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