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> | |
c9482a5b | 38 | #include <linux/magic.h> |
ddbcc7e8 PM |
39 | #include <linux/mm.h> |
40 | #include <linux/mutex.h> | |
41 | #include <linux/mount.h> | |
42 | #include <linux/pagemap.h> | |
a424316c | 43 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
44 | #include <linux/rcupdate.h> |
45 | #include <linux/sched.h> | |
ddbcc7e8 | 46 | #include <linux/slab.h> |
ddbcc7e8 | 47 | #include <linux/spinlock.h> |
1ed13287 | 48 | #include <linux/percpu-rwsem.h> |
ddbcc7e8 | 49 | #include <linux/string.h> |
bbcb81d0 | 50 | #include <linux/sort.h> |
81a6a5cd | 51 | #include <linux/kmod.h> |
846c7bb0 BS |
52 | #include <linux/delayacct.h> |
53 | #include <linux/cgroupstats.h> | |
0ac801fe | 54 | #include <linux/hashtable.h> |
096b7fe0 | 55 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 56 | #include <linux/idr.h> |
d1d9fd33 | 57 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 58 | #include <linux/kthread.h> |
776f02fa | 59 | #include <linux/delay.h> |
846c7bb0 | 60 | |
60063497 | 61 | #include <linux/atomic.h> |
ddbcc7e8 | 62 | |
b1a21367 TH |
63 | /* |
64 | * pidlists linger the following amount before being destroyed. The goal | |
65 | * is avoiding frequent destruction in the middle of consecutive read calls | |
66 | * Expiring in the middle is a performance problem not a correctness one. | |
67 | * 1 sec should be enough. | |
68 | */ | |
69 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
70 | ||
8d7e6fb0 TH |
71 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
72 | MAX_CFTYPE_NAME + 2) | |
73 | ||
e25e2cbb TH |
74 | /* |
75 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
76 | * hierarchy must be performed while holding it. | |
77 | * | |
f0d9a5f1 | 78 | * css_set_lock protects task->cgroups pointer, the list of css_set |
0e1d768f | 79 | * objects, and the chain of tasks off each css_set. |
e25e2cbb | 80 | * |
0e1d768f TH |
81 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
82 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 83 | */ |
2219449a TH |
84 | #ifdef CONFIG_PROVE_RCU |
85 | DEFINE_MUTEX(cgroup_mutex); | |
f0d9a5f1 | 86 | DEFINE_SPINLOCK(css_set_lock); |
0e1d768f | 87 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
f0d9a5f1 | 88 | EXPORT_SYMBOL_GPL(css_set_lock); |
2219449a | 89 | #else |
81a6a5cd | 90 | static DEFINE_MUTEX(cgroup_mutex); |
f0d9a5f1 | 91 | static DEFINE_SPINLOCK(css_set_lock); |
2219449a TH |
92 | #endif |
93 | ||
6fa4918d | 94 | /* |
15a4c835 TH |
95 | * Protects cgroup_idr and css_idr so that IDs can be released without |
96 | * grabbing cgroup_mutex. | |
6fa4918d TH |
97 | */ |
98 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
99 | ||
34c06254 TH |
100 | /* |
101 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
102 | * against file removal/re-creation across css hiding. | |
103 | */ | |
104 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
105 | ||
69e943b7 TH |
106 | /* |
107 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
108 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
109 | */ | |
110 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 111 | |
1ed13287 TH |
112 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
113 | ||
8353da1f | 114 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
115 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
116 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 117 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 118 | |
e5fca243 TH |
119 | /* |
120 | * cgroup destruction makes heavy use of work items and there can be a lot | |
121 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
122 | * destruction work items don't end up filling up max_active of system_wq | |
123 | * which may lead to deadlock. | |
124 | */ | |
125 | static struct workqueue_struct *cgroup_destroy_wq; | |
126 | ||
b1a21367 TH |
127 | /* |
128 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
129 | * separate workqueue as flush domain. | |
130 | */ | |
131 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
132 | ||
3ed80a62 | 133 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 134 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 135 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
136 | #include <linux/cgroup_subsys.h> |
137 | }; | |
073219e9 TH |
138 | #undef SUBSYS |
139 | ||
140 | /* array of cgroup subsystem names */ | |
141 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
142 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
143 | #include <linux/cgroup_subsys.h> |
144 | }; | |
073219e9 | 145 | #undef SUBSYS |
ddbcc7e8 | 146 | |
49d1dc4b TH |
147 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
148 | #define SUBSYS(_x) \ | |
149 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
150 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
151 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
152 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
153 | #include <linux/cgroup_subsys.h> | |
154 | #undef SUBSYS | |
155 | ||
156 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
157 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
158 | #include <linux/cgroup_subsys.h> | |
159 | }; | |
160 | #undef SUBSYS | |
161 | ||
162 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
163 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
164 | #include <linux/cgroup_subsys.h> | |
165 | }; | |
166 | #undef SUBSYS | |
167 | ||
ddbcc7e8 | 168 | /* |
3dd06ffa | 169 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
170 | * unattached - it never has more than a single cgroup, and all tasks are |
171 | * part of that cgroup. | |
ddbcc7e8 | 172 | */ |
a2dd4247 | 173 | struct cgroup_root cgrp_dfl_root; |
d0ec4230 | 174 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 175 | |
a2dd4247 TH |
176 | /* |
177 | * The default hierarchy always exists but is hidden until mounted for the | |
178 | * first time. This is for backward compatibility. | |
179 | */ | |
180 | static bool cgrp_dfl_root_visible; | |
ddbcc7e8 | 181 | |
5533e011 | 182 | /* some controllers are not supported in the default hierarchy */ |
8ab456ac | 183 | static unsigned long cgrp_dfl_root_inhibit_ss_mask; |
5533e011 | 184 | |
ddbcc7e8 PM |
185 | /* The list of hierarchy roots */ |
186 | ||
9871bf95 TH |
187 | static LIST_HEAD(cgroup_roots); |
188 | static int cgroup_root_count; | |
ddbcc7e8 | 189 | |
3417ae1f | 190 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 191 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 192 | |
794611a1 | 193 | /* |
0cb51d71 TH |
194 | * Assign a monotonically increasing serial number to csses. It guarantees |
195 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
196 | * Also, as csses are always appended to the parent's ->children list, it | |
197 | * guarantees that sibling csses are always sorted in the ascending serial | |
198 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 199 | */ |
0cb51d71 | 200 | static u64 css_serial_nr_next = 1; |
794611a1 | 201 | |
cb4a3167 AS |
202 | /* |
203 | * These bitmask flags indicate whether tasks in the fork and exit paths have | |
204 | * fork/exit handlers to call. This avoids us having to do extra work in the | |
205 | * fork/exit path to check which subsystems have fork/exit callbacks. | |
ddbcc7e8 | 206 | */ |
cb4a3167 AS |
207 | static unsigned long have_fork_callback __read_mostly; |
208 | static unsigned long have_exit_callback __read_mostly; | |
afcf6c8b | 209 | static unsigned long have_free_callback __read_mostly; |
ddbcc7e8 | 210 | |
7e47682e AS |
211 | /* Ditto for the can_fork callback. */ |
212 | static unsigned long have_canfork_callback __read_mostly; | |
213 | ||
a14c6874 TH |
214 | static struct cftype cgroup_dfl_base_files[]; |
215 | static struct cftype cgroup_legacy_base_files[]; | |
628f7cd4 | 216 | |
3dd06ffa | 217 | static int rebind_subsystems(struct cgroup_root *dst_root, |
8ab456ac | 218 | unsigned long ss_mask); |
ed27b9f7 | 219 | static void css_task_iter_advance(struct css_task_iter *it); |
42809dd4 | 220 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
f63070d3 TH |
221 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
222 | bool visible); | |
9d755d33 | 223 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 224 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
225 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
226 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 227 | bool is_add); |
42809dd4 | 228 | |
fc5ed1e9 TH |
229 | /** |
230 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
231 | * @ssid: subsys ID of interest | |
232 | * | |
233 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
234 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
235 | * is slower static_key_enabled() based test indexed by @ssid. | |
236 | */ | |
237 | static bool cgroup_ssid_enabled(int ssid) | |
238 | { | |
239 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); | |
240 | } | |
241 | ||
9e10a130 TH |
242 | /** |
243 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
244 | * @cgrp: the cgroup of interest | |
245 | * | |
246 | * The default hierarchy is the v2 interface of cgroup and this function | |
247 | * can be used to test whether a cgroup is on the default hierarchy for | |
248 | * cases where a subsystem should behave differnetly depending on the | |
249 | * interface version. | |
250 | * | |
251 | * The set of behaviors which change on the default hierarchy are still | |
252 | * being determined and the mount option is prefixed with __DEVEL__. | |
253 | * | |
254 | * List of changed behaviors: | |
255 | * | |
256 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
257 | * and "name" are disallowed. | |
258 | * | |
259 | * - When mounting an existing superblock, mount options should match. | |
260 | * | |
261 | * - Remount is disallowed. | |
262 | * | |
263 | * - rename(2) is disallowed. | |
264 | * | |
265 | * - "tasks" is removed. Everything should be at process granularity. Use | |
266 | * "cgroup.procs" instead. | |
267 | * | |
268 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
269 | * recycled inbetween reads. | |
270 | * | |
271 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
272 | * notification mechanism will be implemented. | |
273 | * | |
274 | * - "cgroup.clone_children" is removed. | |
275 | * | |
276 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
277 | * and its descendants contain no task; otherwise, 1. The file also | |
278 | * generates kernfs notification which can be monitored through poll and | |
279 | * [di]notify when the value of the file changes. | |
280 | * | |
281 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
282 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
283 | * moved to an ancestor. | |
284 | * | |
285 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
286 | * masks of ancestors. | |
287 | * | |
288 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
289 | * is not created. | |
290 | * | |
291 | * - blkcg: blk-throttle becomes properly hierarchical. | |
292 | * | |
293 | * - debug: disallowed on the default hierarchy. | |
294 | */ | |
295 | static bool cgroup_on_dfl(const struct cgroup *cgrp) | |
296 | { | |
297 | return cgrp->root == &cgrp_dfl_root; | |
298 | } | |
299 | ||
6fa4918d TH |
300 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
301 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
302 | gfp_t gfp_mask) | |
303 | { | |
304 | int ret; | |
305 | ||
306 | idr_preload(gfp_mask); | |
54504e97 | 307 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 308 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 309 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
310 | idr_preload_end(); |
311 | return ret; | |
312 | } | |
313 | ||
314 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
315 | { | |
316 | void *ret; | |
317 | ||
54504e97 | 318 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 319 | ret = idr_replace(idr, ptr, id); |
54504e97 | 320 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
321 | return ret; |
322 | } | |
323 | ||
324 | static void cgroup_idr_remove(struct idr *idr, int id) | |
325 | { | |
54504e97 | 326 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 327 | idr_remove(idr, id); |
54504e97 | 328 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
329 | } |
330 | ||
d51f39b0 TH |
331 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
332 | { | |
333 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
334 | ||
335 | if (parent_css) | |
336 | return container_of(parent_css, struct cgroup, self); | |
337 | return NULL; | |
338 | } | |
339 | ||
95109b62 TH |
340 | /** |
341 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
342 | * @cgrp: the cgroup of interest | |
9d800df1 | 343 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 344 | * |
ca8bdcaf TH |
345 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
346 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
347 | * the caller is responsible for pinning the returned css if it wants to | |
348 | * keep accessing it outside the said locks. This function may return | |
349 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
350 | */ |
351 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 352 | struct cgroup_subsys *ss) |
95109b62 | 353 | { |
ca8bdcaf | 354 | if (ss) |
aec25020 | 355 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 356 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 357 | else |
9d800df1 | 358 | return &cgrp->self; |
95109b62 | 359 | } |
42809dd4 | 360 | |
aec3dfcb TH |
361 | /** |
362 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
363 | * @cgrp: the cgroup of interest | |
9d800df1 | 364 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 365 | * |
d0f702e6 | 366 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
367 | * as the matching css of the nearest ancestor including self which has @ss |
368 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
369 | * function is guaranteed to return non-NULL css. | |
370 | */ | |
371 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
372 | struct cgroup_subsys *ss) | |
373 | { | |
374 | lockdep_assert_held(&cgroup_mutex); | |
375 | ||
376 | if (!ss) | |
9d800df1 | 377 | return &cgrp->self; |
aec3dfcb TH |
378 | |
379 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
380 | return NULL; | |
381 | ||
eeecbd19 TH |
382 | /* |
383 | * This function is used while updating css associations and thus | |
384 | * can't test the csses directly. Use ->child_subsys_mask. | |
385 | */ | |
d51f39b0 TH |
386 | while (cgroup_parent(cgrp) && |
387 | !(cgroup_parent(cgrp)->child_subsys_mask & (1 << ss->id))) | |
388 | cgrp = cgroup_parent(cgrp); | |
aec3dfcb TH |
389 | |
390 | return cgroup_css(cgrp, ss); | |
95109b62 | 391 | } |
42809dd4 | 392 | |
eeecbd19 TH |
393 | /** |
394 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
395 | * @cgrp: the cgroup of interest | |
396 | * @ss: the subsystem of interest | |
397 | * | |
398 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
399 | * defined as the matching css of the nearest ancestor including self which | |
400 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
401 | * the root css is returned, so this function always returns a valid css. | |
402 | * The returned css must be put using css_put(). | |
403 | */ | |
404 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
405 | struct cgroup_subsys *ss) | |
406 | { | |
407 | struct cgroup_subsys_state *css; | |
408 | ||
409 | rcu_read_lock(); | |
410 | ||
411 | do { | |
412 | css = cgroup_css(cgrp, ss); | |
413 | ||
414 | if (css && css_tryget_online(css)) | |
415 | goto out_unlock; | |
416 | cgrp = cgroup_parent(cgrp); | |
417 | } while (cgrp); | |
418 | ||
419 | css = init_css_set.subsys[ss->id]; | |
420 | css_get(css); | |
421 | out_unlock: | |
422 | rcu_read_unlock(); | |
423 | return css; | |
424 | } | |
425 | ||
ddbcc7e8 | 426 | /* convenient tests for these bits */ |
54766d4a | 427 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 428 | { |
184faf32 | 429 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
430 | } |
431 | ||
052c3f3a TH |
432 | static void cgroup_get(struct cgroup *cgrp) |
433 | { | |
434 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
435 | css_get(&cgrp->self); | |
436 | } | |
437 | ||
438 | static bool cgroup_tryget(struct cgroup *cgrp) | |
439 | { | |
440 | return css_tryget(&cgrp->self); | |
441 | } | |
442 | ||
443 | static void cgroup_put(struct cgroup *cgrp) | |
444 | { | |
445 | css_put(&cgrp->self); | |
446 | } | |
447 | ||
b4168640 | 448 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 449 | { |
2bd59d48 | 450 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 451 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
452 | |
453 | /* | |
454 | * This is open and unprotected implementation of cgroup_css(). | |
455 | * seq_css() is only called from a kernfs file operation which has | |
456 | * an active reference on the file. Because all the subsystem | |
457 | * files are drained before a css is disassociated with a cgroup, | |
458 | * the matching css from the cgroup's subsys table is guaranteed to | |
459 | * be and stay valid until the enclosing operation is complete. | |
460 | */ | |
461 | if (cft->ss) | |
462 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
463 | else | |
9d800df1 | 464 | return &cgrp->self; |
59f5296b | 465 | } |
b4168640 | 466 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 467 | |
78574cf9 LZ |
468 | /** |
469 | * cgroup_is_descendant - test ancestry | |
470 | * @cgrp: the cgroup to be tested | |
471 | * @ancestor: possible ancestor of @cgrp | |
472 | * | |
473 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
474 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
475 | * and @ancestor are accessible. | |
476 | */ | |
477 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
478 | { | |
479 | while (cgrp) { | |
480 | if (cgrp == ancestor) | |
481 | return true; | |
d51f39b0 | 482 | cgrp = cgroup_parent(cgrp); |
78574cf9 LZ |
483 | } |
484 | return false; | |
485 | } | |
ddbcc7e8 | 486 | |
e9685a03 | 487 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 488 | { |
bd89aabc | 489 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
490 | } |
491 | ||
1c6727af TH |
492 | /** |
493 | * for_each_css - iterate all css's of a cgroup | |
494 | * @css: the iteration cursor | |
495 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
496 | * @cgrp: the target cgroup to iterate css's of | |
497 | * | |
aec3dfcb | 498 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
499 | */ |
500 | #define for_each_css(css, ssid, cgrp) \ | |
501 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
502 | if (!((css) = rcu_dereference_check( \ | |
503 | (cgrp)->subsys[(ssid)], \ | |
504 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
505 | else | |
506 | ||
aec3dfcb TH |
507 | /** |
508 | * for_each_e_css - iterate all effective css's of a cgroup | |
509 | * @css: the iteration cursor | |
510 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
511 | * @cgrp: the target cgroup to iterate css's of | |
512 | * | |
513 | * Should be called under cgroup_[tree_]mutex. | |
514 | */ | |
515 | #define for_each_e_css(css, ssid, cgrp) \ | |
516 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
517 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
518 | ; \ | |
519 | else | |
520 | ||
30159ec7 | 521 | /** |
3ed80a62 | 522 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 523 | * @ss: the iteration cursor |
780cd8b3 | 524 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 525 | */ |
780cd8b3 | 526 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
527 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
528 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 529 | |
cb4a3167 AS |
530 | /** |
531 | * for_each_subsys_which - filter for_each_subsys with a bitmask | |
532 | * @ss: the iteration cursor | |
533 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
534 | * @ss_maskp: a pointer to the bitmask | |
535 | * | |
536 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
537 | * mask is set to 1. | |
538 | */ | |
539 | #define for_each_subsys_which(ss, ssid, ss_maskp) \ | |
540 | if (!CGROUP_SUBSYS_COUNT) /* to avoid spurious gcc warning */ \ | |
4a705c5c | 541 | (ssid) = 0; \ |
cb4a3167 AS |
542 | else \ |
543 | for_each_set_bit(ssid, ss_maskp, CGROUP_SUBSYS_COUNT) \ | |
544 | if (((ss) = cgroup_subsys[ssid]) && false) \ | |
545 | break; \ | |
546 | else | |
547 | ||
985ed670 TH |
548 | /* iterate across the hierarchies */ |
549 | #define for_each_root(root) \ | |
5549c497 | 550 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 551 | |
f8f22e53 TH |
552 | /* iterate over child cgrps, lock should be held throughout iteration */ |
553 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 554 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 555 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
556 | cgroup_is_dead(child); })) \ |
557 | ; \ | |
558 | else | |
7ae1bad9 | 559 | |
81a6a5cd | 560 | static void cgroup_release_agent(struct work_struct *work); |
bd89aabc | 561 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 562 | |
69d0206c TH |
563 | /* |
564 | * A cgroup can be associated with multiple css_sets as different tasks may | |
565 | * belong to different cgroups on different hierarchies. In the other | |
566 | * direction, a css_set is naturally associated with multiple cgroups. | |
567 | * This M:N relationship is represented by the following link structure | |
568 | * which exists for each association and allows traversing the associations | |
569 | * from both sides. | |
570 | */ | |
571 | struct cgrp_cset_link { | |
572 | /* the cgroup and css_set this link associates */ | |
573 | struct cgroup *cgrp; | |
574 | struct css_set *cset; | |
575 | ||
576 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
577 | struct list_head cset_link; | |
578 | ||
579 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
580 | struct list_head cgrp_link; | |
817929ec PM |
581 | }; |
582 | ||
172a2c06 TH |
583 | /* |
584 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
585 | * hierarchies being mounted. It contains a pointer to the root state |
586 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
587 | * reference-counted, to improve performance when child cgroups | |
588 | * haven't been created. | |
589 | */ | |
5024ae29 | 590 | struct css_set init_css_set = { |
172a2c06 TH |
591 | .refcount = ATOMIC_INIT(1), |
592 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
593 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
594 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
595 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
596 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
ed27b9f7 | 597 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
172a2c06 | 598 | }; |
817929ec | 599 | |
172a2c06 | 600 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 601 | |
0de0942d TH |
602 | /** |
603 | * css_set_populated - does a css_set contain any tasks? | |
604 | * @cset: target css_set | |
605 | */ | |
606 | static bool css_set_populated(struct css_set *cset) | |
607 | { | |
f0d9a5f1 | 608 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
609 | |
610 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
611 | } | |
612 | ||
842b597e TH |
613 | /** |
614 | * cgroup_update_populated - updated populated count of a cgroup | |
615 | * @cgrp: the target cgroup | |
616 | * @populated: inc or dec populated count | |
617 | * | |
0de0942d TH |
618 | * One of the css_sets associated with @cgrp is either getting its first |
619 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The | |
620 | * count is propagated towards root so that a given cgroup's populated_cnt | |
621 | * is zero iff the cgroup and all its descendants don't contain any tasks. | |
842b597e TH |
622 | * |
623 | * @cgrp's interface file "cgroup.populated" is zero if | |
624 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
625 | * changes from or to zero, userland is notified that the content of the | |
626 | * interface file has changed. This can be used to detect when @cgrp and | |
627 | * its descendants become populated or empty. | |
628 | */ | |
629 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
630 | { | |
f0d9a5f1 | 631 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
632 | |
633 | do { | |
634 | bool trigger; | |
635 | ||
636 | if (populated) | |
637 | trigger = !cgrp->populated_cnt++; | |
638 | else | |
639 | trigger = !--cgrp->populated_cnt; | |
640 | ||
641 | if (!trigger) | |
642 | break; | |
643 | ||
ad2ed2b3 | 644 | check_for_release(cgrp); |
6f60eade TH |
645 | cgroup_file_notify(&cgrp->events_file); |
646 | ||
d51f39b0 | 647 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
648 | } while (cgrp); |
649 | } | |
650 | ||
0de0942d TH |
651 | /** |
652 | * css_set_update_populated - update populated state of a css_set | |
653 | * @cset: target css_set | |
654 | * @populated: whether @cset is populated or depopulated | |
655 | * | |
656 | * @cset is either getting the first task or losing the last. Update the | |
657 | * ->populated_cnt of all associated cgroups accordingly. | |
658 | */ | |
659 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
660 | { | |
661 | struct cgrp_cset_link *link; | |
662 | ||
f0d9a5f1 | 663 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
664 | |
665 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
666 | cgroup_update_populated(link->cgrp, populated); | |
667 | } | |
668 | ||
f6d7d049 TH |
669 | /** |
670 | * css_set_move_task - move a task from one css_set to another | |
671 | * @task: task being moved | |
672 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
673 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
674 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
675 | * | |
676 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
677 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
678 | * instead of moved, @to_cset can be NULL. | |
679 | * | |
ed27b9f7 TH |
680 | * This function automatically handles populated_cnt updates and |
681 | * css_task_iter adjustments but the caller is responsible for managing | |
682 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
683 | */ |
684 | static void css_set_move_task(struct task_struct *task, | |
685 | struct css_set *from_cset, struct css_set *to_cset, | |
686 | bool use_mg_tasks) | |
687 | { | |
f0d9a5f1 | 688 | lockdep_assert_held(&css_set_lock); |
f6d7d049 TH |
689 | |
690 | if (from_cset) { | |
ed27b9f7 TH |
691 | struct css_task_iter *it, *pos; |
692 | ||
f6d7d049 | 693 | WARN_ON_ONCE(list_empty(&task->cg_list)); |
ed27b9f7 TH |
694 | |
695 | /* | |
696 | * @task is leaving, advance task iterators which are | |
697 | * pointing to it so that they can resume at the next | |
698 | * position. Advancing an iterator might remove it from | |
699 | * the list, use safe walk. See css_task_iter_advance*() | |
700 | * for details. | |
701 | */ | |
702 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, | |
703 | iters_node) | |
704 | if (it->task_pos == &task->cg_list) | |
705 | css_task_iter_advance(it); | |
706 | ||
f6d7d049 TH |
707 | list_del_init(&task->cg_list); |
708 | if (!css_set_populated(from_cset)) | |
709 | css_set_update_populated(from_cset, false); | |
710 | } else { | |
711 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
712 | } | |
713 | ||
714 | if (to_cset) { | |
715 | /* | |
716 | * We are synchronized through cgroup_threadgroup_rwsem | |
717 | * against PF_EXITING setting such that we can't race | |
718 | * against cgroup_exit() changing the css_set to | |
719 | * init_css_set and dropping the old one. | |
720 | */ | |
721 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
722 | ||
723 | if (!css_set_populated(to_cset)) | |
724 | css_set_update_populated(to_cset, true); | |
725 | rcu_assign_pointer(task->cgroups, to_cset); | |
726 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : | |
727 | &to_cset->tasks); | |
728 | } | |
729 | } | |
730 | ||
7717f7ba PM |
731 | /* |
732 | * hash table for cgroup groups. This improves the performance to find | |
733 | * an existing css_set. This hash doesn't (currently) take into | |
734 | * account cgroups in empty hierarchies. | |
735 | */ | |
472b1053 | 736 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 737 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 738 | |
0ac801fe | 739 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 740 | { |
0ac801fe | 741 | unsigned long key = 0UL; |
30159ec7 TH |
742 | struct cgroup_subsys *ss; |
743 | int i; | |
472b1053 | 744 | |
30159ec7 | 745 | for_each_subsys(ss, i) |
0ac801fe LZ |
746 | key += (unsigned long)css[i]; |
747 | key = (key >> 16) ^ key; | |
472b1053 | 748 | |
0ac801fe | 749 | return key; |
472b1053 LZ |
750 | } |
751 | ||
a25eb52e | 752 | static void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 753 | { |
69d0206c | 754 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
755 | struct cgroup_subsys *ss; |
756 | int ssid; | |
5abb8855 | 757 | |
f0d9a5f1 | 758 | lockdep_assert_held(&css_set_lock); |
89c5509b TH |
759 | |
760 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 761 | return; |
81a6a5cd | 762 | |
2c6ab6d2 | 763 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
2d8f243a TH |
764 | for_each_subsys(ss, ssid) |
765 | list_del(&cset->e_cset_node[ssid]); | |
5abb8855 | 766 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
767 | css_set_count--; |
768 | ||
69d0206c | 769 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
770 | list_del(&link->cset_link); |
771 | list_del(&link->cgrp_link); | |
2ceb231b TH |
772 | if (cgroup_parent(link->cgrp)) |
773 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 774 | kfree(link); |
81a6a5cd | 775 | } |
2c6ab6d2 | 776 | |
5abb8855 | 777 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
778 | } |
779 | ||
a25eb52e | 780 | static void put_css_set(struct css_set *cset) |
89c5509b TH |
781 | { |
782 | /* | |
783 | * Ensure that the refcount doesn't hit zero while any readers | |
784 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
785 | * rwlock | |
786 | */ | |
787 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
788 | return; | |
789 | ||
f0d9a5f1 | 790 | spin_lock_bh(&css_set_lock); |
a25eb52e | 791 | put_css_set_locked(cset); |
f0d9a5f1 | 792 | spin_unlock_bh(&css_set_lock); |
89c5509b TH |
793 | } |
794 | ||
817929ec PM |
795 | /* |
796 | * refcounted get/put for css_set objects | |
797 | */ | |
5abb8855 | 798 | static inline void get_css_set(struct css_set *cset) |
817929ec | 799 | { |
5abb8855 | 800 | atomic_inc(&cset->refcount); |
817929ec PM |
801 | } |
802 | ||
b326f9d0 | 803 | /** |
7717f7ba | 804 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
805 | * @cset: candidate css_set being tested |
806 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
807 | * @new_cgrp: cgroup that's being entered by the task |
808 | * @template: desired set of css pointers in css_set (pre-calculated) | |
809 | * | |
6f4b7e63 | 810 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
811 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
812 | */ | |
5abb8855 TH |
813 | static bool compare_css_sets(struct css_set *cset, |
814 | struct css_set *old_cset, | |
7717f7ba PM |
815 | struct cgroup *new_cgrp, |
816 | struct cgroup_subsys_state *template[]) | |
817 | { | |
818 | struct list_head *l1, *l2; | |
819 | ||
aec3dfcb TH |
820 | /* |
821 | * On the default hierarchy, there can be csets which are | |
822 | * associated with the same set of cgroups but different csses. | |
823 | * Let's first ensure that csses match. | |
824 | */ | |
825 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 826 | return false; |
7717f7ba PM |
827 | |
828 | /* | |
829 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
830 | * different cgroups in hierarchies. As different cgroups may |
831 | * share the same effective css, this comparison is always | |
832 | * necessary. | |
7717f7ba | 833 | */ |
69d0206c TH |
834 | l1 = &cset->cgrp_links; |
835 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 836 | while (1) { |
69d0206c | 837 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 838 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
839 | |
840 | l1 = l1->next; | |
841 | l2 = l2->next; | |
842 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
843 | if (l1 == &cset->cgrp_links) { |
844 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
845 | break; |
846 | } else { | |
69d0206c | 847 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
848 | } |
849 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
850 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
851 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
852 | cgrp1 = link1->cgrp; | |
853 | cgrp2 = link2->cgrp; | |
7717f7ba | 854 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 855 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
856 | |
857 | /* | |
858 | * If this hierarchy is the hierarchy of the cgroup | |
859 | * that's changing, then we need to check that this | |
860 | * css_set points to the new cgroup; if it's any other | |
861 | * hierarchy, then this css_set should point to the | |
862 | * same cgroup as the old css_set. | |
863 | */ | |
5abb8855 TH |
864 | if (cgrp1->root == new_cgrp->root) { |
865 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
866 | return false; |
867 | } else { | |
5abb8855 | 868 | if (cgrp1 != cgrp2) |
7717f7ba PM |
869 | return false; |
870 | } | |
871 | } | |
872 | return true; | |
873 | } | |
874 | ||
b326f9d0 TH |
875 | /** |
876 | * find_existing_css_set - init css array and find the matching css_set | |
877 | * @old_cset: the css_set that we're using before the cgroup transition | |
878 | * @cgrp: the cgroup that we're moving into | |
879 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 880 | */ |
5abb8855 TH |
881 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
882 | struct cgroup *cgrp, | |
883 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 884 | { |
3dd06ffa | 885 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 886 | struct cgroup_subsys *ss; |
5abb8855 | 887 | struct css_set *cset; |
0ac801fe | 888 | unsigned long key; |
b326f9d0 | 889 | int i; |
817929ec | 890 | |
aae8aab4 BB |
891 | /* |
892 | * Build the set of subsystem state objects that we want to see in the | |
893 | * new css_set. while subsystems can change globally, the entries here | |
894 | * won't change, so no need for locking. | |
895 | */ | |
30159ec7 | 896 | for_each_subsys(ss, i) { |
f392e51c | 897 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
898 | /* |
899 | * @ss is in this hierarchy, so we want the | |
900 | * effective css from @cgrp. | |
901 | */ | |
902 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 903 | } else { |
aec3dfcb TH |
904 | /* |
905 | * @ss is not in this hierarchy, so we don't want | |
906 | * to change the css. | |
907 | */ | |
5abb8855 | 908 | template[i] = old_cset->subsys[i]; |
817929ec PM |
909 | } |
910 | } | |
911 | ||
0ac801fe | 912 | key = css_set_hash(template); |
5abb8855 TH |
913 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
914 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
915 | continue; |
916 | ||
917 | /* This css_set matches what we need */ | |
5abb8855 | 918 | return cset; |
472b1053 | 919 | } |
817929ec PM |
920 | |
921 | /* No existing cgroup group matched */ | |
922 | return NULL; | |
923 | } | |
924 | ||
69d0206c | 925 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 926 | { |
69d0206c | 927 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 928 | |
69d0206c TH |
929 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
930 | list_del(&link->cset_link); | |
36553434 LZ |
931 | kfree(link); |
932 | } | |
933 | } | |
934 | ||
69d0206c TH |
935 | /** |
936 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
937 | * @count: the number of links to allocate | |
938 | * @tmp_links: list_head the allocated links are put on | |
939 | * | |
940 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
941 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 942 | */ |
69d0206c | 943 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 944 | { |
69d0206c | 945 | struct cgrp_cset_link *link; |
817929ec | 946 | int i; |
69d0206c TH |
947 | |
948 | INIT_LIST_HEAD(tmp_links); | |
949 | ||
817929ec | 950 | for (i = 0; i < count; i++) { |
f4f4be2b | 951 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 952 | if (!link) { |
69d0206c | 953 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
954 | return -ENOMEM; |
955 | } | |
69d0206c | 956 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
957 | } |
958 | return 0; | |
959 | } | |
960 | ||
c12f65d4 LZ |
961 | /** |
962 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 963 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 964 | * @cset: the css_set to be linked |
c12f65d4 LZ |
965 | * @cgrp: the destination cgroup |
966 | */ | |
69d0206c TH |
967 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
968 | struct cgroup *cgrp) | |
c12f65d4 | 969 | { |
69d0206c | 970 | struct cgrp_cset_link *link; |
c12f65d4 | 971 | |
69d0206c | 972 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
973 | |
974 | if (cgroup_on_dfl(cgrp)) | |
975 | cset->dfl_cgrp = cgrp; | |
976 | ||
69d0206c TH |
977 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
978 | link->cset = cset; | |
7717f7ba | 979 | link->cgrp = cgrp; |
842b597e | 980 | |
7717f7ba | 981 | /* |
389b9c1b TH |
982 | * Always add links to the tail of the lists so that the lists are |
983 | * in choronological order. | |
7717f7ba | 984 | */ |
389b9c1b | 985 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 986 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
987 | |
988 | if (cgroup_parent(cgrp)) | |
989 | cgroup_get(cgrp); | |
c12f65d4 LZ |
990 | } |
991 | ||
b326f9d0 TH |
992 | /** |
993 | * find_css_set - return a new css_set with one cgroup updated | |
994 | * @old_cset: the baseline css_set | |
995 | * @cgrp: the cgroup to be updated | |
996 | * | |
997 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
998 | * substituted into the appropriate hierarchy. | |
817929ec | 999 | */ |
5abb8855 TH |
1000 | static struct css_set *find_css_set(struct css_set *old_cset, |
1001 | struct cgroup *cgrp) | |
817929ec | 1002 | { |
b326f9d0 | 1003 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 1004 | struct css_set *cset; |
69d0206c TH |
1005 | struct list_head tmp_links; |
1006 | struct cgrp_cset_link *link; | |
2d8f243a | 1007 | struct cgroup_subsys *ss; |
0ac801fe | 1008 | unsigned long key; |
2d8f243a | 1009 | int ssid; |
472b1053 | 1010 | |
b326f9d0 TH |
1011 | lockdep_assert_held(&cgroup_mutex); |
1012 | ||
817929ec PM |
1013 | /* First see if we already have a cgroup group that matches |
1014 | * the desired set */ | |
f0d9a5f1 | 1015 | spin_lock_bh(&css_set_lock); |
5abb8855 TH |
1016 | cset = find_existing_css_set(old_cset, cgrp, template); |
1017 | if (cset) | |
1018 | get_css_set(cset); | |
f0d9a5f1 | 1019 | spin_unlock_bh(&css_set_lock); |
817929ec | 1020 | |
5abb8855 TH |
1021 | if (cset) |
1022 | return cset; | |
817929ec | 1023 | |
f4f4be2b | 1024 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1025 | if (!cset) |
817929ec PM |
1026 | return NULL; |
1027 | ||
69d0206c | 1028 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1029 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1030 | kfree(cset); |
817929ec PM |
1031 | return NULL; |
1032 | } | |
1033 | ||
5abb8855 | 1034 | atomic_set(&cset->refcount, 1); |
69d0206c | 1035 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 1036 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1037 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 1038 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 1039 | INIT_LIST_HEAD(&cset->mg_node); |
ed27b9f7 | 1040 | INIT_LIST_HEAD(&cset->task_iters); |
5abb8855 | 1041 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
1042 | |
1043 | /* Copy the set of subsystem state objects generated in | |
1044 | * find_existing_css_set() */ | |
5abb8855 | 1045 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1046 | |
f0d9a5f1 | 1047 | spin_lock_bh(&css_set_lock); |
817929ec | 1048 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1049 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1050 | struct cgroup *c = link->cgrp; |
69d0206c | 1051 | |
7717f7ba PM |
1052 | if (c->root == cgrp->root) |
1053 | c = cgrp; | |
69d0206c | 1054 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1055 | } |
817929ec | 1056 | |
69d0206c | 1057 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1058 | |
817929ec | 1059 | css_set_count++; |
472b1053 | 1060 | |
2d8f243a | 1061 | /* Add @cset to the hash table */ |
5abb8855 TH |
1062 | key = css_set_hash(cset->subsys); |
1063 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1064 | |
2d8f243a TH |
1065 | for_each_subsys(ss, ssid) |
1066 | list_add_tail(&cset->e_cset_node[ssid], | |
1067 | &cset->subsys[ssid]->cgroup->e_csets[ssid]); | |
1068 | ||
f0d9a5f1 | 1069 | spin_unlock_bh(&css_set_lock); |
817929ec | 1070 | |
5abb8855 | 1071 | return cset; |
b4f48b63 PM |
1072 | } |
1073 | ||
3dd06ffa | 1074 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1075 | { |
3dd06ffa | 1076 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1077 | |
3dd06ffa | 1078 | return root_cgrp->root; |
2bd59d48 TH |
1079 | } |
1080 | ||
3dd06ffa | 1081 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1082 | { |
1083 | int id; | |
1084 | ||
1085 | lockdep_assert_held(&cgroup_mutex); | |
1086 | ||
985ed670 | 1087 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1088 | if (id < 0) |
1089 | return id; | |
1090 | ||
1091 | root->hierarchy_id = id; | |
1092 | return 0; | |
1093 | } | |
1094 | ||
3dd06ffa | 1095 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1096 | { |
1097 | lockdep_assert_held(&cgroup_mutex); | |
1098 | ||
1099 | if (root->hierarchy_id) { | |
1100 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
1101 | root->hierarchy_id = 0; | |
1102 | } | |
1103 | } | |
1104 | ||
3dd06ffa | 1105 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1106 | { |
1107 | if (root) { | |
d0f702e6 | 1108 | /* hierarchy ID should already have been released */ |
f2e85d57 TH |
1109 | WARN_ON_ONCE(root->hierarchy_id); |
1110 | ||
1111 | idr_destroy(&root->cgroup_idr); | |
1112 | kfree(root); | |
1113 | } | |
1114 | } | |
1115 | ||
3dd06ffa | 1116 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1117 | { |
3dd06ffa | 1118 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1119 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1120 | |
2bd59d48 | 1121 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 1122 | |
776f02fa | 1123 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1124 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1125 | |
f2e85d57 | 1126 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 1127 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 1128 | |
7717f7ba | 1129 | /* |
f2e85d57 TH |
1130 | * Release all the links from cset_links to this hierarchy's |
1131 | * root cgroup | |
7717f7ba | 1132 | */ |
f0d9a5f1 | 1133 | spin_lock_bh(&css_set_lock); |
f2e85d57 TH |
1134 | |
1135 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1136 | list_del(&link->cset_link); | |
1137 | list_del(&link->cgrp_link); | |
1138 | kfree(link); | |
1139 | } | |
f0d9a5f1 TH |
1140 | |
1141 | spin_unlock_bh(&css_set_lock); | |
f2e85d57 TH |
1142 | |
1143 | if (!list_empty(&root->root_list)) { | |
1144 | list_del(&root->root_list); | |
1145 | cgroup_root_count--; | |
1146 | } | |
1147 | ||
1148 | cgroup_exit_root_id(root); | |
1149 | ||
1150 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1151 | |
2bd59d48 | 1152 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1153 | cgroup_free_root(root); |
1154 | } | |
1155 | ||
ceb6a081 TH |
1156 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1157 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1158 | struct cgroup_root *root) |
7717f7ba | 1159 | { |
7717f7ba PM |
1160 | struct cgroup *res = NULL; |
1161 | ||
96d365e0 | 1162 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1163 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1164 | |
5abb8855 | 1165 | if (cset == &init_css_set) { |
3dd06ffa | 1166 | res = &root->cgrp; |
7717f7ba | 1167 | } else { |
69d0206c TH |
1168 | struct cgrp_cset_link *link; |
1169 | ||
1170 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1171 | struct cgroup *c = link->cgrp; |
69d0206c | 1172 | |
7717f7ba PM |
1173 | if (c->root == root) { |
1174 | res = c; | |
1175 | break; | |
1176 | } | |
1177 | } | |
1178 | } | |
96d365e0 | 1179 | |
7717f7ba PM |
1180 | BUG_ON(!res); |
1181 | return res; | |
1182 | } | |
1183 | ||
ddbcc7e8 | 1184 | /* |
ceb6a081 | 1185 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1186 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 TH |
1187 | */ |
1188 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 1189 | struct cgroup_root *root) |
ceb6a081 TH |
1190 | { |
1191 | /* | |
1192 | * No need to lock the task - since we hold cgroup_mutex the | |
1193 | * task can't change groups, so the only thing that can happen | |
1194 | * is that it exits and its css is set back to init_css_set. | |
1195 | */ | |
1196 | return cset_cgroup_from_root(task_css_set(task), root); | |
1197 | } | |
1198 | ||
ddbcc7e8 | 1199 | /* |
ddbcc7e8 PM |
1200 | * A task must hold cgroup_mutex to modify cgroups. |
1201 | * | |
1202 | * Any task can increment and decrement the count field without lock. | |
1203 | * So in general, code holding cgroup_mutex can't rely on the count | |
1204 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1205 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1206 | * means that no tasks are currently attached, therefore there is no |
1207 | * way a task attached to that cgroup can fork (the other way to | |
1208 | * increment the count). So code holding cgroup_mutex can safely | |
1209 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1210 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1211 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1212 | * needs that mutex. | |
1213 | * | |
ddbcc7e8 PM |
1214 | * A cgroup can only be deleted if both its 'count' of using tasks |
1215 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1216 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1217 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1218 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1219 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1220 | * |
1221 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1222 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1223 | */ |
1224 | ||
2bd59d48 | 1225 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1226 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1227 | |
8d7e6fb0 TH |
1228 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1229 | char *buf) | |
ddbcc7e8 | 1230 | { |
3e1d2eed TH |
1231 | struct cgroup_subsys *ss = cft->ss; |
1232 | ||
8d7e6fb0 TH |
1233 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1234 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1235 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1236 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1237 | cft->name); | |
8d7e6fb0 TH |
1238 | else |
1239 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1240 | return buf; | |
ddbcc7e8 PM |
1241 | } |
1242 | ||
f2e85d57 TH |
1243 | /** |
1244 | * cgroup_file_mode - deduce file mode of a control file | |
1245 | * @cft: the control file in question | |
1246 | * | |
7dbdb199 | 1247 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1248 | */ |
1249 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1250 | { |
f2e85d57 | 1251 | umode_t mode = 0; |
65dff759 | 1252 | |
f2e85d57 TH |
1253 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1254 | mode |= S_IRUGO; | |
1255 | ||
7dbdb199 TH |
1256 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1257 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1258 | mode |= S_IWUGO; | |
1259 | else | |
1260 | mode |= S_IWUSR; | |
1261 | } | |
f2e85d57 TH |
1262 | |
1263 | return mode; | |
65dff759 LZ |
1264 | } |
1265 | ||
af0ba678 | 1266 | /** |
0f060deb | 1267 | * cgroup_calc_child_subsys_mask - calculate child_subsys_mask |
af0ba678 | 1268 | * @cgrp: the target cgroup |
0f060deb | 1269 | * @subtree_control: the new subtree_control mask to consider |
af0ba678 TH |
1270 | * |
1271 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1272 | * enabled together through its ->depends_on mask. In such cases, more | |
1273 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1274 | * | |
0f060deb TH |
1275 | * This function calculates which subsystems need to be enabled if |
1276 | * @subtree_control is to be applied to @cgrp. The returned mask is always | |
1277 | * a superset of @subtree_control and follows the usual hierarchy rules. | |
af0ba678 | 1278 | */ |
8ab456ac AS |
1279 | static unsigned long cgroup_calc_child_subsys_mask(struct cgroup *cgrp, |
1280 | unsigned long subtree_control) | |
667c2491 | 1281 | { |
af0ba678 | 1282 | struct cgroup *parent = cgroup_parent(cgrp); |
8ab456ac | 1283 | unsigned long cur_ss_mask = subtree_control; |
af0ba678 TH |
1284 | struct cgroup_subsys *ss; |
1285 | int ssid; | |
1286 | ||
1287 | lockdep_assert_held(&cgroup_mutex); | |
1288 | ||
0f060deb TH |
1289 | if (!cgroup_on_dfl(cgrp)) |
1290 | return cur_ss_mask; | |
af0ba678 TH |
1291 | |
1292 | while (true) { | |
8ab456ac | 1293 | unsigned long new_ss_mask = cur_ss_mask; |
af0ba678 | 1294 | |
a966a4ed AS |
1295 | for_each_subsys_which(ss, ssid, &cur_ss_mask) |
1296 | new_ss_mask |= ss->depends_on; | |
af0ba678 TH |
1297 | |
1298 | /* | |
1299 | * Mask out subsystems which aren't available. This can | |
1300 | * happen only if some depended-upon subsystems were bound | |
1301 | * to non-default hierarchies. | |
1302 | */ | |
1303 | if (parent) | |
1304 | new_ss_mask &= parent->child_subsys_mask; | |
1305 | else | |
1306 | new_ss_mask &= cgrp->root->subsys_mask; | |
1307 | ||
1308 | if (new_ss_mask == cur_ss_mask) | |
1309 | break; | |
1310 | cur_ss_mask = new_ss_mask; | |
1311 | } | |
1312 | ||
0f060deb TH |
1313 | return cur_ss_mask; |
1314 | } | |
1315 | ||
1316 | /** | |
1317 | * cgroup_refresh_child_subsys_mask - update child_subsys_mask | |
1318 | * @cgrp: the target cgroup | |
1319 | * | |
1320 | * Update @cgrp->child_subsys_mask according to the current | |
1321 | * @cgrp->subtree_control using cgroup_calc_child_subsys_mask(). | |
1322 | */ | |
1323 | static void cgroup_refresh_child_subsys_mask(struct cgroup *cgrp) | |
1324 | { | |
1325 | cgrp->child_subsys_mask = | |
1326 | cgroup_calc_child_subsys_mask(cgrp, cgrp->subtree_control); | |
667c2491 TH |
1327 | } |
1328 | ||
a9746d8d TH |
1329 | /** |
1330 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1331 | * @kn: the kernfs_node being serviced | |
1332 | * | |
1333 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1334 | * the method finishes if locking succeeded. Note that once this function | |
1335 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1336 | * inaccessible any time. If the caller intends to continue to access the | |
1337 | * cgroup, it should pin it before invoking this function. | |
1338 | */ | |
1339 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1340 | { |
a9746d8d TH |
1341 | struct cgroup *cgrp; |
1342 | ||
1343 | if (kernfs_type(kn) == KERNFS_DIR) | |
1344 | cgrp = kn->priv; | |
1345 | else | |
1346 | cgrp = kn->parent->priv; | |
1347 | ||
1348 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1349 | |
1350 | kernfs_unbreak_active_protection(kn); | |
1351 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1352 | } |
1353 | ||
a9746d8d TH |
1354 | /** |
1355 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1356 | * @kn: the kernfs_node being serviced | |
1357 | * | |
1358 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1359 | * @kn. It breaks the active protection, performs cgroup locking and | |
1360 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1361 | * alive; otherwise, %NULL. A successful return should be undone by a | |
1362 | * matching cgroup_kn_unlock() invocation. | |
1363 | * | |
1364 | * Any cgroup kernfs method implementation which requires locking the | |
1365 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1366 | * locking under kernfs active protection and allows all kernfs operations | |
1367 | * including self-removal. | |
1368 | */ | |
1369 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn) | |
05ef1d7c | 1370 | { |
a9746d8d TH |
1371 | struct cgroup *cgrp; |
1372 | ||
1373 | if (kernfs_type(kn) == KERNFS_DIR) | |
1374 | cgrp = kn->priv; | |
1375 | else | |
1376 | cgrp = kn->parent->priv; | |
05ef1d7c | 1377 | |
2739d3cc | 1378 | /* |
01f6474c | 1379 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1380 | * active_ref. cgroup liveliness check alone provides enough |
1381 | * protection against removal. Ensure @cgrp stays accessible and | |
1382 | * break the active_ref protection. | |
2739d3cc | 1383 | */ |
aa32362f LZ |
1384 | if (!cgroup_tryget(cgrp)) |
1385 | return NULL; | |
a9746d8d TH |
1386 | kernfs_break_active_protection(kn); |
1387 | ||
2bd59d48 | 1388 | mutex_lock(&cgroup_mutex); |
05ef1d7c | 1389 | |
a9746d8d TH |
1390 | if (!cgroup_is_dead(cgrp)) |
1391 | return cgrp; | |
1392 | ||
1393 | cgroup_kn_unlock(kn); | |
1394 | return NULL; | |
ddbcc7e8 | 1395 | } |
05ef1d7c | 1396 | |
2739d3cc | 1397 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1398 | { |
2bd59d48 | 1399 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1400 | |
01f6474c | 1401 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1402 | |
1403 | if (cft->file_offset) { | |
1404 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1405 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1406 | ||
1407 | spin_lock_irq(&cgroup_file_kn_lock); | |
1408 | cfile->kn = NULL; | |
1409 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1410 | } | |
1411 | ||
2bd59d48 | 1412 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1413 | } |
1414 | ||
13af07df | 1415 | /** |
4df8dc90 TH |
1416 | * css_clear_dir - remove subsys files in a cgroup directory |
1417 | * @css: taget css | |
1418 | * @cgrp_override: specify if target cgroup is different from css->cgroup | |
13af07df | 1419 | */ |
4df8dc90 TH |
1420 | static void css_clear_dir(struct cgroup_subsys_state *css, |
1421 | struct cgroup *cgrp_override) | |
05ef1d7c | 1422 | { |
4df8dc90 TH |
1423 | struct cgroup *cgrp = cgrp_override ?: css->cgroup; |
1424 | struct cftype *cfts; | |
05ef1d7c | 1425 | |
4df8dc90 TH |
1426 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1427 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1428 | } |
1429 | ||
ccdca218 | 1430 | /** |
4df8dc90 TH |
1431 | * css_populate_dir - create subsys files in a cgroup directory |
1432 | * @css: target css | |
1433 | * @cgrp_overried: specify if target cgroup is different from css->cgroup | |
ccdca218 TH |
1434 | * |
1435 | * On failure, no file is added. | |
1436 | */ | |
4df8dc90 TH |
1437 | static int css_populate_dir(struct cgroup_subsys_state *css, |
1438 | struct cgroup *cgrp_override) | |
ccdca218 | 1439 | { |
4df8dc90 TH |
1440 | struct cgroup *cgrp = cgrp_override ?: css->cgroup; |
1441 | struct cftype *cfts, *failed_cfts; | |
1442 | int ret; | |
ccdca218 | 1443 | |
4df8dc90 TH |
1444 | if (!css->ss) { |
1445 | if (cgroup_on_dfl(cgrp)) | |
1446 | cfts = cgroup_dfl_base_files; | |
1447 | else | |
1448 | cfts = cgroup_legacy_base_files; | |
ccdca218 | 1449 | |
4df8dc90 TH |
1450 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1451 | } | |
ccdca218 | 1452 | |
4df8dc90 TH |
1453 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1454 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1455 | if (ret < 0) { | |
1456 | failed_cfts = cfts; | |
1457 | goto err; | |
ccdca218 TH |
1458 | } |
1459 | } | |
1460 | return 0; | |
1461 | err: | |
4df8dc90 TH |
1462 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1463 | if (cfts == failed_cfts) | |
1464 | break; | |
1465 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1466 | } | |
ccdca218 TH |
1467 | return ret; |
1468 | } | |
1469 | ||
8ab456ac AS |
1470 | static int rebind_subsystems(struct cgroup_root *dst_root, |
1471 | unsigned long ss_mask) | |
ddbcc7e8 | 1472 | { |
1ada4838 | 1473 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1474 | struct cgroup_subsys *ss; |
8ab456ac | 1475 | unsigned long tmp_ss_mask; |
2d8f243a | 1476 | int ssid, i, ret; |
ddbcc7e8 | 1477 | |
ace2bee8 | 1478 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1479 | |
a966a4ed | 1480 | for_each_subsys_which(ss, ssid, &ss_mask) { |
7fd8c565 TH |
1481 | /* if @ss has non-root csses attached to it, can't move */ |
1482 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1483 | return -EBUSY; |
1d5be6b2 | 1484 | |
5df36032 | 1485 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1486 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1487 | return -EBUSY; |
ddbcc7e8 PM |
1488 | } |
1489 | ||
5533e011 TH |
1490 | /* skip creating root files on dfl_root for inhibited subsystems */ |
1491 | tmp_ss_mask = ss_mask; | |
1492 | if (dst_root == &cgrp_dfl_root) | |
1493 | tmp_ss_mask &= ~cgrp_dfl_root_inhibit_ss_mask; | |
1494 | ||
4df8dc90 TH |
1495 | for_each_subsys_which(ss, ssid, &tmp_ss_mask) { |
1496 | struct cgroup *scgrp = &ss->root->cgrp; | |
1497 | int tssid; | |
1498 | ||
1499 | ret = css_populate_dir(cgroup_css(scgrp, ss), dcgrp); | |
1500 | if (!ret) | |
1501 | continue; | |
ddbcc7e8 | 1502 | |
a2dd4247 TH |
1503 | /* |
1504 | * Rebinding back to the default root is not allowed to | |
1505 | * fail. Using both default and non-default roots should | |
1506 | * be rare. Moving subsystems back and forth even more so. | |
1507 | * Just warn about it and continue. | |
1508 | */ | |
4df8dc90 TH |
1509 | if (dst_root == &cgrp_dfl_root) { |
1510 | if (cgrp_dfl_root_visible) { | |
1511 | pr_warn("failed to create files (%d) while rebinding 0x%lx to default root\n", | |
1512 | ret, ss_mask); | |
1513 | pr_warn("you may retry by moving them to a different hierarchy and unbinding\n"); | |
1514 | } | |
1515 | continue; | |
a2dd4247 | 1516 | } |
4df8dc90 TH |
1517 | |
1518 | for_each_subsys_which(ss, tssid, &tmp_ss_mask) { | |
1519 | if (tssid == ssid) | |
1520 | break; | |
1521 | css_clear_dir(cgroup_css(scgrp, ss), dcgrp); | |
1522 | } | |
1523 | return ret; | |
5df36032 | 1524 | } |
3126121f TH |
1525 | |
1526 | /* | |
1527 | * Nothing can fail from this point on. Remove files for the | |
1528 | * removed subsystems and rebind each subsystem. | |
1529 | */ | |
a966a4ed | 1530 | for_each_subsys_which(ss, ssid, &ss_mask) { |
1ada4838 TH |
1531 | struct cgroup_root *src_root = ss->root; |
1532 | struct cgroup *scgrp = &src_root->cgrp; | |
1533 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1534 | struct css_set *cset; |
a8a648c4 | 1535 | |
1ada4838 | 1536 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1537 | |
4df8dc90 TH |
1538 | css_clear_dir(css, NULL); |
1539 | ||
1ada4838 TH |
1540 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1541 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1542 | ss->root = dst_root; |
1ada4838 | 1543 | css->cgroup = dcgrp; |
73e80ed8 | 1544 | |
f0d9a5f1 | 1545 | spin_lock_bh(&css_set_lock); |
2d8f243a TH |
1546 | hash_for_each(css_set_table, i, cset, hlist) |
1547 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1548 | &dcgrp->e_csets[ss->id]); |
f0d9a5f1 | 1549 | spin_unlock_bh(&css_set_lock); |
2d8f243a | 1550 | |
f392e51c | 1551 | src_root->subsys_mask &= ~(1 << ssid); |
1ada4838 TH |
1552 | scgrp->subtree_control &= ~(1 << ssid); |
1553 | cgroup_refresh_child_subsys_mask(scgrp); | |
f392e51c | 1554 | |
bd53d617 | 1555 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1556 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1557 | if (dst_root == &cgrp_dfl_root) { |
1558 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1559 | } else { | |
1ada4838 TH |
1560 | dcgrp->subtree_control |= 1 << ssid; |
1561 | cgroup_refresh_child_subsys_mask(dcgrp); | |
49d1dc4b | 1562 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1563 | } |
a8a648c4 | 1564 | |
5df36032 TH |
1565 | if (ss->bind) |
1566 | ss->bind(css); | |
ddbcc7e8 | 1567 | } |
ddbcc7e8 | 1568 | |
1ada4838 | 1569 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1570 | return 0; |
1571 | } | |
1572 | ||
2bd59d48 TH |
1573 | static int cgroup_show_options(struct seq_file *seq, |
1574 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1575 | { |
3dd06ffa | 1576 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1577 | struct cgroup_subsys *ss; |
b85d2040 | 1578 | int ssid; |
ddbcc7e8 | 1579 | |
d98817d4 TH |
1580 | if (root != &cgrp_dfl_root) |
1581 | for_each_subsys(ss, ssid) | |
1582 | if (root->subsys_mask & (1 << ssid)) | |
61e57c0c | 1583 | seq_show_option(seq, ss->legacy_name, NULL); |
93438629 | 1584 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1585 | seq_puts(seq, ",noprefix"); |
93438629 | 1586 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1587 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1588 | |
1589 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1590 | if (strlen(root->release_agent_path)) |
a068acf2 KC |
1591 | seq_show_option(seq, "release_agent", |
1592 | root->release_agent_path); | |
69e943b7 TH |
1593 | spin_unlock(&release_agent_path_lock); |
1594 | ||
3dd06ffa | 1595 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1596 | seq_puts(seq, ",clone_children"); |
c6d57f33 | 1597 | if (strlen(root->name)) |
a068acf2 | 1598 | seq_show_option(seq, "name", root->name); |
ddbcc7e8 PM |
1599 | return 0; |
1600 | } | |
1601 | ||
1602 | struct cgroup_sb_opts { | |
8ab456ac | 1603 | unsigned long subsys_mask; |
69dfa00c | 1604 | unsigned int flags; |
81a6a5cd | 1605 | char *release_agent; |
2260e7fc | 1606 | bool cpuset_clone_children; |
c6d57f33 | 1607 | char *name; |
2c6ab6d2 PM |
1608 | /* User explicitly requested empty subsystem */ |
1609 | bool none; | |
ddbcc7e8 PM |
1610 | }; |
1611 | ||
cf5d5941 | 1612 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1613 | { |
32a8cf23 DL |
1614 | char *token, *o = data; |
1615 | bool all_ss = false, one_ss = false; | |
8ab456ac | 1616 | unsigned long mask = -1UL; |
30159ec7 | 1617 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1618 | int nr_opts = 0; |
30159ec7 | 1619 | int i; |
f9ab5b5b LZ |
1620 | |
1621 | #ifdef CONFIG_CPUSETS | |
69dfa00c | 1622 | mask = ~(1U << cpuset_cgrp_id); |
f9ab5b5b | 1623 | #endif |
ddbcc7e8 | 1624 | |
c6d57f33 | 1625 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1626 | |
1627 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1628 | nr_opts++; |
1629 | ||
ddbcc7e8 PM |
1630 | if (!*token) |
1631 | return -EINVAL; | |
32a8cf23 | 1632 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1633 | /* Explicitly have no subsystems */ |
1634 | opts->none = true; | |
32a8cf23 DL |
1635 | continue; |
1636 | } | |
1637 | if (!strcmp(token, "all")) { | |
1638 | /* Mutually exclusive option 'all' + subsystem name */ | |
1639 | if (one_ss) | |
1640 | return -EINVAL; | |
1641 | all_ss = true; | |
1642 | continue; | |
1643 | } | |
873fe09e TH |
1644 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1645 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1646 | continue; | |
1647 | } | |
32a8cf23 | 1648 | if (!strcmp(token, "noprefix")) { |
93438629 | 1649 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1650 | continue; |
1651 | } | |
1652 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1653 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1654 | continue; |
1655 | } | |
03b1cde6 | 1656 | if (!strcmp(token, "xattr")) { |
93438629 | 1657 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1658 | continue; |
1659 | } | |
32a8cf23 | 1660 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1661 | /* Specifying two release agents is forbidden */ |
1662 | if (opts->release_agent) | |
1663 | return -EINVAL; | |
c6d57f33 | 1664 | opts->release_agent = |
e400c285 | 1665 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1666 | if (!opts->release_agent) |
1667 | return -ENOMEM; | |
32a8cf23 DL |
1668 | continue; |
1669 | } | |
1670 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1671 | const char *name = token + 5; |
1672 | /* Can't specify an empty name */ | |
1673 | if (!strlen(name)) | |
1674 | return -EINVAL; | |
1675 | /* Must match [\w.-]+ */ | |
1676 | for (i = 0; i < strlen(name); i++) { | |
1677 | char c = name[i]; | |
1678 | if (isalnum(c)) | |
1679 | continue; | |
1680 | if ((c == '.') || (c == '-') || (c == '_')) | |
1681 | continue; | |
1682 | return -EINVAL; | |
1683 | } | |
1684 | /* Specifying two names is forbidden */ | |
1685 | if (opts->name) | |
1686 | return -EINVAL; | |
1687 | opts->name = kstrndup(name, | |
e400c285 | 1688 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1689 | GFP_KERNEL); |
1690 | if (!opts->name) | |
1691 | return -ENOMEM; | |
32a8cf23 DL |
1692 | |
1693 | continue; | |
1694 | } | |
1695 | ||
30159ec7 | 1696 | for_each_subsys(ss, i) { |
3e1d2eed | 1697 | if (strcmp(token, ss->legacy_name)) |
32a8cf23 | 1698 | continue; |
fc5ed1e9 | 1699 | if (!cgroup_ssid_enabled(i)) |
32a8cf23 DL |
1700 | continue; |
1701 | ||
1702 | /* Mutually exclusive option 'all' + subsystem name */ | |
1703 | if (all_ss) | |
1704 | return -EINVAL; | |
69dfa00c | 1705 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1706 | one_ss = true; |
1707 | ||
1708 | break; | |
1709 | } | |
1710 | if (i == CGROUP_SUBSYS_COUNT) | |
1711 | return -ENOENT; | |
1712 | } | |
1713 | ||
873fe09e | 1714 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
ed3d261b | 1715 | pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); |
7b9a6ba5 TH |
1716 | if (nr_opts != 1) { |
1717 | pr_err("sane_behavior: no other mount options allowed\n"); | |
873fe09e TH |
1718 | return -EINVAL; |
1719 | } | |
7b9a6ba5 | 1720 | return 0; |
873fe09e TH |
1721 | } |
1722 | ||
7b9a6ba5 TH |
1723 | /* |
1724 | * If the 'all' option was specified select all the subsystems, | |
1725 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1726 | * not specified, let's default to 'all' | |
1727 | */ | |
1728 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1729 | for_each_subsys(ss, i) | |
fc5ed1e9 | 1730 | if (cgroup_ssid_enabled(i)) |
7b9a6ba5 TH |
1731 | opts->subsys_mask |= (1 << i); |
1732 | ||
1733 | /* | |
1734 | * We either have to specify by name or by subsystems. (So all | |
1735 | * empty hierarchies must have a name). | |
1736 | */ | |
1737 | if (!opts->subsys_mask && !opts->name) | |
1738 | return -EINVAL; | |
1739 | ||
f9ab5b5b LZ |
1740 | /* |
1741 | * Option noprefix was introduced just for backward compatibility | |
1742 | * with the old cpuset, so we allow noprefix only if mounting just | |
1743 | * the cpuset subsystem. | |
1744 | */ | |
93438629 | 1745 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1746 | return -EINVAL; |
1747 | ||
2c6ab6d2 | 1748 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1749 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1750 | return -EINVAL; |
1751 | ||
ddbcc7e8 PM |
1752 | return 0; |
1753 | } | |
1754 | ||
2bd59d48 | 1755 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1756 | { |
1757 | int ret = 0; | |
3dd06ffa | 1758 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1759 | struct cgroup_sb_opts opts; |
8ab456ac | 1760 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1761 | |
aa6ec29b TH |
1762 | if (root == &cgrp_dfl_root) { |
1763 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1764 | return -EINVAL; |
1765 | } | |
1766 | ||
ddbcc7e8 PM |
1767 | mutex_lock(&cgroup_mutex); |
1768 | ||
1769 | /* See what subsystems are wanted */ | |
1770 | ret = parse_cgroupfs_options(data, &opts); | |
1771 | if (ret) | |
1772 | goto out_unlock; | |
1773 | ||
f392e51c | 1774 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1775 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1776 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1777 | |
f392e51c TH |
1778 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1779 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1780 | |
cf5d5941 | 1781 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1782 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1783 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1784 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1785 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1786 | ret = -EINVAL; |
1787 | goto out_unlock; | |
1788 | } | |
1789 | ||
f172e67c | 1790 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1791 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1792 | ret = -EBUSY; |
0670e08b | 1793 | goto out_unlock; |
cf5d5941 | 1794 | } |
ddbcc7e8 | 1795 | |
5df36032 | 1796 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1797 | if (ret) |
0670e08b | 1798 | goto out_unlock; |
ddbcc7e8 | 1799 | |
3dd06ffa | 1800 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1801 | |
69e943b7 TH |
1802 | if (opts.release_agent) { |
1803 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1804 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1805 | spin_unlock(&release_agent_path_lock); |
1806 | } | |
ddbcc7e8 | 1807 | out_unlock: |
66bdc9cf | 1808 | kfree(opts.release_agent); |
c6d57f33 | 1809 | kfree(opts.name); |
ddbcc7e8 | 1810 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1811 | return ret; |
1812 | } | |
1813 | ||
afeb0f9f TH |
1814 | /* |
1815 | * To reduce the fork() overhead for systems that are not actually using | |
1816 | * their cgroups capability, we don't maintain the lists running through | |
1817 | * each css_set to its tasks until we see the list actually used - in other | |
1818 | * words after the first mount. | |
1819 | */ | |
1820 | static bool use_task_css_set_links __read_mostly; | |
1821 | ||
1822 | static void cgroup_enable_task_cg_lists(void) | |
1823 | { | |
1824 | struct task_struct *p, *g; | |
1825 | ||
f0d9a5f1 | 1826 | spin_lock_bh(&css_set_lock); |
afeb0f9f TH |
1827 | |
1828 | if (use_task_css_set_links) | |
1829 | goto out_unlock; | |
1830 | ||
1831 | use_task_css_set_links = true; | |
1832 | ||
1833 | /* | |
1834 | * We need tasklist_lock because RCU is not safe against | |
1835 | * while_each_thread(). Besides, a forking task that has passed | |
1836 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1837 | * is not guaranteed to have its child immediately visible in the | |
1838 | * tasklist if we walk through it with RCU. | |
1839 | */ | |
1840 | read_lock(&tasklist_lock); | |
1841 | do_each_thread(g, p) { | |
afeb0f9f TH |
1842 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1843 | task_css_set(p) != &init_css_set); | |
1844 | ||
1845 | /* | |
1846 | * We should check if the process is exiting, otherwise | |
1847 | * it will race with cgroup_exit() in that the list | |
1848 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1849 | * Do it while holding siglock so that we don't end up |
1850 | * racing against cgroup_exit(). | |
afeb0f9f | 1851 | */ |
f153ad11 | 1852 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1853 | if (!(p->flags & PF_EXITING)) { |
1854 | struct css_set *cset = task_css_set(p); | |
1855 | ||
0de0942d TH |
1856 | if (!css_set_populated(cset)) |
1857 | css_set_update_populated(cset, true); | |
389b9c1b | 1858 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab TH |
1859 | get_css_set(cset); |
1860 | } | |
f153ad11 | 1861 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1862 | } while_each_thread(g, p); |
1863 | read_unlock(&tasklist_lock); | |
1864 | out_unlock: | |
f0d9a5f1 | 1865 | spin_unlock_bh(&css_set_lock); |
afeb0f9f | 1866 | } |
ddbcc7e8 | 1867 | |
cc31edce PM |
1868 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1869 | { | |
2d8f243a TH |
1870 | struct cgroup_subsys *ss; |
1871 | int ssid; | |
1872 | ||
d5c419b6 TH |
1873 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1874 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1875 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1876 | INIT_LIST_HEAD(&cgrp->pidlists); |
1877 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1878 | cgrp->self.cgroup = cgrp; |
184faf32 | 1879 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1880 | |
1881 | for_each_subsys(ss, ssid) | |
1882 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1883 | |
1884 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1885 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1886 | } |
c6d57f33 | 1887 | |
3dd06ffa | 1888 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1889 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1890 | { |
3dd06ffa | 1891 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1892 | |
ddbcc7e8 | 1893 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1894 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1895 | cgrp->root = root; |
cc31edce | 1896 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1897 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1898 | |
c6d57f33 PM |
1899 | root->flags = opts->flags; |
1900 | if (opts->release_agent) | |
1901 | strcpy(root->release_agent_path, opts->release_agent); | |
1902 | if (opts->name) | |
1903 | strcpy(root->name, opts->name); | |
2260e7fc | 1904 | if (opts->cpuset_clone_children) |
3dd06ffa | 1905 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1906 | } |
1907 | ||
8ab456ac | 1908 | static int cgroup_setup_root(struct cgroup_root *root, unsigned long ss_mask) |
2c6ab6d2 | 1909 | { |
d427dfeb | 1910 | LIST_HEAD(tmp_links); |
3dd06ffa | 1911 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1912 | struct css_set *cset; |
d427dfeb | 1913 | int i, ret; |
2c6ab6d2 | 1914 | |
d427dfeb | 1915 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1916 | |
cf780b7d | 1917 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1918 | if (ret < 0) |
2bd59d48 | 1919 | goto out; |
d427dfeb | 1920 | root_cgrp->id = ret; |
c6d57f33 | 1921 | |
2aad2a86 TH |
1922 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
1923 | GFP_KERNEL); | |
9d755d33 TH |
1924 | if (ret) |
1925 | goto out; | |
1926 | ||
d427dfeb | 1927 | /* |
f0d9a5f1 | 1928 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb TH |
1929 | * but that's OK - it can only be increased by someone holding |
1930 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1931 | * have some link structures left over | |
1932 | */ | |
1933 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1934 | if (ret) | |
9d755d33 | 1935 | goto cancel_ref; |
ddbcc7e8 | 1936 | |
985ed670 | 1937 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1938 | if (ret) |
9d755d33 | 1939 | goto cancel_ref; |
ddbcc7e8 | 1940 | |
2bd59d48 TH |
1941 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1942 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1943 | root_cgrp); | |
1944 | if (IS_ERR(root->kf_root)) { | |
1945 | ret = PTR_ERR(root->kf_root); | |
1946 | goto exit_root_id; | |
1947 | } | |
1948 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1949 | |
4df8dc90 | 1950 | ret = css_populate_dir(&root_cgrp->self, NULL); |
d427dfeb | 1951 | if (ret) |
2bd59d48 | 1952 | goto destroy_root; |
ddbcc7e8 | 1953 | |
5df36032 | 1954 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1955 | if (ret) |
2bd59d48 | 1956 | goto destroy_root; |
ddbcc7e8 | 1957 | |
d427dfeb TH |
1958 | /* |
1959 | * There must be no failure case after here, since rebinding takes | |
1960 | * care of subsystems' refcounts, which are explicitly dropped in | |
1961 | * the failure exit path. | |
1962 | */ | |
1963 | list_add(&root->root_list, &cgroup_roots); | |
1964 | cgroup_root_count++; | |
0df6a63f | 1965 | |
d427dfeb | 1966 | /* |
3dd06ffa | 1967 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1968 | * objects. |
1969 | */ | |
f0d9a5f1 | 1970 | spin_lock_bh(&css_set_lock); |
0de0942d | 1971 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 1972 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
1973 | if (css_set_populated(cset)) |
1974 | cgroup_update_populated(root_cgrp, true); | |
1975 | } | |
f0d9a5f1 | 1976 | spin_unlock_bh(&css_set_lock); |
ddbcc7e8 | 1977 | |
d5c419b6 | 1978 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1979 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1980 | |
2bd59d48 | 1981 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1982 | ret = 0; |
2bd59d48 | 1983 | goto out; |
d427dfeb | 1984 | |
2bd59d48 TH |
1985 | destroy_root: |
1986 | kernfs_destroy_root(root->kf_root); | |
1987 | root->kf_root = NULL; | |
1988 | exit_root_id: | |
d427dfeb | 1989 | cgroup_exit_root_id(root); |
9d755d33 | 1990 | cancel_ref: |
9a1049da | 1991 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 1992 | out: |
d427dfeb TH |
1993 | free_cgrp_cset_links(&tmp_links); |
1994 | return ret; | |
ddbcc7e8 PM |
1995 | } |
1996 | ||
f7e83571 | 1997 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1998 | int flags, const char *unused_dev_name, |
f7e83571 | 1999 | void *data) |
ddbcc7e8 | 2000 | { |
3a32bd72 | 2001 | struct super_block *pinned_sb = NULL; |
970317aa | 2002 | struct cgroup_subsys *ss; |
3dd06ffa | 2003 | struct cgroup_root *root; |
ddbcc7e8 | 2004 | struct cgroup_sb_opts opts; |
2bd59d48 | 2005 | struct dentry *dentry; |
8e30e2b8 | 2006 | int ret; |
970317aa | 2007 | int i; |
c6b3d5bc | 2008 | bool new_sb; |
ddbcc7e8 | 2009 | |
56fde9e0 TH |
2010 | /* |
2011 | * The first time anyone tries to mount a cgroup, enable the list | |
2012 | * linking each css_set to its tasks and fix up all existing tasks. | |
2013 | */ | |
2014 | if (!use_task_css_set_links) | |
2015 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 2016 | |
aae8aab4 | 2017 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
2018 | |
2019 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 2020 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 2021 | if (ret) |
8e30e2b8 | 2022 | goto out_unlock; |
a015edd2 | 2023 | |
2bd59d48 | 2024 | /* look for a matching existing root */ |
7b9a6ba5 | 2025 | if (opts.flags & CGRP_ROOT_SANE_BEHAVIOR) { |
a2dd4247 TH |
2026 | cgrp_dfl_root_visible = true; |
2027 | root = &cgrp_dfl_root; | |
2028 | cgroup_get(&root->cgrp); | |
2029 | ret = 0; | |
2030 | goto out_unlock; | |
ddbcc7e8 PM |
2031 | } |
2032 | ||
970317aa LZ |
2033 | /* |
2034 | * Destruction of cgroup root is asynchronous, so subsystems may | |
2035 | * still be dying after the previous unmount. Let's drain the | |
2036 | * dying subsystems. We just need to ensure that the ones | |
2037 | * unmounted previously finish dying and don't care about new ones | |
2038 | * starting. Testing ref liveliness is good enough. | |
2039 | */ | |
2040 | for_each_subsys(ss, i) { | |
2041 | if (!(opts.subsys_mask & (1 << i)) || | |
2042 | ss->root == &cgrp_dfl_root) | |
2043 | continue; | |
2044 | ||
2045 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
2046 | mutex_unlock(&cgroup_mutex); | |
2047 | msleep(10); | |
2048 | ret = restart_syscall(); | |
2049 | goto out_free; | |
2050 | } | |
2051 | cgroup_put(&ss->root->cgrp); | |
2052 | } | |
2053 | ||
985ed670 | 2054 | for_each_root(root) { |
2bd59d48 | 2055 | bool name_match = false; |
3126121f | 2056 | |
3dd06ffa | 2057 | if (root == &cgrp_dfl_root) |
985ed670 | 2058 | continue; |
3126121f | 2059 | |
cf5d5941 | 2060 | /* |
2bd59d48 TH |
2061 | * If we asked for a name then it must match. Also, if |
2062 | * name matches but sybsys_mask doesn't, we should fail. | |
2063 | * Remember whether name matched. | |
cf5d5941 | 2064 | */ |
2bd59d48 TH |
2065 | if (opts.name) { |
2066 | if (strcmp(opts.name, root->name)) | |
2067 | continue; | |
2068 | name_match = true; | |
2069 | } | |
ddbcc7e8 | 2070 | |
c6d57f33 | 2071 | /* |
2bd59d48 TH |
2072 | * If we asked for subsystems (or explicitly for no |
2073 | * subsystems) then they must match. | |
c6d57f33 | 2074 | */ |
2bd59d48 | 2075 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 2076 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
2077 | if (!name_match) |
2078 | continue; | |
2079 | ret = -EBUSY; | |
2080 | goto out_unlock; | |
2081 | } | |
873fe09e | 2082 | |
7b9a6ba5 TH |
2083 | if (root->flags ^ opts.flags) |
2084 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 2085 | |
776f02fa | 2086 | /* |
3a32bd72 LZ |
2087 | * We want to reuse @root whose lifetime is governed by its |
2088 | * ->cgrp. Let's check whether @root is alive and keep it | |
2089 | * that way. As cgroup_kill_sb() can happen anytime, we | |
2090 | * want to block it by pinning the sb so that @root doesn't | |
2091 | * get killed before mount is complete. | |
2092 | * | |
2093 | * With the sb pinned, tryget_live can reliably indicate | |
2094 | * whether @root can be reused. If it's being killed, | |
2095 | * drain it. We can use wait_queue for the wait but this | |
2096 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 2097 | */ |
3a32bd72 LZ |
2098 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
2099 | if (IS_ERR(pinned_sb) || | |
2100 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 2101 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
2102 | if (!IS_ERR_OR_NULL(pinned_sb)) |
2103 | deactivate_super(pinned_sb); | |
776f02fa | 2104 | msleep(10); |
a015edd2 TH |
2105 | ret = restart_syscall(); |
2106 | goto out_free; | |
776f02fa | 2107 | } |
ddbcc7e8 | 2108 | |
776f02fa | 2109 | ret = 0; |
2bd59d48 | 2110 | goto out_unlock; |
ddbcc7e8 | 2111 | } |
ddbcc7e8 | 2112 | |
817929ec | 2113 | /* |
172a2c06 TH |
2114 | * No such thing, create a new one. name= matching without subsys |
2115 | * specification is allowed for already existing hierarchies but we | |
2116 | * can't create new one without subsys specification. | |
817929ec | 2117 | */ |
172a2c06 TH |
2118 | if (!opts.subsys_mask && !opts.none) { |
2119 | ret = -EINVAL; | |
2120 | goto out_unlock; | |
817929ec | 2121 | } |
817929ec | 2122 | |
172a2c06 TH |
2123 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
2124 | if (!root) { | |
2125 | ret = -ENOMEM; | |
2bd59d48 | 2126 | goto out_unlock; |
839ec545 | 2127 | } |
e5f6a860 | 2128 | |
172a2c06 TH |
2129 | init_cgroup_root(root, &opts); |
2130 | ||
35585573 | 2131 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
2132 | if (ret) |
2133 | cgroup_free_root(root); | |
fa3ca07e | 2134 | |
8e30e2b8 | 2135 | out_unlock: |
ddbcc7e8 | 2136 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 2137 | out_free: |
c6d57f33 PM |
2138 | kfree(opts.release_agent); |
2139 | kfree(opts.name); | |
03b1cde6 | 2140 | |
2bd59d48 | 2141 | if (ret) |
8e30e2b8 | 2142 | return ERR_PTR(ret); |
2bd59d48 | 2143 | |
c9482a5b JZ |
2144 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
2145 | CGROUP_SUPER_MAGIC, &new_sb); | |
c6b3d5bc | 2146 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 2147 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
2148 | |
2149 | /* | |
2150 | * If @pinned_sb, we're reusing an existing root and holding an | |
2151 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
2152 | */ | |
2153 | if (pinned_sb) { | |
2154 | WARN_ON(new_sb); | |
2155 | deactivate_super(pinned_sb); | |
2156 | } | |
2157 | ||
2bd59d48 TH |
2158 | return dentry; |
2159 | } | |
2160 | ||
2161 | static void cgroup_kill_sb(struct super_block *sb) | |
2162 | { | |
2163 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2164 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 2165 | |
9d755d33 TH |
2166 | /* |
2167 | * If @root doesn't have any mounts or children, start killing it. | |
2168 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2169 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2170 | * |
2171 | * And don't kill the default root. | |
9d755d33 | 2172 | */ |
3c606d35 | 2173 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 2174 | root == &cgrp_dfl_root) |
9d755d33 TH |
2175 | cgroup_put(&root->cgrp); |
2176 | else | |
2177 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
2178 | ||
2bd59d48 | 2179 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
2180 | } |
2181 | ||
2182 | static struct file_system_type cgroup_fs_type = { | |
2183 | .name = "cgroup", | |
f7e83571 | 2184 | .mount = cgroup_mount, |
ddbcc7e8 PM |
2185 | .kill_sb = cgroup_kill_sb, |
2186 | }; | |
2187 | ||
857a2beb | 2188 | /** |
913ffdb5 | 2189 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2190 | * @task: target task |
857a2beb TH |
2191 | * @buf: the buffer to write the path into |
2192 | * @buflen: the length of the buffer | |
2193 | * | |
913ffdb5 TH |
2194 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2195 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2196 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2197 | * cgroup controller callbacks. | |
2198 | * | |
e61734c5 | 2199 | * Return value is the same as kernfs_path(). |
857a2beb | 2200 | */ |
e61734c5 | 2201 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2202 | { |
3dd06ffa | 2203 | struct cgroup_root *root; |
913ffdb5 | 2204 | struct cgroup *cgrp; |
e61734c5 TH |
2205 | int hierarchy_id = 1; |
2206 | char *path = NULL; | |
857a2beb TH |
2207 | |
2208 | mutex_lock(&cgroup_mutex); | |
f0d9a5f1 | 2209 | spin_lock_bh(&css_set_lock); |
857a2beb | 2210 | |
913ffdb5 TH |
2211 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2212 | ||
857a2beb TH |
2213 | if (root) { |
2214 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 2215 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
2216 | } else { |
2217 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
2218 | if (strlcpy(buf, "/", buflen) < buflen) |
2219 | path = buf; | |
857a2beb TH |
2220 | } |
2221 | ||
f0d9a5f1 | 2222 | spin_unlock_bh(&css_set_lock); |
857a2beb | 2223 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 2224 | return path; |
857a2beb | 2225 | } |
913ffdb5 | 2226 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2227 | |
b3dc094e | 2228 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 2229 | struct cgroup_taskset { |
b3dc094e TH |
2230 | /* the src and dst cset list running through cset->mg_node */ |
2231 | struct list_head src_csets; | |
2232 | struct list_head dst_csets; | |
2233 | ||
2234 | /* | |
2235 | * Fields for cgroup_taskset_*() iteration. | |
2236 | * | |
2237 | * Before migration is committed, the target migration tasks are on | |
2238 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2239 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2240 | * or ->dst_csets depending on whether migration is committed. | |
2241 | * | |
2242 | * ->cur_csets and ->cur_task point to the current task position | |
2243 | * during iteration. | |
2244 | */ | |
2245 | struct list_head *csets; | |
2246 | struct css_set *cur_cset; | |
2247 | struct task_struct *cur_task; | |
2f7ee569 TH |
2248 | }; |
2249 | ||
adaae5dc TH |
2250 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ |
2251 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ | |
2252 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ | |
2253 | .csets = &tset.src_csets, \ | |
2254 | } | |
2255 | ||
2256 | /** | |
2257 | * cgroup_taskset_add - try to add a migration target task to a taskset | |
2258 | * @task: target task | |
2259 | * @tset: target taskset | |
2260 | * | |
2261 | * Add @task, which is a migration target, to @tset. This function becomes | |
2262 | * noop if @task doesn't need to be migrated. @task's css_set should have | |
2263 | * been added as a migration source and @task->cg_list will be moved from | |
2264 | * the css_set's tasks list to mg_tasks one. | |
2265 | */ | |
2266 | static void cgroup_taskset_add(struct task_struct *task, | |
2267 | struct cgroup_taskset *tset) | |
2268 | { | |
2269 | struct css_set *cset; | |
2270 | ||
f0d9a5f1 | 2271 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2272 | |
2273 | /* @task either already exited or can't exit until the end */ | |
2274 | if (task->flags & PF_EXITING) | |
2275 | return; | |
2276 | ||
2277 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2278 | if (list_empty(&task->cg_list)) | |
2279 | return; | |
2280 | ||
2281 | cset = task_css_set(task); | |
2282 | if (!cset->mg_src_cgrp) | |
2283 | return; | |
2284 | ||
2285 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
2286 | if (list_empty(&cset->mg_node)) | |
2287 | list_add_tail(&cset->mg_node, &tset->src_csets); | |
2288 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2289 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2290 | &tset->dst_csets); | |
2291 | } | |
2292 | ||
2f7ee569 TH |
2293 | /** |
2294 | * cgroup_taskset_first - reset taskset and return the first task | |
2295 | * @tset: taskset of interest | |
2296 | * | |
2297 | * @tset iteration is initialized and the first task is returned. | |
2298 | */ | |
2299 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
2300 | { | |
b3dc094e TH |
2301 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2302 | tset->cur_task = NULL; | |
2303 | ||
2304 | return cgroup_taskset_next(tset); | |
2f7ee569 | 2305 | } |
2f7ee569 TH |
2306 | |
2307 | /** | |
2308 | * cgroup_taskset_next - iterate to the next task in taskset | |
2309 | * @tset: taskset of interest | |
2310 | * | |
2311 | * Return the next task in @tset. Iteration must have been initialized | |
2312 | * with cgroup_taskset_first(). | |
2313 | */ | |
2314 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
2315 | { | |
b3dc094e TH |
2316 | struct css_set *cset = tset->cur_cset; |
2317 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2318 | |
b3dc094e TH |
2319 | while (&cset->mg_node != tset->csets) { |
2320 | if (!task) | |
2321 | task = list_first_entry(&cset->mg_tasks, | |
2322 | struct task_struct, cg_list); | |
2323 | else | |
2324 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2325 | |
b3dc094e TH |
2326 | if (&task->cg_list != &cset->mg_tasks) { |
2327 | tset->cur_cset = cset; | |
2328 | tset->cur_task = task; | |
2329 | return task; | |
2330 | } | |
2f7ee569 | 2331 | |
b3dc094e TH |
2332 | cset = list_next_entry(cset, mg_node); |
2333 | task = NULL; | |
2334 | } | |
2f7ee569 | 2335 | |
b3dc094e | 2336 | return NULL; |
2f7ee569 | 2337 | } |
2f7ee569 | 2338 | |
adaae5dc TH |
2339 | /** |
2340 | * cgroup_taskset_migrate - migrate a taskset to a cgroup | |
2341 | * @tset: taget taskset | |
2342 | * @dst_cgrp: destination cgroup | |
2343 | * | |
2344 | * Migrate tasks in @tset to @dst_cgrp. This function fails iff one of the | |
2345 | * ->can_attach callbacks fails and guarantees that either all or none of | |
2346 | * the tasks in @tset are migrated. @tset is consumed regardless of | |
2347 | * success. | |
2348 | */ | |
2349 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, | |
2350 | struct cgroup *dst_cgrp) | |
2351 | { | |
2352 | struct cgroup_subsys_state *css, *failed_css = NULL; | |
2353 | struct task_struct *task, *tmp_task; | |
2354 | struct css_set *cset, *tmp_cset; | |
2355 | int i, ret; | |
2356 | ||
2357 | /* methods shouldn't be called if no task is actually migrating */ | |
2358 | if (list_empty(&tset->src_csets)) | |
2359 | return 0; | |
2360 | ||
2361 | /* check that we can legitimately attach to the cgroup */ | |
2362 | for_each_e_css(css, i, dst_cgrp) { | |
2363 | if (css->ss->can_attach) { | |
2364 | ret = css->ss->can_attach(css, tset); | |
2365 | if (ret) { | |
2366 | failed_css = css; | |
2367 | goto out_cancel_attach; | |
2368 | } | |
2369 | } | |
2370 | } | |
2371 | ||
2372 | /* | |
2373 | * Now that we're guaranteed success, proceed to move all tasks to | |
2374 | * the new cgroup. There are no failure cases after here, so this | |
2375 | * is the commit point. | |
2376 | */ | |
f0d9a5f1 | 2377 | spin_lock_bh(&css_set_lock); |
adaae5dc | 2378 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2379 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2380 | struct css_set *from_cset = task_css_set(task); | |
2381 | struct css_set *to_cset = cset->mg_dst_cset; | |
2382 | ||
2383 | get_css_set(to_cset); | |
2384 | css_set_move_task(task, from_cset, to_cset, true); | |
2385 | put_css_set_locked(from_cset); | |
2386 | } | |
adaae5dc | 2387 | } |
f0d9a5f1 | 2388 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2389 | |
2390 | /* | |
2391 | * Migration is committed, all target tasks are now on dst_csets. | |
2392 | * Nothing is sensitive to fork() after this point. Notify | |
2393 | * controllers that migration is complete. | |
2394 | */ | |
2395 | tset->csets = &tset->dst_csets; | |
2396 | ||
2397 | for_each_e_css(css, i, dst_cgrp) | |
2398 | if (css->ss->attach) | |
2399 | css->ss->attach(css, tset); | |
2400 | ||
2401 | ret = 0; | |
2402 | goto out_release_tset; | |
2403 | ||
2404 | out_cancel_attach: | |
2405 | for_each_e_css(css, i, dst_cgrp) { | |
2406 | if (css == failed_css) | |
2407 | break; | |
2408 | if (css->ss->cancel_attach) | |
2409 | css->ss->cancel_attach(css, tset); | |
2410 | } | |
2411 | out_release_tset: | |
f0d9a5f1 | 2412 | spin_lock_bh(&css_set_lock); |
adaae5dc TH |
2413 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2414 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2415 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2416 | list_del_init(&cset->mg_node); | |
2417 | } | |
f0d9a5f1 | 2418 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2419 | return ret; |
2420 | } | |
2421 | ||
a043e3b2 | 2422 | /** |
1958d2d5 TH |
2423 | * cgroup_migrate_finish - cleanup after attach |
2424 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2425 | * |
1958d2d5 TH |
2426 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2427 | * those functions for details. | |
74a1166d | 2428 | */ |
1958d2d5 | 2429 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2430 | { |
1958d2d5 | 2431 | struct css_set *cset, *tmp_cset; |
74a1166d | 2432 | |
1958d2d5 TH |
2433 | lockdep_assert_held(&cgroup_mutex); |
2434 | ||
f0d9a5f1 | 2435 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2436 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { |
2437 | cset->mg_src_cgrp = NULL; | |
2438 | cset->mg_dst_cset = NULL; | |
2439 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2440 | put_css_set_locked(cset); |
1958d2d5 | 2441 | } |
f0d9a5f1 | 2442 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2443 | } |
2444 | ||
2445 | /** | |
2446 | * cgroup_migrate_add_src - add a migration source css_set | |
2447 | * @src_cset: the source css_set to add | |
2448 | * @dst_cgrp: the destination cgroup | |
2449 | * @preloaded_csets: list of preloaded css_sets | |
2450 | * | |
2451 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2452 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2453 | * up by cgroup_migrate_finish(). | |
2454 | * | |
1ed13287 TH |
2455 | * This function may be called without holding cgroup_threadgroup_rwsem |
2456 | * even if the target is a process. Threads may be created and destroyed | |
2457 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2458 | * into play and the preloaded css_sets are guaranteed to cover all | |
2459 | * migrations. | |
1958d2d5 TH |
2460 | */ |
2461 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2462 | struct cgroup *dst_cgrp, | |
2463 | struct list_head *preloaded_csets) | |
2464 | { | |
2465 | struct cgroup *src_cgrp; | |
2466 | ||
2467 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2468 | lockdep_assert_held(&css_set_lock); |
1958d2d5 TH |
2469 | |
2470 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
2471 | ||
1958d2d5 TH |
2472 | if (!list_empty(&src_cset->mg_preload_node)) |
2473 | return; | |
2474 | ||
2475 | WARN_ON(src_cset->mg_src_cgrp); | |
2476 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
2477 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2478 | ||
2479 | src_cset->mg_src_cgrp = src_cgrp; | |
2480 | get_css_set(src_cset); | |
2481 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2482 | } | |
2483 | ||
2484 | /** | |
2485 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 2486 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
2487 | * @preloaded_csets: list of preloaded source css_sets |
2488 | * | |
2489 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
2490 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
2491 | * pins all destination css_sets, links each to its source, and append them |
2492 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
2493 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
2494 | * |
2495 | * This function must be called after cgroup_migrate_add_src() has been | |
2496 | * called on each migration source css_set. After migration is performed | |
2497 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2498 | * @preloaded_csets. | |
2499 | */ | |
2500 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
2501 | struct list_head *preloaded_csets) | |
2502 | { | |
2503 | LIST_HEAD(csets); | |
f817de98 | 2504 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2505 | |
2506 | lockdep_assert_held(&cgroup_mutex); | |
2507 | ||
f8f22e53 TH |
2508 | /* |
2509 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2510 | * with tasks so that child cgroups don't compete against tasks. | |
2511 | */ | |
d51f39b0 | 2512 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) && |
f8f22e53 TH |
2513 | dst_cgrp->child_subsys_mask) |
2514 | return -EBUSY; | |
2515 | ||
1958d2d5 | 2516 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 2517 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2518 | struct css_set *dst_cset; |
2519 | ||
f817de98 TH |
2520 | dst_cset = find_css_set(src_cset, |
2521 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
2522 | if (!dst_cset) |
2523 | goto err; | |
2524 | ||
2525 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2526 | |
2527 | /* | |
2528 | * If src cset equals dst, it's noop. Drop the src. | |
2529 | * cgroup_migrate() will skip the cset too. Note that we | |
2530 | * can't handle src == dst as some nodes are used by both. | |
2531 | */ | |
2532 | if (src_cset == dst_cset) { | |
2533 | src_cset->mg_src_cgrp = NULL; | |
2534 | list_del_init(&src_cset->mg_preload_node); | |
a25eb52e ZL |
2535 | put_css_set(src_cset); |
2536 | put_css_set(dst_cset); | |
f817de98 TH |
2537 | continue; |
2538 | } | |
2539 | ||
1958d2d5 TH |
2540 | src_cset->mg_dst_cset = dst_cset; |
2541 | ||
2542 | if (list_empty(&dst_cset->mg_preload_node)) | |
2543 | list_add(&dst_cset->mg_preload_node, &csets); | |
2544 | else | |
a25eb52e | 2545 | put_css_set(dst_cset); |
1958d2d5 TH |
2546 | } |
2547 | ||
f817de98 | 2548 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2549 | return 0; |
2550 | err: | |
2551 | cgroup_migrate_finish(&csets); | |
2552 | return -ENOMEM; | |
2553 | } | |
2554 | ||
2555 | /** | |
2556 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2557 | * @leader: the leader of the process or the task to migrate |
2558 | * @threadgroup: whether @leader points to the whole process or a single task | |
9af2ec45 | 2559 | * @cgrp: the destination cgroup |
1958d2d5 TH |
2560 | * |
2561 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
1ed13287 | 2562 | * process, the caller must be holding cgroup_threadgroup_rwsem. The |
1958d2d5 TH |
2563 | * caller is also responsible for invoking cgroup_migrate_add_src() and |
2564 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2565 | * function and following up with cgroup_migrate_finish(). | |
2566 | * | |
2567 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2568 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2569 | * failure, when migrating multiple targets, the success or failure can be | |
2570 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2571 | * actually starting migrating. | |
2572 | */ | |
9af2ec45 TH |
2573 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
2574 | struct cgroup *cgrp) | |
74a1166d | 2575 | { |
adaae5dc TH |
2576 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
2577 | struct task_struct *task; | |
74a1166d | 2578 | |
fb5d2b4c MSB |
2579 | /* |
2580 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2581 | * already PF_EXITING could be freed from underneath us unless we | |
2582 | * take an rcu_read_lock. | |
2583 | */ | |
f0d9a5f1 | 2584 | spin_lock_bh(&css_set_lock); |
fb5d2b4c | 2585 | rcu_read_lock(); |
9db8de37 | 2586 | task = leader; |
74a1166d | 2587 | do { |
adaae5dc | 2588 | cgroup_taskset_add(task, &tset); |
081aa458 LZ |
2589 | if (!threadgroup) |
2590 | break; | |
9db8de37 | 2591 | } while_each_thread(leader, task); |
fb5d2b4c | 2592 | rcu_read_unlock(); |
f0d9a5f1 | 2593 | spin_unlock_bh(&css_set_lock); |
74a1166d | 2594 | |
adaae5dc | 2595 | return cgroup_taskset_migrate(&tset, cgrp); |
74a1166d BB |
2596 | } |
2597 | ||
1958d2d5 TH |
2598 | /** |
2599 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2600 | * @dst_cgrp: the cgroup to attach to | |
2601 | * @leader: the task or the leader of the threadgroup to be attached | |
2602 | * @threadgroup: attach the whole threadgroup? | |
2603 | * | |
1ed13287 | 2604 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 TH |
2605 | */ |
2606 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2607 | struct task_struct *leader, bool threadgroup) | |
2608 | { | |
2609 | LIST_HEAD(preloaded_csets); | |
2610 | struct task_struct *task; | |
2611 | int ret; | |
2612 | ||
2613 | /* look up all src csets */ | |
f0d9a5f1 | 2614 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2615 | rcu_read_lock(); |
2616 | task = leader; | |
2617 | do { | |
2618 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2619 | &preloaded_csets); | |
2620 | if (!threadgroup) | |
2621 | break; | |
2622 | } while_each_thread(leader, task); | |
2623 | rcu_read_unlock(); | |
f0d9a5f1 | 2624 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2625 | |
2626 | /* prepare dst csets and commit */ | |
2627 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2628 | if (!ret) | |
9af2ec45 | 2629 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp); |
1958d2d5 TH |
2630 | |
2631 | cgroup_migrate_finish(&preloaded_csets); | |
2632 | return ret; | |
74a1166d BB |
2633 | } |
2634 | ||
187fe840 TH |
2635 | static int cgroup_procs_write_permission(struct task_struct *task, |
2636 | struct cgroup *dst_cgrp, | |
2637 | struct kernfs_open_file *of) | |
dedf22e9 TH |
2638 | { |
2639 | const struct cred *cred = current_cred(); | |
2640 | const struct cred *tcred = get_task_cred(task); | |
2641 | int ret = 0; | |
2642 | ||
2643 | /* | |
2644 | * even if we're attaching all tasks in the thread group, we only | |
2645 | * need to check permissions on one of them. | |
2646 | */ | |
2647 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2648 | !uid_eq(cred->euid, tcred->uid) && | |
2649 | !uid_eq(cred->euid, tcred->suid)) | |
2650 | ret = -EACCES; | |
2651 | ||
187fe840 TH |
2652 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
2653 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2654 | struct cgroup *cgrp; | |
2655 | struct inode *inode; | |
2656 | ||
f0d9a5f1 | 2657 | spin_lock_bh(&css_set_lock); |
187fe840 | 2658 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); |
f0d9a5f1 | 2659 | spin_unlock_bh(&css_set_lock); |
187fe840 TH |
2660 | |
2661 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2662 | cgrp = cgroup_parent(cgrp); | |
2663 | ||
2664 | ret = -ENOMEM; | |
6f60eade | 2665 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
187fe840 TH |
2666 | if (inode) { |
2667 | ret = inode_permission(inode, MAY_WRITE); | |
2668 | iput(inode); | |
2669 | } | |
2670 | } | |
2671 | ||
dedf22e9 TH |
2672 | put_cred(tcred); |
2673 | return ret; | |
2674 | } | |
2675 | ||
74a1166d BB |
2676 | /* |
2677 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2678 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2679 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2680 | */ |
acbef755 TH |
2681 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2682 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2683 | { |
bbcb81d0 | 2684 | struct task_struct *tsk; |
e76ecaee | 2685 | struct cgroup *cgrp; |
acbef755 | 2686 | pid_t pid; |
bbcb81d0 PM |
2687 | int ret; |
2688 | ||
acbef755 TH |
2689 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2690 | return -EINVAL; | |
2691 | ||
e76ecaee TH |
2692 | cgrp = cgroup_kn_lock_live(of->kn); |
2693 | if (!cgrp) | |
74a1166d BB |
2694 | return -ENODEV; |
2695 | ||
3014dde7 | 2696 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2697 | rcu_read_lock(); |
bbcb81d0 | 2698 | if (pid) { |
73507f33 | 2699 | tsk = find_task_by_vpid(pid); |
74a1166d | 2700 | if (!tsk) { |
dd4b0a46 | 2701 | ret = -ESRCH; |
3014dde7 | 2702 | goto out_unlock_rcu; |
bbcb81d0 | 2703 | } |
dedf22e9 | 2704 | } else { |
b78949eb | 2705 | tsk = current; |
dedf22e9 | 2706 | } |
cd3d0952 TH |
2707 | |
2708 | if (threadgroup) | |
b78949eb | 2709 | tsk = tsk->group_leader; |
c4c27fbd MG |
2710 | |
2711 | /* | |
14a40ffc | 2712 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2713 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2714 | * with no rt_runtime allocated. Just say no. | |
2715 | */ | |
14a40ffc | 2716 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2717 | ret = -EINVAL; |
3014dde7 | 2718 | goto out_unlock_rcu; |
c4c27fbd MG |
2719 | } |
2720 | ||
b78949eb MSB |
2721 | get_task_struct(tsk); |
2722 | rcu_read_unlock(); | |
2723 | ||
187fe840 | 2724 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2725 | if (!ret) |
2726 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2727 | |
f9f9e7b7 | 2728 | put_task_struct(tsk); |
3014dde7 TH |
2729 | goto out_unlock_threadgroup; |
2730 | ||
2731 | out_unlock_rcu: | |
2732 | rcu_read_unlock(); | |
2733 | out_unlock_threadgroup: | |
2734 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
e76ecaee | 2735 | cgroup_kn_unlock(of->kn); |
acbef755 | 2736 | return ret ?: nbytes; |
bbcb81d0 PM |
2737 | } |
2738 | ||
7ae1bad9 TH |
2739 | /** |
2740 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2741 | * @from: attach to all cgroups of a given task | |
2742 | * @tsk: the task to be attached | |
2743 | */ | |
2744 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2745 | { | |
3dd06ffa | 2746 | struct cgroup_root *root; |
7ae1bad9 TH |
2747 | int retval = 0; |
2748 | ||
47cfcd09 | 2749 | mutex_lock(&cgroup_mutex); |
985ed670 | 2750 | for_each_root(root) { |
96d365e0 TH |
2751 | struct cgroup *from_cgrp; |
2752 | ||
3dd06ffa | 2753 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2754 | continue; |
2755 | ||
f0d9a5f1 | 2756 | spin_lock_bh(&css_set_lock); |
96d365e0 | 2757 | from_cgrp = task_cgroup_from_root(from, root); |
f0d9a5f1 | 2758 | spin_unlock_bh(&css_set_lock); |
7ae1bad9 | 2759 | |
6f4b7e63 | 2760 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2761 | if (retval) |
2762 | break; | |
2763 | } | |
47cfcd09 | 2764 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2765 | |
2766 | return retval; | |
2767 | } | |
2768 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2769 | ||
acbef755 TH |
2770 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2771 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2772 | { |
acbef755 | 2773 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2774 | } |
2775 | ||
acbef755 TH |
2776 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2777 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2778 | { |
acbef755 | 2779 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2780 | } |
2781 | ||
451af504 TH |
2782 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2783 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2784 | { |
e76ecaee | 2785 | struct cgroup *cgrp; |
5f469907 | 2786 | |
e76ecaee | 2787 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 2788 | |
e76ecaee TH |
2789 | cgrp = cgroup_kn_lock_live(of->kn); |
2790 | if (!cgrp) | |
e788e066 | 2791 | return -ENODEV; |
69e943b7 | 2792 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
2793 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
2794 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 2795 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 2796 | cgroup_kn_unlock(of->kn); |
451af504 | 2797 | return nbytes; |
e788e066 PM |
2798 | } |
2799 | ||
2da8ca82 | 2800 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2801 | { |
2da8ca82 | 2802 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2803 | |
46cfeb04 | 2804 | spin_lock(&release_agent_path_lock); |
e788e066 | 2805 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 2806 | spin_unlock(&release_agent_path_lock); |
e788e066 | 2807 | seq_putc(seq, '\n'); |
e788e066 PM |
2808 | return 0; |
2809 | } | |
2810 | ||
2da8ca82 | 2811 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2812 | { |
c1d5d42e | 2813 | seq_puts(seq, "0\n"); |
e788e066 PM |
2814 | return 0; |
2815 | } | |
2816 | ||
8ab456ac | 2817 | static void cgroup_print_ss_mask(struct seq_file *seq, unsigned long ss_mask) |
355e0c48 | 2818 | { |
f8f22e53 TH |
2819 | struct cgroup_subsys *ss; |
2820 | bool printed = false; | |
2821 | int ssid; | |
a742c59d | 2822 | |
a966a4ed AS |
2823 | for_each_subsys_which(ss, ssid, &ss_mask) { |
2824 | if (printed) | |
2825 | seq_putc(seq, ' '); | |
2826 | seq_printf(seq, "%s", ss->name); | |
2827 | printed = true; | |
e73d2c61 | 2828 | } |
f8f22e53 TH |
2829 | if (printed) |
2830 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2831 | } |
2832 | ||
f8f22e53 TH |
2833 | /* show controllers which are currently attached to the default hierarchy */ |
2834 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
db3b1497 | 2835 | { |
f8f22e53 TH |
2836 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2837 | ||
5533e011 TH |
2838 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask & |
2839 | ~cgrp_dfl_root_inhibit_ss_mask); | |
f8f22e53 | 2840 | return 0; |
db3b1497 PM |
2841 | } |
2842 | ||
f8f22e53 TH |
2843 | /* show controllers which are enabled from the parent */ |
2844 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2845 | { |
f8f22e53 TH |
2846 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2847 | ||
667c2491 | 2848 | cgroup_print_ss_mask(seq, cgroup_parent(cgrp)->subtree_control); |
f8f22e53 | 2849 | return 0; |
ddbcc7e8 PM |
2850 | } |
2851 | ||
f8f22e53 TH |
2852 | /* show controllers which are enabled for a given cgroup's children */ |
2853 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2854 | { |
f8f22e53 TH |
2855 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2856 | ||
667c2491 | 2857 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2858 | return 0; |
2859 | } | |
2860 | ||
2861 | /** | |
2862 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2863 | * @cgrp: root of the subtree to update csses for | |
2864 | * | |
2865 | * @cgrp's child_subsys_mask has changed and its subtree's (self excluded) | |
2866 | * css associations need to be updated accordingly. This function looks up | |
2867 | * all css_sets which are attached to the subtree, creates the matching | |
2868 | * updated css_sets and migrates the tasks to the new ones. | |
2869 | */ | |
2870 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2871 | { | |
2872 | LIST_HEAD(preloaded_csets); | |
10265075 | 2873 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
f8f22e53 TH |
2874 | struct cgroup_subsys_state *css; |
2875 | struct css_set *src_cset; | |
2876 | int ret; | |
2877 | ||
f8f22e53 TH |
2878 | lockdep_assert_held(&cgroup_mutex); |
2879 | ||
3014dde7 TH |
2880 | percpu_down_write(&cgroup_threadgroup_rwsem); |
2881 | ||
f8f22e53 | 2882 | /* look up all csses currently attached to @cgrp's subtree */ |
f0d9a5f1 | 2883 | spin_lock_bh(&css_set_lock); |
f8f22e53 TH |
2884 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { |
2885 | struct cgrp_cset_link *link; | |
2886 | ||
2887 | /* self is not affected by child_subsys_mask change */ | |
2888 | if (css->cgroup == cgrp) | |
2889 | continue; | |
2890 | ||
2891 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2892 | cgroup_migrate_add_src(link->cset, cgrp, | |
2893 | &preloaded_csets); | |
2894 | } | |
f0d9a5f1 | 2895 | spin_unlock_bh(&css_set_lock); |
f8f22e53 TH |
2896 | |
2897 | /* NULL dst indicates self on default hierarchy */ | |
2898 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2899 | if (ret) | |
2900 | goto out_finish; | |
2901 | ||
f0d9a5f1 | 2902 | spin_lock_bh(&css_set_lock); |
f8f22e53 | 2903 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { |
10265075 | 2904 | struct task_struct *task, *ntask; |
f8f22e53 TH |
2905 | |
2906 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2907 | if (!src_cset->mg_src_cgrp) | |
2908 | break; | |
2909 | ||
10265075 TH |
2910 | /* all tasks in src_csets need to be migrated */ |
2911 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
2912 | cgroup_taskset_add(task, &tset); | |
f8f22e53 | 2913 | } |
f0d9a5f1 | 2914 | spin_unlock_bh(&css_set_lock); |
f8f22e53 | 2915 | |
10265075 | 2916 | ret = cgroup_taskset_migrate(&tset, cgrp); |
f8f22e53 TH |
2917 | out_finish: |
2918 | cgroup_migrate_finish(&preloaded_csets); | |
3014dde7 | 2919 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
2920 | return ret; |
2921 | } | |
2922 | ||
2923 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
451af504 TH |
2924 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
2925 | char *buf, size_t nbytes, | |
2926 | loff_t off) | |
f8f22e53 | 2927 | { |
8ab456ac AS |
2928 | unsigned long enable = 0, disable = 0; |
2929 | unsigned long css_enable, css_disable, old_sc, new_sc, old_ss, new_ss; | |
a9746d8d | 2930 | struct cgroup *cgrp, *child; |
f8f22e53 | 2931 | struct cgroup_subsys *ss; |
451af504 | 2932 | char *tok; |
f8f22e53 TH |
2933 | int ssid, ret; |
2934 | ||
2935 | /* | |
d37167ab TH |
2936 | * Parse input - space separated list of subsystem names prefixed |
2937 | * with either + or -. | |
f8f22e53 | 2938 | */ |
451af504 TH |
2939 | buf = strstrip(buf); |
2940 | while ((tok = strsep(&buf, " "))) { | |
a966a4ed AS |
2941 | unsigned long tmp_ss_mask = ~cgrp_dfl_root_inhibit_ss_mask; |
2942 | ||
d37167ab TH |
2943 | if (tok[0] == '\0') |
2944 | continue; | |
a966a4ed | 2945 | for_each_subsys_which(ss, ssid, &tmp_ss_mask) { |
fc5ed1e9 TH |
2946 | if (!cgroup_ssid_enabled(ssid) || |
2947 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
2948 | continue; |
2949 | ||
2950 | if (*tok == '+') { | |
7d331fa9 TH |
2951 | enable |= 1 << ssid; |
2952 | disable &= ~(1 << ssid); | |
f8f22e53 | 2953 | } else if (*tok == '-') { |
7d331fa9 TH |
2954 | disable |= 1 << ssid; |
2955 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
2956 | } else { |
2957 | return -EINVAL; | |
2958 | } | |
2959 | break; | |
2960 | } | |
2961 | if (ssid == CGROUP_SUBSYS_COUNT) | |
2962 | return -EINVAL; | |
2963 | } | |
2964 | ||
a9746d8d TH |
2965 | cgrp = cgroup_kn_lock_live(of->kn); |
2966 | if (!cgrp) | |
2967 | return -ENODEV; | |
f8f22e53 TH |
2968 | |
2969 | for_each_subsys(ss, ssid) { | |
2970 | if (enable & (1 << ssid)) { | |
667c2491 | 2971 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
2972 | enable &= ~(1 << ssid); |
2973 | continue; | |
2974 | } | |
2975 | ||
c29adf24 TH |
2976 | /* unavailable or not enabled on the parent? */ |
2977 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
2978 | (cgroup_parent(cgrp) && | |
667c2491 | 2979 | !(cgroup_parent(cgrp)->subtree_control & (1 << ssid)))) { |
c29adf24 TH |
2980 | ret = -ENOENT; |
2981 | goto out_unlock; | |
2982 | } | |
f8f22e53 | 2983 | } else if (disable & (1 << ssid)) { |
667c2491 | 2984 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
2985 | disable &= ~(1 << ssid); |
2986 | continue; | |
2987 | } | |
2988 | ||
2989 | /* a child has it enabled? */ | |
2990 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 2991 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 2992 | ret = -EBUSY; |
ddab2b6e | 2993 | goto out_unlock; |
f8f22e53 TH |
2994 | } |
2995 | } | |
2996 | } | |
2997 | } | |
2998 | ||
2999 | if (!enable && !disable) { | |
3000 | ret = 0; | |
ddab2b6e | 3001 | goto out_unlock; |
f8f22e53 TH |
3002 | } |
3003 | ||
3004 | /* | |
667c2491 | 3005 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
3006 | * with tasks so that child cgroups don't compete against tasks. |
3007 | */ | |
d51f39b0 | 3008 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
3009 | ret = -EBUSY; |
3010 | goto out_unlock; | |
3011 | } | |
3012 | ||
3013 | /* | |
f63070d3 TH |
3014 | * Update subsys masks and calculate what needs to be done. More |
3015 | * subsystems than specified may need to be enabled or disabled | |
3016 | * depending on subsystem dependencies. | |
3017 | */ | |
755bf5ee TH |
3018 | old_sc = cgrp->subtree_control; |
3019 | old_ss = cgrp->child_subsys_mask; | |
3020 | new_sc = (old_sc | enable) & ~disable; | |
3021 | new_ss = cgroup_calc_child_subsys_mask(cgrp, new_sc); | |
f63070d3 | 3022 | |
755bf5ee TH |
3023 | css_enable = ~old_ss & new_ss; |
3024 | css_disable = old_ss & ~new_ss; | |
f63070d3 TH |
3025 | enable |= css_enable; |
3026 | disable |= css_disable; | |
c29adf24 | 3027 | |
db6e3053 TH |
3028 | /* |
3029 | * Because css offlining is asynchronous, userland might try to | |
3030 | * re-enable the same controller while the previous instance is | |
3031 | * still around. In such cases, wait till it's gone using | |
3032 | * offline_waitq. | |
3033 | */ | |
a966a4ed | 3034 | for_each_subsys_which(ss, ssid, &css_enable) { |
db6e3053 TH |
3035 | cgroup_for_each_live_child(child, cgrp) { |
3036 | DEFINE_WAIT(wait); | |
3037 | ||
3038 | if (!cgroup_css(child, ss)) | |
3039 | continue; | |
3040 | ||
3041 | cgroup_get(child); | |
3042 | prepare_to_wait(&child->offline_waitq, &wait, | |
3043 | TASK_UNINTERRUPTIBLE); | |
3044 | cgroup_kn_unlock(of->kn); | |
3045 | schedule(); | |
3046 | finish_wait(&child->offline_waitq, &wait); | |
3047 | cgroup_put(child); | |
3048 | ||
3049 | return restart_syscall(); | |
3050 | } | |
3051 | } | |
3052 | ||
755bf5ee TH |
3053 | cgrp->subtree_control = new_sc; |
3054 | cgrp->child_subsys_mask = new_ss; | |
3055 | ||
f63070d3 TH |
3056 | /* |
3057 | * Create new csses or make the existing ones visible. A css is | |
3058 | * created invisible if it's being implicitly enabled through | |
3059 | * dependency. An invisible css is made visible when the userland | |
3060 | * explicitly enables it. | |
f8f22e53 TH |
3061 | */ |
3062 | for_each_subsys(ss, ssid) { | |
3063 | if (!(enable & (1 << ssid))) | |
3064 | continue; | |
3065 | ||
3066 | cgroup_for_each_live_child(child, cgrp) { | |
f63070d3 TH |
3067 | if (css_enable & (1 << ssid)) |
3068 | ret = create_css(child, ss, | |
3069 | cgrp->subtree_control & (1 << ssid)); | |
3070 | else | |
4df8dc90 TH |
3071 | ret = css_populate_dir(cgroup_css(child, ss), |
3072 | NULL); | |
f8f22e53 TH |
3073 | if (ret) |
3074 | goto err_undo_css; | |
3075 | } | |
3076 | } | |
3077 | ||
c29adf24 TH |
3078 | /* |
3079 | * At this point, cgroup_e_css() results reflect the new csses | |
3080 | * making the following cgroup_update_dfl_csses() properly update | |
3081 | * css associations of all tasks in the subtree. | |
3082 | */ | |
f8f22e53 TH |
3083 | ret = cgroup_update_dfl_csses(cgrp); |
3084 | if (ret) | |
3085 | goto err_undo_css; | |
3086 | ||
f63070d3 TH |
3087 | /* |
3088 | * All tasks are migrated out of disabled csses. Kill or hide | |
3089 | * them. A css is hidden when the userland requests it to be | |
b4536f0c TH |
3090 | * disabled while other subsystems are still depending on it. The |
3091 | * css must not actively control resources and be in the vanilla | |
3092 | * state if it's made visible again later. Controllers which may | |
3093 | * be depended upon should provide ->css_reset() for this purpose. | |
f63070d3 | 3094 | */ |
f8f22e53 TH |
3095 | for_each_subsys(ss, ssid) { |
3096 | if (!(disable & (1 << ssid))) | |
3097 | continue; | |
3098 | ||
f63070d3 | 3099 | cgroup_for_each_live_child(child, cgrp) { |
b4536f0c TH |
3100 | struct cgroup_subsys_state *css = cgroup_css(child, ss); |
3101 | ||
3102 | if (css_disable & (1 << ssid)) { | |
3103 | kill_css(css); | |
3104 | } else { | |
4df8dc90 | 3105 | css_clear_dir(css, NULL); |
b4536f0c TH |
3106 | if (ss->css_reset) |
3107 | ss->css_reset(css); | |
3108 | } | |
f63070d3 | 3109 | } |
f8f22e53 TH |
3110 | } |
3111 | ||
56c807ba TH |
3112 | /* |
3113 | * The effective csses of all the descendants (excluding @cgrp) may | |
3114 | * have changed. Subsystems can optionally subscribe to this event | |
3115 | * by implementing ->css_e_css_changed() which is invoked if any of | |
3116 | * the effective csses seen from the css's cgroup may have changed. | |
3117 | */ | |
3118 | for_each_subsys(ss, ssid) { | |
3119 | struct cgroup_subsys_state *this_css = cgroup_css(cgrp, ss); | |
3120 | struct cgroup_subsys_state *css; | |
3121 | ||
3122 | if (!ss->css_e_css_changed || !this_css) | |
3123 | continue; | |
3124 | ||
3125 | css_for_each_descendant_pre(css, this_css) | |
3126 | if (css != this_css) | |
3127 | ss->css_e_css_changed(css); | |
3128 | } | |
3129 | ||
f8f22e53 TH |
3130 | kernfs_activate(cgrp->kn); |
3131 | ret = 0; | |
3132 | out_unlock: | |
a9746d8d | 3133 | cgroup_kn_unlock(of->kn); |
451af504 | 3134 | return ret ?: nbytes; |
f8f22e53 TH |
3135 | |
3136 | err_undo_css: | |
755bf5ee TH |
3137 | cgrp->subtree_control = old_sc; |
3138 | cgrp->child_subsys_mask = old_ss; | |
f8f22e53 TH |
3139 | |
3140 | for_each_subsys(ss, ssid) { | |
3141 | if (!(enable & (1 << ssid))) | |
3142 | continue; | |
3143 | ||
3144 | cgroup_for_each_live_child(child, cgrp) { | |
3145 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
f63070d3 TH |
3146 | |
3147 | if (!css) | |
3148 | continue; | |
3149 | ||
3150 | if (css_enable & (1 << ssid)) | |
f8f22e53 | 3151 | kill_css(css); |
f63070d3 | 3152 | else |
4df8dc90 | 3153 | css_clear_dir(css, NULL); |
f8f22e53 TH |
3154 | } |
3155 | } | |
3156 | goto out_unlock; | |
3157 | } | |
3158 | ||
4a07c222 | 3159 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3160 | { |
4a07c222 | 3161 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 3162 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
3163 | return 0; |
3164 | } | |
3165 | ||
2bd59d48 TH |
3166 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3167 | size_t nbytes, loff_t off) | |
355e0c48 | 3168 | { |
2bd59d48 TH |
3169 | struct cgroup *cgrp = of->kn->parent->priv; |
3170 | struct cftype *cft = of->kn->priv; | |
3171 | struct cgroup_subsys_state *css; | |
a742c59d | 3172 | int ret; |
355e0c48 | 3173 | |
b4168640 TH |
3174 | if (cft->write) |
3175 | return cft->write(of, buf, nbytes, off); | |
3176 | ||
2bd59d48 TH |
3177 | /* |
3178 | * kernfs guarantees that a file isn't deleted with operations in | |
3179 | * flight, which means that the matching css is and stays alive and | |
3180 | * doesn't need to be pinned. The RCU locking is not necessary | |
3181 | * either. It's just for the convenience of using cgroup_css(). | |
3182 | */ | |
3183 | rcu_read_lock(); | |
3184 | css = cgroup_css(cgrp, cft->ss); | |
3185 | rcu_read_unlock(); | |
a742c59d | 3186 | |
451af504 | 3187 | if (cft->write_u64) { |
a742c59d TH |
3188 | unsigned long long v; |
3189 | ret = kstrtoull(buf, 0, &v); | |
3190 | if (!ret) | |
3191 | ret = cft->write_u64(css, cft, v); | |
3192 | } else if (cft->write_s64) { | |
3193 | long long v; | |
3194 | ret = kstrtoll(buf, 0, &v); | |
3195 | if (!ret) | |
3196 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3197 | } else { |
a742c59d | 3198 | ret = -EINVAL; |
e73d2c61 | 3199 | } |
2bd59d48 | 3200 | |
a742c59d | 3201 | return ret ?: nbytes; |
355e0c48 PM |
3202 | } |
3203 | ||
6612f05b | 3204 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3205 | { |
2bd59d48 | 3206 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3207 | } |
3208 | ||
6612f05b | 3209 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3210 | { |
2bd59d48 | 3211 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3212 | } |
3213 | ||
6612f05b | 3214 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3215 | { |
2bd59d48 | 3216 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
3217 | } |
3218 | ||
91796569 | 3219 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3220 | { |
7da11279 TH |
3221 | struct cftype *cft = seq_cft(m); |
3222 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3223 | |
2da8ca82 TH |
3224 | if (cft->seq_show) |
3225 | return cft->seq_show(m, arg); | |
e73d2c61 | 3226 | |
f4c753b7 | 3227 | if (cft->read_u64) |
896f5199 TH |
3228 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3229 | else if (cft->read_s64) | |
3230 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3231 | else | |
3232 | return -EINVAL; | |
3233 | return 0; | |
91796569 PM |
3234 | } |
3235 | ||
2bd59d48 TH |
3236 | static struct kernfs_ops cgroup_kf_single_ops = { |
3237 | .atomic_write_len = PAGE_SIZE, | |
3238 | .write = cgroup_file_write, | |
3239 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3240 | }; |
3241 | ||
2bd59d48 TH |
3242 | static struct kernfs_ops cgroup_kf_ops = { |
3243 | .atomic_write_len = PAGE_SIZE, | |
3244 | .write = cgroup_file_write, | |
3245 | .seq_start = cgroup_seqfile_start, | |
3246 | .seq_next = cgroup_seqfile_next, | |
3247 | .seq_stop = cgroup_seqfile_stop, | |
3248 | .seq_show = cgroup_seqfile_show, | |
3249 | }; | |
ddbcc7e8 PM |
3250 | |
3251 | /* | |
3252 | * cgroup_rename - Only allow simple rename of directories in place. | |
3253 | */ | |
2bd59d48 TH |
3254 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
3255 | const char *new_name_str) | |
ddbcc7e8 | 3256 | { |
2bd59d48 | 3257 | struct cgroup *cgrp = kn->priv; |
65dff759 | 3258 | int ret; |
65dff759 | 3259 | |
2bd59d48 | 3260 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 3261 | return -ENOTDIR; |
2bd59d48 | 3262 | if (kn->parent != new_parent) |
ddbcc7e8 | 3263 | return -EIO; |
65dff759 | 3264 | |
6db8e85c TH |
3265 | /* |
3266 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 3267 | * limited. Disallow on the default hierarchy. |
6db8e85c | 3268 | */ |
aa6ec29b | 3269 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 3270 | return -EPERM; |
099fca32 | 3271 | |
e1b2dc17 | 3272 | /* |
8353da1f | 3273 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3274 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3275 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3276 | */ |
3277 | kernfs_break_active_protection(new_parent); | |
3278 | kernfs_break_active_protection(kn); | |
099fca32 | 3279 | |
2bd59d48 | 3280 | mutex_lock(&cgroup_mutex); |
099fca32 | 3281 | |
2bd59d48 | 3282 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3283 | |
2bd59d48 | 3284 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3285 | |
3286 | kernfs_unbreak_active_protection(kn); | |
3287 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3288 | return ret; |
099fca32 LZ |
3289 | } |
3290 | ||
49957f8e TH |
3291 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3292 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3293 | { | |
3294 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3295 | .ia_uid = current_fsuid(), | |
3296 | .ia_gid = current_fsgid(), }; | |
3297 | ||
3298 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3299 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3300 | return 0; | |
3301 | ||
3302 | return kernfs_setattr(kn, &iattr); | |
3303 | } | |
3304 | ||
4df8dc90 TH |
3305 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3306 | struct cftype *cft) | |
ddbcc7e8 | 3307 | { |
8d7e6fb0 | 3308 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3309 | struct kernfs_node *kn; |
3310 | struct lock_class_key *key = NULL; | |
49957f8e | 3311 | int ret; |
05ef1d7c | 3312 | |
2bd59d48 TH |
3313 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3314 | key = &cft->lockdep_key; | |
3315 | #endif | |
3316 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3317 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3318 | NULL, key); |
49957f8e TH |
3319 | if (IS_ERR(kn)) |
3320 | return PTR_ERR(kn); | |
3321 | ||
3322 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3323 | if (ret) { |
49957f8e | 3324 | kernfs_remove(kn); |
f8f22e53 TH |
3325 | return ret; |
3326 | } | |
3327 | ||
6f60eade TH |
3328 | if (cft->file_offset) { |
3329 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3330 | ||
34c06254 | 3331 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3332 | cfile->kn = kn; |
34c06254 | 3333 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3334 | } |
3335 | ||
f8f22e53 | 3336 | return 0; |
ddbcc7e8 PM |
3337 | } |
3338 | ||
b1f28d31 TH |
3339 | /** |
3340 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3341 | * @css: the target css |
3342 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3343 | * @cfts: array of cftypes to be added |
3344 | * @is_add: whether to add or remove | |
3345 | * | |
3346 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3347 | * For removals, this function never fails. |
b1f28d31 | 3348 | */ |
4df8dc90 TH |
3349 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3350 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3351 | bool is_add) |
ddbcc7e8 | 3352 | { |
6732ed85 | 3353 | struct cftype *cft, *cft_end = NULL; |
b1f28d31 TH |
3354 | int ret; |
3355 | ||
01f6474c | 3356 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3357 | |
6732ed85 TH |
3358 | restart: |
3359 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3360 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3361 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3362 | continue; |
05ebb6e6 | 3363 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3364 | continue; |
d51f39b0 | 3365 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3366 | continue; |
d51f39b0 | 3367 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3368 | continue; |
3369 | ||
2739d3cc | 3370 | if (is_add) { |
4df8dc90 | 3371 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3372 | if (ret) { |
ed3d261b JP |
3373 | pr_warn("%s: failed to add %s, err=%d\n", |
3374 | __func__, cft->name, ret); | |
6732ed85 TH |
3375 | cft_end = cft; |
3376 | is_add = false; | |
3377 | goto restart; | |
b1f28d31 | 3378 | } |
2739d3cc LZ |
3379 | } else { |
3380 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3381 | } |
ddbcc7e8 | 3382 | } |
b1f28d31 | 3383 | return 0; |
ddbcc7e8 PM |
3384 | } |
3385 | ||
21a2d343 | 3386 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3387 | { |
3388 | LIST_HEAD(pending); | |
2bb566cb | 3389 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3390 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3391 | struct cgroup_subsys_state *css; |
9ccece80 | 3392 | int ret = 0; |
8e3f6541 | 3393 | |
01f6474c | 3394 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3395 | |
e8c82d20 | 3396 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3397 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3398 | struct cgroup *cgrp = css->cgroup; |
3399 | ||
e8c82d20 LZ |
3400 | if (cgroup_is_dead(cgrp)) |
3401 | continue; | |
3402 | ||
4df8dc90 | 3403 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3404 | if (ret) |
3405 | break; | |
8e3f6541 | 3406 | } |
21a2d343 TH |
3407 | |
3408 | if (is_add && !ret) | |
3409 | kernfs_activate(root->kn); | |
9ccece80 | 3410 | return ret; |
8e3f6541 TH |
3411 | } |
3412 | ||
2da440a2 | 3413 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3414 | { |
2bb566cb | 3415 | struct cftype *cft; |
8e3f6541 | 3416 | |
2bd59d48 TH |
3417 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3418 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3419 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3420 | kfree(cft->kf_ops); | |
3421 | cft->kf_ops = NULL; | |
2da440a2 | 3422 | cft->ss = NULL; |
a8ddc821 TH |
3423 | |
3424 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3425 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3426 | } |
2da440a2 TH |
3427 | } |
3428 | ||
2bd59d48 | 3429 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3430 | { |
3431 | struct cftype *cft; | |
3432 | ||
2bd59d48 TH |
3433 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3434 | struct kernfs_ops *kf_ops; | |
3435 | ||
0adb0704 TH |
3436 | WARN_ON(cft->ss || cft->kf_ops); |
3437 | ||
2bd59d48 TH |
3438 | if (cft->seq_start) |
3439 | kf_ops = &cgroup_kf_ops; | |
3440 | else | |
3441 | kf_ops = &cgroup_kf_single_ops; | |
3442 | ||
3443 | /* | |
3444 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3445 | * make a copy of kf_ops to set its atomic_write_len. | |
3446 | */ | |
3447 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3448 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3449 | if (!kf_ops) { | |
3450 | cgroup_exit_cftypes(cfts); | |
3451 | return -ENOMEM; | |
3452 | } | |
3453 | kf_ops->atomic_write_len = cft->max_write_len; | |
3454 | } | |
8e3f6541 | 3455 | |
2bd59d48 | 3456 | cft->kf_ops = kf_ops; |
2bb566cb | 3457 | cft->ss = ss; |
2bd59d48 | 3458 | } |
2bb566cb | 3459 | |
2bd59d48 | 3460 | return 0; |
2da440a2 TH |
3461 | } |
3462 | ||
21a2d343 TH |
3463 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3464 | { | |
01f6474c | 3465 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3466 | |
3467 | if (!cfts || !cfts[0].ss) | |
3468 | return -ENOENT; | |
3469 | ||
3470 | list_del(&cfts->node); | |
3471 | cgroup_apply_cftypes(cfts, false); | |
3472 | cgroup_exit_cftypes(cfts); | |
3473 | return 0; | |
8e3f6541 | 3474 | } |
8e3f6541 | 3475 | |
79578621 TH |
3476 | /** |
3477 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3478 | * @cfts: zero-length name terminated array of cftypes |
3479 | * | |
2bb566cb TH |
3480 | * Unregister @cfts. Files described by @cfts are removed from all |
3481 | * existing cgroups and all future cgroups won't have them either. This | |
3482 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3483 | * |
3484 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3485 | * registered. |
79578621 | 3486 | */ |
2bb566cb | 3487 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3488 | { |
21a2d343 | 3489 | int ret; |
79578621 | 3490 | |
01f6474c | 3491 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3492 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3493 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3494 | return ret; |
80b13586 TH |
3495 | } |
3496 | ||
8e3f6541 TH |
3497 | /** |
3498 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3499 | * @ss: target cgroup subsystem | |
3500 | * @cfts: zero-length name terminated array of cftypes | |
3501 | * | |
3502 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3503 | * existing cgroups to which @ss is attached and all future cgroups will | |
3504 | * have them too. This function can be called anytime whether @ss is | |
3505 | * attached or not. | |
3506 | * | |
3507 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3508 | * function currently returns 0 as long as @cfts registration is successful | |
3509 | * even if some file creation attempts on existing cgroups fail. | |
3510 | */ | |
2cf669a5 | 3511 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3512 | { |
9ccece80 | 3513 | int ret; |
8e3f6541 | 3514 | |
fc5ed1e9 | 3515 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3516 | return 0; |
3517 | ||
dc5736ed LZ |
3518 | if (!cfts || cfts[0].name[0] == '\0') |
3519 | return 0; | |
2bb566cb | 3520 | |
2bd59d48 TH |
3521 | ret = cgroup_init_cftypes(ss, cfts); |
3522 | if (ret) | |
3523 | return ret; | |
79578621 | 3524 | |
01f6474c | 3525 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3526 | |
0adb0704 | 3527 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3528 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3529 | if (ret) |
21a2d343 | 3530 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3531 | |
01f6474c | 3532 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3533 | return ret; |
79578621 TH |
3534 | } |
3535 | ||
a8ddc821 TH |
3536 | /** |
3537 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3538 | * @ss: target cgroup subsystem | |
3539 | * @cfts: zero-length name terminated array of cftypes | |
3540 | * | |
3541 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3542 | * the default hierarchy. | |
3543 | */ | |
3544 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3545 | { | |
3546 | struct cftype *cft; | |
3547 | ||
3548 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3549 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3550 | return cgroup_add_cftypes(ss, cfts); |
3551 | } | |
3552 | ||
3553 | /** | |
3554 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3555 | * @ss: target cgroup subsystem | |
3556 | * @cfts: zero-length name terminated array of cftypes | |
3557 | * | |
3558 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3559 | * the legacy hierarchies. | |
3560 | */ | |
2cf669a5 TH |
3561 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3562 | { | |
a8ddc821 TH |
3563 | struct cftype *cft; |
3564 | ||
e4b7037c TH |
3565 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3566 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3567 | return cgroup_add_cftypes(ss, cfts); |
3568 | } | |
3569 | ||
34c06254 TH |
3570 | /** |
3571 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3572 | * @cfile: target cgroup_file | |
3573 | * | |
3574 | * @cfile must have been obtained by setting cftype->file_offset. | |
3575 | */ | |
3576 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3577 | { | |
3578 | unsigned long flags; | |
3579 | ||
3580 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3581 | if (cfile->kn) | |
3582 | kernfs_notify(cfile->kn); | |
3583 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3584 | } | |
3585 | ||
a043e3b2 LZ |
3586 | /** |
3587 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3588 | * @cgrp: the cgroup in question | |
3589 | * | |
3590 | * Return the number of tasks in the cgroup. | |
3591 | */ | |
07bc356e | 3592 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3593 | { |
3594 | int count = 0; | |
69d0206c | 3595 | struct cgrp_cset_link *link; |
817929ec | 3596 | |
f0d9a5f1 | 3597 | spin_lock_bh(&css_set_lock); |
69d0206c TH |
3598 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3599 | count += atomic_read(&link->cset->refcount); | |
f0d9a5f1 | 3600 | spin_unlock_bh(&css_set_lock); |
bbcb81d0 PM |
3601 | return count; |
3602 | } | |
3603 | ||
53fa5261 | 3604 | /** |
492eb21b | 3605 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3606 | * @pos: the current position (%NULL to initiate traversal) |
3607 | * @parent: css whose children to walk | |
53fa5261 | 3608 | * |
c2931b70 | 3609 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3610 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3611 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3612 | * be returned regardless of their states. | |
3613 | * | |
3614 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3615 | * css which finished ->css_online() is guaranteed to be visible in the | |
3616 | * future iterations and will stay visible until the last reference is put. | |
3617 | * A css which hasn't finished ->css_online() or already finished | |
3618 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3619 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3620 | */ |
c2931b70 TH |
3621 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3622 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3623 | { |
c2931b70 | 3624 | struct cgroup_subsys_state *next; |
53fa5261 | 3625 | |
8353da1f | 3626 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3627 | |
3628 | /* | |
de3f0341 TH |
3629 | * @pos could already have been unlinked from the sibling list. |
3630 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3631 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3632 | * @pos is taken off list, at which time its next pointer is valid, | |
3633 | * and, as releases are serialized, the one pointed to by the next | |
3634 | * pointer is guaranteed to not have started release yet. This | |
3635 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3636 | * critical section, the one pointed to by its next pointer is | |
3637 | * guaranteed to not have finished its RCU grace period even if we | |
3638 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3639 | * |
de3f0341 TH |
3640 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3641 | * dereferenced; however, as each css is given a monotonically | |
3642 | * increasing unique serial number and always appended to the | |
3643 | * sibling list, the next one can be found by walking the parent's | |
3644 | * children until the first css with higher serial number than | |
3645 | * @pos's. While this path can be slower, it happens iff iteration | |
3646 | * races against release and the race window is very small. | |
53fa5261 | 3647 | */ |
3b287a50 | 3648 | if (!pos) { |
c2931b70 TH |
3649 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3650 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3651 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3652 | } else { |
c2931b70 | 3653 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3654 | if (next->serial_nr > pos->serial_nr) |
3655 | break; | |
53fa5261 TH |
3656 | } |
3657 | ||
3b281afb TH |
3658 | /* |
3659 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3660 | * the next sibling. |
3b281afb | 3661 | */ |
c2931b70 TH |
3662 | if (&next->sibling != &parent->children) |
3663 | return next; | |
3b281afb | 3664 | return NULL; |
53fa5261 | 3665 | } |
53fa5261 | 3666 | |
574bd9f7 | 3667 | /** |
492eb21b | 3668 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3669 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3670 | * @root: css whose descendants to walk |
574bd9f7 | 3671 | * |
492eb21b | 3672 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3673 | * to visit for pre-order traversal of @root's descendants. @root is |
3674 | * included in the iteration and the first node to be visited. | |
75501a6d | 3675 | * |
87fb54f1 TH |
3676 | * While this function requires cgroup_mutex or RCU read locking, it |
3677 | * doesn't require the whole traversal to be contained in a single critical | |
3678 | * section. This function will return the correct next descendant as long | |
3679 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3680 | * |
3681 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3682 | * css which finished ->css_online() is guaranteed to be visible in the | |
3683 | * future iterations and will stay visible until the last reference is put. | |
3684 | * A css which hasn't finished ->css_online() or already finished | |
3685 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3686 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3687 | */ |
492eb21b TH |
3688 | struct cgroup_subsys_state * |
3689 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3690 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3691 | { |
492eb21b | 3692 | struct cgroup_subsys_state *next; |
574bd9f7 | 3693 | |
8353da1f | 3694 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3695 | |
bd8815a6 | 3696 | /* if first iteration, visit @root */ |
7805d000 | 3697 | if (!pos) |
bd8815a6 | 3698 | return root; |
574bd9f7 TH |
3699 | |
3700 | /* visit the first child if exists */ | |
492eb21b | 3701 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3702 | if (next) |
3703 | return next; | |
3704 | ||
3705 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3706 | while (pos != root) { |
5c9d535b | 3707 | next = css_next_child(pos, pos->parent); |
75501a6d | 3708 | if (next) |
574bd9f7 | 3709 | return next; |
5c9d535b | 3710 | pos = pos->parent; |
7805d000 | 3711 | } |
574bd9f7 TH |
3712 | |
3713 | return NULL; | |
3714 | } | |
574bd9f7 | 3715 | |
12a9d2fe | 3716 | /** |
492eb21b TH |
3717 | * css_rightmost_descendant - return the rightmost descendant of a css |
3718 | * @pos: css of interest | |
12a9d2fe | 3719 | * |
492eb21b TH |
3720 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3721 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3722 | * subtree of @pos. |
75501a6d | 3723 | * |
87fb54f1 TH |
3724 | * While this function requires cgroup_mutex or RCU read locking, it |
3725 | * doesn't require the whole traversal to be contained in a single critical | |
3726 | * section. This function will return the correct rightmost descendant as | |
3727 | * long as @pos is accessible. | |
12a9d2fe | 3728 | */ |
492eb21b TH |
3729 | struct cgroup_subsys_state * |
3730 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3731 | { |
492eb21b | 3732 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3733 | |
8353da1f | 3734 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3735 | |
3736 | do { | |
3737 | last = pos; | |
3738 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3739 | pos = NULL; | |
492eb21b | 3740 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3741 | pos = tmp; |
3742 | } while (pos); | |
3743 | ||
3744 | return last; | |
3745 | } | |
12a9d2fe | 3746 | |
492eb21b TH |
3747 | static struct cgroup_subsys_state * |
3748 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3749 | { |
492eb21b | 3750 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3751 | |
3752 | do { | |
3753 | last = pos; | |
492eb21b | 3754 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3755 | } while (pos); |
3756 | ||
3757 | return last; | |
3758 | } | |
3759 | ||
3760 | /** | |
492eb21b | 3761 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3762 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3763 | * @root: css whose descendants to walk |
574bd9f7 | 3764 | * |
492eb21b | 3765 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3766 | * to visit for post-order traversal of @root's descendants. @root is |
3767 | * included in the iteration and the last node to be visited. | |
75501a6d | 3768 | * |
87fb54f1 TH |
3769 | * While this function requires cgroup_mutex or RCU read locking, it |
3770 | * doesn't require the whole traversal to be contained in a single critical | |
3771 | * section. This function will return the correct next descendant as long | |
3772 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3773 | * @cgroup. | |
c2931b70 TH |
3774 | * |
3775 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3776 | * css which finished ->css_online() is guaranteed to be visible in the | |
3777 | * future iterations and will stay visible until the last reference is put. | |
3778 | * A css which hasn't finished ->css_online() or already finished | |
3779 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3780 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3781 | */ |
492eb21b TH |
3782 | struct cgroup_subsys_state * |
3783 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3784 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3785 | { |
492eb21b | 3786 | struct cgroup_subsys_state *next; |
574bd9f7 | 3787 | |
8353da1f | 3788 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3789 | |
58b79a91 TH |
3790 | /* if first iteration, visit leftmost descendant which may be @root */ |
3791 | if (!pos) | |
3792 | return css_leftmost_descendant(root); | |
574bd9f7 | 3793 | |
bd8815a6 TH |
3794 | /* if we visited @root, we're done */ |
3795 | if (pos == root) | |
3796 | return NULL; | |
3797 | ||
574bd9f7 | 3798 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 3799 | next = css_next_child(pos, pos->parent); |
75501a6d | 3800 | if (next) |
492eb21b | 3801 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3802 | |
3803 | /* no sibling left, visit parent */ | |
5c9d535b | 3804 | return pos->parent; |
574bd9f7 | 3805 | } |
574bd9f7 | 3806 | |
f3d46500 TH |
3807 | /** |
3808 | * css_has_online_children - does a css have online children | |
3809 | * @css: the target css | |
3810 | * | |
3811 | * Returns %true if @css has any online children; otherwise, %false. This | |
3812 | * function can be called from any context but the caller is responsible | |
3813 | * for synchronizing against on/offlining as necessary. | |
3814 | */ | |
3815 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 3816 | { |
f3d46500 TH |
3817 | struct cgroup_subsys_state *child; |
3818 | bool ret = false; | |
cbc125ef TH |
3819 | |
3820 | rcu_read_lock(); | |
f3d46500 | 3821 | css_for_each_child(child, css) { |
99bae5f9 | 3822 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
3823 | ret = true; |
3824 | break; | |
cbc125ef TH |
3825 | } |
3826 | } | |
3827 | rcu_read_unlock(); | |
f3d46500 | 3828 | return ret; |
574bd9f7 | 3829 | } |
574bd9f7 | 3830 | |
0942eeee | 3831 | /** |
ecb9d535 | 3832 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
3833 | * @it: the iterator to advance |
3834 | * | |
3835 | * Advance @it to the next css_set to walk. | |
d515876e | 3836 | */ |
ecb9d535 | 3837 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 3838 | { |
0f0a2b4f | 3839 | struct list_head *l = it->cset_pos; |
d515876e TH |
3840 | struct cgrp_cset_link *link; |
3841 | struct css_set *cset; | |
3842 | ||
f0d9a5f1 | 3843 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 3844 | |
d515876e TH |
3845 | /* Advance to the next non-empty css_set */ |
3846 | do { | |
3847 | l = l->next; | |
0f0a2b4f TH |
3848 | if (l == it->cset_head) { |
3849 | it->cset_pos = NULL; | |
ecb9d535 | 3850 | it->task_pos = NULL; |
d515876e TH |
3851 | return; |
3852 | } | |
3ebb2b6e TH |
3853 | |
3854 | if (it->ss) { | |
3855 | cset = container_of(l, struct css_set, | |
3856 | e_cset_node[it->ss->id]); | |
3857 | } else { | |
3858 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3859 | cset = link->cset; | |
3860 | } | |
0de0942d | 3861 | } while (!css_set_populated(cset)); |
c7561128 | 3862 | |
0f0a2b4f | 3863 | it->cset_pos = l; |
c7561128 TH |
3864 | |
3865 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3866 | it->task_pos = cset->tasks.next; |
c7561128 | 3867 | else |
0f0a2b4f TH |
3868 | it->task_pos = cset->mg_tasks.next; |
3869 | ||
3870 | it->tasks_head = &cset->tasks; | |
3871 | it->mg_tasks_head = &cset->mg_tasks; | |
ed27b9f7 TH |
3872 | |
3873 | /* | |
3874 | * We don't keep css_sets locked across iteration steps and thus | |
3875 | * need to take steps to ensure that iteration can be resumed after | |
3876 | * the lock is re-acquired. Iteration is performed at two levels - | |
3877 | * css_sets and tasks in them. | |
3878 | * | |
3879 | * Once created, a css_set never leaves its cgroup lists, so a | |
3880 | * pinned css_set is guaranteed to stay put and we can resume | |
3881 | * iteration afterwards. | |
3882 | * | |
3883 | * Tasks may leave @cset across iteration steps. This is resolved | |
3884 | * by registering each iterator with the css_set currently being | |
3885 | * walked and making css_set_move_task() advance iterators whose | |
3886 | * next task is leaving. | |
3887 | */ | |
3888 | if (it->cur_cset) { | |
3889 | list_del(&it->iters_node); | |
3890 | put_css_set_locked(it->cur_cset); | |
3891 | } | |
3892 | get_css_set(cset); | |
3893 | it->cur_cset = cset; | |
3894 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
3895 | } |
3896 | ||
ecb9d535 TH |
3897 | static void css_task_iter_advance(struct css_task_iter *it) |
3898 | { | |
3899 | struct list_head *l = it->task_pos; | |
3900 | ||
f0d9a5f1 | 3901 | lockdep_assert_held(&css_set_lock); |
ecb9d535 TH |
3902 | WARN_ON_ONCE(!l); |
3903 | ||
3904 | /* | |
3905 | * Advance iterator to find next entry. cset->tasks is consumed | |
3906 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3907 | * next cset. | |
3908 | */ | |
3909 | l = l->next; | |
3910 | ||
3911 | if (l == it->tasks_head) | |
3912 | l = it->mg_tasks_head->next; | |
3913 | ||
3914 | if (l == it->mg_tasks_head) | |
3915 | css_task_iter_advance_css_set(it); | |
3916 | else | |
3917 | it->task_pos = l; | |
3918 | } | |
3919 | ||
0942eeee | 3920 | /** |
72ec7029 TH |
3921 | * css_task_iter_start - initiate task iteration |
3922 | * @css: the css to walk tasks of | |
0942eeee TH |
3923 | * @it: the task iterator to use |
3924 | * | |
72ec7029 TH |
3925 | * Initiate iteration through the tasks of @css. The caller can call |
3926 | * css_task_iter_next() to walk through the tasks until the function | |
3927 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3928 | * called. | |
0942eeee | 3929 | */ |
72ec7029 TH |
3930 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3931 | struct css_task_iter *it) | |
817929ec | 3932 | { |
56fde9e0 TH |
3933 | /* no one should try to iterate before mounting cgroups */ |
3934 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3935 | |
ed27b9f7 TH |
3936 | memset(it, 0, sizeof(*it)); |
3937 | ||
f0d9a5f1 | 3938 | spin_lock_bh(&css_set_lock); |
c59cd3d8 | 3939 | |
3ebb2b6e TH |
3940 | it->ss = css->ss; |
3941 | ||
3942 | if (it->ss) | |
3943 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3944 | else | |
3945 | it->cset_pos = &css->cgroup->cset_links; | |
3946 | ||
0f0a2b4f | 3947 | it->cset_head = it->cset_pos; |
c59cd3d8 | 3948 | |
ecb9d535 | 3949 | css_task_iter_advance_css_set(it); |
ed27b9f7 | 3950 | |
f0d9a5f1 | 3951 | spin_unlock_bh(&css_set_lock); |
817929ec PM |
3952 | } |
3953 | ||
0942eeee | 3954 | /** |
72ec7029 | 3955 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3956 | * @it: the task iterator being iterated |
3957 | * | |
3958 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3959 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3960 | * reaches the end. | |
0942eeee | 3961 | */ |
72ec7029 | 3962 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 3963 | { |
d5745675 | 3964 | if (it->cur_task) { |
ed27b9f7 | 3965 | put_task_struct(it->cur_task); |
d5745675 TH |
3966 | it->cur_task = NULL; |
3967 | } | |
ed27b9f7 | 3968 | |
f0d9a5f1 | 3969 | spin_lock_bh(&css_set_lock); |
ed27b9f7 | 3970 | |
d5745675 TH |
3971 | if (it->task_pos) { |
3972 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
3973 | cg_list); | |
3974 | get_task_struct(it->cur_task); | |
3975 | css_task_iter_advance(it); | |
3976 | } | |
ed27b9f7 | 3977 | |
f0d9a5f1 | 3978 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
3979 | |
3980 | return it->cur_task; | |
817929ec PM |
3981 | } |
3982 | ||
0942eeee | 3983 | /** |
72ec7029 | 3984 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3985 | * @it: the task iterator to finish |
3986 | * | |
72ec7029 | 3987 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3988 | */ |
72ec7029 | 3989 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 3990 | { |
ed27b9f7 | 3991 | if (it->cur_cset) { |
f0d9a5f1 | 3992 | spin_lock_bh(&css_set_lock); |
ed27b9f7 TH |
3993 | list_del(&it->iters_node); |
3994 | put_css_set_locked(it->cur_cset); | |
f0d9a5f1 | 3995 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
3996 | } |
3997 | ||
3998 | if (it->cur_task) | |
3999 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4000 | } |
4001 | ||
4002 | /** | |
8cc99345 TH |
4003 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
4004 | * @to: cgroup to which the tasks will be moved | |
4005 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 4006 | * |
eaf797ab TH |
4007 | * Locking rules between cgroup_post_fork() and the migration path |
4008 | * guarantee that, if a task is forking while being migrated, the new child | |
4009 | * is guaranteed to be either visible in the source cgroup after the | |
4010 | * parent's migration is complete or put into the target cgroup. No task | |
4011 | * can slip out of migration through forking. | |
31a7df01 | 4012 | */ |
8cc99345 | 4013 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 4014 | { |
952aaa12 TH |
4015 | LIST_HEAD(preloaded_csets); |
4016 | struct cgrp_cset_link *link; | |
72ec7029 | 4017 | struct css_task_iter it; |
e406d1cf | 4018 | struct task_struct *task; |
952aaa12 | 4019 | int ret; |
31a7df01 | 4020 | |
952aaa12 | 4021 | mutex_lock(&cgroup_mutex); |
31a7df01 | 4022 | |
952aaa12 | 4023 | /* all tasks in @from are being moved, all csets are source */ |
f0d9a5f1 | 4024 | spin_lock_bh(&css_set_lock); |
952aaa12 TH |
4025 | list_for_each_entry(link, &from->cset_links, cset_link) |
4026 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
f0d9a5f1 | 4027 | spin_unlock_bh(&css_set_lock); |
31a7df01 | 4028 | |
952aaa12 TH |
4029 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
4030 | if (ret) | |
4031 | goto out_err; | |
8cc99345 | 4032 | |
952aaa12 TH |
4033 | /* |
4034 | * Migrate tasks one-by-one until @form is empty. This fails iff | |
4035 | * ->can_attach() fails. | |
4036 | */ | |
e406d1cf | 4037 | do { |
9d800df1 | 4038 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
4039 | task = css_task_iter_next(&it); |
4040 | if (task) | |
4041 | get_task_struct(task); | |
4042 | css_task_iter_end(&it); | |
4043 | ||
4044 | if (task) { | |
9af2ec45 | 4045 | ret = cgroup_migrate(task, false, to); |
e406d1cf TH |
4046 | put_task_struct(task); |
4047 | } | |
4048 | } while (task && !ret); | |
952aaa12 TH |
4049 | out_err: |
4050 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 4051 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 4052 | return ret; |
8cc99345 TH |
4053 | } |
4054 | ||
bbcb81d0 | 4055 | /* |
102a775e | 4056 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
4057 | * |
4058 | * Reading this file can return large amounts of data if a cgroup has | |
4059 | * *lots* of attached tasks. So it may need several calls to read(), | |
4060 | * but we cannot guarantee that the information we produce is correct | |
4061 | * unless we produce it entirely atomically. | |
4062 | * | |
bbcb81d0 | 4063 | */ |
bbcb81d0 | 4064 | |
24528255 LZ |
4065 | /* which pidlist file are we talking about? */ |
4066 | enum cgroup_filetype { | |
4067 | CGROUP_FILE_PROCS, | |
4068 | CGROUP_FILE_TASKS, | |
4069 | }; | |
4070 | ||
4071 | /* | |
4072 | * A pidlist is a list of pids that virtually represents the contents of one | |
4073 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
4074 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
4075 | * to the cgroup. | |
4076 | */ | |
4077 | struct cgroup_pidlist { | |
4078 | /* | |
4079 | * used to find which pidlist is wanted. doesn't change as long as | |
4080 | * this particular list stays in the list. | |
4081 | */ | |
4082 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
4083 | /* array of xids */ | |
4084 | pid_t *list; | |
4085 | /* how many elements the above list has */ | |
4086 | int length; | |
24528255 LZ |
4087 | /* each of these stored in a list by its cgroup */ |
4088 | struct list_head links; | |
4089 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
4090 | struct cgroup *owner; | |
b1a21367 TH |
4091 | /* for delayed destruction */ |
4092 | struct delayed_work destroy_dwork; | |
24528255 LZ |
4093 | }; |
4094 | ||
d1d9fd33 BB |
4095 | /* |
4096 | * The following two functions "fix" the issue where there are more pids | |
4097 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
4098 | * TODO: replace with a kernel-wide solution to this problem | |
4099 | */ | |
4100 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
4101 | static void *pidlist_allocate(int count) | |
4102 | { | |
4103 | if (PIDLIST_TOO_LARGE(count)) | |
4104 | return vmalloc(count * sizeof(pid_t)); | |
4105 | else | |
4106 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
4107 | } | |
b1a21367 | 4108 | |
d1d9fd33 BB |
4109 | static void pidlist_free(void *p) |
4110 | { | |
58794514 | 4111 | kvfree(p); |
d1d9fd33 | 4112 | } |
d1d9fd33 | 4113 | |
b1a21367 TH |
4114 | /* |
4115 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
4116 | * should be left afterwards. | |
4117 | */ | |
4118 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
4119 | { | |
4120 | struct cgroup_pidlist *l, *tmp_l; | |
4121 | ||
4122 | mutex_lock(&cgrp->pidlist_mutex); | |
4123 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
4124 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
4125 | mutex_unlock(&cgrp->pidlist_mutex); | |
4126 | ||
4127 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
4128 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
4129 | } | |
4130 | ||
4131 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
4132 | { | |
4133 | struct delayed_work *dwork = to_delayed_work(work); | |
4134 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
4135 | destroy_dwork); | |
4136 | struct cgroup_pidlist *tofree = NULL; | |
4137 | ||
4138 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
4139 | |
4140 | /* | |
04502365 TH |
4141 | * Destroy iff we didn't get queued again. The state won't change |
4142 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 4143 | */ |
04502365 | 4144 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
4145 | list_del(&l->links); |
4146 | pidlist_free(l->list); | |
4147 | put_pid_ns(l->key.ns); | |
4148 | tofree = l; | |
4149 | } | |
4150 | ||
b1a21367 TH |
4151 | mutex_unlock(&l->owner->pidlist_mutex); |
4152 | kfree(tofree); | |
4153 | } | |
4154 | ||
bbcb81d0 | 4155 | /* |
102a775e | 4156 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 4157 | * Returns the number of unique elements. |
bbcb81d0 | 4158 | */ |
6ee211ad | 4159 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 4160 | { |
102a775e | 4161 | int src, dest = 1; |
102a775e BB |
4162 | |
4163 | /* | |
4164 | * we presume the 0th element is unique, so i starts at 1. trivial | |
4165 | * edge cases first; no work needs to be done for either | |
4166 | */ | |
4167 | if (length == 0 || length == 1) | |
4168 | return length; | |
4169 | /* src and dest walk down the list; dest counts unique elements */ | |
4170 | for (src = 1; src < length; src++) { | |
4171 | /* find next unique element */ | |
4172 | while (list[src] == list[src-1]) { | |
4173 | src++; | |
4174 | if (src == length) | |
4175 | goto after; | |
4176 | } | |
4177 | /* dest always points to where the next unique element goes */ | |
4178 | list[dest] = list[src]; | |
4179 | dest++; | |
4180 | } | |
4181 | after: | |
102a775e BB |
4182 | return dest; |
4183 | } | |
4184 | ||
afb2bc14 TH |
4185 | /* |
4186 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
4187 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
4188 | * different per namespace, each namespace needs differently sorted list, | |
4189 | * making it impossible to use, for example, single rbtree of member tasks | |
4190 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
4191 | * per open file is dangerous, so cgroup had to implement shared pool of | |
4192 | * pidlists keyed by cgroup and namespace. | |
4193 | * | |
4194 | * All this extra complexity was caused by the original implementation | |
4195 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
4196 | * want to do away with it. Explicitly scramble sort order if on the |
4197 | * default hierarchy so that no such expectation exists in the new | |
4198 | * interface. | |
afb2bc14 TH |
4199 | * |
4200 | * Scrambling is done by swapping every two consecutive bits, which is | |
4201 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
4202 | */ | |
4203 | static pid_t pid_fry(pid_t pid) | |
4204 | { | |
4205 | unsigned a = pid & 0x55555555; | |
4206 | unsigned b = pid & 0xAAAAAAAA; | |
4207 | ||
4208 | return (a << 1) | (b >> 1); | |
4209 | } | |
4210 | ||
4211 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
4212 | { | |
aa6ec29b | 4213 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4214 | return pid_fry(pid); |
4215 | else | |
4216 | return pid; | |
4217 | } | |
4218 | ||
102a775e BB |
4219 | static int cmppid(const void *a, const void *b) |
4220 | { | |
4221 | return *(pid_t *)a - *(pid_t *)b; | |
4222 | } | |
4223 | ||
afb2bc14 TH |
4224 | static int fried_cmppid(const void *a, const void *b) |
4225 | { | |
4226 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
4227 | } | |
4228 | ||
e6b81710 TH |
4229 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
4230 | enum cgroup_filetype type) | |
4231 | { | |
4232 | struct cgroup_pidlist *l; | |
4233 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4234 | struct pid_namespace *ns = task_active_pid_ns(current); | |
4235 | ||
4236 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4237 | ||
4238 | list_for_each_entry(l, &cgrp->pidlists, links) | |
4239 | if (l->key.type == type && l->key.ns == ns) | |
4240 | return l; | |
4241 | return NULL; | |
4242 | } | |
4243 | ||
72a8cb30 BB |
4244 | /* |
4245 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
4246 | * returns with the lock on that pidlist already held, and takes care | |
4247 | * of the use count, or returns NULL with no locks held if we're out of | |
4248 | * memory. | |
4249 | */ | |
e6b81710 TH |
4250 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
4251 | enum cgroup_filetype type) | |
72a8cb30 BB |
4252 | { |
4253 | struct cgroup_pidlist *l; | |
b70cc5fd | 4254 | |
e6b81710 TH |
4255 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4256 | ||
4257 | l = cgroup_pidlist_find(cgrp, type); | |
4258 | if (l) | |
4259 | return l; | |
4260 | ||
72a8cb30 | 4261 | /* entry not found; create a new one */ |
f4f4be2b | 4262 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 4263 | if (!l) |
72a8cb30 | 4264 | return l; |
e6b81710 | 4265 | |
b1a21367 | 4266 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 4267 | l->key.type = type; |
e6b81710 TH |
4268 | /* don't need task_nsproxy() if we're looking at ourself */ |
4269 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
4270 | l->owner = cgrp; |
4271 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
4272 | return l; |
4273 | } | |
4274 | ||
102a775e BB |
4275 | /* |
4276 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4277 | */ | |
72a8cb30 BB |
4278 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
4279 | struct cgroup_pidlist **lp) | |
102a775e BB |
4280 | { |
4281 | pid_t *array; | |
4282 | int length; | |
4283 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 4284 | struct css_task_iter it; |
817929ec | 4285 | struct task_struct *tsk; |
102a775e BB |
4286 | struct cgroup_pidlist *l; |
4287 | ||
4bac00d1 TH |
4288 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4289 | ||
102a775e BB |
4290 | /* |
4291 | * If cgroup gets more users after we read count, we won't have | |
4292 | * enough space - tough. This race is indistinguishable to the | |
4293 | * caller from the case that the additional cgroup users didn't | |
4294 | * show up until sometime later on. | |
4295 | */ | |
4296 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 4297 | array = pidlist_allocate(length); |
102a775e BB |
4298 | if (!array) |
4299 | return -ENOMEM; | |
4300 | /* now, populate the array */ | |
9d800df1 | 4301 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4302 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 4303 | if (unlikely(n == length)) |
817929ec | 4304 | break; |
102a775e | 4305 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
4306 | if (type == CGROUP_FILE_PROCS) |
4307 | pid = task_tgid_vnr(tsk); | |
4308 | else | |
4309 | pid = task_pid_vnr(tsk); | |
102a775e BB |
4310 | if (pid > 0) /* make sure to only use valid results */ |
4311 | array[n++] = pid; | |
817929ec | 4312 | } |
72ec7029 | 4313 | css_task_iter_end(&it); |
102a775e BB |
4314 | length = n; |
4315 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 4316 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4317 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
4318 | else | |
4319 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 4320 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 4321 | length = pidlist_uniq(array, length); |
e6b81710 | 4322 | |
e6b81710 | 4323 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 4324 | if (!l) { |
d1d9fd33 | 4325 | pidlist_free(array); |
72a8cb30 | 4326 | return -ENOMEM; |
102a775e | 4327 | } |
e6b81710 TH |
4328 | |
4329 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 4330 | pidlist_free(l->list); |
102a775e BB |
4331 | l->list = array; |
4332 | l->length = length; | |
72a8cb30 | 4333 | *lp = l; |
102a775e | 4334 | return 0; |
bbcb81d0 PM |
4335 | } |
4336 | ||
846c7bb0 | 4337 | /** |
a043e3b2 | 4338 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4339 | * @stats: cgroupstats to fill information into |
4340 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4341 | * been requested. | |
a043e3b2 LZ |
4342 | * |
4343 | * Build and fill cgroupstats so that taskstats can export it to user | |
4344 | * space. | |
846c7bb0 BS |
4345 | */ |
4346 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4347 | { | |
2bd59d48 | 4348 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4349 | struct cgroup *cgrp; |
72ec7029 | 4350 | struct css_task_iter it; |
846c7bb0 | 4351 | struct task_struct *tsk; |
33d283be | 4352 | |
2bd59d48 TH |
4353 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4354 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4355 | kernfs_type(kn) != KERNFS_DIR) | |
4356 | return -EINVAL; | |
4357 | ||
bad34660 LZ |
4358 | mutex_lock(&cgroup_mutex); |
4359 | ||
846c7bb0 | 4360 | /* |
2bd59d48 | 4361 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4362 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4363 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4364 | */ |
2bd59d48 TH |
4365 | rcu_read_lock(); |
4366 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4367 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4368 | rcu_read_unlock(); |
bad34660 | 4369 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4370 | return -ENOENT; |
4371 | } | |
bad34660 | 4372 | rcu_read_unlock(); |
846c7bb0 | 4373 | |
9d800df1 | 4374 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4375 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4376 | switch (tsk->state) { |
4377 | case TASK_RUNNING: | |
4378 | stats->nr_running++; | |
4379 | break; | |
4380 | case TASK_INTERRUPTIBLE: | |
4381 | stats->nr_sleeping++; | |
4382 | break; | |
4383 | case TASK_UNINTERRUPTIBLE: | |
4384 | stats->nr_uninterruptible++; | |
4385 | break; | |
4386 | case TASK_STOPPED: | |
4387 | stats->nr_stopped++; | |
4388 | break; | |
4389 | default: | |
4390 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4391 | stats->nr_io_wait++; | |
4392 | break; | |
4393 | } | |
4394 | } | |
72ec7029 | 4395 | css_task_iter_end(&it); |
846c7bb0 | 4396 | |
bad34660 | 4397 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4398 | return 0; |
846c7bb0 BS |
4399 | } |
4400 | ||
8f3ff208 | 4401 | |
bbcb81d0 | 4402 | /* |
102a775e | 4403 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4404 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4405 | * in the cgroup->l->list array. |
bbcb81d0 | 4406 | */ |
cc31edce | 4407 | |
102a775e | 4408 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4409 | { |
cc31edce PM |
4410 | /* |
4411 | * Initially we receive a position value that corresponds to | |
4412 | * one more than the last pid shown (or 0 on the first call or | |
4413 | * after a seek to the start). Use a binary-search to find the | |
4414 | * next pid to display, if any | |
4415 | */ | |
2bd59d48 | 4416 | struct kernfs_open_file *of = s->private; |
7da11279 | 4417 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4418 | struct cgroup_pidlist *l; |
7da11279 | 4419 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4420 | int index = 0, pid = *pos; |
4bac00d1 TH |
4421 | int *iter, ret; |
4422 | ||
4423 | mutex_lock(&cgrp->pidlist_mutex); | |
4424 | ||
4425 | /* | |
5d22444f | 4426 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4427 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4428 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4429 | * could already have been destroyed. |
4430 | */ | |
5d22444f TH |
4431 | if (of->priv) |
4432 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4433 | |
4434 | /* | |
4435 | * Either this is the first start() after open or the matching | |
4436 | * pidlist has been destroyed inbetween. Create a new one. | |
4437 | */ | |
5d22444f TH |
4438 | if (!of->priv) { |
4439 | ret = pidlist_array_load(cgrp, type, | |
4440 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4441 | if (ret) |
4442 | return ERR_PTR(ret); | |
4443 | } | |
5d22444f | 4444 | l = of->priv; |
cc31edce | 4445 | |
cc31edce | 4446 | if (pid) { |
102a775e | 4447 | int end = l->length; |
20777766 | 4448 | |
cc31edce PM |
4449 | while (index < end) { |
4450 | int mid = (index + end) / 2; | |
afb2bc14 | 4451 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4452 | index = mid; |
4453 | break; | |
afb2bc14 | 4454 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4455 | index = mid + 1; |
4456 | else | |
4457 | end = mid; | |
4458 | } | |
4459 | } | |
4460 | /* If we're off the end of the array, we're done */ | |
102a775e | 4461 | if (index >= l->length) |
cc31edce PM |
4462 | return NULL; |
4463 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4464 | iter = l->list + index; |
afb2bc14 | 4465 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4466 | return iter; |
4467 | } | |
4468 | ||
102a775e | 4469 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4470 | { |
2bd59d48 | 4471 | struct kernfs_open_file *of = s->private; |
5d22444f | 4472 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4473 | |
5d22444f TH |
4474 | if (l) |
4475 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4476 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4477 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4478 | } |
4479 | ||
102a775e | 4480 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4481 | { |
2bd59d48 | 4482 | struct kernfs_open_file *of = s->private; |
5d22444f | 4483 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4484 | pid_t *p = v; |
4485 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4486 | /* |
4487 | * Advance to the next pid in the array. If this goes off the | |
4488 | * end, we're done | |
4489 | */ | |
4490 | p++; | |
4491 | if (p >= end) { | |
4492 | return NULL; | |
4493 | } else { | |
7da11279 | 4494 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4495 | return p; |
4496 | } | |
4497 | } | |
4498 | ||
102a775e | 4499 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce | 4500 | { |
94ff212d JP |
4501 | seq_printf(s, "%d\n", *(int *)v); |
4502 | ||
4503 | return 0; | |
cc31edce | 4504 | } |
bbcb81d0 | 4505 | |
182446d0 TH |
4506 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4507 | struct cftype *cft) | |
81a6a5cd | 4508 | { |
182446d0 | 4509 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4510 | } |
4511 | ||
182446d0 TH |
4512 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4513 | struct cftype *cft, u64 val) | |
6379c106 | 4514 | { |
6379c106 | 4515 | if (val) |
182446d0 | 4516 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4517 | else |
182446d0 | 4518 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4519 | return 0; |
4520 | } | |
4521 | ||
182446d0 TH |
4522 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4523 | struct cftype *cft) | |
97978e6d | 4524 | { |
182446d0 | 4525 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4526 | } |
4527 | ||
182446d0 TH |
4528 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4529 | struct cftype *cft, u64 val) | |
97978e6d DL |
4530 | { |
4531 | if (val) | |
182446d0 | 4532 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4533 | else |
182446d0 | 4534 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4535 | return 0; |
4536 | } | |
4537 | ||
a14c6874 TH |
4538 | /* cgroup core interface files for the default hierarchy */ |
4539 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4540 | { |
d5c56ced | 4541 | .name = "cgroup.procs", |
6f60eade | 4542 | .file_offset = offsetof(struct cgroup, procs_file), |
6612f05b TH |
4543 | .seq_start = cgroup_pidlist_start, |
4544 | .seq_next = cgroup_pidlist_next, | |
4545 | .seq_stop = cgroup_pidlist_stop, | |
4546 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4547 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4548 | .write = cgroup_procs_write, |
102a775e | 4549 | }, |
f8f22e53 TH |
4550 | { |
4551 | .name = "cgroup.controllers", | |
a14c6874 | 4552 | .flags = CFTYPE_ONLY_ON_ROOT, |
f8f22e53 TH |
4553 | .seq_show = cgroup_root_controllers_show, |
4554 | }, | |
4555 | { | |
4556 | .name = "cgroup.controllers", | |
a14c6874 | 4557 | .flags = CFTYPE_NOT_ON_ROOT, |
f8f22e53 TH |
4558 | .seq_show = cgroup_controllers_show, |
4559 | }, | |
4560 | { | |
4561 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4562 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4563 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4564 | }, |
842b597e | 4565 | { |
4a07c222 | 4566 | .name = "cgroup.events", |
a14c6874 | 4567 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4568 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4569 | .seq_show = cgroup_events_show, |
842b597e | 4570 | }, |
a14c6874 TH |
4571 | { } /* terminate */ |
4572 | }; | |
d5c56ced | 4573 | |
a14c6874 TH |
4574 | /* cgroup core interface files for the legacy hierarchies */ |
4575 | static struct cftype cgroup_legacy_base_files[] = { | |
4576 | { | |
4577 | .name = "cgroup.procs", | |
4578 | .seq_start = cgroup_pidlist_start, | |
4579 | .seq_next = cgroup_pidlist_next, | |
4580 | .seq_stop = cgroup_pidlist_stop, | |
4581 | .seq_show = cgroup_pidlist_show, | |
4582 | .private = CGROUP_FILE_PROCS, | |
4583 | .write = cgroup_procs_write, | |
a14c6874 TH |
4584 | }, |
4585 | { | |
4586 | .name = "cgroup.clone_children", | |
4587 | .read_u64 = cgroup_clone_children_read, | |
4588 | .write_u64 = cgroup_clone_children_write, | |
4589 | }, | |
4590 | { | |
4591 | .name = "cgroup.sane_behavior", | |
4592 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4593 | .seq_show = cgroup_sane_behavior_show, | |
4594 | }, | |
d5c56ced TH |
4595 | { |
4596 | .name = "tasks", | |
6612f05b TH |
4597 | .seq_start = cgroup_pidlist_start, |
4598 | .seq_next = cgroup_pidlist_next, | |
4599 | .seq_stop = cgroup_pidlist_stop, | |
4600 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4601 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4602 | .write = cgroup_tasks_write, |
d5c56ced TH |
4603 | }, |
4604 | { | |
4605 | .name = "notify_on_release", | |
d5c56ced TH |
4606 | .read_u64 = cgroup_read_notify_on_release, |
4607 | .write_u64 = cgroup_write_notify_on_release, | |
4608 | }, | |
6e6ff25b TH |
4609 | { |
4610 | .name = "release_agent", | |
a14c6874 | 4611 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4612 | .seq_show = cgroup_release_agent_show, |
451af504 | 4613 | .write = cgroup_release_agent_write, |
5f469907 | 4614 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4615 | }, |
db0416b6 | 4616 | { } /* terminate */ |
bbcb81d0 PM |
4617 | }; |
4618 | ||
0c21ead1 TH |
4619 | /* |
4620 | * css destruction is four-stage process. | |
4621 | * | |
4622 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4623 | * Implemented in kill_css(). | |
4624 | * | |
4625 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4626 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4627 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4628 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4629 | * |
4630 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4631 | * accessors are inside RCU read sections. css_release() schedules the | |
4632 | * RCU callback. | |
4633 | * | |
4634 | * 4. After the grace period, the css can be freed. Implemented in | |
4635 | * css_free_work_fn(). | |
4636 | * | |
4637 | * It is actually hairier because both step 2 and 4 require process context | |
4638 | * and thus involve punting to css->destroy_work adding two additional | |
4639 | * steps to the already complex sequence. | |
4640 | */ | |
35ef10da | 4641 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4642 | { |
4643 | struct cgroup_subsys_state *css = | |
35ef10da | 4644 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4645 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4646 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4647 | |
9a1049da TH |
4648 | percpu_ref_exit(&css->refcnt); |
4649 | ||
01e58659 | 4650 | if (ss) { |
9d755d33 | 4651 | /* css free path */ |
01e58659 VD |
4652 | int id = css->id; |
4653 | ||
9d755d33 TH |
4654 | if (css->parent) |
4655 | css_put(css->parent); | |
0ae78e0b | 4656 | |
01e58659 VD |
4657 | ss->css_free(css); |
4658 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 TH |
4659 | cgroup_put(cgrp); |
4660 | } else { | |
4661 | /* cgroup free path */ | |
4662 | atomic_dec(&cgrp->root->nr_cgrps); | |
4663 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 4664 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4665 | |
d51f39b0 | 4666 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4667 | /* |
4668 | * We get a ref to the parent, and put the ref when | |
4669 | * this cgroup is being freed, so it's guaranteed | |
4670 | * that the parent won't be destroyed before its | |
4671 | * children. | |
4672 | */ | |
d51f39b0 | 4673 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4674 | kernfs_put(cgrp->kn); |
4675 | kfree(cgrp); | |
4676 | } else { | |
4677 | /* | |
4678 | * This is root cgroup's refcnt reaching zero, | |
4679 | * which indicates that the root should be | |
4680 | * released. | |
4681 | */ | |
4682 | cgroup_destroy_root(cgrp->root); | |
4683 | } | |
4684 | } | |
48ddbe19 TH |
4685 | } |
4686 | ||
0c21ead1 | 4687 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4688 | { |
4689 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4690 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4691 | |
35ef10da | 4692 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4693 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4694 | } |
4695 | ||
25e15d83 | 4696 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4697 | { |
4698 | struct cgroup_subsys_state *css = | |
25e15d83 | 4699 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4700 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4701 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4702 | |
1fed1b2e TH |
4703 | mutex_lock(&cgroup_mutex); |
4704 | ||
de3f0341 | 4705 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4706 | list_del_rcu(&css->sibling); |
4707 | ||
9d755d33 TH |
4708 | if (ss) { |
4709 | /* css release path */ | |
01e58659 | 4710 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
4711 | if (ss->css_released) |
4712 | ss->css_released(css); | |
9d755d33 TH |
4713 | } else { |
4714 | /* cgroup release path */ | |
9d755d33 TH |
4715 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4716 | cgrp->id = -1; | |
a4189487 LZ |
4717 | |
4718 | /* | |
4719 | * There are two control paths which try to determine | |
4720 | * cgroup from dentry without going through kernfs - | |
4721 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4722 | * Those are supported by RCU protecting clearing of | |
4723 | * cgrp->kn->priv backpointer. | |
4724 | */ | |
4725 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); | |
9d755d33 | 4726 | } |
d3daf28d | 4727 | |
1fed1b2e TH |
4728 | mutex_unlock(&cgroup_mutex); |
4729 | ||
0c21ead1 | 4730 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4731 | } |
4732 | ||
d3daf28d TH |
4733 | static void css_release(struct percpu_ref *ref) |
4734 | { | |
4735 | struct cgroup_subsys_state *css = | |
4736 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4737 | ||
25e15d83 TH |
4738 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
4739 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
4740 | } |
4741 | ||
ddfcadab TH |
4742 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4743 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4744 | { |
0cb51d71 TH |
4745 | lockdep_assert_held(&cgroup_mutex); |
4746 | ||
ddfcadab TH |
4747 | cgroup_get(cgrp); |
4748 | ||
d5c419b6 | 4749 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4750 | css->cgroup = cgrp; |
72c97e54 | 4751 | css->ss = ss; |
d5c419b6 TH |
4752 | INIT_LIST_HEAD(&css->sibling); |
4753 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 4754 | css->serial_nr = css_serial_nr_next++; |
0ae78e0b | 4755 | |
d51f39b0 TH |
4756 | if (cgroup_parent(cgrp)) { |
4757 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4758 | css_get(css->parent); |
ddfcadab | 4759 | } |
48ddbe19 | 4760 | |
ca8bdcaf | 4761 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4762 | } |
4763 | ||
2a4ac633 | 4764 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4765 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4766 | { |
623f926b | 4767 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4768 | int ret = 0; |
4769 | ||
a31f2d3f TH |
4770 | lockdep_assert_held(&cgroup_mutex); |
4771 | ||
92fb9748 | 4772 | if (ss->css_online) |
eb95419b | 4773 | ret = ss->css_online(css); |
ae7f164a | 4774 | if (!ret) { |
eb95419b | 4775 | css->flags |= CSS_ONLINE; |
aec25020 | 4776 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 4777 | } |
b1929db4 | 4778 | return ret; |
a31f2d3f TH |
4779 | } |
4780 | ||
2a4ac633 | 4781 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4782 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4783 | { |
623f926b | 4784 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4785 | |
4786 | lockdep_assert_held(&cgroup_mutex); | |
4787 | ||
4788 | if (!(css->flags & CSS_ONLINE)) | |
4789 | return; | |
4790 | ||
d7eeac19 | 4791 | if (ss->css_offline) |
eb95419b | 4792 | ss->css_offline(css); |
a31f2d3f | 4793 | |
eb95419b | 4794 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4795 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4796 | |
4797 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4798 | } |
4799 | ||
c81c925a TH |
4800 | /** |
4801 | * create_css - create a cgroup_subsys_state | |
4802 | * @cgrp: the cgroup new css will be associated with | |
4803 | * @ss: the subsys of new css | |
f63070d3 | 4804 | * @visible: whether to create control knobs for the new css or not |
c81c925a TH |
4805 | * |
4806 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
f63070d3 TH |
4807 | * css is online and installed in @cgrp with all interface files created if |
4808 | * @visible. Returns 0 on success, -errno on failure. | |
c81c925a | 4809 | */ |
f63070d3 TH |
4810 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
4811 | bool visible) | |
c81c925a | 4812 | { |
d51f39b0 | 4813 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4814 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4815 | struct cgroup_subsys_state *css; |
4816 | int err; | |
4817 | ||
c81c925a TH |
4818 | lockdep_assert_held(&cgroup_mutex); |
4819 | ||
1fed1b2e | 4820 | css = ss->css_alloc(parent_css); |
c81c925a TH |
4821 | if (IS_ERR(css)) |
4822 | return PTR_ERR(css); | |
4823 | ||
ddfcadab | 4824 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4825 | |
2aad2a86 | 4826 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 4827 | if (err) |
3eb59ec6 | 4828 | goto err_free_css; |
c81c925a | 4829 | |
cf780b7d | 4830 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 TH |
4831 | if (err < 0) |
4832 | goto err_free_percpu_ref; | |
4833 | css->id = err; | |
c81c925a | 4834 | |
f63070d3 | 4835 | if (visible) { |
4df8dc90 | 4836 | err = css_populate_dir(css, NULL); |
f63070d3 TH |
4837 | if (err) |
4838 | goto err_free_id; | |
4839 | } | |
15a4c835 TH |
4840 | |
4841 | /* @css is ready to be brought online now, make it visible */ | |
1fed1b2e | 4842 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4843 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4844 | |
4845 | err = online_css(css); | |
4846 | if (err) | |
1fed1b2e | 4847 | goto err_list_del; |
94419627 | 4848 | |
c81c925a | 4849 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4850 | cgroup_parent(parent)) { |
ed3d261b | 4851 | 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 | 4852 | current->comm, current->pid, ss->name); |
c81c925a | 4853 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4854 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4855 | ss->warned_broken_hierarchy = true; |
4856 | } | |
4857 | ||
4858 | return 0; | |
4859 | ||
1fed1b2e TH |
4860 | err_list_del: |
4861 | list_del_rcu(&css->sibling); | |
4df8dc90 | 4862 | css_clear_dir(css, NULL); |
15a4c835 TH |
4863 | err_free_id: |
4864 | cgroup_idr_remove(&ss->css_idr, css->id); | |
3eb59ec6 | 4865 | err_free_percpu_ref: |
9a1049da | 4866 | percpu_ref_exit(&css->refcnt); |
3eb59ec6 | 4867 | err_free_css: |
a2bed820 | 4868 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
c81c925a TH |
4869 | return err; |
4870 | } | |
4871 | ||
b3bfd983 TH |
4872 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
4873 | umode_t mode) | |
ddbcc7e8 | 4874 | { |
a9746d8d TH |
4875 | struct cgroup *parent, *cgrp; |
4876 | struct cgroup_root *root; | |
ddbcc7e8 | 4877 | struct cgroup_subsys *ss; |
2bd59d48 | 4878 | struct kernfs_node *kn; |
b3bfd983 | 4879 | int ssid, ret; |
ddbcc7e8 | 4880 | |
71b1fb5c AC |
4881 | /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable. |
4882 | */ | |
4883 | if (strchr(name, '\n')) | |
4884 | return -EINVAL; | |
4885 | ||
a9746d8d TH |
4886 | parent = cgroup_kn_lock_live(parent_kn); |
4887 | if (!parent) | |
4888 | return -ENODEV; | |
4889 | root = parent->root; | |
ddbcc7e8 | 4890 | |
0a950f65 | 4891 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc | 4892 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
ba0f4d76 TH |
4893 | if (!cgrp) { |
4894 | ret = -ENOMEM; | |
4895 | goto out_unlock; | |
0ab02ca8 LZ |
4896 | } |
4897 | ||
2aad2a86 | 4898 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
4899 | if (ret) |
4900 | goto out_free_cgrp; | |
4901 | ||
0ab02ca8 LZ |
4902 | /* |
4903 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4904 | * a half-baked cgroup. | |
4905 | */ | |
cf780b7d | 4906 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 4907 | if (cgrp->id < 0) { |
ba0f4d76 | 4908 | ret = -ENOMEM; |
9d755d33 | 4909 | goto out_cancel_ref; |
976c06bc TH |
4910 | } |
4911 | ||
cc31edce | 4912 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4913 | |
9d800df1 | 4914 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 4915 | cgrp->root = root; |
ddbcc7e8 | 4916 | |
b6abdb0e LZ |
4917 | if (notify_on_release(parent)) |
4918 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4919 | ||
2260e7fc TH |
4920 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4921 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4922 | |
2bd59d48 | 4923 | /* create the directory */ |
e61734c5 | 4924 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 | 4925 | if (IS_ERR(kn)) { |
ba0f4d76 TH |
4926 | ret = PTR_ERR(kn); |
4927 | goto out_free_id; | |
2bd59d48 TH |
4928 | } |
4929 | cgrp->kn = kn; | |
ddbcc7e8 | 4930 | |
4e139afc | 4931 | /* |
6f30558f TH |
4932 | * This extra ref will be put in cgroup_free_fn() and guarantees |
4933 | * that @cgrp->kn is always accessible. | |
4e139afc | 4934 | */ |
6f30558f | 4935 | kernfs_get(kn); |
ddbcc7e8 | 4936 | |
0cb51d71 | 4937 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 4938 | |
4e139afc | 4939 | /* allocation complete, commit to creation */ |
d5c419b6 | 4940 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 4941 | atomic_inc(&root->nr_cgrps); |
59f5296b | 4942 | cgroup_get(parent); |
415cf07a | 4943 | |
0d80255e TH |
4944 | /* |
4945 | * @cgrp is now fully operational. If something fails after this | |
4946 | * point, it'll be released via the normal destruction path. | |
4947 | */ | |
6fa4918d | 4948 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 4949 | |
ba0f4d76 TH |
4950 | ret = cgroup_kn_set_ugid(kn); |
4951 | if (ret) | |
4952 | goto out_destroy; | |
49957f8e | 4953 | |
4df8dc90 | 4954 | ret = css_populate_dir(&cgrp->self, NULL); |
ba0f4d76 TH |
4955 | if (ret) |
4956 | goto out_destroy; | |
628f7cd4 | 4957 | |
9d403e99 | 4958 | /* let's create and online css's */ |
b85d2040 | 4959 | for_each_subsys(ss, ssid) { |
f392e51c | 4960 | if (parent->child_subsys_mask & (1 << ssid)) { |
f63070d3 TH |
4961 | ret = create_css(cgrp, ss, |
4962 | parent->subtree_control & (1 << ssid)); | |
ba0f4d76 TH |
4963 | if (ret) |
4964 | goto out_destroy; | |
b85d2040 | 4965 | } |
a8638030 | 4966 | } |
ddbcc7e8 | 4967 | |
bd53d617 TH |
4968 | /* |
4969 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 4970 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 4971 | */ |
667c2491 TH |
4972 | if (!cgroup_on_dfl(cgrp)) { |
4973 | cgrp->subtree_control = parent->subtree_control; | |
4974 | cgroup_refresh_child_subsys_mask(cgrp); | |
4975 | } | |
2bd59d48 | 4976 | |
2bd59d48 | 4977 | kernfs_activate(kn); |
ddbcc7e8 | 4978 | |
ba0f4d76 TH |
4979 | ret = 0; |
4980 | goto out_unlock; | |
ddbcc7e8 | 4981 | |
ba0f4d76 | 4982 | out_free_id: |
6fa4918d | 4983 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); |
9d755d33 | 4984 | out_cancel_ref: |
9a1049da | 4985 | percpu_ref_exit(&cgrp->self.refcnt); |
ba0f4d76 | 4986 | out_free_cgrp: |
bd89aabc | 4987 | kfree(cgrp); |
ba0f4d76 | 4988 | out_unlock: |
a9746d8d | 4989 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 4990 | return ret; |
4b8b47eb | 4991 | |
ba0f4d76 | 4992 | out_destroy: |
4b8b47eb | 4993 | cgroup_destroy_locked(cgrp); |
ba0f4d76 | 4994 | goto out_unlock; |
ddbcc7e8 PM |
4995 | } |
4996 | ||
223dbc38 TH |
4997 | /* |
4998 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
4999 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5000 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5001 | */ |
5002 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5003 | { |
223dbc38 TH |
5004 | struct cgroup_subsys_state *css = |
5005 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5006 | |
f20104de | 5007 | mutex_lock(&cgroup_mutex); |
09a503ea | 5008 | offline_css(css); |
f20104de | 5009 | mutex_unlock(&cgroup_mutex); |
09a503ea | 5010 | |
09a503ea | 5011 | css_put(css); |
d3daf28d TH |
5012 | } |
5013 | ||
223dbc38 TH |
5014 | /* css kill confirmation processing requires process context, bounce */ |
5015 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5016 | { |
5017 | struct cgroup_subsys_state *css = | |
5018 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5019 | ||
223dbc38 | 5020 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 5021 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
5022 | } |
5023 | ||
f392e51c TH |
5024 | /** |
5025 | * kill_css - destroy a css | |
5026 | * @css: css to destroy | |
5027 | * | |
5028 | * This function initiates destruction of @css by removing cgroup interface | |
5029 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5030 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5031 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5032 | */ |
5033 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5034 | { |
01f6474c | 5035 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5036 | |
2bd59d48 TH |
5037 | /* |
5038 | * This must happen before css is disassociated with its cgroup. | |
5039 | * See seq_css() for details. | |
5040 | */ | |
4df8dc90 | 5041 | css_clear_dir(css, NULL); |
3c14f8b4 | 5042 | |
edae0c33 TH |
5043 | /* |
5044 | * Killing would put the base ref, but we need to keep it alive | |
5045 | * until after ->css_offline(). | |
5046 | */ | |
5047 | css_get(css); | |
5048 | ||
5049 | /* | |
5050 | * cgroup core guarantees that, by the time ->css_offline() is | |
5051 | * invoked, no new css reference will be given out via | |
ec903c0c | 5052 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5053 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5054 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5055 | * | |
5056 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5057 | * css is confirmed to be seen as killed on all CPUs. | |
5058 | */ | |
5059 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5060 | } |
5061 | ||
5062 | /** | |
5063 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5064 | * @cgrp: cgroup to be destroyed | |
5065 | * | |
5066 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5067 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5068 | * guarantee that css_tryget_online() won't succeed by the time |
5069 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5070 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5071 | * |
5072 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5073 | * userland visible parts and start killing the percpu refcnts of | |
5074 | * css's. Set up so that the next stage will be kicked off once all | |
5075 | * the percpu refcnts are confirmed to be killed. | |
5076 | * | |
5077 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5078 | * rest of destruction. Once all cgroup references are gone, the | |
5079 | * cgroup is RCU-freed. | |
5080 | * | |
5081 | * This function implements s1. After this step, @cgrp is gone as far as | |
5082 | * the userland is concerned and a new cgroup with the same name may be | |
5083 | * created. As cgroup doesn't care about the names internally, this | |
5084 | * doesn't cause any problem. | |
5085 | */ | |
42809dd4 TH |
5086 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5087 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5088 | { |
2bd59d48 | 5089 | struct cgroup_subsys_state *css; |
1c6727af | 5090 | int ssid; |
ddbcc7e8 | 5091 | |
42809dd4 TH |
5092 | lockdep_assert_held(&cgroup_mutex); |
5093 | ||
91486f61 TH |
5094 | /* |
5095 | * Only migration can raise populated from zero and we're already | |
5096 | * holding cgroup_mutex. | |
5097 | */ | |
5098 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5099 | return -EBUSY; |
a043e3b2 | 5100 | |
bb78a92f | 5101 | /* |
d5c419b6 TH |
5102 | * Make sure there's no live children. We can't test emptiness of |
5103 | * ->self.children as dead children linger on it while being | |
5104 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5105 | */ |
f3d46500 | 5106 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5107 | return -EBUSY; |
5108 | ||
455050d2 TH |
5109 | /* |
5110 | * Mark @cgrp dead. This prevents further task migration and child | |
de3f0341 | 5111 | * creation by disabling cgroup_lock_live_group(). |
455050d2 | 5112 | */ |
184faf32 | 5113 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5114 | |
249f3468 | 5115 | /* initiate massacre of all css's */ |
1c6727af TH |
5116 | for_each_css(css, ssid, cgrp) |
5117 | kill_css(css); | |
455050d2 | 5118 | |
455050d2 | 5119 | /* |
01f6474c TH |
5120 | * Remove @cgrp directory along with the base files. @cgrp has an |
5121 | * extra ref on its kn. | |
f20104de | 5122 | */ |
01f6474c | 5123 | kernfs_remove(cgrp->kn); |
f20104de | 5124 | |
d51f39b0 | 5125 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 5126 | |
249f3468 | 5127 | /* put the base reference */ |
9d755d33 | 5128 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5129 | |
ea15f8cc TH |
5130 | return 0; |
5131 | }; | |
5132 | ||
2bd59d48 | 5133 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5134 | { |
a9746d8d | 5135 | struct cgroup *cgrp; |
2bd59d48 | 5136 | int ret = 0; |
42809dd4 | 5137 | |
a9746d8d TH |
5138 | cgrp = cgroup_kn_lock_live(kn); |
5139 | if (!cgrp) | |
5140 | return 0; | |
42809dd4 | 5141 | |
a9746d8d | 5142 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 5143 | |
a9746d8d | 5144 | cgroup_kn_unlock(kn); |
42809dd4 | 5145 | return ret; |
8e3f6541 TH |
5146 | } |
5147 | ||
2bd59d48 TH |
5148 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5149 | .remount_fs = cgroup_remount, | |
5150 | .show_options = cgroup_show_options, | |
5151 | .mkdir = cgroup_mkdir, | |
5152 | .rmdir = cgroup_rmdir, | |
5153 | .rename = cgroup_rename, | |
5154 | }; | |
5155 | ||
15a4c835 | 5156 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5157 | { |
ddbcc7e8 | 5158 | struct cgroup_subsys_state *css; |
cfe36bde DC |
5159 | |
5160 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 5161 | |
648bb56d TH |
5162 | mutex_lock(&cgroup_mutex); |
5163 | ||
15a4c835 | 5164 | idr_init(&ss->css_idr); |
0adb0704 | 5165 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5166 | |
3dd06ffa TH |
5167 | /* Create the root cgroup state for this subsystem */ |
5168 | ss->root = &cgrp_dfl_root; | |
5169 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5170 | /* We don't handle early failures gracefully */ |
5171 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5172 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5173 | |
5174 | /* | |
5175 | * Root csses are never destroyed and we can't initialize | |
5176 | * percpu_ref during early init. Disable refcnting. | |
5177 | */ | |
5178 | css->flags |= CSS_NO_REF; | |
5179 | ||
15a4c835 | 5180 | if (early) { |
9395a450 | 5181 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5182 | css->id = 1; |
5183 | } else { | |
5184 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5185 | BUG_ON(css->id < 0); | |
5186 | } | |
ddbcc7e8 | 5187 | |
e8d55fde | 5188 | /* Update the init_css_set to contain a subsys |
817929ec | 5189 | * pointer to this state - since the subsystem is |
e8d55fde | 5190 | * newly registered, all tasks and hence the |
3dd06ffa | 5191 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5192 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5193 | |
cb4a3167 AS |
5194 | have_fork_callback |= (bool)ss->fork << ss->id; |
5195 | have_exit_callback |= (bool)ss->exit << ss->id; | |
afcf6c8b | 5196 | have_free_callback |= (bool)ss->free << ss->id; |
7e47682e | 5197 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5198 | |
e8d55fde LZ |
5199 | /* At system boot, before all subsystems have been |
5200 | * registered, no tasks have been forked, so we don't | |
5201 | * need to invoke fork callbacks here. */ | |
5202 | BUG_ON(!list_empty(&init_task.tasks)); | |
5203 | ||
ae7f164a | 5204 | BUG_ON(online_css(css)); |
a8638030 | 5205 | |
cf5d5941 BB |
5206 | mutex_unlock(&cgroup_mutex); |
5207 | } | |
cf5d5941 | 5208 | |
ddbcc7e8 | 5209 | /** |
a043e3b2 LZ |
5210 | * cgroup_init_early - cgroup initialization at system boot |
5211 | * | |
5212 | * Initialize cgroups at system boot, and initialize any | |
5213 | * subsystems that request early init. | |
ddbcc7e8 PM |
5214 | */ |
5215 | int __init cgroup_init_early(void) | |
5216 | { | |
7b9a6ba5 | 5217 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5218 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5219 | int i; |
30159ec7 | 5220 | |
3dd06ffa | 5221 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5222 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5223 | ||
a4ea1cc9 | 5224 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5225 | |
3ed80a62 | 5226 | for_each_subsys(ss, i) { |
aec25020 | 5227 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
5228 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
5229 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 5230 | ss->id, ss->name); |
073219e9 TH |
5231 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5232 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5233 | ||
aec25020 | 5234 | ss->id = i; |
073219e9 | 5235 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5236 | if (!ss->legacy_name) |
5237 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5238 | |
5239 | if (ss->early_init) | |
15a4c835 | 5240 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5241 | } |
5242 | return 0; | |
5243 | } | |
5244 | ||
a3e72739 TH |
5245 | static unsigned long cgroup_disable_mask __initdata; |
5246 | ||
ddbcc7e8 | 5247 | /** |
a043e3b2 LZ |
5248 | * cgroup_init - cgroup initialization |
5249 | * | |
5250 | * Register cgroup filesystem and /proc file, and initialize | |
5251 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5252 | */ |
5253 | int __init cgroup_init(void) | |
5254 | { | |
30159ec7 | 5255 | struct cgroup_subsys *ss; |
0ac801fe | 5256 | unsigned long key; |
035f4f51 | 5257 | int ssid; |
ddbcc7e8 | 5258 | |
1ed13287 | 5259 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
a14c6874 TH |
5260 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
5261 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 5262 | |
54e7b4eb | 5263 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5264 | |
82fe9b0d TH |
5265 | /* Add init_css_set to the hash table */ |
5266 | key = css_set_hash(init_css_set.subsys); | |
5267 | hash_add(css_set_table, &init_css_set.hlist, key); | |
5268 | ||
3dd06ffa | 5269 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5270 | |
54e7b4eb TH |
5271 | mutex_unlock(&cgroup_mutex); |
5272 | ||
172a2c06 | 5273 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5274 | if (ss->early_init) { |
5275 | struct cgroup_subsys_state *css = | |
5276 | init_css_set.subsys[ss->id]; | |
5277 | ||
5278 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5279 | GFP_KERNEL); | |
5280 | BUG_ON(css->id < 0); | |
5281 | } else { | |
5282 | cgroup_init_subsys(ss, false); | |
5283 | } | |
172a2c06 | 5284 | |
2d8f243a TH |
5285 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5286 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5287 | |
5288 | /* | |
c731ae1d LZ |
5289 | * Setting dfl_root subsys_mask needs to consider the |
5290 | * disabled flag and cftype registration needs kmalloc, | |
5291 | * both of which aren't available during early_init. | |
172a2c06 | 5292 | */ |
a3e72739 TH |
5293 | if (cgroup_disable_mask & (1 << ssid)) { |
5294 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5295 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5296 | ss->name); | |
a8ddc821 | 5297 | continue; |
a3e72739 | 5298 | } |
a8ddc821 TH |
5299 | |
5300 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; | |
5301 | ||
5de4fa13 TH |
5302 | if (!ss->dfl_cftypes) |
5303 | cgrp_dfl_root_inhibit_ss_mask |= 1 << ss->id; | |
5304 | ||
a8ddc821 TH |
5305 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5306 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5307 | } else { | |
5308 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5309 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5310 | } |
295458e6 VD |
5311 | |
5312 | if (ss->bind) | |
5313 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GK |
5314 | } |
5315 | ||
035f4f51 TH |
5316 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5317 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
5318 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); | |
ddbcc7e8 | 5319 | |
2bd59d48 | 5320 | return 0; |
ddbcc7e8 | 5321 | } |
b4f48b63 | 5322 | |
e5fca243 TH |
5323 | static int __init cgroup_wq_init(void) |
5324 | { | |
5325 | /* | |
5326 | * There isn't much point in executing destruction path in | |
5327 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5328 | * Use 1 for @max_active. |
e5fca243 TH |
5329 | * |
5330 | * We would prefer to do this in cgroup_init() above, but that | |
5331 | * is called before init_workqueues(): so leave this until after. | |
5332 | */ | |
1a11533f | 5333 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5334 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5335 | |
5336 | /* | |
5337 | * Used to destroy pidlists and separate to serve as flush domain. | |
5338 | * Cap @max_active to 1 too. | |
5339 | */ | |
5340 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5341 | 0, 1); | |
5342 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5343 | ||
e5fca243 TH |
5344 | return 0; |
5345 | } | |
5346 | core_initcall(cgroup_wq_init); | |
5347 | ||
a424316c PM |
5348 | /* |
5349 | * proc_cgroup_show() | |
5350 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5351 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5352 | */ |
006f4ac4 ZL |
5353 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5354 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5355 | { |
e61734c5 | 5356 | char *buf, *path; |
a424316c | 5357 | int retval; |
3dd06ffa | 5358 | struct cgroup_root *root; |
a424316c PM |
5359 | |
5360 | retval = -ENOMEM; | |
e61734c5 | 5361 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5362 | if (!buf) |
5363 | goto out; | |
5364 | ||
a424316c | 5365 | mutex_lock(&cgroup_mutex); |
f0d9a5f1 | 5366 | spin_lock_bh(&css_set_lock); |
a424316c | 5367 | |
985ed670 | 5368 | for_each_root(root) { |
a424316c | 5369 | struct cgroup_subsys *ss; |
bd89aabc | 5370 | struct cgroup *cgrp; |
b85d2040 | 5371 | int ssid, count = 0; |
a424316c | 5372 | |
a2dd4247 | 5373 | if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) |
985ed670 TH |
5374 | continue; |
5375 | ||
2c6ab6d2 | 5376 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5377 | if (root != &cgrp_dfl_root) |
5378 | for_each_subsys(ss, ssid) | |
5379 | if (root->subsys_mask & (1 << ssid)) | |
5380 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5381 | ss->legacy_name); |
c6d57f33 PM |
5382 | if (strlen(root->name)) |
5383 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5384 | root->name); | |
a424316c | 5385 | seq_putc(m, ':'); |
2e91fa7f | 5386 | |
7717f7ba | 5387 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5388 | |
5389 | /* | |
5390 | * On traditional hierarchies, all zombie tasks show up as | |
5391 | * belonging to the root cgroup. On the default hierarchy, | |
5392 | * while a zombie doesn't show up in "cgroup.procs" and | |
5393 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5394 | * reporting the cgroup it belonged to before exiting. If | |
5395 | * the cgroup is removed before the zombie is reaped, | |
5396 | * " (deleted)" is appended to the cgroup path. | |
5397 | */ | |
5398 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
5399 | path = cgroup_path(cgrp, buf, PATH_MAX); | |
5400 | if (!path) { | |
5401 | retval = -ENAMETOOLONG; | |
5402 | goto out_unlock; | |
5403 | } | |
5404 | } else { | |
5405 | path = "/"; | |
e61734c5 | 5406 | } |
2e91fa7f | 5407 | |
e61734c5 | 5408 | seq_puts(m, path); |
2e91fa7f TH |
5409 | |
5410 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) | |
5411 | seq_puts(m, " (deleted)\n"); | |
5412 | else | |
5413 | seq_putc(m, '\n'); | |
a424316c PM |
5414 | } |
5415 | ||
006f4ac4 | 5416 | retval = 0; |
a424316c | 5417 | out_unlock: |
f0d9a5f1 | 5418 | spin_unlock_bh(&css_set_lock); |
a424316c | 5419 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5420 | kfree(buf); |
5421 | out: | |
5422 | return retval; | |
5423 | } | |
5424 | ||
a424316c PM |
5425 | /* Display information about each subsystem and each hierarchy */ |
5426 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5427 | { | |
30159ec7 | 5428 | struct cgroup_subsys *ss; |
a424316c | 5429 | int i; |
a424316c | 5430 | |
8bab8dde | 5431 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5432 | /* |
5433 | * ideally we don't want subsystems moving around while we do this. | |
5434 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5435 | * subsys/hierarchy state. | |
5436 | */ | |
a424316c | 5437 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5438 | |
5439 | for_each_subsys(ss, i) | |
2c6ab6d2 | 5440 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3e1d2eed | 5441 | ss->legacy_name, ss->root->hierarchy_id, |
fc5ed1e9 TH |
5442 | atomic_read(&ss->root->nr_cgrps), |
5443 | cgroup_ssid_enabled(i)); | |
30159ec7 | 5444 | |
a424316c PM |
5445 | mutex_unlock(&cgroup_mutex); |
5446 | return 0; | |
5447 | } | |
5448 | ||
5449 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5450 | { | |
9dce07f1 | 5451 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5452 | } |
5453 | ||
828c0950 | 5454 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5455 | .open = cgroupstats_open, |
5456 | .read = seq_read, | |
5457 | .llseek = seq_lseek, | |
5458 | .release = single_release, | |
5459 | }; | |
5460 | ||
7e47682e AS |
5461 | static void **subsys_canfork_priv_p(void *ss_priv[CGROUP_CANFORK_COUNT], int i) |
5462 | { | |
5463 | if (CGROUP_CANFORK_START <= i && i < CGROUP_CANFORK_END) | |
5464 | return &ss_priv[i - CGROUP_CANFORK_START]; | |
5465 | return NULL; | |
5466 | } | |
5467 | ||
5468 | static void *subsys_canfork_priv(void *ss_priv[CGROUP_CANFORK_COUNT], int i) | |
5469 | { | |
5470 | void **private = subsys_canfork_priv_p(ss_priv, i); | |
5471 | return private ? *private : NULL; | |
5472 | } | |
5473 | ||
b4f48b63 | 5474 | /** |
eaf797ab | 5475 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5476 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5477 | * |
eaf797ab TH |
5478 | * A task is associated with the init_css_set until cgroup_post_fork() |
5479 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5480 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5481 | */ |
5482 | void cgroup_fork(struct task_struct *child) | |
5483 | { | |
eaf797ab | 5484 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5485 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5486 | } |
5487 | ||
7e47682e AS |
5488 | /** |
5489 | * cgroup_can_fork - called on a new task before the process is exposed | |
5490 | * @child: the task in question. | |
5491 | * | |
5492 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5493 | * returns an error, the fork aborts with that error code. This allows for | |
5494 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5495 | */ | |
5496 | int cgroup_can_fork(struct task_struct *child, | |
5497 | void *ss_priv[CGROUP_CANFORK_COUNT]) | |
5498 | { | |
5499 | struct cgroup_subsys *ss; | |
5500 | int i, j, ret; | |
5501 | ||
5502 | for_each_subsys_which(ss, i, &have_canfork_callback) { | |
5503 | ret = ss->can_fork(child, subsys_canfork_priv_p(ss_priv, i)); | |
5504 | if (ret) | |
5505 | goto out_revert; | |
5506 | } | |
5507 | ||
5508 | return 0; | |
5509 | ||
5510 | out_revert: | |
5511 | for_each_subsys(ss, j) { | |
5512 | if (j >= i) | |
5513 | break; | |
5514 | if (ss->cancel_fork) | |
5515 | ss->cancel_fork(child, subsys_canfork_priv(ss_priv, j)); | |
5516 | } | |
5517 | ||
5518 | return ret; | |
5519 | } | |
5520 | ||
5521 | /** | |
5522 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5523 | * @child: the task in question | |
5524 | * | |
5525 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5526 | * cgroup_can_fork() succeded. | |
5527 | */ | |
5528 | void cgroup_cancel_fork(struct task_struct *child, | |
5529 | void *ss_priv[CGROUP_CANFORK_COUNT]) | |
5530 | { | |
5531 | struct cgroup_subsys *ss; | |
5532 | int i; | |
5533 | ||
5534 | for_each_subsys(ss, i) | |
5535 | if (ss->cancel_fork) | |
5536 | ss->cancel_fork(child, subsys_canfork_priv(ss_priv, i)); | |
5537 | } | |
5538 | ||
817929ec | 5539 | /** |
a043e3b2 LZ |
5540 | * cgroup_post_fork - called on a new task after adding it to the task list |
5541 | * @child: the task in question | |
5542 | * | |
5edee61e TH |
5543 | * Adds the task to the list running through its css_set if necessary and |
5544 | * call the subsystem fork() callbacks. Has to be after the task is | |
5545 | * visible on the task list in case we race with the first call to | |
0942eeee | 5546 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5547 | * list. |
a043e3b2 | 5548 | */ |
7e47682e AS |
5549 | void cgroup_post_fork(struct task_struct *child, |
5550 | void *old_ss_priv[CGROUP_CANFORK_COUNT]) | |
817929ec | 5551 | { |
30159ec7 | 5552 | struct cgroup_subsys *ss; |
5edee61e TH |
5553 | int i; |
5554 | ||
3ce3230a | 5555 | /* |
251f8c03 | 5556 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5557 | * function sets use_task_css_set_links before grabbing |
5558 | * tasklist_lock and we just went through tasklist_lock to add | |
5559 | * @child, it's guaranteed that either we see the set | |
5560 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5561 | * @child during its iteration. | |
5562 | * | |
5563 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 5564 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
5565 | * association is stable, and, on completion of the parent's |
5566 | * migration, @child is visible in the source of migration or | |
5567 | * already in the destination cgroup. This guarantee is necessary | |
5568 | * when implementing operations which need to migrate all tasks of | |
5569 | * a cgroup to another. | |
5570 | * | |
251f8c03 | 5571 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5572 | * will remain in init_css_set. This is safe because all tasks are |
5573 | * in the init_css_set before cg_links is enabled and there's no | |
5574 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5575 | */ |
817929ec | 5576 | if (use_task_css_set_links) { |
eaf797ab TH |
5577 | struct css_set *cset; |
5578 | ||
f0d9a5f1 | 5579 | spin_lock_bh(&css_set_lock); |
0e1d768f | 5580 | cset = task_css_set(current); |
eaf797ab | 5581 | if (list_empty(&child->cg_list)) { |
eaf797ab | 5582 | get_css_set(cset); |
f6d7d049 | 5583 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 5584 | } |
f0d9a5f1 | 5585 | spin_unlock_bh(&css_set_lock); |
817929ec | 5586 | } |
5edee61e TH |
5587 | |
5588 | /* | |
5589 | * Call ss->fork(). This must happen after @child is linked on | |
5590 | * css_set; otherwise, @child might change state between ->fork() | |
5591 | * and addition to css_set. | |
5592 | */ | |
cb4a3167 | 5593 | for_each_subsys_which(ss, i, &have_fork_callback) |
7e47682e | 5594 | ss->fork(child, subsys_canfork_priv(old_ss_priv, i)); |
817929ec | 5595 | } |
5edee61e | 5596 | |
b4f48b63 PM |
5597 | /** |
5598 | * cgroup_exit - detach cgroup from exiting task | |
5599 | * @tsk: pointer to task_struct of exiting process | |
5600 | * | |
5601 | * Description: Detach cgroup from @tsk and release it. | |
5602 | * | |
5603 | * Note that cgroups marked notify_on_release force every task in | |
5604 | * them to take the global cgroup_mutex mutex when exiting. | |
5605 | * This could impact scaling on very large systems. Be reluctant to | |
5606 | * use notify_on_release cgroups where very high task exit scaling | |
5607 | * is required on large systems. | |
5608 | * | |
0e1d768f TH |
5609 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5610 | * call cgroup_exit() while the task is still competent to handle | |
5611 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5612 | * each hierarchy for the remainder of its exit. No need to bother with | |
5613 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5614 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5615 | */ |
1ec41830 | 5616 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5617 | { |
30159ec7 | 5618 | struct cgroup_subsys *ss; |
5abb8855 | 5619 | struct css_set *cset; |
d41d5a01 | 5620 | int i; |
817929ec PM |
5621 | |
5622 | /* | |
0e1d768f | 5623 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 5624 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 5625 | */ |
0de0942d TH |
5626 | cset = task_css_set(tsk); |
5627 | ||
817929ec | 5628 | if (!list_empty(&tsk->cg_list)) { |
f0d9a5f1 | 5629 | spin_lock_bh(&css_set_lock); |
f6d7d049 | 5630 | css_set_move_task(tsk, cset, NULL, false); |
f0d9a5f1 | 5631 | spin_unlock_bh(&css_set_lock); |
2e91fa7f TH |
5632 | } else { |
5633 | get_css_set(cset); | |
817929ec PM |
5634 | } |
5635 | ||
cb4a3167 | 5636 | /* see cgroup_post_fork() for details */ |
2e91fa7f TH |
5637 | for_each_subsys_which(ss, i, &have_exit_callback) |
5638 | ss->exit(tsk); | |
5639 | } | |
30159ec7 | 5640 | |
2e91fa7f TH |
5641 | void cgroup_free(struct task_struct *task) |
5642 | { | |
5643 | struct css_set *cset = task_css_set(task); | |
afcf6c8b TH |
5644 | struct cgroup_subsys *ss; |
5645 | int ssid; | |
5646 | ||
5647 | for_each_subsys_which(ss, ssid, &have_free_callback) | |
5648 | ss->free(task); | |
d41d5a01 | 5649 | |
2e91fa7f | 5650 | put_css_set(cset); |
b4f48b63 | 5651 | } |
697f4161 | 5652 | |
bd89aabc | 5653 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5654 | { |
27bd4dbb | 5655 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && |
971ff493 ZL |
5656 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
5657 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
5658 | } |
5659 | ||
81a6a5cd PM |
5660 | /* |
5661 | * Notify userspace when a cgroup is released, by running the | |
5662 | * configured release agent with the name of the cgroup (path | |
5663 | * relative to the root of cgroup file system) as the argument. | |
5664 | * | |
5665 | * Most likely, this user command will try to rmdir this cgroup. | |
5666 | * | |
5667 | * This races with the possibility that some other task will be | |
5668 | * attached to this cgroup before it is removed, or that some other | |
5669 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5670 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5671 | * unused, and this cgroup will be reprieved from its death sentence, | |
5672 | * to continue to serve a useful existence. Next time it's released, | |
5673 | * we will get notified again, if it still has 'notify_on_release' set. | |
5674 | * | |
5675 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5676 | * means only wait until the task is successfully execve()'d. The | |
5677 | * separate release agent task is forked by call_usermodehelper(), | |
5678 | * then control in this thread returns here, without waiting for the | |
5679 | * release agent task. We don't bother to wait because the caller of | |
5680 | * this routine has no use for the exit status of the release agent | |
5681 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5682 | */ |
81a6a5cd PM |
5683 | static void cgroup_release_agent(struct work_struct *work) |
5684 | { | |
971ff493 ZL |
5685 | struct cgroup *cgrp = |
5686 | container_of(work, struct cgroup, release_agent_work); | |
5687 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
5688 | char *argv[3], *envp[3]; | |
5689 | ||
81a6a5cd | 5690 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
5691 | |
5692 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
5693 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5694 | if (!pathbuf || !agentbuf) | |
5695 | goto out; | |
5696 | ||
5697 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); | |
5698 | if (!path) | |
5699 | goto out; | |
5700 | ||
5701 | argv[0] = agentbuf; | |
5702 | argv[1] = path; | |
5703 | argv[2] = NULL; | |
5704 | ||
5705 | /* minimal command environment */ | |
5706 | envp[0] = "HOME=/"; | |
5707 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5708 | envp[2] = NULL; | |
5709 | ||
81a6a5cd | 5710 | mutex_unlock(&cgroup_mutex); |
971ff493 | 5711 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 5712 | goto out_free; |
971ff493 | 5713 | out: |
81a6a5cd | 5714 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 5715 | out_free: |
971ff493 ZL |
5716 | kfree(agentbuf); |
5717 | kfree(pathbuf); | |
81a6a5cd | 5718 | } |
8bab8dde PM |
5719 | |
5720 | static int __init cgroup_disable(char *str) | |
5721 | { | |
30159ec7 | 5722 | struct cgroup_subsys *ss; |
8bab8dde | 5723 | char *token; |
30159ec7 | 5724 | int i; |
8bab8dde PM |
5725 | |
5726 | while ((token = strsep(&str, ",")) != NULL) { | |
5727 | if (!*token) | |
5728 | continue; | |
be45c900 | 5729 | |
3ed80a62 | 5730 | for_each_subsys(ss, i) { |
3e1d2eed TH |
5731 | if (strcmp(token, ss->name) && |
5732 | strcmp(token, ss->legacy_name)) | |
5733 | continue; | |
a3e72739 | 5734 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
5735 | } |
5736 | } | |
5737 | return 1; | |
5738 | } | |
5739 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5740 | |
b77d7b60 | 5741 | /** |
ec903c0c | 5742 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5743 | * @dentry: directory dentry of interest |
5744 | * @ss: subsystem of interest | |
b77d7b60 | 5745 | * |
5a17f543 TH |
5746 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5747 | * to get the corresponding css and return it. If such css doesn't exist | |
5748 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5749 | */ |
ec903c0c TH |
5750 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5751 | struct cgroup_subsys *ss) | |
e5d1367f | 5752 | { |
2bd59d48 TH |
5753 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
5754 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 5755 | struct cgroup *cgrp; |
e5d1367f | 5756 | |
35cf0836 | 5757 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
5758 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
5759 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5760 | return ERR_PTR(-EBADF); |
5761 | ||
5a17f543 TH |
5762 | rcu_read_lock(); |
5763 | ||
2bd59d48 TH |
5764 | /* |
5765 | * This path doesn't originate from kernfs and @kn could already | |
5766 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 5767 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
5768 | */ |
5769 | cgrp = rcu_dereference(kn->priv); | |
5770 | if (cgrp) | |
5771 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5772 | |
ec903c0c | 5773 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5774 | css = ERR_PTR(-ENOENT); |
5775 | ||
5776 | rcu_read_unlock(); | |
5777 | return css; | |
e5d1367f | 5778 | } |
e5d1367f | 5779 | |
1cb650b9 LZ |
5780 | /** |
5781 | * css_from_id - lookup css by id | |
5782 | * @id: the cgroup id | |
5783 | * @ss: cgroup subsys to be looked into | |
5784 | * | |
5785 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5786 | * Should be called under rcu_read_lock(). | |
5787 | */ | |
5788 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5789 | { | |
6fa4918d | 5790 | WARN_ON_ONCE(!rcu_read_lock_held()); |
adbe427b | 5791 | return id > 0 ? idr_find(&ss->css_idr, id) : NULL; |
e5d1367f SE |
5792 | } |
5793 | ||
fe693435 | 5794 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5795 | static struct cgroup_subsys_state * |
5796 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5797 | { |
5798 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5799 | ||
5800 | if (!css) | |
5801 | return ERR_PTR(-ENOMEM); | |
5802 | ||
5803 | return css; | |
5804 | } | |
5805 | ||
eb95419b | 5806 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5807 | { |
eb95419b | 5808 | kfree(css); |
fe693435 PM |
5809 | } |
5810 | ||
182446d0 TH |
5811 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5812 | struct cftype *cft) | |
fe693435 | 5813 | { |
182446d0 | 5814 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5815 | } |
5816 | ||
182446d0 TH |
5817 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5818 | struct cftype *cft) | |
fe693435 PM |
5819 | { |
5820 | return (u64)(unsigned long)current->cgroups; | |
5821 | } | |
5822 | ||
182446d0 | 5823 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5824 | struct cftype *cft) |
fe693435 PM |
5825 | { |
5826 | u64 count; | |
5827 | ||
5828 | rcu_read_lock(); | |
a8ad805c | 5829 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5830 | rcu_read_unlock(); |
5831 | return count; | |
5832 | } | |
5833 | ||
2da8ca82 | 5834 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5835 | { |
69d0206c | 5836 | struct cgrp_cset_link *link; |
5abb8855 | 5837 | struct css_set *cset; |
e61734c5 TH |
5838 | char *name_buf; |
5839 | ||
5840 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
5841 | if (!name_buf) | |
5842 | return -ENOMEM; | |
7717f7ba | 5843 | |
f0d9a5f1 | 5844 | spin_lock_bh(&css_set_lock); |
7717f7ba | 5845 | rcu_read_lock(); |
5abb8855 | 5846 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5847 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 5848 | struct cgroup *c = link->cgrp; |
7717f7ba | 5849 | |
a2dd4247 | 5850 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 5851 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 5852 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
5853 | } |
5854 | rcu_read_unlock(); | |
f0d9a5f1 | 5855 | spin_unlock_bh(&css_set_lock); |
e61734c5 | 5856 | kfree(name_buf); |
7717f7ba PM |
5857 | return 0; |
5858 | } | |
5859 | ||
5860 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 5861 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5862 | { |
2da8ca82 | 5863 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 5864 | struct cgrp_cset_link *link; |
7717f7ba | 5865 | |
f0d9a5f1 | 5866 | spin_lock_bh(&css_set_lock); |
182446d0 | 5867 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5868 | struct css_set *cset = link->cset; |
7717f7ba PM |
5869 | struct task_struct *task; |
5870 | int count = 0; | |
c7561128 | 5871 | |
5abb8855 | 5872 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 5873 | |
5abb8855 | 5874 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
5875 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
5876 | goto overflow; | |
5877 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
5878 | } | |
5879 | ||
5880 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
5881 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
5882 | goto overflow; | |
5883 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 5884 | } |
c7561128 TH |
5885 | continue; |
5886 | overflow: | |
5887 | seq_puts(seq, " ...\n"); | |
7717f7ba | 5888 | } |
f0d9a5f1 | 5889 | spin_unlock_bh(&css_set_lock); |
7717f7ba PM |
5890 | return 0; |
5891 | } | |
5892 | ||
182446d0 | 5893 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5894 | { |
27bd4dbb | 5895 | return (!cgroup_is_populated(css->cgroup) && |
a25eb52e | 5896 | !css_has_online_children(&css->cgroup->self)); |
fe693435 PM |
5897 | } |
5898 | ||
5899 | static struct cftype debug_files[] = { | |
fe693435 PM |
5900 | { |
5901 | .name = "taskcount", | |
5902 | .read_u64 = debug_taskcount_read, | |
5903 | }, | |
5904 | ||
5905 | { | |
5906 | .name = "current_css_set", | |
5907 | .read_u64 = current_css_set_read, | |
5908 | }, | |
5909 | ||
5910 | { | |
5911 | .name = "current_css_set_refcount", | |
5912 | .read_u64 = current_css_set_refcount_read, | |
5913 | }, | |
5914 | ||
7717f7ba PM |
5915 | { |
5916 | .name = "current_css_set_cg_links", | |
2da8ca82 | 5917 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
5918 | }, |
5919 | ||
5920 | { | |
5921 | .name = "cgroup_css_links", | |
2da8ca82 | 5922 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
5923 | }, |
5924 | ||
fe693435 PM |
5925 | { |
5926 | .name = "releasable", | |
5927 | .read_u64 = releasable_read, | |
5928 | }, | |
fe693435 | 5929 | |
4baf6e33 TH |
5930 | { } /* terminate */ |
5931 | }; | |
fe693435 | 5932 | |
073219e9 | 5933 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
5934 | .css_alloc = debug_css_alloc, |
5935 | .css_free = debug_css_free, | |
5577964e | 5936 | .legacy_cftypes = debug_files, |
fe693435 PM |
5937 | }; |
5938 | #endif /* CONFIG_CGROUP_DEBUG */ |