5941 | 4699 | |
ddbcc7e8 | 4700 | /** |
a043e3b2 LZ |
4701 | * cgroup_init_early - cgroup initialization at system boot |
4702 | * | |
4703 | * Initialize cgroups at system boot, and initialize any | |
4704 | * subsystems that request early init. | |
ddbcc7e8 PM |
4705 | */ |
4706 | int __init cgroup_init_early(void) | |
4707 | { | |
a2dd4247 TH |
4708 | static struct cgroup_sb_opts __initdata opts = |
4709 | { .flags = CGRP_ROOT_SANE_BEHAVIOR }; | |
30159ec7 | 4710 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4711 | int i; |
30159ec7 | 4712 | |
3dd06ffa | 4713 | init_cgroup_root(&cgrp_dfl_root, &opts); |
a4ea1cc9 | 4714 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4715 | |
3ed80a62 | 4716 | for_each_subsys(ss, i) { |
aec25020 | 4717 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
4718 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
4719 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 4720 | ss->id, ss->name); |
073219e9 TH |
4721 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4722 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4723 | ||
aec25020 | 4724 | ss->id = i; |
073219e9 | 4725 | ss->name = cgroup_subsys_name[i]; |
ddbcc7e8 PM |
4726 | |
4727 | if (ss->early_init) | |
15a4c835 | 4728 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
4729 | } |
4730 | return 0; | |
4731 | } | |
4732 | ||
4733 | /** | |
a043e3b2 LZ |
4734 | * cgroup_init - cgroup initialization |
4735 | * | |
4736 | * Register cgroup filesystem and /proc file, and initialize | |
4737 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4738 | */ |
4739 | int __init cgroup_init(void) | |
4740 | { | |
30159ec7 | 4741 | struct cgroup_subsys *ss; |
0ac801fe | 4742 | unsigned long key; |
172a2c06 | 4743 | int ssid, err; |
ddbcc7e8 | 4744 | |
2bd59d48 | 4745 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
ddbcc7e8 | 4746 | |
985ed670 | 4747 | mutex_lock(&cgroup_tree_mutex); |
54e7b4eb | 4748 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4749 | |
82fe9b0d TH |
4750 | /* Add init_css_set to the hash table */ |
4751 | key = css_set_hash(init_css_set.subsys); | |
4752 | hash_add(css_set_table, &init_css_set.hlist, key); | |
4753 | ||
3dd06ffa | 4754 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 4755 | |
54e7b4eb | 4756 | mutex_unlock(&cgroup_mutex); |
985ed670 | 4757 | mutex_unlock(&cgroup_tree_mutex); |
54e7b4eb | 4758 | |
172a2c06 | 4759 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
4760 | if (ss->early_init) { |
4761 | struct cgroup_subsys_state *css = | |
4762 | init_css_set.subsys[ss->id]; | |
4763 | ||
4764 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
4765 | GFP_KERNEL); | |
4766 | BUG_ON(css->id < 0); | |
4767 | } else { | |
4768 | cgroup_init_subsys(ss, false); | |
4769 | } | |
172a2c06 | 4770 | |
2d8f243a TH |
4771 | list_add_tail(&init_css_set.e_cset_node[ssid], |
4772 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
4773 | ||
172a2c06 TH |
4774 | /* |
4775 | * cftype registration needs kmalloc and can't be done | |
4776 | * during early_init. Register base cftypes separately. | |
4777 | */ | |
4778 | if (ss->base_cftypes) | |
4779 | WARN_ON(cgroup_add_cftypes(ss, ss->base_cftypes)); | |
676db4af GK |
4780 | } |
4781 | ||
676db4af | 4782 | cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj); |
2bd59d48 TH |
4783 | if (!cgroup_kobj) |
4784 | return -ENOMEM; | |
676db4af | 4785 | |
ddbcc7e8 | 4786 | err = register_filesystem(&cgroup_fs_type); |
676db4af GK |
4787 | if (err < 0) { |
4788 | kobject_put(cgroup_kobj); | |
2bd59d48 | 4789 | return err; |
676db4af | 4790 | } |
ddbcc7e8 | 4791 | |
46ae220b | 4792 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 4793 | return 0; |
ddbcc7e8 | 4794 | } |
b4f48b63 | 4795 | |
e5fca243 TH |
4796 | static int __init cgroup_wq_init(void) |
4797 | { | |
4798 | /* | |
4799 | * There isn't much point in executing destruction path in | |
4800 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 4801 | * Use 1 for @max_active. |
e5fca243 TH |
4802 | * |
4803 | * We would prefer to do this in cgroup_init() above, but that | |
4804 | * is called before init_workqueues(): so leave this until after. | |
4805 | */ | |
1a11533f | 4806 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 4807 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
4808 | |
4809 | /* | |
4810 | * Used to destroy pidlists and separate to serve as flush domain. | |
4811 | * Cap @max_active to 1 too. | |
4812 | */ | |
4813 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
4814 | 0, 1); | |
4815 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
4816 | ||
e5fca243 TH |
4817 | return 0; |
4818 | } | |
4819 | core_initcall(cgroup_wq_init); | |
4820 | ||
a424316c PM |
4821 | /* |
4822 | * proc_cgroup_show() | |
4823 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4824 | * - Used for /proc/<pid>/cgroup. | |
a424316c PM |
4825 | */ |
4826 | ||
4827 | /* TODO: Use a proper seq_file iterator */ | |
8d8b97ba | 4828 | int proc_cgroup_show(struct seq_file *m, void *v) |
a424316c PM |
4829 | { |
4830 | struct pid *pid; | |
4831 | struct task_struct *tsk; | |
e61734c5 | 4832 | char *buf, *path; |
a424316c | 4833 | int retval; |
3dd06ffa | 4834 | struct cgroup_root *root; |
a424316c PM |
4835 | |
4836 | retval = -ENOMEM; | |
e61734c5 | 4837 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4838 | if (!buf) |
4839 | goto out; | |
4840 | ||
4841 | retval = -ESRCH; | |
4842 | pid = m->private; | |
4843 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
4844 | if (!tsk) | |
4845 | goto out_free; | |
4846 | ||
4847 | retval = 0; | |
4848 | ||
4849 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 4850 | down_read(&css_set_rwsem); |
a424316c | 4851 | |
985ed670 | 4852 | for_each_root(root) { |
a424316c | 4853 | struct cgroup_subsys *ss; |
bd89aabc | 4854 | struct cgroup *cgrp; |
b85d2040 | 4855 | int ssid, count = 0; |
a424316c | 4856 | |
a2dd4247 | 4857 | if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) |
985ed670 TH |
4858 | continue; |
4859 | ||
2c6ab6d2 | 4860 | seq_printf(m, "%d:", root->hierarchy_id); |
b85d2040 | 4861 | for_each_subsys(ss, ssid) |
f392e51c | 4862 | if (root->subsys_mask & (1 << ssid)) |
b85d2040 | 4863 | seq_printf(m, "%s%s", count++ ? "," : "", ss->name); |
c6d57f33 PM |
4864 | if (strlen(root->name)) |
4865 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4866 | root->name); | |
a424316c | 4867 | seq_putc(m, ':'); |
7717f7ba | 4868 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
4869 | path = cgroup_path(cgrp, buf, PATH_MAX); |
4870 | if (!path) { | |
4871 | retval = -ENAMETOOLONG; | |
a424316c | 4872 | goto out_unlock; |
e61734c5 TH |
4873 | } |
4874 | seq_puts(m, path); | |
a424316c PM |
4875 | seq_putc(m, '\n'); |
4876 | } | |
4877 | ||
4878 | out_unlock: | |
96d365e0 | 4879 | up_read(&css_set_rwsem); |
a424316c PM |
4880 | mutex_unlock(&cgroup_mutex); |
4881 | put_task_struct(tsk); | |
4882 | out_free: | |
4883 | kfree(buf); | |
4884 | out: | |
4885 | return retval; | |
4886 | } | |
4887 | ||
a424316c PM |
4888 | /* Display information about each subsystem and each hierarchy */ |
4889 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
4890 | { | |
30159ec7 | 4891 | struct cgroup_subsys *ss; |
a424316c | 4892 | int i; |
a424316c | 4893 | |
8bab8dde | 4894 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
4895 | /* |
4896 | * ideally we don't want subsystems moving around while we do this. | |
4897 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
4898 | * subsys/hierarchy state. | |
4899 | */ | |
a424316c | 4900 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
4901 | |
4902 | for_each_subsys(ss, i) | |
2c6ab6d2 PM |
4903 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
4904 | ss->name, ss->root->hierarchy_id, | |
3c9c825b | 4905 | atomic_read(&ss->root->nr_cgrps), !ss->disabled); |
30159ec7 | 4906 | |
a424316c PM |
4907 | mutex_unlock(&cgroup_mutex); |
4908 | return 0; | |
4909 | } | |
4910 | ||
4911 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
4912 | { | |
9dce07f1 | 4913 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
4914 | } |
4915 | ||
828c0950 | 4916 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
4917 | .open = cgroupstats_open, |
4918 | .read = seq_read, | |
4919 | .llseek = seq_lseek, | |
4920 | .release = single_release, | |
4921 | }; | |
4922 | ||
b4f48b63 | 4923 | /** |
eaf797ab | 4924 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 4925 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 4926 | * |
eaf797ab TH |
4927 | * A task is associated with the init_css_set until cgroup_post_fork() |
4928 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
4929 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
4930 | */ |
4931 | void cgroup_fork(struct task_struct *child) | |
4932 | { | |
eaf797ab | 4933 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 4934 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4935 | } |
4936 | ||
817929ec | 4937 | /** |
a043e3b2 LZ |
4938 | * cgroup_post_fork - called on a new task after adding it to the task list |
4939 | * @child: the task in question | |
4940 | * | |
5edee61e TH |
4941 | * Adds the task to the list running through its css_set if necessary and |
4942 | * call the subsystem fork() callbacks. Has to be after the task is | |
4943 | * visible on the task list in case we race with the first call to | |
0942eeee | 4944 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4945 | * list. |
a043e3b2 | 4946 | */ |
817929ec PM |
4947 | void cgroup_post_fork(struct task_struct *child) |
4948 | { | |
30159ec7 | 4949 | struct cgroup_subsys *ss; |
5edee61e TH |
4950 | int i; |
4951 | ||
3ce3230a | 4952 | /* |
eaf797ab TH |
4953 | * This may race against cgroup_enable_task_cg_links(). As that |
4954 | * function sets use_task_css_set_links before grabbing | |
4955 | * tasklist_lock and we just went through tasklist_lock to add | |
4956 | * @child, it's guaranteed that either we see the set | |
4957 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
4958 | * @child during its iteration. | |
4959 | * | |
4960 | * If we won the race, @child is associated with %current's | |
4961 | * css_set. Grabbing css_set_rwsem guarantees both that the | |
4962 | * association is stable, and, on completion of the parent's | |
4963 | * migration, @child is visible in the source of migration or | |
4964 | * already in the destination cgroup. This guarantee is necessary | |
4965 | * when implementing operations which need to migrate all tasks of | |
4966 | * a cgroup to another. | |
4967 | * | |
4968 | * Note that if we lose to cgroup_enable_task_cg_links(), @child | |
4969 | * will remain in init_css_set. This is safe because all tasks are | |
4970 | * in the init_css_set before cg_links is enabled and there's no | |
4971 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 4972 | */ |
817929ec | 4973 | if (use_task_css_set_links) { |
eaf797ab TH |
4974 | struct css_set *cset; |
4975 | ||
96d365e0 | 4976 | down_write(&css_set_rwsem); |
0e1d768f | 4977 | cset = task_css_set(current); |
eaf797ab TH |
4978 | if (list_empty(&child->cg_list)) { |
4979 | rcu_assign_pointer(child->cgroups, cset); | |
4980 | list_add(&child->cg_list, &cset->tasks); | |
4981 | get_css_set(cset); | |
4982 | } | |
96d365e0 | 4983 | up_write(&css_set_rwsem); |
817929ec | 4984 | } |
5edee61e TH |
4985 | |
4986 | /* | |
4987 | * Call ss->fork(). This must happen after @child is linked on | |
4988 | * css_set; otherwise, @child might change state between ->fork() | |
4989 | * and addition to css_set. | |
4990 | */ | |
4991 | if (need_forkexit_callback) { | |
3ed80a62 | 4992 | for_each_subsys(ss, i) |
5edee61e TH |
4993 | if (ss->fork) |
4994 | ss->fork(child); | |
5edee61e | 4995 | } |
817929ec | 4996 | } |
5edee61e | 4997 | |
b4f48b63 PM |
4998 | /** |
4999 | * cgroup_exit - detach cgroup from exiting task | |
5000 | * @tsk: pointer to task_struct of exiting process | |
5001 | * | |
5002 | * Description: Detach cgroup from @tsk and release it. | |
5003 | * | |
5004 | * Note that cgroups marked notify_on_release force every task in | |
5005 | * them to take the global cgroup_mutex mutex when exiting. | |
5006 | * This could impact scaling on very large systems. Be reluctant to | |
5007 | * use notify_on_release cgroups where very high task exit scaling | |
5008 | * is required on large systems. | |
5009 | * | |
0e1d768f TH |
5010 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5011 | * call cgroup_exit() while the task is still competent to handle | |
5012 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5013 | * each hierarchy for the remainder of its exit. No need to bother with | |
5014 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5015 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5016 | */ |
1ec41830 | 5017 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5018 | { |
30159ec7 | 5019 | struct cgroup_subsys *ss; |
5abb8855 | 5020 | struct css_set *cset; |
eaf797ab | 5021 | bool put_cset = false; |
d41d5a01 | 5022 | int i; |
817929ec PM |
5023 | |
5024 | /* | |
0e1d768f TH |
5025 | * Unlink from @tsk from its css_set. As migration path can't race |
5026 | * with us, we can check cg_list without grabbing css_set_rwsem. | |
817929ec PM |
5027 | */ |
5028 | if (!list_empty(&tsk->cg_list)) { | |
96d365e0 | 5029 | down_write(&css_set_rwsem); |
0e1d768f | 5030 | list_del_init(&tsk->cg_list); |
96d365e0 | 5031 | up_write(&css_set_rwsem); |
0e1d768f | 5032 | put_cset = true; |
817929ec PM |
5033 | } |
5034 | ||
b4f48b63 | 5035 | /* Reassign the task to the init_css_set. */ |
a8ad805c TH |
5036 | cset = task_css_set(tsk); |
5037 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); | |
d41d5a01 | 5038 | |
1ec41830 | 5039 | if (need_forkexit_callback) { |
3ed80a62 TH |
5040 | /* see cgroup_post_fork() for details */ |
5041 | for_each_subsys(ss, i) { | |
d41d5a01 | 5042 | if (ss->exit) { |
eb95419b TH |
5043 | struct cgroup_subsys_state *old_css = cset->subsys[i]; |
5044 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 5045 | |
eb95419b | 5046 | ss->exit(css, old_css, tsk); |
d41d5a01 PZ |
5047 | } |
5048 | } | |
5049 | } | |
d41d5a01 | 5050 | |
eaf797ab TH |
5051 | if (put_cset) |
5052 | put_css_set(cset, true); | |
b4f48b63 | 5053 | } |
697f4161 | 5054 | |
bd89aabc | 5055 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5056 | { |
f50daa70 | 5057 | if (cgroup_is_releasable(cgrp) && |
6f3d828f | 5058 | list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) { |
f50daa70 LZ |
5059 | /* |
5060 | * Control Group is currently removeable. If it's not | |
81a6a5cd | 5061 | * already queued for a userspace notification, queue |
f50daa70 LZ |
5062 | * it now |
5063 | */ | |
81a6a5cd | 5064 | int need_schedule_work = 0; |
f50daa70 | 5065 | |
cdcc136f | 5066 | raw_spin_lock(&release_list_lock); |
54766d4a | 5067 | if (!cgroup_is_dead(cgrp) && |
bd89aabc PM |
5068 | list_empty(&cgrp->release_list)) { |
5069 | list_add(&cgrp->release_list, &release_list); | |
81a6a5cd PM |
5070 | need_schedule_work = 1; |
5071 | } | |
cdcc136f | 5072 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5073 | if (need_schedule_work) |
5074 | schedule_work(&release_agent_work); | |
5075 | } | |
5076 | } | |
5077 | ||
81a6a5cd PM |
5078 | /* |
5079 | * Notify userspace when a cgroup is released, by running the | |
5080 | * configured release agent with the name of the cgroup (path | |
5081 | * relative to the root of cgroup file system) as the argument. | |
5082 | * | |
5083 | * Most likely, this user command will try to rmdir this cgroup. | |
5084 | * | |
5085 | * This races with the possibility that some other task will be | |
5086 | * attached to this cgroup before it is removed, or that some other | |
5087 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5088 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5089 | * unused, and this cgroup will be reprieved from its death sentence, | |
5090 | * to continue to serve a useful existence. Next time it's released, | |
5091 | * we will get notified again, if it still has 'notify_on_release' set. | |
5092 | * | |
5093 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5094 | * means only wait until the task is successfully execve()'d. The | |
5095 | * separate release agent task is forked by call_usermodehelper(), | |
5096 | * then control in this thread returns here, without waiting for the | |
5097 | * release agent task. We don't bother to wait because the caller of | |
5098 | * this routine has no use for the exit status of the release agent | |
5099 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5100 | */ |
81a6a5cd PM |
5101 | static void cgroup_release_agent(struct work_struct *work) |
5102 | { | |
5103 | BUG_ON(work != &release_agent_work); | |
5104 | mutex_lock(&cgroup_mutex); | |
cdcc136f | 5105 | raw_spin_lock(&release_list_lock); |
81a6a5cd PM |
5106 | while (!list_empty(&release_list)) { |
5107 | char *argv[3], *envp[3]; | |
5108 | int i; | |
e61734c5 | 5109 | char *pathbuf = NULL, *agentbuf = NULL, *path; |
bd89aabc | 5110 | struct cgroup *cgrp = list_entry(release_list.next, |
81a6a5cd PM |
5111 | struct cgroup, |
5112 | release_list); | |
bd89aabc | 5113 | list_del_init(&cgrp->release_list); |
cdcc136f | 5114 | raw_spin_unlock(&release_list_lock); |
e61734c5 | 5115 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); |
e788e066 PM |
5116 | if (!pathbuf) |
5117 | goto continue_free; | |
e61734c5 TH |
5118 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); |
5119 | if (!path) | |
e788e066 PM |
5120 | goto continue_free; |
5121 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5122 | if (!agentbuf) | |
5123 | goto continue_free; | |
81a6a5cd PM |
5124 | |
5125 | i = 0; | |
e788e066 | 5126 | argv[i++] = agentbuf; |
e61734c5 | 5127 | argv[i++] = path; |
81a6a5cd PM |
5128 | argv[i] = NULL; |
5129 | ||
5130 | i = 0; | |
5131 | /* minimal command environment */ | |
5132 | envp[i++] = "HOME=/"; | |
5133 | envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5134 | envp[i] = NULL; | |
5135 | ||
5136 | /* Drop the lock while we invoke the usermode helper, | |
5137 | * since the exec could involve hitting disk and hence | |
5138 | * be a slow process */ | |
5139 | mutex_unlock(&cgroup_mutex); | |
5140 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); | |
81a6a5cd | 5141 | mutex_lock(&cgroup_mutex); |
e788e066 PM |
5142 | continue_free: |
5143 | kfree(pathbuf); | |
5144 | kfree(agentbuf); | |
cdcc136f | 5145 | raw_spin_lock(&release_list_lock); |
81a6a5cd | 5146 | } |
cdcc136f | 5147 | raw_spin_unlock(&release_list_lock); |
81a6a5cd PM |
5148 | mutex_unlock(&cgroup_mutex); |
5149 | } | |
8bab8dde PM |
5150 | |
5151 | static int __init cgroup_disable(char *str) | |
5152 | { | |
30159ec7 | 5153 | struct cgroup_subsys *ss; |
8bab8dde | 5154 | char *token; |
30159ec7 | 5155 | int i; |
8bab8dde PM |
5156 | |
5157 | while ((token = strsep(&str, ",")) != NULL) { | |
5158 | if (!*token) | |
5159 | continue; | |
be45c900 | 5160 | |
3ed80a62 | 5161 | for_each_subsys(ss, i) { |
8bab8dde PM |
5162 | if (!strcmp(token, ss->name)) { |
5163 | ss->disabled = 1; | |
5164 | printk(KERN_INFO "Disabling %s control group" | |
5165 | " subsystem\n", ss->name); | |
5166 | break; | |
5167 | } | |
5168 | } | |
5169 | } | |
5170 | return 1; | |
5171 | } | |
5172 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5173 | |
b77d7b60 | 5174 | /** |
ec903c0c | 5175 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5176 | * @dentry: directory dentry of interest |
5177 | * @ss: subsystem of interest | |
b77d7b60 | 5178 | * |
5a17f543 TH |
5179 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5180 | * to get the corresponding css and return it. If such css doesn't exist | |
5181 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5182 | */ |
ec903c0c TH |
5183 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5184 | struct cgroup_subsys *ss) | |
e5d1367f | 5185 | { |
2bd59d48 TH |
5186 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
5187 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 5188 | struct cgroup *cgrp; |
e5d1367f | 5189 | |
35cf0836 | 5190 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
5191 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
5192 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5193 | return ERR_PTR(-EBADF); |
5194 | ||
5a17f543 TH |
5195 | rcu_read_lock(); |
5196 | ||
2bd59d48 TH |
5197 | /* |
5198 | * This path doesn't originate from kernfs and @kn could already | |
5199 | * have been or be removed at any point. @kn->priv is RCU | |
5200 | * protected for this access. See destroy_locked() for details. | |
5201 | */ | |
5202 | cgrp = rcu_dereference(kn->priv); | |
5203 | if (cgrp) | |
5204 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5205 | |
ec903c0c | 5206 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5207 | css = ERR_PTR(-ENOENT); |
5208 | ||
5209 | rcu_read_unlock(); | |
5210 | return css; | |
e5d1367f | 5211 | } |
e5d1367f | 5212 | |
1cb650b9 LZ |
5213 | /** |
5214 | * css_from_id - lookup css by id | |
5215 | * @id: the cgroup id | |
5216 | * @ss: cgroup subsys to be looked into | |
5217 | * | |
5218 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5219 | * Should be called under rcu_read_lock(). | |
5220 | */ | |
5221 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5222 | { | |
6fa4918d | 5223 | WARN_ON_ONCE(!rcu_read_lock_held()); |
15a4c835 | 5224 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
5225 | } |
5226 | ||
fe693435 | 5227 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5228 | static struct cgroup_subsys_state * |
5229 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5230 | { |
5231 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5232 | ||
5233 | if (!css) | |
5234 | return ERR_PTR(-ENOMEM); | |
5235 | ||
5236 | return css; | |
5237 | } | |
5238 | ||
eb95419b | 5239 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5240 | { |
eb95419b | 5241 | kfree(css); |
fe693435 PM |
5242 | } |
5243 | ||
182446d0 TH |
5244 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5245 | struct cftype *cft) | |
fe693435 | 5246 | { |
182446d0 | 5247 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5248 | } |
5249 | ||
182446d0 TH |
5250 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5251 | struct cftype *cft) | |
fe693435 PM |
5252 | { |
5253 | return (u64)(unsigned long)current->cgroups; | |
5254 | } | |
5255 | ||
182446d0 | 5256 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5257 | struct cftype *cft) |
fe693435 PM |
5258 | { |
5259 | u64 count; | |
5260 | ||
5261 | rcu_read_lock(); | |
a8ad805c | 5262 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5263 | rcu_read_unlock(); |
5264 | return count; | |
5265 | } | |
5266 | ||
2da8ca82 | 5267 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5268 | { |
69d0206c | 5269 | struct cgrp_cset_link *link; |
5abb8855 | 5270 | struct css_set *cset; |
e61734c5 TH |
5271 | char *name_buf; |
5272 | ||
5273 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
5274 | if (!name_buf) | |
5275 | return -ENOMEM; | |
7717f7ba | 5276 | |
96d365e0 | 5277 | down_read(&css_set_rwsem); |
7717f7ba | 5278 | rcu_read_lock(); |
5abb8855 | 5279 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5280 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 5281 | struct cgroup *c = link->cgrp; |
7717f7ba | 5282 | |
a2dd4247 | 5283 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 5284 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 5285 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
5286 | } |
5287 | rcu_read_unlock(); | |
96d365e0 | 5288 | up_read(&css_set_rwsem); |
e61734c5 | 5289 | kfree(name_buf); |
7717f7ba PM |
5290 | return 0; |
5291 | } | |
5292 | ||
5293 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 5294 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5295 | { |
2da8ca82 | 5296 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 5297 | struct cgrp_cset_link *link; |
7717f7ba | 5298 | |
96d365e0 | 5299 | down_read(&css_set_rwsem); |
182446d0 | 5300 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5301 | struct css_set *cset = link->cset; |
7717f7ba PM |
5302 | struct task_struct *task; |
5303 | int count = 0; | |
c7561128 | 5304 | |
5abb8855 | 5305 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 5306 | |
5abb8855 | 5307 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
5308 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
5309 | goto overflow; | |
5310 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
5311 | } | |
5312 | ||
5313 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
5314 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
5315 | goto overflow; | |
5316 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 5317 | } |
c7561128 TH |
5318 | continue; |
5319 | overflow: | |
5320 | seq_puts(seq, " ...\n"); | |
7717f7ba | 5321 | } |
96d365e0 | 5322 | up_read(&css_set_rwsem); |
7717f7ba PM |
5323 | return 0; |
5324 | } | |
5325 | ||
182446d0 | 5326 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5327 | { |
182446d0 | 5328 | return test_bit(CGRP_RELEASABLE, &css->cgroup->flags); |
fe693435 PM |
5329 | } |
5330 | ||
5331 | static struct cftype debug_files[] = { | |
fe693435 PM |
5332 | { |
5333 | .name = "taskcount", | |
5334 | .read_u64 = debug_taskcount_read, | |
5335 | }, | |
5336 | ||
5337 | { | |
5338 | .name = "current_css_set", | |
5339 | .read_u64 = current_css_set_read, | |
5340 | }, | |
5341 | ||
5342 | { | |
5343 | .name = "current_css_set_refcount", | |
5344 | .read_u64 = current_css_set_refcount_read, | |
5345 | }, | |
5346 | ||
7717f7ba PM |
5347 | { |
5348 | .name = "current_css_set_cg_links", | |
2da8ca82 | 5349 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
5350 | }, |
5351 | ||
5352 | { | |
5353 | .name = "cgroup_css_links", | |
2da8ca82 | 5354 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
5355 | }, |
5356 | ||
fe693435 PM |
5357 | { |
5358 | .name = "releasable", | |
5359 | .read_u64 = releasable_read, | |
5360 | }, | |
fe693435 | 5361 | |
4baf6e33 TH |
5362 | { } /* terminate */ |
5363 | }; | |
fe693435 | 5364 | |
073219e9 | 5365 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
5366 | .css_alloc = debug_css_alloc, |
5367 | .css_free = debug_css_free, | |
4baf6e33 | 5368 | .base_cftypes = debug_files, |
fe693435 PM |
5369 | }; |
5370 | #endif /* CONFIG_CGROUP_DEBUG */ |