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 | ||
0a268dbd TH |
31 | #include "cgroup-internal.h" |
32 | ||
2ce9738b | 33 | #include <linux/cred.h> |
ddbcc7e8 | 34 | #include <linux/errno.h> |
2ce9738b | 35 | #include <linux/init_task.h> |
ddbcc7e8 | 36 | #include <linux/kernel.h> |
c9482a5b | 37 | #include <linux/magic.h> |
ddbcc7e8 PM |
38 | #include <linux/mutex.h> |
39 | #include <linux/mount.h> | |
40 | #include <linux/pagemap.h> | |
a424316c | 41 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
42 | #include <linux/rcupdate.h> |
43 | #include <linux/sched.h> | |
29930025 | 44 | #include <linux/sched/task.h> |
ddbcc7e8 | 45 | #include <linux/slab.h> |
ddbcc7e8 | 46 | #include <linux/spinlock.h> |
1ed13287 | 47 | #include <linux/percpu-rwsem.h> |
ddbcc7e8 | 48 | #include <linux/string.h> |
0ac801fe | 49 | #include <linux/hashtable.h> |
2c6ab6d2 | 50 | #include <linux/idr.h> |
c4c27fbd | 51 | #include <linux/kthread.h> |
60063497 | 52 | #include <linux/atomic.h> |
e93ad19d | 53 | #include <linux/cpuset.h> |
a79a908f AK |
54 | #include <linux/proc_ns.h> |
55 | #include <linux/nsproxy.h> | |
1f3fe7eb | 56 | #include <linux/file.h> |
d4ff749b | 57 | #include <linux/sched/cputime.h> |
2ce7135a | 58 | #include <linux/psi.h> |
bd1060a1 | 59 | #include <net/sock.h> |
ddbcc7e8 | 60 | |
ed1777de TH |
61 | #define CREATE_TRACE_POINTS |
62 | #include <trace/events/cgroup.h> | |
63 | ||
8d7e6fb0 TH |
64 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
65 | MAX_CFTYPE_NAME + 2) | |
b12e3583 TH |
66 | /* let's not notify more than 100 times per second */ |
67 | #define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100) | |
8d7e6fb0 | 68 | |
e25e2cbb TH |
69 | /* |
70 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
71 | * hierarchy must be performed while holding it. | |
72 | * | |
f0d9a5f1 | 73 | * css_set_lock protects task->cgroups pointer, the list of css_set |
0e1d768f | 74 | * objects, and the chain of tasks off each css_set. |
e25e2cbb | 75 | * |
0e1d768f TH |
76 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
77 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 78 | */ |
2219449a | 79 | DEFINE_MUTEX(cgroup_mutex); |
f0d9a5f1 | 80 | DEFINE_SPINLOCK(css_set_lock); |
0a268dbd TH |
81 | |
82 | #ifdef CONFIG_PROVE_RCU | |
0e1d768f | 83 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
f0d9a5f1 | 84 | EXPORT_SYMBOL_GPL(css_set_lock); |
2219449a TH |
85 | #endif |
86 | ||
e4f8d81c SRV |
87 | DEFINE_SPINLOCK(trace_cgroup_path_lock); |
88 | char trace_cgroup_path[TRACE_CGROUP_PATH_LEN]; | |
5cf8114d | 89 | bool cgroup_debug __read_mostly; |
e4f8d81c | 90 | |
6fa4918d | 91 | /* |
15a4c835 TH |
92 | * Protects cgroup_idr and css_idr so that IDs can be released without |
93 | * grabbing cgroup_mutex. | |
6fa4918d TH |
94 | */ |
95 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
96 | ||
34c06254 TH |
97 | /* |
98 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
99 | * against file removal/re-creation across css hiding. | |
100 | */ | |
101 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
102 | ||
1ed13287 TH |
103 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
104 | ||
8353da1f | 105 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
106 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
107 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 108 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 109 | |
e5fca243 TH |
110 | /* |
111 | * cgroup destruction makes heavy use of work items and there can be a lot | |
112 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
113 | * destruction work items don't end up filling up max_active of system_wq | |
114 | * which may lead to deadlock. | |
115 | */ | |
116 | static struct workqueue_struct *cgroup_destroy_wq; | |
117 | ||
3ed80a62 | 118 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 119 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
0a268dbd | 120 | struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
121 | #include <linux/cgroup_subsys.h> |
122 | }; | |
073219e9 TH |
123 | #undef SUBSYS |
124 | ||
125 | /* array of cgroup subsystem names */ | |
126 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
127 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
128 | #include <linux/cgroup_subsys.h> |
129 | }; | |
073219e9 | 130 | #undef SUBSYS |
ddbcc7e8 | 131 | |
49d1dc4b TH |
132 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
133 | #define SUBSYS(_x) \ | |
134 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
135 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
136 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
137 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
138 | #include <linux/cgroup_subsys.h> | |
139 | #undef SUBSYS | |
140 | ||
141 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
142 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
143 | #include <linux/cgroup_subsys.h> | |
144 | }; | |
145 | #undef SUBSYS | |
146 | ||
147 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
148 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
149 | #include <linux/cgroup_subsys.h> | |
150 | }; | |
151 | #undef SUBSYS | |
152 | ||
c58632b3 | 153 | static DEFINE_PER_CPU(struct cgroup_rstat_cpu, cgrp_dfl_root_rstat_cpu); |
041cd640 | 154 | |
ddbcc7e8 | 155 | /* |
3dd06ffa | 156 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
157 | * unattached - it never has more than a single cgroup, and all tasks are |
158 | * part of that cgroup. | |
ddbcc7e8 | 159 | */ |
c58632b3 | 160 | struct cgroup_root cgrp_dfl_root = { .cgrp.rstat_cpu = &cgrp_dfl_root_rstat_cpu }; |
d0ec4230 | 161 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 162 | |
a2dd4247 TH |
163 | /* |
164 | * The default hierarchy always exists but is hidden until mounted for the | |
165 | * first time. This is for backward compatibility. | |
166 | */ | |
a7165264 | 167 | static bool cgrp_dfl_visible; |
ddbcc7e8 | 168 | |
5533e011 | 169 | /* some controllers are not supported in the default hierarchy */ |
a7165264 | 170 | static u16 cgrp_dfl_inhibit_ss_mask; |
5533e011 | 171 | |
f6d635ad | 172 | /* some controllers are implicitly enabled on the default hierarchy */ |
b807421a | 173 | static u16 cgrp_dfl_implicit_ss_mask; |
f6d635ad | 174 | |
8cfd8147 TH |
175 | /* some controllers can be threaded on the default hierarchy */ |
176 | static u16 cgrp_dfl_threaded_ss_mask; | |
177 | ||
ddbcc7e8 | 178 | /* The list of hierarchy roots */ |
0a268dbd | 179 | LIST_HEAD(cgroup_roots); |
9871bf95 | 180 | static int cgroup_root_count; |
ddbcc7e8 | 181 | |
3417ae1f | 182 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 183 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 184 | |
794611a1 | 185 | /* |
0cb51d71 TH |
186 | * Assign a monotonically increasing serial number to csses. It guarantees |
187 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
188 | * Also, as csses are always appended to the parent's ->children list, it | |
189 | * guarantees that sibling csses are always sorted in the ascending serial | |
190 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 191 | */ |
0cb51d71 | 192 | static u64 css_serial_nr_next = 1; |
794611a1 | 193 | |
cb4a3167 | 194 | /* |
b807421a TH |
195 | * These bitmasks identify subsystems with specific features to avoid |
196 | * having to do iterative checks repeatedly. | |
ddbcc7e8 | 197 | */ |
6e5c8307 TH |
198 | static u16 have_fork_callback __read_mostly; |
199 | static u16 have_exit_callback __read_mostly; | |
200 | static u16 have_free_callback __read_mostly; | |
b807421a | 201 | static u16 have_canfork_callback __read_mostly; |
ddbcc7e8 | 202 | |
a79a908f AK |
203 | /* cgroup namespace for init task */ |
204 | struct cgroup_namespace init_cgroup_ns = { | |
387ad967 | 205 | .count = REFCOUNT_INIT(2), |
a79a908f AK |
206 | .user_ns = &init_user_ns, |
207 | .ns.ops = &cgroupns_operations, | |
208 | .ns.inum = PROC_CGROUP_INIT_INO, | |
209 | .root_cset = &init_css_set, | |
210 | }; | |
211 | ||
67e9c74b | 212 | static struct file_system_type cgroup2_fs_type; |
d62beb7f | 213 | static struct cftype cgroup_base_files[]; |
628f7cd4 | 214 | |
334c3679 TH |
215 | static int cgroup_apply_control(struct cgroup *cgrp); |
216 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret); | |
ed27b9f7 | 217 | static void css_task_iter_advance(struct css_task_iter *it); |
42809dd4 | 218 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
6cd0f5bb TH |
219 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
220 | struct cgroup_subsys *ss); | |
9d755d33 | 221 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 222 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
223 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
224 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 225 | bool is_add); |
42809dd4 | 226 | |
fc5ed1e9 TH |
227 | /** |
228 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
229 | * @ssid: subsys ID of interest | |
230 | * | |
231 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
232 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
233 | * is slower static_key_enabled() based test indexed by @ssid. | |
234 | */ | |
0a268dbd | 235 | bool cgroup_ssid_enabled(int ssid) |
fc5ed1e9 | 236 | { |
cfe02a8a AB |
237 | if (CGROUP_SUBSYS_COUNT == 0) |
238 | return false; | |
239 | ||
fc5ed1e9 TH |
240 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); |
241 | } | |
242 | ||
9e10a130 TH |
243 | /** |
244 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
245 | * @cgrp: the cgroup of interest | |
246 | * | |
247 | * The default hierarchy is the v2 interface of cgroup and this function | |
248 | * can be used to test whether a cgroup is on the default hierarchy for | |
249 | * cases where a subsystem should behave differnetly depending on the | |
250 | * interface version. | |
251 | * | |
252 | * The set of behaviors which change on the default hierarchy are still | |
253 | * being determined and the mount option is prefixed with __DEVEL__. | |
254 | * | |
255 | * List of changed behaviors: | |
256 | * | |
257 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
258 | * and "name" are disallowed. | |
259 | * | |
260 | * - When mounting an existing superblock, mount options should match. | |
261 | * | |
262 | * - Remount is disallowed. | |
263 | * | |
264 | * - rename(2) is disallowed. | |
265 | * | |
266 | * - "tasks" is removed. Everything should be at process granularity. Use | |
267 | * "cgroup.procs" instead. | |
268 | * | |
269 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
270 | * recycled inbetween reads. | |
271 | * | |
272 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
273 | * notification mechanism will be implemented. | |
274 | * | |
275 | * - "cgroup.clone_children" is removed. | |
276 | * | |
277 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
278 | * and its descendants contain no task; otherwise, 1. The file also | |
279 | * generates kernfs notification which can be monitored through poll and | |
280 | * [di]notify when the value of the file changes. | |
281 | * | |
282 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
283 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
284 | * moved to an ancestor. | |
285 | * | |
286 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
287 | * masks of ancestors. | |
288 | * | |
289 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
290 | * is not created. | |
291 | * | |
292 | * - blkcg: blk-throttle becomes properly hierarchical. | |
293 | * | |
294 | * - debug: disallowed on the default hierarchy. | |
295 | */ | |
0a268dbd | 296 | bool cgroup_on_dfl(const struct cgroup *cgrp) |
9e10a130 TH |
297 | { |
298 | return cgrp->root == &cgrp_dfl_root; | |
299 | } | |
300 | ||
6fa4918d TH |
301 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
302 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
303 | gfp_t gfp_mask) | |
304 | { | |
305 | int ret; | |
306 | ||
307 | idr_preload(gfp_mask); | |
54504e97 | 308 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 309 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 310 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
311 | idr_preload_end(); |
312 | return ret; | |
313 | } | |
314 | ||
315 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
316 | { | |
317 | void *ret; | |
318 | ||
54504e97 | 319 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 320 | ret = idr_replace(idr, ptr, id); |
54504e97 | 321 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
322 | return ret; |
323 | } | |
324 | ||
325 | static void cgroup_idr_remove(struct idr *idr, int id) | |
326 | { | |
54504e97 | 327 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 328 | idr_remove(idr, id); |
54504e97 | 329 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
330 | } |
331 | ||
27f26753 | 332 | static bool cgroup_has_tasks(struct cgroup *cgrp) |
d51f39b0 | 333 | { |
27f26753 TH |
334 | return cgrp->nr_populated_csets; |
335 | } | |
d51f39b0 | 336 | |
7a0cf0e7 | 337 | bool cgroup_is_threaded(struct cgroup *cgrp) |
454000ad TH |
338 | { |
339 | return cgrp->dom_cgrp != cgrp; | |
340 | } | |
341 | ||
8cfd8147 TH |
342 | /* can @cgrp host both domain and threaded children? */ |
343 | static bool cgroup_is_mixable(struct cgroup *cgrp) | |
344 | { | |
345 | /* | |
346 | * Root isn't under domain level resource control exempting it from | |
347 | * the no-internal-process constraint, so it can serve as a thread | |
348 | * root and a parent of resource domains at the same time. | |
349 | */ | |
350 | return !cgroup_parent(cgrp); | |
351 | } | |
352 | ||
353 | /* can @cgrp become a thread root? should always be true for a thread root */ | |
354 | static bool cgroup_can_be_thread_root(struct cgroup *cgrp) | |
355 | { | |
356 | /* mixables don't care */ | |
357 | if (cgroup_is_mixable(cgrp)) | |
358 | return true; | |
359 | ||
360 | /* domain roots can't be nested under threaded */ | |
361 | if (cgroup_is_threaded(cgrp)) | |
362 | return false; | |
363 | ||
364 | /* can only have either domain or threaded children */ | |
365 | if (cgrp->nr_populated_domain_children) | |
366 | return false; | |
367 | ||
368 | /* and no domain controllers can be enabled */ | |
369 | if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask) | |
370 | return false; | |
371 | ||
372 | return true; | |
373 | } | |
374 | ||
375 | /* is @cgrp root of a threaded subtree? */ | |
7a0cf0e7 | 376 | bool cgroup_is_thread_root(struct cgroup *cgrp) |
8cfd8147 TH |
377 | { |
378 | /* thread root should be a domain */ | |
379 | if (cgroup_is_threaded(cgrp)) | |
380 | return false; | |
381 | ||
382 | /* a domain w/ threaded children is a thread root */ | |
383 | if (cgrp->nr_threaded_children) | |
384 | return true; | |
385 | ||
386 | /* | |
387 | * A domain which has tasks and explicit threaded controllers | |
388 | * enabled is a thread root. | |
389 | */ | |
390 | if (cgroup_has_tasks(cgrp) && | |
391 | (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask)) | |
392 | return true; | |
393 | ||
394 | return false; | |
395 | } | |
396 | ||
397 | /* a domain which isn't connected to the root w/o brekage can't be used */ | |
398 | static bool cgroup_is_valid_domain(struct cgroup *cgrp) | |
399 | { | |
400 | /* the cgroup itself can be a thread root */ | |
401 | if (cgroup_is_threaded(cgrp)) | |
402 | return false; | |
403 | ||
404 | /* but the ancestors can't be unless mixable */ | |
405 | while ((cgrp = cgroup_parent(cgrp))) { | |
406 | if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp)) | |
407 | return false; | |
408 | if (cgroup_is_threaded(cgrp)) | |
409 | return false; | |
410 | } | |
411 | ||
412 | return true; | |
d51f39b0 TH |
413 | } |
414 | ||
5531dc91 TH |
415 | /* subsystems visibly enabled on a cgroup */ |
416 | static u16 cgroup_control(struct cgroup *cgrp) | |
417 | { | |
418 | struct cgroup *parent = cgroup_parent(cgrp); | |
419 | u16 root_ss_mask = cgrp->root->subsys_mask; | |
420 | ||
8cfd8147 TH |
421 | if (parent) { |
422 | u16 ss_mask = parent->subtree_control; | |
423 | ||
424 | /* threaded cgroups can only have threaded controllers */ | |
425 | if (cgroup_is_threaded(cgrp)) | |
426 | ss_mask &= cgrp_dfl_threaded_ss_mask; | |
427 | return ss_mask; | |
428 | } | |
5531dc91 TH |
429 | |
430 | if (cgroup_on_dfl(cgrp)) | |
f6d635ad TH |
431 | root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask | |
432 | cgrp_dfl_implicit_ss_mask); | |
5531dc91 TH |
433 | return root_ss_mask; |
434 | } | |
435 | ||
436 | /* subsystems enabled on a cgroup */ | |
437 | static u16 cgroup_ss_mask(struct cgroup *cgrp) | |
438 | { | |
439 | struct cgroup *parent = cgroup_parent(cgrp); | |
440 | ||
8cfd8147 TH |
441 | if (parent) { |
442 | u16 ss_mask = parent->subtree_ss_mask; | |
443 | ||
444 | /* threaded cgroups can only have threaded controllers */ | |
445 | if (cgroup_is_threaded(cgrp)) | |
446 | ss_mask &= cgrp_dfl_threaded_ss_mask; | |
447 | return ss_mask; | |
448 | } | |
5531dc91 TH |
449 | |
450 | return cgrp->root->subsys_mask; | |
451 | } | |
452 | ||
95109b62 TH |
453 | /** |
454 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
455 | * @cgrp: the cgroup of interest | |
9d800df1 | 456 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 457 | * |
ca8bdcaf TH |
458 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
459 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
460 | * the caller is responsible for pinning the returned css if it wants to | |
461 | * keep accessing it outside the said locks. This function may return | |
462 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
463 | */ |
464 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 465 | struct cgroup_subsys *ss) |
95109b62 | 466 | { |
ca8bdcaf | 467 | if (ss) |
aec25020 | 468 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 469 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 470 | else |
9d800df1 | 471 | return &cgrp->self; |
95109b62 | 472 | } |
42809dd4 | 473 | |
d41bf8c9 TH |
474 | /** |
475 | * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem | |
476 | * @cgrp: the cgroup of interest | |
477 | * @ss: the subsystem of interest | |
478 | * | |
479 | * Find and get @cgrp's css assocaited with @ss. If the css doesn't exist | |
480 | * or is offline, %NULL is returned. | |
481 | */ | |
482 | static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp, | |
483 | struct cgroup_subsys *ss) | |
484 | { | |
485 | struct cgroup_subsys_state *css; | |
486 | ||
487 | rcu_read_lock(); | |
488 | css = cgroup_css(cgrp, ss); | |
489 | if (!css || !css_tryget_online(css)) | |
490 | css = NULL; | |
491 | rcu_read_unlock(); | |
492 | ||
493 | return css; | |
494 | } | |
495 | ||
aec3dfcb | 496 | /** |
fc5a828b | 497 | * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss |
aec3dfcb | 498 | * @cgrp: the cgroup of interest |
9d800df1 | 499 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 500 | * |
d0f702e6 | 501 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
502 | * as the matching css of the nearest ancestor including self which has @ss |
503 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
504 | * function is guaranteed to return non-NULL css. | |
505 | */ | |
fc5a828b DZ |
506 | static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp, |
507 | struct cgroup_subsys *ss) | |
aec3dfcb TH |
508 | { |
509 | lockdep_assert_held(&cgroup_mutex); | |
510 | ||
511 | if (!ss) | |
9d800df1 | 512 | return &cgrp->self; |
aec3dfcb | 513 | |
eeecbd19 TH |
514 | /* |
515 | * This function is used while updating css associations and thus | |
5531dc91 | 516 | * can't test the csses directly. Test ss_mask. |
eeecbd19 | 517 | */ |
5531dc91 | 518 | while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) { |
d51f39b0 | 519 | cgrp = cgroup_parent(cgrp); |
5531dc91 TH |
520 | if (!cgrp) |
521 | return NULL; | |
522 | } | |
aec3dfcb TH |
523 | |
524 | return cgroup_css(cgrp, ss); | |
95109b62 | 525 | } |
42809dd4 | 526 | |
fc5a828b DZ |
527 | /** |
528 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
529 | * @cgrp: the cgroup of interest | |
530 | * @ss: the subsystem of interest | |
531 | * | |
532 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
533 | * defined as the matching css of the nearest ancestor including self which | |
534 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
535 | * the root css is returned, so this function always returns a valid css. | |
536 | * | |
537 | * The returned css is not guaranteed to be online, and therefore it is the | |
538 | * callers responsiblity to tryget a reference for it. | |
539 | */ | |
540 | struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
541 | struct cgroup_subsys *ss) | |
542 | { | |
543 | struct cgroup_subsys_state *css; | |
544 | ||
545 | do { | |
546 | css = cgroup_css(cgrp, ss); | |
547 | ||
548 | if (css) | |
549 | return css; | |
550 | cgrp = cgroup_parent(cgrp); | |
551 | } while (cgrp); | |
552 | ||
553 | return init_css_set.subsys[ss->id]; | |
554 | } | |
555 | ||
eeecbd19 TH |
556 | /** |
557 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
558 | * @cgrp: the cgroup of interest | |
559 | * @ss: the subsystem of interest | |
560 | * | |
561 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
562 | * defined as the matching css of the nearest ancestor including self which | |
563 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
564 | * the root css is returned, so this function always returns a valid css. | |
565 | * The returned css must be put using css_put(). | |
566 | */ | |
567 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
568 | struct cgroup_subsys *ss) | |
569 | { | |
570 | struct cgroup_subsys_state *css; | |
571 | ||
572 | rcu_read_lock(); | |
573 | ||
574 | do { | |
575 | css = cgroup_css(cgrp, ss); | |
576 | ||
577 | if (css && css_tryget_online(css)) | |
578 | goto out_unlock; | |
579 | cgrp = cgroup_parent(cgrp); | |
580 | } while (cgrp); | |
581 | ||
582 | css = init_css_set.subsys[ss->id]; | |
583 | css_get(css); | |
584 | out_unlock: | |
585 | rcu_read_unlock(); | |
586 | return css; | |
587 | } | |
588 | ||
a590b90d | 589 | static void cgroup_get_live(struct cgroup *cgrp) |
052c3f3a TH |
590 | { |
591 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
592 | css_get(&cgrp->self); | |
593 | } | |
594 | ||
b4168640 | 595 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 596 | { |
2bd59d48 | 597 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 598 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
599 | |
600 | /* | |
601 | * This is open and unprotected implementation of cgroup_css(). | |
602 | * seq_css() is only called from a kernfs file operation which has | |
603 | * an active reference on the file. Because all the subsystem | |
604 | * files are drained before a css is disassociated with a cgroup, | |
605 | * the matching css from the cgroup's subsys table is guaranteed to | |
606 | * be and stay valid until the enclosing operation is complete. | |
607 | */ | |
608 | if (cft->ss) | |
609 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
610 | else | |
9d800df1 | 611 | return &cgrp->self; |
59f5296b | 612 | } |
b4168640 | 613 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 614 | |
1c6727af TH |
615 | /** |
616 | * for_each_css - iterate all css's of a cgroup | |
617 | * @css: the iteration cursor | |
618 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
619 | * @cgrp: the target cgroup to iterate css's of | |
620 | * | |
aec3dfcb | 621 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
622 | */ |
623 | #define for_each_css(css, ssid, cgrp) \ | |
624 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
625 | if (!((css) = rcu_dereference_check( \ | |
626 | (cgrp)->subsys[(ssid)], \ | |
627 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
628 | else | |
629 | ||
aec3dfcb TH |
630 | /** |
631 | * for_each_e_css - iterate all effective css's of a cgroup | |
632 | * @css: the iteration cursor | |
633 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
634 | * @cgrp: the target cgroup to iterate css's of | |
635 | * | |
636 | * Should be called under cgroup_[tree_]mutex. | |
637 | */ | |
fc5a828b DZ |
638 | #define for_each_e_css(css, ssid, cgrp) \ |
639 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
640 | if (!((css) = cgroup_e_css_by_mask(cgrp, \ | |
641 | cgroup_subsys[(ssid)]))) \ | |
642 | ; \ | |
aec3dfcb TH |
643 | else |
644 | ||
cb4a3167 | 645 | /** |
b4e0eeaf | 646 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
cb4a3167 AS |
647 | * @ss: the iteration cursor |
648 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
b4e0eeaf | 649 | * @ss_mask: the bitmask |
cb4a3167 AS |
650 | * |
651 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
b4e0eeaf | 652 | * @ss_mask is set. |
cb4a3167 | 653 | */ |
b4e0eeaf TH |
654 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
655 | unsigned long __ss_mask = (ss_mask); \ | |
656 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ | |
4a705c5c | 657 | (ssid) = 0; \ |
b4e0eeaf TH |
658 | break; \ |
659 | } \ | |
660 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
661 | (ss) = cgroup_subsys[ssid]; \ | |
662 | { | |
663 | ||
664 | #define while_each_subsys_mask() \ | |
665 | } \ | |
666 | } \ | |
667 | } while (false) | |
cb4a3167 | 668 | |
f8f22e53 TH |
669 | /* iterate over child cgrps, lock should be held throughout iteration */ |
670 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 671 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 672 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
673 | cgroup_is_dead(child); })) \ |
674 | ; \ | |
675 | else | |
7ae1bad9 | 676 | |
ce3f1d9d TH |
677 | /* walk live descendants in preorder */ |
678 | #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \ | |
679 | css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \ | |
680 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
681 | (dsct) = (d_css)->cgroup; \ | |
682 | cgroup_is_dead(dsct); })) \ | |
683 | ; \ | |
684 | else | |
685 | ||
686 | /* walk live descendants in postorder */ | |
687 | #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \ | |
688 | css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \ | |
689 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
690 | (dsct) = (d_css)->cgroup; \ | |
691 | cgroup_is_dead(dsct); })) \ | |
692 | ; \ | |
693 | else | |
694 | ||
172a2c06 TH |
695 | /* |
696 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
697 | * hierarchies being mounted. It contains a pointer to the root state |
698 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
699 | * reference-counted, to improve performance when child cgroups | |
700 | * haven't been created. | |
701 | */ | |
5024ae29 | 702 | struct css_set init_css_set = { |
4b9502e6 | 703 | .refcount = REFCOUNT_INIT(1), |
454000ad | 704 | .dom_cset = &init_css_set, |
172a2c06 TH |
705 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), |
706 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
5f617ebb | 707 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
454000ad | 708 | .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets), |
5f617ebb | 709 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), |
172a2c06 TH |
710 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), |
711 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
38683148 TH |
712 | |
713 | /* | |
714 | * The following field is re-initialized when this cset gets linked | |
715 | * in cgroup_init(). However, let's initialize the field | |
716 | * statically too so that the default cgroup can be accessed safely | |
717 | * early during boot. | |
718 | */ | |
719 | .dfl_cgrp = &cgrp_dfl_root.cgrp, | |
172a2c06 | 720 | }; |
817929ec | 721 | |
172a2c06 | 722 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 723 | |
454000ad TH |
724 | static bool css_set_threaded(struct css_set *cset) |
725 | { | |
726 | return cset->dom_cset != cset; | |
727 | } | |
728 | ||
0de0942d TH |
729 | /** |
730 | * css_set_populated - does a css_set contain any tasks? | |
731 | * @cset: target css_set | |
73a7242a WL |
732 | * |
733 | * css_set_populated() should be the same as !!cset->nr_tasks at steady | |
734 | * state. However, css_set_populated() can be called while a task is being | |
735 | * added to or removed from the linked list before the nr_tasks is | |
736 | * properly updated. Hence, we can't just look at ->nr_tasks here. | |
0de0942d TH |
737 | */ |
738 | static bool css_set_populated(struct css_set *cset) | |
739 | { | |
f0d9a5f1 | 740 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
741 | |
742 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
743 | } | |
744 | ||
842b597e | 745 | /** |
788b950c | 746 | * cgroup_update_populated - update the populated count of a cgroup |
842b597e TH |
747 | * @cgrp: the target cgroup |
748 | * @populated: inc or dec populated count | |
749 | * | |
0de0942d | 750 | * One of the css_sets associated with @cgrp is either getting its first |
788b950c TH |
751 | * task or losing the last. Update @cgrp->nr_populated_* accordingly. The |
752 | * count is propagated towards root so that a given cgroup's | |
753 | * nr_populated_children is zero iff none of its descendants contain any | |
754 | * tasks. | |
842b597e | 755 | * |
788b950c TH |
756 | * @cgrp's interface file "cgroup.populated" is zero if both |
757 | * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and | |
758 | * 1 otherwise. When the sum changes from or to zero, userland is notified | |
759 | * that the content of the interface file has changed. This can be used to | |
760 | * detect when @cgrp and its descendants become populated or empty. | |
842b597e TH |
761 | */ |
762 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
763 | { | |
788b950c TH |
764 | struct cgroup *child = NULL; |
765 | int adj = populated ? 1 : -1; | |
766 | ||
f0d9a5f1 | 767 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
768 | |
769 | do { | |
788b950c | 770 | bool was_populated = cgroup_is_populated(cgrp); |
842b597e | 771 | |
454000ad | 772 | if (!child) { |
788b950c | 773 | cgrp->nr_populated_csets += adj; |
454000ad TH |
774 | } else { |
775 | if (cgroup_is_threaded(child)) | |
776 | cgrp->nr_populated_threaded_children += adj; | |
777 | else | |
778 | cgrp->nr_populated_domain_children += adj; | |
779 | } | |
842b597e | 780 | |
788b950c | 781 | if (was_populated == cgroup_is_populated(cgrp)) |
842b597e TH |
782 | break; |
783 | ||
d62beb7f | 784 | cgroup1_check_for_release(cgrp); |
6f60eade TH |
785 | cgroup_file_notify(&cgrp->events_file); |
786 | ||
788b950c | 787 | child = cgrp; |
d51f39b0 | 788 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
789 | } while (cgrp); |
790 | } | |
791 | ||
0de0942d TH |
792 | /** |
793 | * css_set_update_populated - update populated state of a css_set | |
794 | * @cset: target css_set | |
795 | * @populated: whether @cset is populated or depopulated | |
796 | * | |
797 | * @cset is either getting the first task or losing the last. Update the | |
788b950c | 798 | * populated counters of all associated cgroups accordingly. |
0de0942d TH |
799 | */ |
800 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
801 | { | |
802 | struct cgrp_cset_link *link; | |
803 | ||
f0d9a5f1 | 804 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
805 | |
806 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
807 | cgroup_update_populated(link->cgrp, populated); | |
808 | } | |
809 | ||
f6d7d049 TH |
810 | /** |
811 | * css_set_move_task - move a task from one css_set to another | |
812 | * @task: task being moved | |
813 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
814 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
815 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
816 | * | |
817 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
818 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
819 | * instead of moved, @to_cset can be NULL. | |
820 | * | |
788b950c | 821 | * This function automatically handles populated counter updates and |
ed27b9f7 TH |
822 | * css_task_iter adjustments but the caller is responsible for managing |
823 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
824 | */ |
825 | static void css_set_move_task(struct task_struct *task, | |
826 | struct css_set *from_cset, struct css_set *to_cset, | |
827 | bool use_mg_tasks) | |
828 | { | |
f0d9a5f1 | 829 | lockdep_assert_held(&css_set_lock); |
f6d7d049 | 830 | |
20b454a6 TH |
831 | if (to_cset && !css_set_populated(to_cset)) |
832 | css_set_update_populated(to_cset, true); | |
833 | ||
f6d7d049 | 834 | if (from_cset) { |
ed27b9f7 TH |
835 | struct css_task_iter *it, *pos; |
836 | ||
f6d7d049 | 837 | WARN_ON_ONCE(list_empty(&task->cg_list)); |
ed27b9f7 TH |
838 | |
839 | /* | |
840 | * @task is leaving, advance task iterators which are | |
841 | * pointing to it so that they can resume at the next | |
842 | * position. Advancing an iterator might remove it from | |
843 | * the list, use safe walk. See css_task_iter_advance*() | |
844 | * for details. | |
845 | */ | |
846 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, | |
847 | iters_node) | |
848 | if (it->task_pos == &task->cg_list) | |
849 | css_task_iter_advance(it); | |
850 | ||
f6d7d049 TH |
851 | list_del_init(&task->cg_list); |
852 | if (!css_set_populated(from_cset)) | |
853 | css_set_update_populated(from_cset, false); | |
854 | } else { | |
855 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
856 | } | |
857 | ||
858 | if (to_cset) { | |
859 | /* | |
860 | * We are synchronized through cgroup_threadgroup_rwsem | |
861 | * against PF_EXITING setting such that we can't race | |
862 | * against cgroup_exit() changing the css_set to | |
863 | * init_css_set and dropping the old one. | |
864 | */ | |
865 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
866 | ||
2ce7135a | 867 | cgroup_move_task(task, to_cset); |
f6d7d049 TH |
868 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : |
869 | &to_cset->tasks); | |
870 | } | |
871 | } | |
872 | ||
7717f7ba PM |
873 | /* |
874 | * hash table for cgroup groups. This improves the performance to find | |
875 | * an existing css_set. This hash doesn't (currently) take into | |
876 | * account cgroups in empty hierarchies. | |
877 | */ | |
472b1053 | 878 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 879 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 880 | |
0ac801fe | 881 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 882 | { |
0ac801fe | 883 | unsigned long key = 0UL; |
30159ec7 TH |
884 | struct cgroup_subsys *ss; |
885 | int i; | |
472b1053 | 886 | |
30159ec7 | 887 | for_each_subsys(ss, i) |
0ac801fe LZ |
888 | key += (unsigned long)css[i]; |
889 | key = (key >> 16) ^ key; | |
472b1053 | 890 | |
0ac801fe | 891 | return key; |
472b1053 LZ |
892 | } |
893 | ||
dcfe149b | 894 | void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 895 | { |
69d0206c | 896 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
897 | struct cgroup_subsys *ss; |
898 | int ssid; | |
5abb8855 | 899 | |
f0d9a5f1 | 900 | lockdep_assert_held(&css_set_lock); |
89c5509b | 901 | |
4b9502e6 | 902 | if (!refcount_dec_and_test(&cset->refcount)) |
146aa1bd | 903 | return; |
81a6a5cd | 904 | |
454000ad TH |
905 | WARN_ON_ONCE(!list_empty(&cset->threaded_csets)); |
906 | ||
53254f90 TH |
907 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
908 | for_each_subsys(ss, ssid) { | |
2d8f243a | 909 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
910 | css_put(cset->subsys[ssid]); |
911 | } | |
5abb8855 | 912 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
913 | css_set_count--; |
914 | ||
69d0206c | 915 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
916 | list_del(&link->cset_link); |
917 | list_del(&link->cgrp_link); | |
2ceb231b TH |
918 | if (cgroup_parent(link->cgrp)) |
919 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 920 | kfree(link); |
81a6a5cd | 921 | } |
2c6ab6d2 | 922 | |
454000ad TH |
923 | if (css_set_threaded(cset)) { |
924 | list_del(&cset->threaded_csets_node); | |
925 | put_css_set_locked(cset->dom_cset); | |
926 | } | |
927 | ||
5abb8855 | 928 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
929 | } |
930 | ||
b326f9d0 | 931 | /** |
7717f7ba | 932 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
933 | * @cset: candidate css_set being tested |
934 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
935 | * @new_cgrp: cgroup that's being entered by the task |
936 | * @template: desired set of css pointers in css_set (pre-calculated) | |
937 | * | |
6f4b7e63 | 938 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
939 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
940 | */ | |
5abb8855 TH |
941 | static bool compare_css_sets(struct css_set *cset, |
942 | struct css_set *old_cset, | |
7717f7ba PM |
943 | struct cgroup *new_cgrp, |
944 | struct cgroup_subsys_state *template[]) | |
945 | { | |
454000ad | 946 | struct cgroup *new_dfl_cgrp; |
7717f7ba PM |
947 | struct list_head *l1, *l2; |
948 | ||
aec3dfcb TH |
949 | /* |
950 | * On the default hierarchy, there can be csets which are | |
951 | * associated with the same set of cgroups but different csses. | |
952 | * Let's first ensure that csses match. | |
953 | */ | |
954 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 955 | return false; |
7717f7ba | 956 | |
454000ad TH |
957 | |
958 | /* @cset's domain should match the default cgroup's */ | |
959 | if (cgroup_on_dfl(new_cgrp)) | |
960 | new_dfl_cgrp = new_cgrp; | |
961 | else | |
962 | new_dfl_cgrp = old_cset->dfl_cgrp; | |
963 | ||
964 | if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp) | |
965 | return false; | |
966 | ||
7717f7ba PM |
967 | /* |
968 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
969 | * different cgroups in hierarchies. As different cgroups may |
970 | * share the same effective css, this comparison is always | |
971 | * necessary. | |
7717f7ba | 972 | */ |
69d0206c TH |
973 | l1 = &cset->cgrp_links; |
974 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 975 | while (1) { |
69d0206c | 976 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 977 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
978 | |
979 | l1 = l1->next; | |
980 | l2 = l2->next; | |
981 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
982 | if (l1 == &cset->cgrp_links) { |
983 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
984 | break; |
985 | } else { | |
69d0206c | 986 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
987 | } |
988 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
989 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
990 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
991 | cgrp1 = link1->cgrp; | |
992 | cgrp2 = link2->cgrp; | |
7717f7ba | 993 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 994 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
995 | |
996 | /* | |
997 | * If this hierarchy is the hierarchy of the cgroup | |
998 | * that's changing, then we need to check that this | |
999 | * css_set points to the new cgroup; if it's any other | |
1000 | * hierarchy, then this css_set should point to the | |
1001 | * same cgroup as the old css_set. | |
1002 | */ | |
5abb8855 TH |
1003 | if (cgrp1->root == new_cgrp->root) { |
1004 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
1005 | return false; |
1006 | } else { | |
5abb8855 | 1007 | if (cgrp1 != cgrp2) |
7717f7ba PM |
1008 | return false; |
1009 | } | |
1010 | } | |
1011 | return true; | |
1012 | } | |
1013 | ||
b326f9d0 TH |
1014 | /** |
1015 | * find_existing_css_set - init css array and find the matching css_set | |
1016 | * @old_cset: the css_set that we're using before the cgroup transition | |
1017 | * @cgrp: the cgroup that we're moving into | |
1018 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 1019 | */ |
5abb8855 TH |
1020 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
1021 | struct cgroup *cgrp, | |
1022 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 1023 | { |
3dd06ffa | 1024 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 1025 | struct cgroup_subsys *ss; |
5abb8855 | 1026 | struct css_set *cset; |
0ac801fe | 1027 | unsigned long key; |
b326f9d0 | 1028 | int i; |
817929ec | 1029 | |
aae8aab4 BB |
1030 | /* |
1031 | * Build the set of subsystem state objects that we want to see in the | |
1032 | * new css_set. while subsystems can change globally, the entries here | |
1033 | * won't change, so no need for locking. | |
1034 | */ | |
30159ec7 | 1035 | for_each_subsys(ss, i) { |
f392e51c | 1036 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
1037 | /* |
1038 | * @ss is in this hierarchy, so we want the | |
1039 | * effective css from @cgrp. | |
1040 | */ | |
fc5a828b | 1041 | template[i] = cgroup_e_css_by_mask(cgrp, ss); |
817929ec | 1042 | } else { |
aec3dfcb TH |
1043 | /* |
1044 | * @ss is not in this hierarchy, so we don't want | |
1045 | * to change the css. | |
1046 | */ | |
5abb8855 | 1047 | template[i] = old_cset->subsys[i]; |
817929ec PM |
1048 | } |
1049 | } | |
1050 | ||
0ac801fe | 1051 | key = css_set_hash(template); |
5abb8855 TH |
1052 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
1053 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
1054 | continue; |
1055 | ||
1056 | /* This css_set matches what we need */ | |
5abb8855 | 1057 | return cset; |
472b1053 | 1058 | } |
817929ec PM |
1059 | |
1060 | /* No existing cgroup group matched */ | |
1061 | return NULL; | |
1062 | } | |
1063 | ||
69d0206c | 1064 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 1065 | { |
69d0206c | 1066 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 1067 | |
69d0206c TH |
1068 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
1069 | list_del(&link->cset_link); | |
36553434 LZ |
1070 | kfree(link); |
1071 | } | |
1072 | } | |
1073 | ||
69d0206c TH |
1074 | /** |
1075 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
1076 | * @count: the number of links to allocate | |
1077 | * @tmp_links: list_head the allocated links are put on | |
1078 | * | |
1079 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
1080 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 1081 | */ |
69d0206c | 1082 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 1083 | { |
69d0206c | 1084 | struct cgrp_cset_link *link; |
817929ec | 1085 | int i; |
69d0206c TH |
1086 | |
1087 | INIT_LIST_HEAD(tmp_links); | |
1088 | ||
817929ec | 1089 | for (i = 0; i < count; i++) { |
f4f4be2b | 1090 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 1091 | if (!link) { |
69d0206c | 1092 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
1093 | return -ENOMEM; |
1094 | } | |
69d0206c | 1095 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
1096 | } |
1097 | return 0; | |
1098 | } | |
1099 | ||
c12f65d4 LZ |
1100 | /** |
1101 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 1102 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 1103 | * @cset: the css_set to be linked |
c12f65d4 LZ |
1104 | * @cgrp: the destination cgroup |
1105 | */ | |
69d0206c TH |
1106 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
1107 | struct cgroup *cgrp) | |
c12f65d4 | 1108 | { |
69d0206c | 1109 | struct cgrp_cset_link *link; |
c12f65d4 | 1110 | |
69d0206c | 1111 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
1112 | |
1113 | if (cgroup_on_dfl(cgrp)) | |
1114 | cset->dfl_cgrp = cgrp; | |
1115 | ||
69d0206c TH |
1116 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
1117 | link->cset = cset; | |
7717f7ba | 1118 | link->cgrp = cgrp; |
842b597e | 1119 | |
7717f7ba | 1120 | /* |
389b9c1b TH |
1121 | * Always add links to the tail of the lists so that the lists are |
1122 | * in choronological order. | |
7717f7ba | 1123 | */ |
389b9c1b | 1124 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 1125 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
1126 | |
1127 | if (cgroup_parent(cgrp)) | |
a590b90d | 1128 | cgroup_get_live(cgrp); |
c12f65d4 LZ |
1129 | } |
1130 | ||
b326f9d0 TH |
1131 | /** |
1132 | * find_css_set - return a new css_set with one cgroup updated | |
1133 | * @old_cset: the baseline css_set | |
1134 | * @cgrp: the cgroup to be updated | |
1135 | * | |
1136 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
1137 | * substituted into the appropriate hierarchy. | |
817929ec | 1138 | */ |
5abb8855 TH |
1139 | static struct css_set *find_css_set(struct css_set *old_cset, |
1140 | struct cgroup *cgrp) | |
817929ec | 1141 | { |
b326f9d0 | 1142 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 1143 | struct css_set *cset; |
69d0206c TH |
1144 | struct list_head tmp_links; |
1145 | struct cgrp_cset_link *link; | |
2d8f243a | 1146 | struct cgroup_subsys *ss; |
0ac801fe | 1147 | unsigned long key; |
2d8f243a | 1148 | int ssid; |
472b1053 | 1149 | |
b326f9d0 TH |
1150 | lockdep_assert_held(&cgroup_mutex); |
1151 | ||
817929ec PM |
1152 | /* First see if we already have a cgroup group that matches |
1153 | * the desired set */ | |
82d6489d | 1154 | spin_lock_irq(&css_set_lock); |
5abb8855 TH |
1155 | cset = find_existing_css_set(old_cset, cgrp, template); |
1156 | if (cset) | |
1157 | get_css_set(cset); | |
82d6489d | 1158 | spin_unlock_irq(&css_set_lock); |
817929ec | 1159 | |
5abb8855 TH |
1160 | if (cset) |
1161 | return cset; | |
817929ec | 1162 | |
f4f4be2b | 1163 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1164 | if (!cset) |
817929ec PM |
1165 | return NULL; |
1166 | ||
69d0206c | 1167 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1168 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1169 | kfree(cset); |
817929ec PM |
1170 | return NULL; |
1171 | } | |
1172 | ||
4b9502e6 | 1173 | refcount_set(&cset->refcount, 1); |
454000ad | 1174 | cset->dom_cset = cset; |
5abb8855 | 1175 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1176 | INIT_LIST_HEAD(&cset->mg_tasks); |
ed27b9f7 | 1177 | INIT_LIST_HEAD(&cset->task_iters); |
454000ad | 1178 | INIT_LIST_HEAD(&cset->threaded_csets); |
5abb8855 | 1179 | INIT_HLIST_NODE(&cset->hlist); |
5f617ebb TH |
1180 | INIT_LIST_HEAD(&cset->cgrp_links); |
1181 | INIT_LIST_HEAD(&cset->mg_preload_node); | |
1182 | INIT_LIST_HEAD(&cset->mg_node); | |
817929ec PM |
1183 | |
1184 | /* Copy the set of subsystem state objects generated in | |
1185 | * find_existing_css_set() */ | |
5abb8855 | 1186 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1187 | |
82d6489d | 1188 | spin_lock_irq(&css_set_lock); |
817929ec | 1189 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1190 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1191 | struct cgroup *c = link->cgrp; |
69d0206c | 1192 | |
7717f7ba PM |
1193 | if (c->root == cgrp->root) |
1194 | c = cgrp; | |
69d0206c | 1195 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1196 | } |
817929ec | 1197 | |
69d0206c | 1198 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1199 | |
817929ec | 1200 | css_set_count++; |
472b1053 | 1201 | |
2d8f243a | 1202 | /* Add @cset to the hash table */ |
5abb8855 TH |
1203 | key = css_set_hash(cset->subsys); |
1204 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1205 | |
53254f90 TH |
1206 | for_each_subsys(ss, ssid) { |
1207 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1208 | ||
2d8f243a | 1209 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1210 | &css->cgroup->e_csets[ssid]); |
1211 | css_get(css); | |
1212 | } | |
2d8f243a | 1213 | |
82d6489d | 1214 | spin_unlock_irq(&css_set_lock); |
817929ec | 1215 | |
454000ad TH |
1216 | /* |
1217 | * If @cset should be threaded, look up the matching dom_cset and | |
1218 | * link them up. We first fully initialize @cset then look for the | |
1219 | * dom_cset. It's simpler this way and safe as @cset is guaranteed | |
1220 | * to stay empty until we return. | |
1221 | */ | |
1222 | if (cgroup_is_threaded(cset->dfl_cgrp)) { | |
1223 | struct css_set *dcset; | |
1224 | ||
1225 | dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp); | |
1226 | if (!dcset) { | |
1227 | put_css_set(cset); | |
1228 | return NULL; | |
1229 | } | |
1230 | ||
1231 | spin_lock_irq(&css_set_lock); | |
1232 | cset->dom_cset = dcset; | |
1233 | list_add_tail(&cset->threaded_csets_node, | |
1234 | &dcset->threaded_csets); | |
1235 | spin_unlock_irq(&css_set_lock); | |
1236 | } | |
1237 | ||
5abb8855 | 1238 | return cset; |
b4f48b63 PM |
1239 | } |
1240 | ||
0a268dbd | 1241 | struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1242 | { |
3dd06ffa | 1243 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1244 | |
3dd06ffa | 1245 | return root_cgrp->root; |
2bd59d48 TH |
1246 | } |
1247 | ||
3dd06ffa | 1248 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1249 | { |
1250 | int id; | |
1251 | ||
1252 | lockdep_assert_held(&cgroup_mutex); | |
1253 | ||
985ed670 | 1254 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1255 | if (id < 0) |
1256 | return id; | |
1257 | ||
1258 | root->hierarchy_id = id; | |
1259 | return 0; | |
1260 | } | |
1261 | ||
3dd06ffa | 1262 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1263 | { |
1264 | lockdep_assert_held(&cgroup_mutex); | |
1265 | ||
8c8a5502 | 1266 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
f2e85d57 TH |
1267 | } |
1268 | ||
1592c9b2 | 1269 | void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1270 | { |
1271 | if (root) { | |
f2e85d57 TH |
1272 | idr_destroy(&root->cgroup_idr); |
1273 | kfree(root); | |
1274 | } | |
1275 | } | |
1276 | ||
3dd06ffa | 1277 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1278 | { |
3dd06ffa | 1279 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1280 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1281 | |
ed1777de TH |
1282 | trace_cgroup_destroy_root(root); |
1283 | ||
334c3679 | 1284 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
f2e85d57 | 1285 | |
776f02fa | 1286 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1287 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1288 | |
f2e85d57 | 1289 | /* Rebind all subsystems back to the default hierarchy */ |
334c3679 | 1290 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
7717f7ba | 1291 | |
7717f7ba | 1292 | /* |
f2e85d57 TH |
1293 | * Release all the links from cset_links to this hierarchy's |
1294 | * root cgroup | |
7717f7ba | 1295 | */ |
82d6489d | 1296 | spin_lock_irq(&css_set_lock); |
f2e85d57 TH |
1297 | |
1298 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1299 | list_del(&link->cset_link); | |
1300 | list_del(&link->cgrp_link); | |
1301 | kfree(link); | |
1302 | } | |
f0d9a5f1 | 1303 | |
82d6489d | 1304 | spin_unlock_irq(&css_set_lock); |
f2e85d57 TH |
1305 | |
1306 | if (!list_empty(&root->root_list)) { | |
1307 | list_del(&root->root_list); | |
1308 | cgroup_root_count--; | |
1309 | } | |
1310 | ||
1311 | cgroup_exit_root_id(root); | |
1312 | ||
1313 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1314 | |
2bd59d48 | 1315 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1316 | cgroup_free_root(root); |
1317 | } | |
1318 | ||
4f41fc59 SH |
1319 | /* |
1320 | * look up cgroup associated with current task's cgroup namespace on the | |
1321 | * specified hierarchy | |
1322 | */ | |
1323 | static struct cgroup * | |
1324 | current_cgns_cgroup_from_root(struct cgroup_root *root) | |
1325 | { | |
1326 | struct cgroup *res = NULL; | |
1327 | struct css_set *cset; | |
1328 | ||
1329 | lockdep_assert_held(&css_set_lock); | |
1330 | ||
1331 | rcu_read_lock(); | |
1332 | ||
1333 | cset = current->nsproxy->cgroup_ns->root_cset; | |
1334 | if (cset == &init_css_set) { | |
1335 | res = &root->cgrp; | |
1336 | } else { | |
1337 | struct cgrp_cset_link *link; | |
1338 | ||
1339 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1340 | struct cgroup *c = link->cgrp; | |
1341 | ||
1342 | if (c->root == root) { | |
1343 | res = c; | |
1344 | break; | |
1345 | } | |
1346 | } | |
1347 | } | |
1348 | rcu_read_unlock(); | |
1349 | ||
1350 | BUG_ON(!res); | |
1351 | return res; | |
1352 | } | |
1353 | ||
ceb6a081 TH |
1354 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1355 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1356 | struct cgroup_root *root) |
7717f7ba | 1357 | { |
7717f7ba PM |
1358 | struct cgroup *res = NULL; |
1359 | ||
96d365e0 | 1360 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1361 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1362 | |
5abb8855 | 1363 | if (cset == &init_css_set) { |
3dd06ffa | 1364 | res = &root->cgrp; |
13d82fb7 TH |
1365 | } else if (root == &cgrp_dfl_root) { |
1366 | res = cset->dfl_cgrp; | |
7717f7ba | 1367 | } else { |
69d0206c TH |
1368 | struct cgrp_cset_link *link; |
1369 | ||
1370 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1371 | struct cgroup *c = link->cgrp; |
69d0206c | 1372 | |
7717f7ba PM |
1373 | if (c->root == root) { |
1374 | res = c; | |
1375 | break; | |
1376 | } | |
1377 | } | |
1378 | } | |
96d365e0 | 1379 | |
7717f7ba PM |
1380 | BUG_ON(!res); |
1381 | return res; | |
1382 | } | |
1383 | ||
ddbcc7e8 | 1384 | /* |
ceb6a081 | 1385 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1386 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 | 1387 | */ |
0a268dbd TH |
1388 | struct cgroup *task_cgroup_from_root(struct task_struct *task, |
1389 | struct cgroup_root *root) | |
ceb6a081 TH |
1390 | { |
1391 | /* | |
1392 | * No need to lock the task - since we hold cgroup_mutex the | |
1393 | * task can't change groups, so the only thing that can happen | |
1394 | * is that it exits and its css is set back to init_css_set. | |
1395 | */ | |
1396 | return cset_cgroup_from_root(task_css_set(task), root); | |
1397 | } | |
1398 | ||
ddbcc7e8 | 1399 | /* |
ddbcc7e8 PM |
1400 | * A task must hold cgroup_mutex to modify cgroups. |
1401 | * | |
1402 | * Any task can increment and decrement the count field without lock. | |
1403 | * So in general, code holding cgroup_mutex can't rely on the count | |
1404 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1405 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1406 | * means that no tasks are currently attached, therefore there is no |
1407 | * way a task attached to that cgroup can fork (the other way to | |
1408 | * increment the count). So code holding cgroup_mutex can safely | |
1409 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1410 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1411 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1412 | * needs that mutex. | |
1413 | * | |
ddbcc7e8 PM |
1414 | * A cgroup can only be deleted if both its 'count' of using tasks |
1415 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1416 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1417 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1418 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1419 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1420 | * |
1421 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1422 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1423 | */ |
1424 | ||
2bd59d48 | 1425 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
a424316c | 1426 | |
8d7e6fb0 TH |
1427 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1428 | char *buf) | |
ddbcc7e8 | 1429 | { |
3e1d2eed TH |
1430 | struct cgroup_subsys *ss = cft->ss; |
1431 | ||
8d7e6fb0 | 1432 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
c1bbd933 TH |
1433 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) { |
1434 | const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : ""; | |
1435 | ||
1436 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s", | |
1437 | dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, | |
3e1d2eed | 1438 | cft->name); |
c1bbd933 | 1439 | } else { |
08a77676 | 1440 | strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX); |
c1bbd933 | 1441 | } |
8d7e6fb0 | 1442 | return buf; |
ddbcc7e8 PM |
1443 | } |
1444 | ||
f2e85d57 TH |
1445 | /** |
1446 | * cgroup_file_mode - deduce file mode of a control file | |
1447 | * @cft: the control file in question | |
1448 | * | |
7dbdb199 | 1449 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1450 | */ |
1451 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1452 | { |
f2e85d57 | 1453 | umode_t mode = 0; |
65dff759 | 1454 | |
f2e85d57 TH |
1455 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1456 | mode |= S_IRUGO; | |
1457 | ||
7dbdb199 TH |
1458 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1459 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1460 | mode |= S_IWUGO; | |
1461 | else | |
1462 | mode |= S_IWUSR; | |
1463 | } | |
f2e85d57 TH |
1464 | |
1465 | return mode; | |
65dff759 LZ |
1466 | } |
1467 | ||
af0ba678 | 1468 | /** |
8699b776 | 1469 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
0f060deb | 1470 | * @subtree_control: the new subtree_control mask to consider |
5ced2518 | 1471 | * @this_ss_mask: available subsystems |
af0ba678 TH |
1472 | * |
1473 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1474 | * enabled together through its ->depends_on mask. In such cases, more | |
1475 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1476 | * | |
0f060deb | 1477 | * This function calculates which subsystems need to be enabled if |
5ced2518 | 1478 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
af0ba678 | 1479 | */ |
5ced2518 | 1480 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
667c2491 | 1481 | { |
6e5c8307 | 1482 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1483 | struct cgroup_subsys *ss; |
1484 | int ssid; | |
1485 | ||
1486 | lockdep_assert_held(&cgroup_mutex); | |
1487 | ||
f6d635ad TH |
1488 | cur_ss_mask |= cgrp_dfl_implicit_ss_mask; |
1489 | ||
af0ba678 | 1490 | while (true) { |
6e5c8307 | 1491 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1492 | |
b4e0eeaf | 1493 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1494 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1495 | } while_each_subsys_mask(); |
af0ba678 TH |
1496 | |
1497 | /* | |
1498 | * Mask out subsystems which aren't available. This can | |
1499 | * happen only if some depended-upon subsystems were bound | |
1500 | * to non-default hierarchies. | |
1501 | */ | |
5ced2518 | 1502 | new_ss_mask &= this_ss_mask; |
af0ba678 TH |
1503 | |
1504 | if (new_ss_mask == cur_ss_mask) | |
1505 | break; | |
1506 | cur_ss_mask = new_ss_mask; | |
1507 | } | |
1508 | ||
0f060deb TH |
1509 | return cur_ss_mask; |
1510 | } | |
1511 | ||
a9746d8d TH |
1512 | /** |
1513 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1514 | * @kn: the kernfs_node being serviced | |
1515 | * | |
1516 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1517 | * the method finishes if locking succeeded. Note that once this function | |
1518 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1519 | * inaccessible any time. If the caller intends to continue to access the | |
1520 | * cgroup, it should pin it before invoking this function. | |
1521 | */ | |
0a268dbd | 1522 | void cgroup_kn_unlock(struct kernfs_node *kn) |
ddbcc7e8 | 1523 | { |
a9746d8d TH |
1524 | struct cgroup *cgrp; |
1525 | ||
1526 | if (kernfs_type(kn) == KERNFS_DIR) | |
1527 | cgrp = kn->priv; | |
1528 | else | |
1529 | cgrp = kn->parent->priv; | |
1530 | ||
1531 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1532 | |
1533 | kernfs_unbreak_active_protection(kn); | |
1534 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1535 | } |
1536 | ||
a9746d8d TH |
1537 | /** |
1538 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1539 | * @kn: the kernfs_node being serviced | |
945ba199 | 1540 | * @drain_offline: perform offline draining on the cgroup |
a9746d8d TH |
1541 | * |
1542 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1543 | * @kn. It breaks the active protection, performs cgroup locking and | |
1544 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1545 | * alive; otherwise, %NULL. A successful return should be undone by a | |
945ba199 TH |
1546 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
1547 | * cgroup is drained of offlining csses before return. | |
a9746d8d TH |
1548 | * |
1549 | * Any cgroup kernfs method implementation which requires locking the | |
1550 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1551 | * locking under kernfs active protection and allows all kernfs operations | |
1552 | * including self-removal. | |
1553 | */ | |
0a268dbd | 1554 | struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline) |
05ef1d7c | 1555 | { |
a9746d8d TH |
1556 | struct cgroup *cgrp; |
1557 | ||
1558 | if (kernfs_type(kn) == KERNFS_DIR) | |
1559 | cgrp = kn->priv; | |
1560 | else | |
1561 | cgrp = kn->parent->priv; | |
05ef1d7c | 1562 | |
2739d3cc | 1563 | /* |
01f6474c | 1564 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1565 | * active_ref. cgroup liveliness check alone provides enough |
1566 | * protection against removal. Ensure @cgrp stays accessible and | |
1567 | * break the active_ref protection. | |
2739d3cc | 1568 | */ |
aa32362f LZ |
1569 | if (!cgroup_tryget(cgrp)) |
1570 | return NULL; | |
a9746d8d TH |
1571 | kernfs_break_active_protection(kn); |
1572 | ||
945ba199 TH |
1573 | if (drain_offline) |
1574 | cgroup_lock_and_drain_offline(cgrp); | |
1575 | else | |
1576 | mutex_lock(&cgroup_mutex); | |
05ef1d7c | 1577 | |
a9746d8d TH |
1578 | if (!cgroup_is_dead(cgrp)) |
1579 | return cgrp; | |
1580 | ||
1581 | cgroup_kn_unlock(kn); | |
1582 | return NULL; | |
ddbcc7e8 | 1583 | } |
05ef1d7c | 1584 | |
2739d3cc | 1585 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1586 | { |
2bd59d48 | 1587 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1588 | |
01f6474c | 1589 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1590 | |
1591 | if (cft->file_offset) { | |
1592 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1593 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1594 | ||
1595 | spin_lock_irq(&cgroup_file_kn_lock); | |
1596 | cfile->kn = NULL; | |
1597 | spin_unlock_irq(&cgroup_file_kn_lock); | |
b12e3583 TH |
1598 | |
1599 | del_timer_sync(&cfile->notify_timer); | |
34c06254 TH |
1600 | } |
1601 | ||
2bd59d48 | 1602 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1603 | } |
1604 | ||
13af07df | 1605 | /** |
4df8dc90 TH |
1606 | * css_clear_dir - remove subsys files in a cgroup directory |
1607 | * @css: taget css | |
13af07df | 1608 | */ |
334c3679 | 1609 | static void css_clear_dir(struct cgroup_subsys_state *css) |
05ef1d7c | 1610 | { |
334c3679 | 1611 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 | 1612 | struct cftype *cfts; |
05ef1d7c | 1613 | |
88cb04b9 TH |
1614 | if (!(css->flags & CSS_VISIBLE)) |
1615 | return; | |
1616 | ||
1617 | css->flags &= ~CSS_VISIBLE; | |
1618 | ||
5faaf05f TH |
1619 | if (!css->ss) { |
1620 | if (cgroup_on_dfl(cgrp)) | |
1621 | cfts = cgroup_base_files; | |
1622 | else | |
1623 | cfts = cgroup1_base_files; | |
1624 | ||
4df8dc90 | 1625 | cgroup_addrm_files(css, cgrp, cfts, false); |
5faaf05f TH |
1626 | } else { |
1627 | list_for_each_entry(cfts, &css->ss->cfts, node) | |
1628 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1629 | } | |
ddbcc7e8 PM |
1630 | } |
1631 | ||
ccdca218 | 1632 | /** |
4df8dc90 TH |
1633 | * css_populate_dir - create subsys files in a cgroup directory |
1634 | * @css: target css | |
ccdca218 TH |
1635 | * |
1636 | * On failure, no file is added. | |
1637 | */ | |
334c3679 | 1638 | static int css_populate_dir(struct cgroup_subsys_state *css) |
ccdca218 | 1639 | { |
334c3679 | 1640 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 TH |
1641 | struct cftype *cfts, *failed_cfts; |
1642 | int ret; | |
ccdca218 | 1643 | |
03970d3c | 1644 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
88cb04b9 TH |
1645 | return 0; |
1646 | ||
4df8dc90 TH |
1647 | if (!css->ss) { |
1648 | if (cgroup_on_dfl(cgrp)) | |
d62beb7f | 1649 | cfts = cgroup_base_files; |
4df8dc90 | 1650 | else |
d62beb7f | 1651 | cfts = cgroup1_base_files; |
ccdca218 | 1652 | |
5faaf05f TH |
1653 | ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1654 | if (ret < 0) | |
1655 | return ret; | |
1656 | } else { | |
1657 | list_for_each_entry(cfts, &css->ss->cfts, node) { | |
1658 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1659 | if (ret < 0) { | |
1660 | failed_cfts = cfts; | |
1661 | goto err; | |
1662 | } | |
ccdca218 TH |
1663 | } |
1664 | } | |
88cb04b9 TH |
1665 | |
1666 | css->flags |= CSS_VISIBLE; | |
1667 | ||
ccdca218 TH |
1668 | return 0; |
1669 | err: | |
4df8dc90 TH |
1670 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1671 | if (cfts == failed_cfts) | |
1672 | break; | |
1673 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1674 | } | |
ccdca218 TH |
1675 | return ret; |
1676 | } | |
1677 | ||
0a268dbd | 1678 | int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1679 | { |
1ada4838 | 1680 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1681 | struct cgroup_subsys *ss; |
2d8f243a | 1682 | int ssid, i, ret; |
ddbcc7e8 | 1683 | |
ace2bee8 | 1684 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1685 | |
b4e0eeaf | 1686 | do_each_subsys_mask(ss, ssid, ss_mask) { |
f6d635ad TH |
1687 | /* |
1688 | * If @ss has non-root csses attached to it, can't move. | |
1689 | * If @ss is an implicit controller, it is exempt from this | |
1690 | * rule and can be stolen. | |
1691 | */ | |
1692 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) && | |
1693 | !ss->implicit_on_dfl) | |
3ed80a62 | 1694 | return -EBUSY; |
1d5be6b2 | 1695 | |
5df36032 | 1696 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1697 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1698 | return -EBUSY; |
b4e0eeaf | 1699 | } while_each_subsys_mask(); |
ddbcc7e8 | 1700 | |
b4e0eeaf | 1701 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1702 | struct cgroup_root *src_root = ss->root; |
1703 | struct cgroup *scgrp = &src_root->cgrp; | |
1704 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1705 | struct css_set *cset; |
a8a648c4 | 1706 | |
1ada4838 | 1707 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1708 | |
334c3679 TH |
1709 | /* disable from the source */ |
1710 | src_root->subsys_mask &= ~(1 << ssid); | |
1711 | WARN_ON(cgroup_apply_control(scgrp)); | |
1712 | cgroup_finalize_control(scgrp, 0); | |
4df8dc90 | 1713 | |
334c3679 | 1714 | /* rebind */ |
1ada4838 TH |
1715 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1716 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1717 | ss->root = dst_root; |
1ada4838 | 1718 | css->cgroup = dcgrp; |
73e80ed8 | 1719 | |
82d6489d | 1720 | spin_lock_irq(&css_set_lock); |
2d8f243a TH |
1721 | hash_for_each(css_set_table, i, cset, hlist) |
1722 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1723 | &dcgrp->e_csets[ss->id]); |
82d6489d | 1724 | spin_unlock_irq(&css_set_lock); |
2d8f243a | 1725 | |
bd53d617 | 1726 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1727 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1728 | if (dst_root == &cgrp_dfl_root) { |
1729 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1730 | } else { | |
1ada4838 | 1731 | dcgrp->subtree_control |= 1 << ssid; |
49d1dc4b | 1732 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1733 | } |
a8a648c4 | 1734 | |
334c3679 TH |
1735 | ret = cgroup_apply_control(dcgrp); |
1736 | if (ret) | |
1737 | pr_warn("partial failure to rebind %s controller (err=%d)\n", | |
1738 | ss->name, ret); | |
1739 | ||
5df36032 TH |
1740 | if (ss->bind) |
1741 | ss->bind(css); | |
b4e0eeaf | 1742 | } while_each_subsys_mask(); |
ddbcc7e8 | 1743 | |
1ada4838 | 1744 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1745 | return 0; |
1746 | } | |
1747 | ||
1592c9b2 TH |
1748 | int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node, |
1749 | struct kernfs_root *kf_root) | |
4f41fc59 | 1750 | { |
09be4c82 | 1751 | int len = 0; |
4f41fc59 SH |
1752 | char *buf = NULL; |
1753 | struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root); | |
1754 | struct cgroup *ns_cgroup; | |
1755 | ||
1756 | buf = kmalloc(PATH_MAX, GFP_KERNEL); | |
1757 | if (!buf) | |
1758 | return -ENOMEM; | |
1759 | ||
82d6489d | 1760 | spin_lock_irq(&css_set_lock); |
4f41fc59 SH |
1761 | ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot); |
1762 | len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX); | |
82d6489d | 1763 | spin_unlock_irq(&css_set_lock); |
4f41fc59 SH |
1764 | |
1765 | if (len >= PATH_MAX) | |
1766 | len = -ERANGE; | |
1767 | else if (len > 0) { | |
1768 | seq_escape(sf, buf, " \t\n\\"); | |
1769 | len = 0; | |
1770 | } | |
1771 | kfree(buf); | |
1772 | return len; | |
1773 | } | |
1774 | ||
5136f636 TH |
1775 | static int parse_cgroup_root_flags(char *data, unsigned int *root_flags) |
1776 | { | |
1777 | char *token; | |
1778 | ||
1779 | *root_flags = 0; | |
1780 | ||
e250d91d | 1781 | if (!data || *data == '\0') |
5136f636 TH |
1782 | return 0; |
1783 | ||
1784 | while ((token = strsep(&data, ",")) != NULL) { | |
1785 | if (!strcmp(token, "nsdelegate")) { | |
1786 | *root_flags |= CGRP_ROOT_NS_DELEGATE; | |
1787 | continue; | |
1788 | } | |
1789 | ||
1790 | pr_err("cgroup2: unknown option \"%s\"\n", token); | |
1791 | return -EINVAL; | |
1792 | } | |
1793 | ||
1794 | return 0; | |
1795 | } | |
1796 | ||
1797 | static void apply_cgroup_root_flags(unsigned int root_flags) | |
1798 | { | |
1799 | if (current->nsproxy->cgroup_ns == &init_cgroup_ns) { | |
1800 | if (root_flags & CGRP_ROOT_NS_DELEGATE) | |
1801 | cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE; | |
1802 | else | |
1803 | cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE; | |
1804 | } | |
1805 | } | |
1806 | ||
1807 | static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root) | |
1808 | { | |
1809 | if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) | |
1810 | seq_puts(seq, ",nsdelegate"); | |
1811 | return 0; | |
1812 | } | |
1813 | ||
2bd59d48 | 1814 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 | 1815 | { |
5136f636 TH |
1816 | unsigned int root_flags; |
1817 | int ret; | |
1818 | ||
1819 | ret = parse_cgroup_root_flags(data, &root_flags); | |
1820 | if (ret) | |
1821 | return ret; | |
1822 | ||
1823 | apply_cgroup_root_flags(root_flags); | |
1824 | return 0; | |
ddbcc7e8 PM |
1825 | } |
1826 | ||
afeb0f9f TH |
1827 | /* |
1828 | * To reduce the fork() overhead for systems that are not actually using | |
1829 | * their cgroups capability, we don't maintain the lists running through | |
1830 | * each css_set to its tasks until we see the list actually used - in other | |
1831 | * words after the first mount. | |
1832 | */ | |
1833 | static bool use_task_css_set_links __read_mostly; | |
1834 | ||
1835 | static void cgroup_enable_task_cg_lists(void) | |
1836 | { | |
1837 | struct task_struct *p, *g; | |
1838 | ||
afeb0f9f TH |
1839 | /* |
1840 | * We need tasklist_lock because RCU is not safe against | |
1841 | * while_each_thread(). Besides, a forking task that has passed | |
1842 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1843 | * is not guaranteed to have its child immediately visible in the | |
1844 | * tasklist if we walk through it with RCU. | |
1845 | */ | |
1846 | read_lock(&tasklist_lock); | |
d8742e22 TH |
1847 | spin_lock_irq(&css_set_lock); |
1848 | ||
1849 | if (use_task_css_set_links) | |
1850 | goto out_unlock; | |
1851 | ||
1852 | use_task_css_set_links = true; | |
1853 | ||
afeb0f9f | 1854 | do_each_thread(g, p) { |
afeb0f9f TH |
1855 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1856 | task_css_set(p) != &init_css_set); | |
1857 | ||
1858 | /* | |
1859 | * We should check if the process is exiting, otherwise | |
1860 | * it will race with cgroup_exit() in that the list | |
1861 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1862 | * Do it while holding siglock so that we don't end up |
1863 | * racing against cgroup_exit(). | |
82d6489d DBO |
1864 | * |
1865 | * Interrupts were already disabled while acquiring | |
1866 | * the css_set_lock, so we do not need to disable it | |
1867 | * again when acquiring the sighand->siglock here. | |
afeb0f9f | 1868 | */ |
82d6489d | 1869 | spin_lock(&p->sighand->siglock); |
eaf797ab TH |
1870 | if (!(p->flags & PF_EXITING)) { |
1871 | struct css_set *cset = task_css_set(p); | |
1872 | ||
0de0942d TH |
1873 | if (!css_set_populated(cset)) |
1874 | css_set_update_populated(cset, true); | |
389b9c1b | 1875 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab | 1876 | get_css_set(cset); |
73a7242a | 1877 | cset->nr_tasks++; |
eaf797ab | 1878 | } |
82d6489d | 1879 | spin_unlock(&p->sighand->siglock); |
afeb0f9f | 1880 | } while_each_thread(g, p); |
afeb0f9f | 1881 | out_unlock: |
82d6489d | 1882 | spin_unlock_irq(&css_set_lock); |
d8742e22 | 1883 | read_unlock(&tasklist_lock); |
afeb0f9f | 1884 | } |
ddbcc7e8 | 1885 | |
cc31edce PM |
1886 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1887 | { | |
2d8f243a TH |
1888 | struct cgroup_subsys *ss; |
1889 | int ssid; | |
1890 | ||
d5c419b6 TH |
1891 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1892 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1893 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1894 | INIT_LIST_HEAD(&cgrp->pidlists); |
1895 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1896 | cgrp->self.cgroup = cgrp; |
184faf32 | 1897 | cgrp->self.flags |= CSS_ONLINE; |
454000ad | 1898 | cgrp->dom_cgrp = cgrp; |
1a926e0b RG |
1899 | cgrp->max_descendants = INT_MAX; |
1900 | cgrp->max_depth = INT_MAX; | |
8f53470b | 1901 | INIT_LIST_HEAD(&cgrp->rstat_css_list); |
d4ff749b | 1902 | prev_cputime_init(&cgrp->prev_cputime); |
2d8f243a TH |
1903 | |
1904 | for_each_subsys(ss, ssid) | |
1905 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1906 | |
1907 | init_waitqueue_head(&cgrp->offline_waitq); | |
d62beb7f | 1908 | INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent); |
cc31edce | 1909 | } |
c6d57f33 | 1910 | |
1592c9b2 | 1911 | void init_cgroup_root(struct cgroup_root *root, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1912 | { |
3dd06ffa | 1913 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1914 | |
ddbcc7e8 | 1915 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1916 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1917 | cgrp->root = root; |
cc31edce | 1918 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1919 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1920 | |
c6d57f33 PM |
1921 | root->flags = opts->flags; |
1922 | if (opts->release_agent) | |
08a77676 | 1923 | strscpy(root->release_agent_path, opts->release_agent, PATH_MAX); |
c6d57f33 | 1924 | if (opts->name) |
08a77676 | 1925 | strscpy(root->name, opts->name, MAX_CGROUP_ROOT_NAMELEN); |
2260e7fc | 1926 | if (opts->cpuset_clone_children) |
3dd06ffa | 1927 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1928 | } |
1929 | ||
35ac1184 | 1930 | int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) |
2c6ab6d2 | 1931 | { |
d427dfeb | 1932 | LIST_HEAD(tmp_links); |
3dd06ffa | 1933 | struct cgroup *root_cgrp = &root->cgrp; |
fa069904 | 1934 | struct kernfs_syscall_ops *kf_sops; |
d427dfeb | 1935 | struct css_set *cset; |
d427dfeb | 1936 | int i, ret; |
2c6ab6d2 | 1937 | |
d427dfeb | 1938 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1939 | |
cf780b7d | 1940 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1941 | if (ret < 0) |
2bd59d48 | 1942 | goto out; |
d427dfeb | 1943 | root_cgrp->id = ret; |
b11cfb58 | 1944 | root_cgrp->ancestor_ids[0] = ret; |
c6d57f33 | 1945 | |
9732adc5 | 1946 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, |
35ac1184 | 1947 | 0, GFP_KERNEL); |
9d755d33 TH |
1948 | if (ret) |
1949 | goto out; | |
1950 | ||
d427dfeb | 1951 | /* |
f0d9a5f1 | 1952 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb | 1953 | * but that's OK - it can only be increased by someone holding |
04313591 TH |
1954 | * cgroup_lock, and that's us. Later rebinding may disable |
1955 | * controllers on the default hierarchy and thus create new csets, | |
1956 | * which can't be more than the existing ones. Allocate 2x. | |
d427dfeb | 1957 | */ |
04313591 | 1958 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
d427dfeb | 1959 | if (ret) |
9d755d33 | 1960 | goto cancel_ref; |
ddbcc7e8 | 1961 | |
985ed670 | 1962 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1963 | if (ret) |
9d755d33 | 1964 | goto cancel_ref; |
ddbcc7e8 | 1965 | |
fa069904 TH |
1966 | kf_sops = root == &cgrp_dfl_root ? |
1967 | &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops; | |
1968 | ||
1969 | root->kf_root = kernfs_create_root(kf_sops, | |
aa818825 SL |
1970 | KERNFS_ROOT_CREATE_DEACTIVATED | |
1971 | KERNFS_ROOT_SUPPORT_EXPORTOP, | |
2bd59d48 TH |
1972 | root_cgrp); |
1973 | if (IS_ERR(root->kf_root)) { | |
1974 | ret = PTR_ERR(root->kf_root); | |
1975 | goto exit_root_id; | |
1976 | } | |
1977 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1978 | |
334c3679 | 1979 | ret = css_populate_dir(&root_cgrp->self); |
d427dfeb | 1980 | if (ret) |
2bd59d48 | 1981 | goto destroy_root; |
ddbcc7e8 | 1982 | |
5df36032 | 1983 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1984 | if (ret) |
2bd59d48 | 1985 | goto destroy_root; |
ddbcc7e8 | 1986 | |
324bda9e AS |
1987 | ret = cgroup_bpf_inherit(root_cgrp); |
1988 | WARN_ON_ONCE(ret); | |
1989 | ||
ed1777de TH |
1990 | trace_cgroup_setup_root(root); |
1991 | ||
d427dfeb TH |
1992 | /* |
1993 | * There must be no failure case after here, since rebinding takes | |
1994 | * care of subsystems' refcounts, which are explicitly dropped in | |
1995 | * the failure exit path. | |
1996 | */ | |
1997 | list_add(&root->root_list, &cgroup_roots); | |
1998 | cgroup_root_count++; | |
0df6a63f | 1999 | |
d427dfeb | 2000 | /* |
3dd06ffa | 2001 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
2002 | * objects. |
2003 | */ | |
82d6489d | 2004 | spin_lock_irq(&css_set_lock); |
0de0942d | 2005 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 2006 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
2007 | if (css_set_populated(cset)) |
2008 | cgroup_update_populated(root_cgrp, true); | |
2009 | } | |
82d6489d | 2010 | spin_unlock_irq(&css_set_lock); |
ddbcc7e8 | 2011 | |
d5c419b6 | 2012 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 2013 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 2014 | |
2bd59d48 | 2015 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 2016 | ret = 0; |
2bd59d48 | 2017 | goto out; |
d427dfeb | 2018 | |
2bd59d48 TH |
2019 | destroy_root: |
2020 | kernfs_destroy_root(root->kf_root); | |
2021 | root->kf_root = NULL; | |
2022 | exit_root_id: | |
d427dfeb | 2023 | cgroup_exit_root_id(root); |
9d755d33 | 2024 | cancel_ref: |
9a1049da | 2025 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 2026 | out: |
d427dfeb TH |
2027 | free_cgrp_cset_links(&tmp_links); |
2028 | return ret; | |
ddbcc7e8 PM |
2029 | } |
2030 | ||
1592c9b2 TH |
2031 | struct dentry *cgroup_do_mount(struct file_system_type *fs_type, int flags, |
2032 | struct cgroup_root *root, unsigned long magic, | |
2033 | struct cgroup_namespace *ns) | |
ddbcc7e8 | 2034 | { |
2bd59d48 | 2035 | struct dentry *dentry; |
399504e2 | 2036 | bool new_sb = false; |
ddbcc7e8 | 2037 | |
633feee3 | 2038 | dentry = kernfs_mount(fs_type, flags, root->kf_root, magic, &new_sb); |
ed82571b | 2039 | |
56fde9e0 | 2040 | /* |
633feee3 TH |
2041 | * In non-init cgroup namespace, instead of root cgroup's dentry, |
2042 | * we return the dentry corresponding to the cgroupns->root_cgrp. | |
56fde9e0 | 2043 | */ |
633feee3 TH |
2044 | if (!IS_ERR(dentry) && ns != &init_cgroup_ns) { |
2045 | struct dentry *nsdentry; | |
399504e2 | 2046 | struct super_block *sb = dentry->d_sb; |
633feee3 | 2047 | struct cgroup *cgrp; |
e37a06f1 | 2048 | |
633feee3 TH |
2049 | mutex_lock(&cgroup_mutex); |
2050 | spin_lock_irq(&css_set_lock); | |
2051 | ||
2052 | cgrp = cset_cgroup_from_root(ns->root_cset, root); | |
2053 | ||
2054 | spin_unlock_irq(&css_set_lock); | |
2055 | mutex_unlock(&cgroup_mutex); | |
2056 | ||
399504e2 | 2057 | nsdentry = kernfs_node_dentry(cgrp->kn, sb); |
633feee3 | 2058 | dput(dentry); |
399504e2 AV |
2059 | if (IS_ERR(nsdentry)) |
2060 | deactivate_locked_super(sb); | |
633feee3 | 2061 | dentry = nsdentry; |
67e9c74b TH |
2062 | } |
2063 | ||
399504e2 | 2064 | if (!new_sb) |
633feee3 TH |
2065 | cgroup_put(&root->cgrp); |
2066 | ||
2067 | return dentry; | |
2068 | } | |
2069 | ||
f7e83571 | 2070 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 2071 | int flags, const char *unused_dev_name, |
f7e83571 | 2072 | void *data) |
ddbcc7e8 | 2073 | { |
ed82571b | 2074 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; |
2bd59d48 | 2075 | struct dentry *dentry; |
5136f636 | 2076 | int ret; |
ddbcc7e8 | 2077 | |
ed82571b SH |
2078 | get_cgroup_ns(ns); |
2079 | ||
2080 | /* Check if the caller has permission to mount. */ | |
2081 | if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) { | |
2082 | put_cgroup_ns(ns); | |
2083 | return ERR_PTR(-EPERM); | |
2084 | } | |
2085 | ||
56fde9e0 TH |
2086 | /* |
2087 | * The first time anyone tries to mount a cgroup, enable the list | |
2088 | * linking each css_set to its tasks and fix up all existing tasks. | |
2089 | */ | |
2090 | if (!use_task_css_set_links) | |
2091 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 2092 | |
633feee3 | 2093 | if (fs_type == &cgroup2_fs_type) { |
5136f636 TH |
2094 | unsigned int root_flags; |
2095 | ||
2096 | ret = parse_cgroup_root_flags(data, &root_flags); | |
2097 | if (ret) { | |
ed82571b | 2098 | put_cgroup_ns(ns); |
5136f636 | 2099 | return ERR_PTR(ret); |
67e9c74b | 2100 | } |
5136f636 | 2101 | |
a7165264 | 2102 | cgrp_dfl_visible = true; |
a590b90d | 2103 | cgroup_get_live(&cgrp_dfl_root.cgrp); |
633feee3 TH |
2104 | |
2105 | dentry = cgroup_do_mount(&cgroup2_fs_type, flags, &cgrp_dfl_root, | |
2106 | CGROUP2_SUPER_MAGIC, ns); | |
5136f636 TH |
2107 | if (!IS_ERR(dentry)) |
2108 | apply_cgroup_root_flags(root_flags); | |
633feee3 TH |
2109 | } else { |
2110 | dentry = cgroup1_mount(&cgroup_fs_type, flags, data, | |
2111 | CGROUP_SUPER_MAGIC, ns); | |
67e9c74b TH |
2112 | } |
2113 | ||
ed82571b | 2114 | put_cgroup_ns(ns); |
2bd59d48 TH |
2115 | return dentry; |
2116 | } | |
8e30e2b8 | 2117 | |
2bd59d48 TH |
2118 | static void cgroup_kill_sb(struct super_block *sb) |
2119 | { | |
2120 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2121 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
a015edd2 | 2122 | |
970317aa | 2123 | /* |
35ac1184 | 2124 | * If @root doesn't have any children, start killing it. |
9d755d33 TH |
2125 | * This prevents new mounts by disabling percpu_ref_tryget_live(). |
2126 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2127 | * |
2128 | * And don't kill the default root. | |
970317aa | 2129 | */ |
35ac1184 AV |
2130 | if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root && |
2131 | !percpu_ref_is_dying(&root->cgrp.self.refcnt)) | |
9d755d33 | 2132 | percpu_ref_kill(&root->cgrp.self.refcnt); |
35ac1184 | 2133 | cgroup_put(&root->cgrp); |
2bd59d48 | 2134 | kernfs_kill_sb(sb); |
ddbcc7e8 | 2135 | } |
970317aa | 2136 | |
0a268dbd | 2137 | struct file_system_type cgroup_fs_type = { |
ddbcc7e8 | 2138 | .name = "cgroup", |
f7e83571 | 2139 | .mount = cgroup_mount, |
ddbcc7e8 | 2140 | .kill_sb = cgroup_kill_sb, |
1c53753e | 2141 | .fs_flags = FS_USERNS_MOUNT, |
ddbcc7e8 | 2142 | }; |
3126121f | 2143 | |
67e9c74b TH |
2144 | static struct file_system_type cgroup2_fs_type = { |
2145 | .name = "cgroup2", | |
2146 | .mount = cgroup_mount, | |
2147 | .kill_sb = cgroup_kill_sb, | |
1c53753e | 2148 | .fs_flags = FS_USERNS_MOUNT, |
67e9c74b | 2149 | }; |
3126121f | 2150 | |
0a268dbd TH |
2151 | int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen, |
2152 | struct cgroup_namespace *ns) | |
a79a908f AK |
2153 | { |
2154 | struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root); | |
a79a908f | 2155 | |
4c737b41 | 2156 | return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen); |
a79a908f AK |
2157 | } |
2158 | ||
4c737b41 TH |
2159 | int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, |
2160 | struct cgroup_namespace *ns) | |
a79a908f | 2161 | { |
4c737b41 | 2162 | int ret; |
a79a908f AK |
2163 | |
2164 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2165 | spin_lock_irq(&css_set_lock); |
a79a908f AK |
2166 | |
2167 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns); | |
2168 | ||
82d6489d | 2169 | spin_unlock_irq(&css_set_lock); |
a79a908f AK |
2170 | mutex_unlock(&cgroup_mutex); |
2171 | ||
2172 | return ret; | |
2173 | } | |
2174 | EXPORT_SYMBOL_GPL(cgroup_path_ns); | |
2175 | ||
857a2beb | 2176 | /** |
913ffdb5 | 2177 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2178 | * @task: target task |
857a2beb TH |
2179 | * @buf: the buffer to write the path into |
2180 | * @buflen: the length of the buffer | |
2181 | * | |
913ffdb5 TH |
2182 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2183 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2184 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2185 | * cgroup controller callbacks. | |
2186 | * | |
e61734c5 | 2187 | * Return value is the same as kernfs_path(). |
857a2beb | 2188 | */ |
4c737b41 | 2189 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2190 | { |
3dd06ffa | 2191 | struct cgroup_root *root; |
913ffdb5 | 2192 | struct cgroup *cgrp; |
e61734c5 | 2193 | int hierarchy_id = 1; |
4c737b41 | 2194 | int ret; |
857a2beb TH |
2195 | |
2196 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2197 | spin_lock_irq(&css_set_lock); |
857a2beb | 2198 | |
913ffdb5 TH |
2199 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2200 | ||
857a2beb TH |
2201 | if (root) { |
2202 | cgrp = task_cgroup_from_root(task, root); | |
4c737b41 | 2203 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns); |
913ffdb5 TH |
2204 | } else { |
2205 | /* if no hierarchy exists, everyone is in "/" */ | |
4c737b41 | 2206 | ret = strlcpy(buf, "/", buflen); |
857a2beb TH |
2207 | } |
2208 | ||
82d6489d | 2209 | spin_unlock_irq(&css_set_lock); |
857a2beb | 2210 | mutex_unlock(&cgroup_mutex); |
4c737b41 | 2211 | return ret; |
857a2beb | 2212 | } |
913ffdb5 | 2213 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2214 | |
adaae5dc | 2215 | /** |
e595cd70 | 2216 | * cgroup_migrate_add_task - add a migration target task to a migration context |
adaae5dc | 2217 | * @task: target task |
e595cd70 | 2218 | * @mgctx: target migration context |
adaae5dc | 2219 | * |
e595cd70 TH |
2220 | * Add @task, which is a migration target, to @mgctx->tset. This function |
2221 | * becomes noop if @task doesn't need to be migrated. @task's css_set | |
2222 | * should have been added as a migration source and @task->cg_list will be | |
2223 | * moved from the css_set's tasks list to mg_tasks one. | |
adaae5dc | 2224 | */ |
e595cd70 TH |
2225 | static void cgroup_migrate_add_task(struct task_struct *task, |
2226 | struct cgroup_mgctx *mgctx) | |
adaae5dc TH |
2227 | { |
2228 | struct css_set *cset; | |
2229 | ||
f0d9a5f1 | 2230 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2231 | |
2232 | /* @task either already exited or can't exit until the end */ | |
2233 | if (task->flags & PF_EXITING) | |
2234 | return; | |
2235 | ||
2236 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2237 | if (list_empty(&task->cg_list)) | |
2238 | return; | |
2239 | ||
2240 | cset = task_css_set(task); | |
2241 | if (!cset->mg_src_cgrp) | |
2242 | return; | |
2243 | ||
61046727 TH |
2244 | mgctx->tset.nr_tasks++; |
2245 | ||
adaae5dc TH |
2246 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2247 | if (list_empty(&cset->mg_node)) | |
e595cd70 TH |
2248 | list_add_tail(&cset->mg_node, |
2249 | &mgctx->tset.src_csets); | |
adaae5dc | 2250 | if (list_empty(&cset->mg_dst_cset->mg_node)) |
d8ebf519 | 2251 | list_add_tail(&cset->mg_dst_cset->mg_node, |
e595cd70 | 2252 | &mgctx->tset.dst_csets); |
adaae5dc TH |
2253 | } |
2254 | ||
2f7ee569 TH |
2255 | /** |
2256 | * cgroup_taskset_first - reset taskset and return the first task | |
2257 | * @tset: taskset of interest | |
1f7dd3e5 | 2258 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2259 | * |
2260 | * @tset iteration is initialized and the first task is returned. | |
2261 | */ | |
1f7dd3e5 TH |
2262 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2263 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2264 | { |
b3dc094e TH |
2265 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2266 | tset->cur_task = NULL; | |
2267 | ||
1f7dd3e5 | 2268 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2269 | } |
2f7ee569 TH |
2270 | |
2271 | /** | |
2272 | * cgroup_taskset_next - iterate to the next task in taskset | |
2273 | * @tset: taskset of interest | |
1f7dd3e5 | 2274 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2275 | * |
2276 | * Return the next task in @tset. Iteration must have been initialized | |
2277 | * with cgroup_taskset_first(). | |
2278 | */ | |
1f7dd3e5 TH |
2279 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2280 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2281 | { |
b3dc094e TH |
2282 | struct css_set *cset = tset->cur_cset; |
2283 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2284 | |
b3dc094e TH |
2285 | while (&cset->mg_node != tset->csets) { |
2286 | if (!task) | |
2287 | task = list_first_entry(&cset->mg_tasks, | |
2288 | struct task_struct, cg_list); | |
2289 | else | |
2290 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2291 | |
b3dc094e TH |
2292 | if (&task->cg_list != &cset->mg_tasks) { |
2293 | tset->cur_cset = cset; | |
2294 | tset->cur_task = task; | |
1f7dd3e5 TH |
2295 | |
2296 | /* | |
2297 | * This function may be called both before and | |
2298 | * after cgroup_taskset_migrate(). The two cases | |
2299 | * can be distinguished by looking at whether @cset | |
2300 | * has its ->mg_dst_cset set. | |
2301 | */ | |
2302 | if (cset->mg_dst_cset) | |
2303 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2304 | else | |
2305 | *dst_cssp = cset->subsys[tset->ssid]; | |
2306 | ||
b3dc094e TH |
2307 | return task; |
2308 | } | |
2f7ee569 | 2309 | |
b3dc094e TH |
2310 | cset = list_next_entry(cset, mg_node); |
2311 | task = NULL; | |
2312 | } | |
2f7ee569 | 2313 | |
b3dc094e | 2314 | return NULL; |
2f7ee569 | 2315 | } |
2f7ee569 | 2316 | |
adaae5dc | 2317 | /** |
37ff9f8f | 2318 | * cgroup_taskset_migrate - migrate a taskset |
e595cd70 | 2319 | * @mgctx: migration context |
adaae5dc | 2320 | * |
e595cd70 | 2321 | * Migrate tasks in @mgctx as setup by migration preparation functions. |
37ff9f8f | 2322 | * This function fails iff one of the ->can_attach callbacks fails and |
e595cd70 TH |
2323 | * guarantees that either all or none of the tasks in @mgctx are migrated. |
2324 | * @mgctx is consumed regardless of success. | |
adaae5dc | 2325 | */ |
bfc2cf6f | 2326 | static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx) |
adaae5dc | 2327 | { |
e595cd70 | 2328 | struct cgroup_taskset *tset = &mgctx->tset; |
37ff9f8f | 2329 | struct cgroup_subsys *ss; |
adaae5dc TH |
2330 | struct task_struct *task, *tmp_task; |
2331 | struct css_set *cset, *tmp_cset; | |
37ff9f8f | 2332 | int ssid, failed_ssid, ret; |
adaae5dc | 2333 | |
adaae5dc | 2334 | /* check that we can legitimately attach to the cgroup */ |
61046727 TH |
2335 | if (tset->nr_tasks) { |
2336 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2337 | if (ss->can_attach) { | |
2338 | tset->ssid = ssid; | |
2339 | ret = ss->can_attach(tset); | |
2340 | if (ret) { | |
2341 | failed_ssid = ssid; | |
2342 | goto out_cancel_attach; | |
2343 | } | |
adaae5dc | 2344 | } |
61046727 TH |
2345 | } while_each_subsys_mask(); |
2346 | } | |
adaae5dc TH |
2347 | |
2348 | /* | |
2349 | * Now that we're guaranteed success, proceed to move all tasks to | |
2350 | * the new cgroup. There are no failure cases after here, so this | |
2351 | * is the commit point. | |
2352 | */ | |
82d6489d | 2353 | spin_lock_irq(&css_set_lock); |
adaae5dc | 2354 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2355 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2356 | struct css_set *from_cset = task_css_set(task); | |
2357 | struct css_set *to_cset = cset->mg_dst_cset; | |
2358 | ||
2359 | get_css_set(to_cset); | |
73a7242a | 2360 | to_cset->nr_tasks++; |
f6d7d049 TH |
2361 | css_set_move_task(task, from_cset, to_cset, true); |
2362 | put_css_set_locked(from_cset); | |
73a7242a | 2363 | from_cset->nr_tasks--; |
f6d7d049 | 2364 | } |
adaae5dc | 2365 | } |
82d6489d | 2366 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2367 | |
2368 | /* | |
2369 | * Migration is committed, all target tasks are now on dst_csets. | |
2370 | * Nothing is sensitive to fork() after this point. Notify | |
2371 | * controllers that migration is complete. | |
2372 | */ | |
2373 | tset->csets = &tset->dst_csets; | |
2374 | ||
61046727 TH |
2375 | if (tset->nr_tasks) { |
2376 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2377 | if (ss->attach) { | |
2378 | tset->ssid = ssid; | |
2379 | ss->attach(tset); | |
2380 | } | |
2381 | } while_each_subsys_mask(); | |
2382 | } | |
adaae5dc TH |
2383 | |
2384 | ret = 0; | |
2385 | goto out_release_tset; | |
2386 | ||
2387 | out_cancel_attach: | |
61046727 TH |
2388 | if (tset->nr_tasks) { |
2389 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2390 | if (ssid == failed_ssid) | |
2391 | break; | |
2392 | if (ss->cancel_attach) { | |
2393 | tset->ssid = ssid; | |
2394 | ss->cancel_attach(tset); | |
2395 | } | |
2396 | } while_each_subsys_mask(); | |
2397 | } | |
adaae5dc | 2398 | out_release_tset: |
82d6489d | 2399 | spin_lock_irq(&css_set_lock); |
adaae5dc TH |
2400 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2401 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2402 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2403 | list_del_init(&cset->mg_node); | |
2404 | } | |
82d6489d | 2405 | spin_unlock_irq(&css_set_lock); |
c4fa6c43 WL |
2406 | |
2407 | /* | |
2408 | * Re-initialize the cgroup_taskset structure in case it is reused | |
2409 | * again in another cgroup_migrate_add_task()/cgroup_migrate_execute() | |
2410 | * iteration. | |
2411 | */ | |
2412 | tset->nr_tasks = 0; | |
2413 | tset->csets = &tset->src_csets; | |
adaae5dc TH |
2414 | return ret; |
2415 | } | |
2416 | ||
6c694c88 | 2417 | /** |
8cfd8147 | 2418 | * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination |
6c694c88 TH |
2419 | * @dst_cgrp: destination cgroup to test |
2420 | * | |
8cfd8147 TH |
2421 | * On the default hierarchy, except for the mixable, (possible) thread root |
2422 | * and threaded cgroups, subtree_control must be zero for migration | |
2423 | * destination cgroups with tasks so that child cgroups don't compete | |
2424 | * against tasks. | |
6c694c88 | 2425 | */ |
8cfd8147 | 2426 | int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp) |
6c694c88 | 2427 | { |
8cfd8147 TH |
2428 | /* v1 doesn't have any restriction */ |
2429 | if (!cgroup_on_dfl(dst_cgrp)) | |
2430 | return 0; | |
2431 | ||
2432 | /* verify @dst_cgrp can host resources */ | |
2433 | if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp)) | |
2434 | return -EOPNOTSUPP; | |
2435 | ||
2436 | /* mixables don't care */ | |
2437 | if (cgroup_is_mixable(dst_cgrp)) | |
2438 | return 0; | |
2439 | ||
2440 | /* | |
2441 | * If @dst_cgrp is already or can become a thread root or is | |
2442 | * threaded, it doesn't matter. | |
2443 | */ | |
2444 | if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp)) | |
2445 | return 0; | |
2446 | ||
2447 | /* apply no-internal-process constraint */ | |
2448 | if (dst_cgrp->subtree_control) | |
2449 | return -EBUSY; | |
2450 | ||
2451 | return 0; | |
6c694c88 TH |
2452 | } |
2453 | ||
a043e3b2 | 2454 | /** |
1958d2d5 | 2455 | * cgroup_migrate_finish - cleanup after attach |
e595cd70 | 2456 | * @mgctx: migration context |
74a1166d | 2457 | * |
1958d2d5 TH |
2458 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2459 | * those functions for details. | |
74a1166d | 2460 | */ |
e595cd70 | 2461 | void cgroup_migrate_finish(struct cgroup_mgctx *mgctx) |
74a1166d | 2462 | { |
e595cd70 | 2463 | LIST_HEAD(preloaded); |
1958d2d5 | 2464 | struct css_set *cset, *tmp_cset; |
74a1166d | 2465 | |
1958d2d5 TH |
2466 | lockdep_assert_held(&cgroup_mutex); |
2467 | ||
82d6489d | 2468 | spin_lock_irq(&css_set_lock); |
e595cd70 TH |
2469 | |
2470 | list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded); | |
2471 | list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded); | |
2472 | ||
2473 | list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) { | |
1958d2d5 | 2474 | cset->mg_src_cgrp = NULL; |
e4857982 | 2475 | cset->mg_dst_cgrp = NULL; |
1958d2d5 TH |
2476 | cset->mg_dst_cset = NULL; |
2477 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2478 | put_css_set_locked(cset); |
1958d2d5 | 2479 | } |
e595cd70 | 2480 | |
82d6489d | 2481 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2482 | } |
2483 | ||
2484 | /** | |
2485 | * cgroup_migrate_add_src - add a migration source css_set | |
2486 | * @src_cset: the source css_set to add | |
2487 | * @dst_cgrp: the destination cgroup | |
e595cd70 | 2488 | * @mgctx: migration context |
1958d2d5 TH |
2489 | * |
2490 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
e595cd70 | 2491 | * @src_cset and add it to @mgctx->src_csets, which should later be cleaned |
1958d2d5 TH |
2492 | * up by cgroup_migrate_finish(). |
2493 | * | |
1ed13287 TH |
2494 | * This function may be called without holding cgroup_threadgroup_rwsem |
2495 | * even if the target is a process. Threads may be created and destroyed | |
2496 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2497 | * into play and the preloaded css_sets are guaranteed to cover all | |
2498 | * migrations. | |
1958d2d5 | 2499 | */ |
0a268dbd TH |
2500 | void cgroup_migrate_add_src(struct css_set *src_cset, |
2501 | struct cgroup *dst_cgrp, | |
e595cd70 | 2502 | struct cgroup_mgctx *mgctx) |
1958d2d5 TH |
2503 | { |
2504 | struct cgroup *src_cgrp; | |
2505 | ||
2506 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2507 | lockdep_assert_held(&css_set_lock); |
1958d2d5 | 2508 | |
2b021cbf TH |
2509 | /* |
2510 | * If ->dead, @src_set is associated with one or more dead cgroups | |
2511 | * and doesn't contain any migratable tasks. Ignore it early so | |
2512 | * that the rest of migration path doesn't get confused by it. | |
2513 | */ | |
2514 | if (src_cset->dead) | |
2515 | return; | |
2516 | ||
1958d2d5 TH |
2517 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); |
2518 | ||
1958d2d5 TH |
2519 | if (!list_empty(&src_cset->mg_preload_node)) |
2520 | return; | |
2521 | ||
2522 | WARN_ON(src_cset->mg_src_cgrp); | |
e4857982 | 2523 | WARN_ON(src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2524 | WARN_ON(!list_empty(&src_cset->mg_tasks)); |
2525 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2526 | ||
2527 | src_cset->mg_src_cgrp = src_cgrp; | |
e4857982 | 2528 | src_cset->mg_dst_cgrp = dst_cgrp; |
1958d2d5 | 2529 | get_css_set(src_cset); |
e595cd70 | 2530 | list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets); |
1958d2d5 TH |
2531 | } |
2532 | ||
2533 | /** | |
2534 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
e595cd70 | 2535 | * @mgctx: migration context |
1958d2d5 | 2536 | * |
e4857982 | 2537 | * Tasks are about to be moved and all the source css_sets have been |
e595cd70 TH |
2538 | * preloaded to @mgctx->preloaded_src_csets. This function looks up and |
2539 | * pins all destination css_sets, links each to its source, and append them | |
2540 | * to @mgctx->preloaded_dst_csets. | |
1958d2d5 TH |
2541 | * |
2542 | * This function must be called after cgroup_migrate_add_src() has been | |
2543 | * called on each migration source css_set. After migration is performed | |
2544 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
e595cd70 | 2545 | * @mgctx. |
1958d2d5 | 2546 | */ |
e595cd70 | 2547 | int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx) |
1958d2d5 | 2548 | { |
f817de98 | 2549 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2550 | |
2551 | lockdep_assert_held(&cgroup_mutex); | |
2552 | ||
2553 | /* look up the dst cset for each src cset and link it to src */ | |
e595cd70 TH |
2554 | list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets, |
2555 | mg_preload_node) { | |
1958d2d5 | 2556 | struct css_set *dst_cset; |
bfc2cf6f TH |
2557 | struct cgroup_subsys *ss; |
2558 | int ssid; | |
1958d2d5 | 2559 | |
e4857982 | 2560 | dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2561 | if (!dst_cset) |
2562 | goto err; | |
2563 | ||
2564 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2565 | |
2566 | /* | |
2567 | * If src cset equals dst, it's noop. Drop the src. | |
2568 | * cgroup_migrate() will skip the cset too. Note that we | |
2569 | * can't handle src == dst as some nodes are used by both. | |
2570 | */ | |
2571 | if (src_cset == dst_cset) { | |
2572 | src_cset->mg_src_cgrp = NULL; | |
e4857982 | 2573 | src_cset->mg_dst_cgrp = NULL; |
f817de98 | 2574 | list_del_init(&src_cset->mg_preload_node); |
a25eb52e ZL |
2575 | put_css_set(src_cset); |
2576 | put_css_set(dst_cset); | |
f817de98 TH |
2577 | continue; |
2578 | } | |
2579 | ||
1958d2d5 TH |
2580 | src_cset->mg_dst_cset = dst_cset; |
2581 | ||
2582 | if (list_empty(&dst_cset->mg_preload_node)) | |
e595cd70 TH |
2583 | list_add_tail(&dst_cset->mg_preload_node, |
2584 | &mgctx->preloaded_dst_csets); | |
1958d2d5 | 2585 | else |
a25eb52e | 2586 | put_css_set(dst_cset); |
bfc2cf6f TH |
2587 | |
2588 | for_each_subsys(ss, ssid) | |
2589 | if (src_cset->subsys[ssid] != dst_cset->subsys[ssid]) | |
2590 | mgctx->ss_mask |= 1 << ssid; | |
1958d2d5 TH |
2591 | } |
2592 | ||
1958d2d5 TH |
2593 | return 0; |
2594 | err: | |
e595cd70 | 2595 | cgroup_migrate_finish(mgctx); |
1958d2d5 TH |
2596 | return -ENOMEM; |
2597 | } | |
2598 | ||
2599 | /** | |
2600 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2601 | * @leader: the leader of the process or the task to migrate |
2602 | * @threadgroup: whether @leader points to the whole process or a single task | |
e595cd70 | 2603 | * @mgctx: migration context |
1958d2d5 | 2604 | * |
37ff9f8f TH |
2605 | * Migrate a process or task denoted by @leader. If migrating a process, |
2606 | * the caller must be holding cgroup_threadgroup_rwsem. The caller is also | |
2607 | * responsible for invoking cgroup_migrate_add_src() and | |
1958d2d5 TH |
2608 | * cgroup_migrate_prepare_dst() on the targets before invoking this |
2609 | * function and following up with cgroup_migrate_finish(). | |
2610 | * | |
2611 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2612 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2613 | * failure, when migrating multiple targets, the success or failure can be | |
2614 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2615 | * actually starting migrating. | |
2616 | */ | |
0a268dbd | 2617 | int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
bfc2cf6f | 2618 | struct cgroup_mgctx *mgctx) |
74a1166d | 2619 | { |
adaae5dc | 2620 | struct task_struct *task; |
74a1166d | 2621 | |
fb5d2b4c MSB |
2622 | /* |
2623 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2624 | * already PF_EXITING could be freed from underneath us unless we | |
2625 | * take an rcu_read_lock. | |
2626 | */ | |
82d6489d | 2627 | spin_lock_irq(&css_set_lock); |
fb5d2b4c | 2628 | rcu_read_lock(); |
9db8de37 | 2629 | task = leader; |
74a1166d | 2630 | do { |
e595cd70 | 2631 | cgroup_migrate_add_task(task, mgctx); |
081aa458 LZ |
2632 | if (!threadgroup) |
2633 | break; | |
9db8de37 | 2634 | } while_each_thread(leader, task); |
fb5d2b4c | 2635 | rcu_read_unlock(); |
82d6489d | 2636 | spin_unlock_irq(&css_set_lock); |
74a1166d | 2637 | |
bfc2cf6f | 2638 | return cgroup_migrate_execute(mgctx); |
74a1166d BB |
2639 | } |
2640 | ||
1958d2d5 TH |
2641 | /** |
2642 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2643 | * @dst_cgrp: the cgroup to attach to | |
2644 | * @leader: the task or the leader of the threadgroup to be attached | |
2645 | * @threadgroup: attach the whole threadgroup? | |
2646 | * | |
1ed13287 | 2647 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 | 2648 | */ |
0a268dbd TH |
2649 | int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader, |
2650 | bool threadgroup) | |
1958d2d5 | 2651 | { |
e595cd70 | 2652 | DEFINE_CGROUP_MGCTX(mgctx); |
1958d2d5 TH |
2653 | struct task_struct *task; |
2654 | int ret; | |
2655 | ||
8cfd8147 TH |
2656 | ret = cgroup_migrate_vet_dst(dst_cgrp); |
2657 | if (ret) | |
2658 | return ret; | |
6c694c88 | 2659 | |
1958d2d5 | 2660 | /* look up all src csets */ |
82d6489d | 2661 | spin_lock_irq(&css_set_lock); |
1958d2d5 TH |
2662 | rcu_read_lock(); |
2663 | task = leader; | |
2664 | do { | |
e595cd70 | 2665 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx); |
1958d2d5 TH |
2666 | if (!threadgroup) |
2667 | break; | |
2668 | } while_each_thread(leader, task); | |
2669 | rcu_read_unlock(); | |
82d6489d | 2670 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2671 | |
2672 | /* prepare dst csets and commit */ | |
e595cd70 | 2673 | ret = cgroup_migrate_prepare_dst(&mgctx); |
1958d2d5 | 2674 | if (!ret) |
bfc2cf6f | 2675 | ret = cgroup_migrate(leader, threadgroup, &mgctx); |
1958d2d5 | 2676 | |
e595cd70 | 2677 | cgroup_migrate_finish(&mgctx); |
ed1777de TH |
2678 | |
2679 | if (!ret) | |
e4f8d81c | 2680 | TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup); |
ed1777de | 2681 | |
1958d2d5 | 2682 | return ret; |
74a1166d BB |
2683 | } |
2684 | ||
715c809d TH |
2685 | struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup) |
2686 | __acquires(&cgroup_threadgroup_rwsem) | |
bbcb81d0 | 2687 | { |
bbcb81d0 | 2688 | struct task_struct *tsk; |
acbef755 | 2689 | pid_t pid; |
bbcb81d0 | 2690 | |
acbef755 | 2691 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
715c809d | 2692 | return ERR_PTR(-EINVAL); |
74a1166d | 2693 | |
3014dde7 | 2694 | percpu_down_write(&cgroup_threadgroup_rwsem); |
715c809d | 2695 | |
b78949eb | 2696 | rcu_read_lock(); |
bbcb81d0 | 2697 | if (pid) { |
73507f33 | 2698 | tsk = find_task_by_vpid(pid); |
74a1166d | 2699 | if (!tsk) { |
715c809d TH |
2700 | tsk = ERR_PTR(-ESRCH); |
2701 | goto out_unlock_threadgroup; | |
bbcb81d0 | 2702 | } |
dedf22e9 | 2703 | } else { |
b78949eb | 2704 | tsk = current; |
dedf22e9 | 2705 | } |
cd3d0952 TH |
2706 | |
2707 | if (threadgroup) | |
b78949eb | 2708 | tsk = tsk->group_leader; |
c4c27fbd MG |
2709 | |
2710 | /* | |
77f88796 TH |
2711 | * kthreads may acquire PF_NO_SETAFFINITY during initialization. |
2712 | * If userland migrates such a kthread to a non-root cgroup, it can | |
2713 | * become trapped in a cpuset, or RT kthread may be born in a | |
2714 | * cgroup with no rt_runtime allocated. Just say no. | |
c4c27fbd | 2715 | */ |
77f88796 | 2716 | if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) { |
715c809d TH |
2717 | tsk = ERR_PTR(-EINVAL); |
2718 | goto out_unlock_threadgroup; | |
c4c27fbd MG |
2719 | } |
2720 | ||
b78949eb | 2721 | get_task_struct(tsk); |
715c809d TH |
2722 | goto out_unlock_rcu; |
2723 | ||
2724 | out_unlock_threadgroup: | |
2725 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
2726 | out_unlock_rcu: | |
b78949eb | 2727 | rcu_read_unlock(); |
715c809d TH |
2728 | return tsk; |
2729 | } | |
b78949eb | 2730 | |
715c809d TH |
2731 | void cgroup_procs_write_finish(struct task_struct *task) |
2732 | __releases(&cgroup_threadgroup_rwsem) | |
2733 | { | |
2734 | struct cgroup_subsys *ss; | |
2735 | int ssid; | |
081aa458 | 2736 | |
715c809d TH |
2737 | /* release reference from cgroup_procs_write_start() */ |
2738 | put_task_struct(task); | |
3014dde7 | 2739 | |
3014dde7 | 2740 | percpu_up_write(&cgroup_threadgroup_rwsem); |
5cf1cacb TH |
2741 | for_each_subsys(ss, ssid) |
2742 | if (ss->post_attach) | |
2743 | ss->post_attach(); | |
af351026 PM |
2744 | } |
2745 | ||
6e5c8307 | 2746 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 2747 | { |
f8f22e53 TH |
2748 | struct cgroup_subsys *ss; |
2749 | bool printed = false; | |
2750 | int ssid; | |
a742c59d | 2751 | |
b4e0eeaf | 2752 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
2753 | if (printed) |
2754 | seq_putc(seq, ' '); | |
2755 | seq_printf(seq, "%s", ss->name); | |
2756 | printed = true; | |
b4e0eeaf | 2757 | } while_each_subsys_mask(); |
f8f22e53 TH |
2758 | if (printed) |
2759 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2760 | } |
2761 | ||
f8f22e53 TH |
2762 | /* show controllers which are enabled from the parent */ |
2763 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2764 | { |
f8f22e53 TH |
2765 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2766 | ||
5531dc91 | 2767 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
f8f22e53 | 2768 | return 0; |
ddbcc7e8 PM |
2769 | } |
2770 | ||
f8f22e53 TH |
2771 | /* show controllers which are enabled for a given cgroup's children */ |
2772 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2773 | { |
f8f22e53 TH |
2774 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2775 | ||
667c2491 | 2776 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2777 | return 0; |
2778 | } | |
2779 | ||
2780 | /** | |
2781 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2782 | * @cgrp: root of the subtree to update csses for | |
2783 | * | |
54962604 TH |
2784 | * @cgrp's control masks have changed and its subtree's css associations |
2785 | * need to be updated accordingly. This function looks up all css_sets | |
2786 | * which are attached to the subtree, creates the matching updated css_sets | |
2787 | * and migrates the tasks to the new ones. | |
f8f22e53 TH |
2788 | */ |
2789 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2790 | { | |
e595cd70 | 2791 | DEFINE_CGROUP_MGCTX(mgctx); |
54962604 TH |
2792 | struct cgroup_subsys_state *d_css; |
2793 | struct cgroup *dsct; | |
f8f22e53 TH |
2794 | struct css_set *src_cset; |
2795 | int ret; | |
2796 | ||
f8f22e53 TH |
2797 | lockdep_assert_held(&cgroup_mutex); |
2798 | ||
3014dde7 TH |
2799 | percpu_down_write(&cgroup_threadgroup_rwsem); |
2800 | ||
f8f22e53 | 2801 | /* look up all csses currently attached to @cgrp's subtree */ |
82d6489d | 2802 | spin_lock_irq(&css_set_lock); |
54962604 | 2803 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
f8f22e53 TH |
2804 | struct cgrp_cset_link *link; |
2805 | ||
54962604 | 2806 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
e595cd70 | 2807 | cgroup_migrate_add_src(link->cset, dsct, &mgctx); |
f8f22e53 | 2808 | } |
82d6489d | 2809 | spin_unlock_irq(&css_set_lock); |
f8f22e53 TH |
2810 | |
2811 | /* NULL dst indicates self on default hierarchy */ | |
e595cd70 | 2812 | ret = cgroup_migrate_prepare_dst(&mgctx); |
f8f22e53 TH |
2813 | if (ret) |
2814 | goto out_finish; | |
2815 | ||
82d6489d | 2816 | spin_lock_irq(&css_set_lock); |
e595cd70 | 2817 | list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) { |
10265075 | 2818 | struct task_struct *task, *ntask; |
f8f22e53 | 2819 | |
10265075 TH |
2820 | /* all tasks in src_csets need to be migrated */ |
2821 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
e595cd70 | 2822 | cgroup_migrate_add_task(task, &mgctx); |
f8f22e53 | 2823 | } |
82d6489d | 2824 | spin_unlock_irq(&css_set_lock); |
f8f22e53 | 2825 | |
bfc2cf6f | 2826 | ret = cgroup_migrate_execute(&mgctx); |
f8f22e53 | 2827 | out_finish: |
e595cd70 | 2828 | cgroup_migrate_finish(&mgctx); |
3014dde7 | 2829 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
2830 | return ret; |
2831 | } | |
2832 | ||
1b9b96a1 | 2833 | /** |
945ba199 | 2834 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
ce3f1d9d | 2835 | * @cgrp: root of the target subtree |
1b9b96a1 TH |
2836 | * |
2837 | * Because css offlining is asynchronous, userland may try to re-enable a | |
945ba199 TH |
2838 | * controller while the previous css is still around. This function grabs |
2839 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. | |
1b9b96a1 | 2840 | */ |
0a268dbd | 2841 | void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
945ba199 | 2842 | __acquires(&cgroup_mutex) |
1b9b96a1 TH |
2843 | { |
2844 | struct cgroup *dsct; | |
ce3f1d9d | 2845 | struct cgroup_subsys_state *d_css; |
1b9b96a1 TH |
2846 | struct cgroup_subsys *ss; |
2847 | int ssid; | |
2848 | ||
945ba199 TH |
2849 | restart: |
2850 | mutex_lock(&cgroup_mutex); | |
1b9b96a1 | 2851 | |
ce3f1d9d | 2852 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
1b9b96a1 TH |
2853 | for_each_subsys(ss, ssid) { |
2854 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
2855 | DEFINE_WAIT(wait); | |
2856 | ||
ce3f1d9d | 2857 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
1b9b96a1 TH |
2858 | continue; |
2859 | ||
a590b90d | 2860 | cgroup_get_live(dsct); |
1b9b96a1 TH |
2861 | prepare_to_wait(&dsct->offline_waitq, &wait, |
2862 | TASK_UNINTERRUPTIBLE); | |
2863 | ||
2864 | mutex_unlock(&cgroup_mutex); | |
2865 | schedule(); | |
2866 | finish_wait(&dsct->offline_waitq, &wait); | |
1b9b96a1 TH |
2867 | |
2868 | cgroup_put(dsct); | |
945ba199 | 2869 | goto restart; |
1b9b96a1 TH |
2870 | } |
2871 | } | |
1b9b96a1 TH |
2872 | } |
2873 | ||
15a27c36 | 2874 | /** |
479adb89 | 2875 | * cgroup_save_control - save control masks and dom_cgrp of a subtree |
15a27c36 TH |
2876 | * @cgrp: root of the target subtree |
2877 | * | |
479adb89 TH |
2878 | * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the |
2879 | * respective old_ prefixed fields for @cgrp's subtree including @cgrp | |
2880 | * itself. | |
15a27c36 TH |
2881 | */ |
2882 | static void cgroup_save_control(struct cgroup *cgrp) | |
2883 | { | |
2884 | struct cgroup *dsct; | |
2885 | struct cgroup_subsys_state *d_css; | |
2886 | ||
2887 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
2888 | dsct->old_subtree_control = dsct->subtree_control; | |
2889 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; | |
479adb89 | 2890 | dsct->old_dom_cgrp = dsct->dom_cgrp; |
15a27c36 TH |
2891 | } |
2892 | } | |
2893 | ||
2894 | /** | |
2895 | * cgroup_propagate_control - refresh control masks of a subtree | |
2896 | * @cgrp: root of the target subtree | |
2897 | * | |
2898 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches | |
2899 | * ->subtree_control and propagate controller availability through the | |
2900 | * subtree so that descendants don't have unavailable controllers enabled. | |
2901 | */ | |
2902 | static void cgroup_propagate_control(struct cgroup *cgrp) | |
2903 | { | |
2904 | struct cgroup *dsct; | |
2905 | struct cgroup_subsys_state *d_css; | |
2906 | ||
2907 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
2908 | dsct->subtree_control &= cgroup_control(dsct); | |
5ced2518 TH |
2909 | dsct->subtree_ss_mask = |
2910 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, | |
2911 | cgroup_ss_mask(dsct)); | |
15a27c36 TH |
2912 | } |
2913 | } | |
2914 | ||
2915 | /** | |
479adb89 | 2916 | * cgroup_restore_control - restore control masks and dom_cgrp of a subtree |
15a27c36 TH |
2917 | * @cgrp: root of the target subtree |
2918 | * | |
479adb89 TH |
2919 | * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the |
2920 | * respective old_ prefixed fields for @cgrp's subtree including @cgrp | |
2921 | * itself. | |
15a27c36 TH |
2922 | */ |
2923 | static void cgroup_restore_control(struct cgroup *cgrp) | |
2924 | { | |
2925 | struct cgroup *dsct; | |
2926 | struct cgroup_subsys_state *d_css; | |
2927 | ||
2928 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { | |
2929 | dsct->subtree_control = dsct->old_subtree_control; | |
2930 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; | |
479adb89 | 2931 | dsct->dom_cgrp = dsct->old_dom_cgrp; |
15a27c36 TH |
2932 | } |
2933 | } | |
2934 | ||
f6d635ad TH |
2935 | static bool css_visible(struct cgroup_subsys_state *css) |
2936 | { | |
2937 | struct cgroup_subsys *ss = css->ss; | |
2938 | struct cgroup *cgrp = css->cgroup; | |
2939 | ||
2940 | if (cgroup_control(cgrp) & (1 << ss->id)) | |
2941 | return true; | |
2942 | if (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) | |
2943 | return false; | |
2944 | return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl; | |
2945 | } | |
2946 | ||
bdb53bd7 TH |
2947 | /** |
2948 | * cgroup_apply_control_enable - enable or show csses according to control | |
ce3f1d9d | 2949 | * @cgrp: root of the target subtree |
bdb53bd7 | 2950 | * |
ce3f1d9d | 2951 | * Walk @cgrp's subtree and create new csses or make the existing ones |
bdb53bd7 TH |
2952 | * visible. A css is created invisible if it's being implicitly enabled |
2953 | * through dependency. An invisible css is made visible when the userland | |
2954 | * explicitly enables it. | |
2955 | * | |
2956 | * Returns 0 on success, -errno on failure. On failure, csses which have | |
2957 | * been processed already aren't cleaned up. The caller is responsible for | |
8a1115ff | 2958 | * cleaning up with cgroup_apply_control_disable(). |
bdb53bd7 TH |
2959 | */ |
2960 | static int cgroup_apply_control_enable(struct cgroup *cgrp) | |
2961 | { | |
2962 | struct cgroup *dsct; | |
ce3f1d9d | 2963 | struct cgroup_subsys_state *d_css; |
bdb53bd7 TH |
2964 | struct cgroup_subsys *ss; |
2965 | int ssid, ret; | |
2966 | ||
ce3f1d9d | 2967 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
bdb53bd7 TH |
2968 | for_each_subsys(ss, ssid) { |
2969 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
2970 | ||
945ba199 TH |
2971 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
2972 | ||
bdb53bd7 TH |
2973 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) |
2974 | continue; | |
2975 | ||
2976 | if (!css) { | |
2977 | css = css_create(dsct, ss); | |
2978 | if (IS_ERR(css)) | |
2979 | return PTR_ERR(css); | |
2980 | } | |
2981 | ||
f6d635ad | 2982 | if (css_visible(css)) { |
334c3679 | 2983 | ret = css_populate_dir(css); |
bdb53bd7 TH |
2984 | if (ret) |
2985 | return ret; | |
2986 | } | |
2987 | } | |
2988 | } | |
2989 | ||
2990 | return 0; | |
2991 | } | |
2992 | ||
12b3bb6a TH |
2993 | /** |
2994 | * cgroup_apply_control_disable - kill or hide csses according to control | |
ce3f1d9d | 2995 | * @cgrp: root of the target subtree |
12b3bb6a | 2996 | * |
ce3f1d9d | 2997 | * Walk @cgrp's subtree and kill and hide csses so that they match |
12b3bb6a TH |
2998 | * cgroup_ss_mask() and cgroup_visible_mask(). |
2999 | * | |
3000 | * A css is hidden when the userland requests it to be disabled while other | |
3001 | * subsystems are still depending on it. The css must not actively control | |
3002 | * resources and be in the vanilla state if it's made visible again later. | |
3003 | * Controllers which may be depended upon should provide ->css_reset() for | |
3004 | * this purpose. | |
3005 | */ | |
3006 | static void cgroup_apply_control_disable(struct cgroup *cgrp) | |
3007 | { | |
3008 | struct cgroup *dsct; | |
ce3f1d9d | 3009 | struct cgroup_subsys_state *d_css; |
12b3bb6a TH |
3010 | struct cgroup_subsys *ss; |
3011 | int ssid; | |
3012 | ||
ce3f1d9d | 3013 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
12b3bb6a TH |
3014 | for_each_subsys(ss, ssid) { |
3015 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3016 | ||
945ba199 TH |
3017 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3018 | ||
12b3bb6a TH |
3019 | if (!css) |
3020 | continue; | |
3021 | ||
334c3679 TH |
3022 | if (css->parent && |
3023 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { | |
12b3bb6a | 3024 | kill_css(css); |
f6d635ad | 3025 | } else if (!css_visible(css)) { |
334c3679 | 3026 | css_clear_dir(css); |
12b3bb6a TH |
3027 | if (ss->css_reset) |
3028 | ss->css_reset(css); | |
3029 | } | |
3030 | } | |
3031 | } | |
3032 | } | |
3033 | ||
f7b2814b TH |
3034 | /** |
3035 | * cgroup_apply_control - apply control mask updates to the subtree | |
3036 | * @cgrp: root of the target subtree | |
3037 | * | |
3038 | * subsystems can be enabled and disabled in a subtree using the following | |
3039 | * steps. | |
3040 | * | |
3041 | * 1. Call cgroup_save_control() to stash the current state. | |
3042 | * 2. Update ->subtree_control masks in the subtree as desired. | |
3043 | * 3. Call cgroup_apply_control() to apply the changes. | |
3044 | * 4. Optionally perform other related operations. | |
3045 | * 5. Call cgroup_finalize_control() to finish up. | |
3046 | * | |
3047 | * This function implements step 3 and propagates the mask changes | |
3048 | * throughout @cgrp's subtree, updates csses accordingly and perform | |
3049 | * process migrations. | |
3050 | */ | |
3051 | static int cgroup_apply_control(struct cgroup *cgrp) | |
3052 | { | |
3053 | int ret; | |
3054 | ||
3055 | cgroup_propagate_control(cgrp); | |
3056 | ||
3057 | ret = cgroup_apply_control_enable(cgrp); | |
3058 | if (ret) | |
3059 | return ret; | |
3060 | ||
3061 | /* | |
fc5a828b | 3062 | * At this point, cgroup_e_css_by_mask() results reflect the new csses |
f7b2814b TH |
3063 | * making the following cgroup_update_dfl_csses() properly update |
3064 | * css associations of all tasks in the subtree. | |
3065 | */ | |
3066 | ret = cgroup_update_dfl_csses(cgrp); | |
3067 | if (ret) | |
3068 | return ret; | |
3069 | ||
3070 | return 0; | |
3071 | } | |
3072 | ||
3073 | /** | |
3074 | * cgroup_finalize_control - finalize control mask update | |
3075 | * @cgrp: root of the target subtree | |
3076 | * @ret: the result of the update | |
3077 | * | |
3078 | * Finalize control mask update. See cgroup_apply_control() for more info. | |
3079 | */ | |
3080 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) | |
3081 | { | |
3082 | if (ret) { | |
3083 | cgroup_restore_control(cgrp); | |
3084 | cgroup_propagate_control(cgrp); | |
3085 | } | |
3086 | ||
3087 | cgroup_apply_control_disable(cgrp); | |
3088 | } | |
3089 | ||
8cfd8147 TH |
3090 | static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable) |
3091 | { | |
3092 | u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask; | |
3093 | ||
3094 | /* if nothing is getting enabled, nothing to worry about */ | |
3095 | if (!enable) | |
3096 | return 0; | |
3097 | ||
3098 | /* can @cgrp host any resources? */ | |
3099 | if (!cgroup_is_valid_domain(cgrp->dom_cgrp)) | |
3100 | return -EOPNOTSUPP; | |
3101 | ||
3102 | /* mixables don't care */ | |
3103 | if (cgroup_is_mixable(cgrp)) | |
3104 | return 0; | |
3105 | ||
3106 | if (domain_enable) { | |
3107 | /* can't enable domain controllers inside a thread subtree */ | |
3108 | if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp)) | |
3109 | return -EOPNOTSUPP; | |
3110 | } else { | |
3111 | /* | |
3112 | * Threaded controllers can handle internal competitions | |
3113 | * and are always allowed inside a (prospective) thread | |
3114 | * subtree. | |
3115 | */ | |
3116 | if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp)) | |
3117 | return 0; | |
3118 | } | |
3119 | ||
3120 | /* | |
3121 | * Controllers can't be enabled for a cgroup with tasks to avoid | |
3122 | * child cgroups competing against tasks. | |
3123 | */ | |
3124 | if (cgroup_has_tasks(cgrp)) | |
3125 | return -EBUSY; | |
3126 | ||
3127 | return 0; | |
3128 | } | |
3129 | ||
f8f22e53 | 3130 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
451af504 TH |
3131 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
3132 | char *buf, size_t nbytes, | |
3133 | loff_t off) | |
f8f22e53 | 3134 | { |
6e5c8307 | 3135 | u16 enable = 0, disable = 0; |
a9746d8d | 3136 | struct cgroup *cgrp, *child; |
f8f22e53 | 3137 | struct cgroup_subsys *ss; |
451af504 | 3138 | char *tok; |
f8f22e53 TH |
3139 | int ssid, ret; |
3140 | ||
3141 | /* | |
d37167ab TH |
3142 | * Parse input - space separated list of subsystem names prefixed |
3143 | * with either + or -. | |
f8f22e53 | 3144 | */ |
451af504 TH |
3145 | buf = strstrip(buf); |
3146 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
3147 | if (tok[0] == '\0') |
3148 | continue; | |
a7165264 | 3149 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
3150 | if (!cgroup_ssid_enabled(ssid) || |
3151 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
3152 | continue; |
3153 | ||
3154 | if (*tok == '+') { | |
7d331fa9 TH |
3155 | enable |= 1 << ssid; |
3156 | disable &= ~(1 << ssid); | |
f8f22e53 | 3157 | } else if (*tok == '-') { |
7d331fa9 TH |
3158 | disable |= 1 << ssid; |
3159 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
3160 | } else { |
3161 | return -EINVAL; | |
3162 | } | |
3163 | break; | |
b4e0eeaf | 3164 | } while_each_subsys_mask(); |
f8f22e53 TH |
3165 | if (ssid == CGROUP_SUBSYS_COUNT) |
3166 | return -EINVAL; | |
3167 | } | |
3168 | ||
945ba199 | 3169 | cgrp = cgroup_kn_lock_live(of->kn, true); |
a9746d8d TH |
3170 | if (!cgrp) |
3171 | return -ENODEV; | |
f8f22e53 TH |
3172 | |
3173 | for_each_subsys(ss, ssid) { | |
3174 | if (enable & (1 << ssid)) { | |
667c2491 | 3175 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
3176 | enable &= ~(1 << ssid); |
3177 | continue; | |
3178 | } | |
3179 | ||
5531dc91 | 3180 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
c29adf24 TH |
3181 | ret = -ENOENT; |
3182 | goto out_unlock; | |
3183 | } | |
f8f22e53 | 3184 | } else if (disable & (1 << ssid)) { |
667c2491 | 3185 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3186 | disable &= ~(1 << ssid); |
3187 | continue; | |
3188 | } | |
3189 | ||
3190 | /* a child has it enabled? */ | |
3191 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3192 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3193 | ret = -EBUSY; |
ddab2b6e | 3194 | goto out_unlock; |
f8f22e53 TH |
3195 | } |
3196 | } | |
3197 | } | |
3198 | } | |
3199 | ||
3200 | if (!enable && !disable) { | |
3201 | ret = 0; | |
ddab2b6e | 3202 | goto out_unlock; |
f8f22e53 TH |
3203 | } |
3204 | ||
8cfd8147 TH |
3205 | ret = cgroup_vet_subtree_control_enable(cgrp, enable); |
3206 | if (ret) | |
27f26753 | 3207 | goto out_unlock; |
f8f22e53 | 3208 | |
15a27c36 TH |
3209 | /* save and update control masks and prepare csses */ |
3210 | cgroup_save_control(cgrp); | |
f63070d3 | 3211 | |
15a27c36 TH |
3212 | cgrp->subtree_control |= enable; |
3213 | cgrp->subtree_control &= ~disable; | |
c29adf24 | 3214 | |
f7b2814b | 3215 | ret = cgroup_apply_control(cgrp); |
f7b2814b | 3216 | cgroup_finalize_control(cgrp, ret); |
3c745417 TH |
3217 | if (ret) |
3218 | goto out_unlock; | |
f8f22e53 TH |
3219 | |
3220 | kernfs_activate(cgrp->kn); | |
f8f22e53 | 3221 | out_unlock: |
a9746d8d | 3222 | cgroup_kn_unlock(of->kn); |
451af504 | 3223 | return ret ?: nbytes; |
f8f22e53 TH |
3224 | } |
3225 | ||
c705a00d TH |
3226 | /** |
3227 | * cgroup_enable_threaded - make @cgrp threaded | |
3228 | * @cgrp: the target cgroup | |
3229 | * | |
3230 | * Called when "threaded" is written to the cgroup.type interface file and | |
3231 | * tries to make @cgrp threaded and join the parent's resource domain. | |
3232 | * This function is never called on the root cgroup as cgroup.type doesn't | |
3233 | * exist on it. | |
3234 | */ | |
8cfd8147 TH |
3235 | static int cgroup_enable_threaded(struct cgroup *cgrp) |
3236 | { | |
3237 | struct cgroup *parent = cgroup_parent(cgrp); | |
3238 | struct cgroup *dom_cgrp = parent->dom_cgrp; | |
479adb89 TH |
3239 | struct cgroup *dsct; |
3240 | struct cgroup_subsys_state *d_css; | |
8cfd8147 TH |
3241 | int ret; |
3242 | ||
3243 | lockdep_assert_held(&cgroup_mutex); | |
3244 | ||
3245 | /* noop if already threaded */ | |
3246 | if (cgroup_is_threaded(cgrp)) | |
3247 | return 0; | |
3248 | ||
d1897c95 TH |
3249 | /* |
3250 | * If @cgroup is populated or has domain controllers enabled, it | |
3251 | * can't be switched. While the below cgroup_can_be_thread_root() | |
3252 | * test can catch the same conditions, that's only when @parent is | |
3253 | * not mixable, so let's check it explicitly. | |
3254 | */ | |
3255 | if (cgroup_is_populated(cgrp) || | |
3256 | cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask) | |
3257 | return -EOPNOTSUPP; | |
3258 | ||
8cfd8147 TH |
3259 | /* we're joining the parent's domain, ensure its validity */ |
3260 | if (!cgroup_is_valid_domain(dom_cgrp) || | |
3261 | !cgroup_can_be_thread_root(dom_cgrp)) | |
3262 | return -EOPNOTSUPP; | |
3263 | ||
8cfd8147 TH |
3264 | /* |
3265 | * The following shouldn't cause actual migrations and should | |
3266 | * always succeed. | |
3267 | */ | |
3268 | cgroup_save_control(cgrp); | |
3269 | ||
479adb89 TH |
3270 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) |
3271 | if (dsct == cgrp || cgroup_is_threaded(dsct)) | |
3272 | dsct->dom_cgrp = dom_cgrp; | |
3273 | ||
8cfd8147 TH |
3274 | ret = cgroup_apply_control(cgrp); |
3275 | if (!ret) | |
3276 | parent->nr_threaded_children++; | |
8cfd8147 TH |
3277 | |
3278 | cgroup_finalize_control(cgrp, ret); | |
3279 | return ret; | |
3280 | } | |
3281 | ||
3282 | static int cgroup_type_show(struct seq_file *seq, void *v) | |
3283 | { | |
3284 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3285 | ||
3286 | if (cgroup_is_threaded(cgrp)) | |
3287 | seq_puts(seq, "threaded\n"); | |
3288 | else if (!cgroup_is_valid_domain(cgrp)) | |
3289 | seq_puts(seq, "domain invalid\n"); | |
3290 | else if (cgroup_is_thread_root(cgrp)) | |
3291 | seq_puts(seq, "domain threaded\n"); | |
3292 | else | |
3293 | seq_puts(seq, "domain\n"); | |
3294 | ||
3295 | return 0; | |
3296 | } | |
3297 | ||
3298 | static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf, | |
3299 | size_t nbytes, loff_t off) | |
3300 | { | |
3301 | struct cgroup *cgrp; | |
3302 | int ret; | |
3303 | ||
3304 | /* only switching to threaded mode is supported */ | |
3305 | if (strcmp(strstrip(buf), "threaded")) | |
3306 | return -EINVAL; | |
3307 | ||
3308 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3309 | if (!cgrp) | |
3310 | return -ENOENT; | |
3311 | ||
3312 | /* threaded can only be enabled */ | |
3313 | ret = cgroup_enable_threaded(cgrp); | |
3314 | ||
3315 | cgroup_kn_unlock(of->kn); | |
3316 | return ret ?: nbytes; | |
3317 | } | |
3318 | ||
1a926e0b RG |
3319 | static int cgroup_max_descendants_show(struct seq_file *seq, void *v) |
3320 | { | |
3321 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3322 | int descendants = READ_ONCE(cgrp->max_descendants); | |
3323 | ||
3324 | if (descendants == INT_MAX) | |
3325 | seq_puts(seq, "max\n"); | |
3326 | else | |
3327 | seq_printf(seq, "%d\n", descendants); | |
3328 | ||
3329 | return 0; | |
3330 | } | |
3331 | ||
3332 | static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of, | |
3333 | char *buf, size_t nbytes, loff_t off) | |
3334 | { | |
3335 | struct cgroup *cgrp; | |
3336 | int descendants; | |
3337 | ssize_t ret; | |
3338 | ||
3339 | buf = strstrip(buf); | |
3340 | if (!strcmp(buf, "max")) { | |
3341 | descendants = INT_MAX; | |
3342 | } else { | |
3343 | ret = kstrtoint(buf, 0, &descendants); | |
3344 | if (ret) | |
3345 | return ret; | |
3346 | } | |
3347 | ||
696b98f2 | 3348 | if (descendants < 0) |
1a926e0b RG |
3349 | return -ERANGE; |
3350 | ||
3351 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3352 | if (!cgrp) | |
3353 | return -ENOENT; | |
3354 | ||
3355 | cgrp->max_descendants = descendants; | |
3356 | ||
3357 | cgroup_kn_unlock(of->kn); | |
3358 | ||
3359 | return nbytes; | |
3360 | } | |
3361 | ||
3362 | static int cgroup_max_depth_show(struct seq_file *seq, void *v) | |
3363 | { | |
3364 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3365 | int depth = READ_ONCE(cgrp->max_depth); | |
3366 | ||
3367 | if (depth == INT_MAX) | |
3368 | seq_puts(seq, "max\n"); | |
3369 | else | |
3370 | seq_printf(seq, "%d\n", depth); | |
3371 | ||
3372 | return 0; | |
3373 | } | |
3374 | ||
3375 | static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of, | |
3376 | char *buf, size_t nbytes, loff_t off) | |
3377 | { | |
3378 | struct cgroup *cgrp; | |
3379 | ssize_t ret; | |
3380 | int depth; | |
3381 | ||
3382 | buf = strstrip(buf); | |
3383 | if (!strcmp(buf, "max")) { | |
3384 | depth = INT_MAX; | |
3385 | } else { | |
3386 | ret = kstrtoint(buf, 0, &depth); | |
3387 | if (ret) | |
3388 | return ret; | |
3389 | } | |
3390 | ||
696b98f2 | 3391 | if (depth < 0) |
1a926e0b RG |
3392 | return -ERANGE; |
3393 | ||
3394 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3395 | if (!cgrp) | |
3396 | return -ENOENT; | |
3397 | ||
3398 | cgrp->max_depth = depth; | |
3399 | ||
3400 | cgroup_kn_unlock(of->kn); | |
3401 | ||
3402 | return nbytes; | |
3403 | } | |
3404 | ||
4a07c222 | 3405 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3406 | { |
4a07c222 | 3407 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 3408 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
3409 | return 0; |
3410 | } | |
3411 | ||
3e48930c | 3412 | static int cgroup_stat_show(struct seq_file *seq, void *v) |
ec39225c RG |
3413 | { |
3414 | struct cgroup *cgroup = seq_css(seq)->cgroup; | |
3415 | ||
3416 | seq_printf(seq, "nr_descendants %d\n", | |
3417 | cgroup->nr_descendants); | |
3418 | seq_printf(seq, "nr_dying_descendants %d\n", | |
3419 | cgroup->nr_dying_descendants); | |
3420 | ||
3421 | return 0; | |
3422 | } | |
3423 | ||
d41bf8c9 TH |
3424 | static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq, |
3425 | struct cgroup *cgrp, int ssid) | |
3426 | { | |
3427 | struct cgroup_subsys *ss = cgroup_subsys[ssid]; | |
3428 | struct cgroup_subsys_state *css; | |
3429 | int ret; | |
3430 | ||
3431 | if (!ss->css_extra_stat_show) | |
3432 | return 0; | |
3433 | ||
3434 | css = cgroup_tryget_css(cgrp, ss); | |
3435 | if (!css) | |
3436 | return 0; | |
3437 | ||
3438 | ret = ss->css_extra_stat_show(seq, css); | |
3439 | css_put(css); | |
3440 | return ret; | |
3441 | } | |
3442 | ||
3443 | static int cpu_stat_show(struct seq_file *seq, void *v) | |
3444 | { | |
c3ba1329 | 3445 | struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup; |
d41bf8c9 TH |
3446 | int ret = 0; |
3447 | ||
d4ff749b | 3448 | cgroup_base_stat_cputime_show(seq); |
d41bf8c9 TH |
3449 | #ifdef CONFIG_CGROUP_SCHED |
3450 | ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id); | |
3451 | #endif | |
3452 | return ret; | |
3453 | } | |
3454 | ||
2ce7135a JW |
3455 | #ifdef CONFIG_PSI |
3456 | static int cgroup_io_pressure_show(struct seq_file *seq, void *v) | |
3457 | { | |
3458 | return psi_show(seq, &seq_css(seq)->cgroup->psi, PSI_IO); | |
3459 | } | |
3460 | static int cgroup_memory_pressure_show(struct seq_file *seq, void *v) | |
3461 | { | |
3462 | return psi_show(seq, &seq_css(seq)->cgroup->psi, PSI_MEM); | |
3463 | } | |
3464 | static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v) | |
3465 | { | |
3466 | return psi_show(seq, &seq_css(seq)->cgroup->psi, PSI_CPU); | |
3467 | } | |
3468 | #endif | |
3469 | ||
e90cbebc TH |
3470 | static int cgroup_file_open(struct kernfs_open_file *of) |
3471 | { | |
3472 | struct cftype *cft = of->kn->priv; | |
3473 | ||
3474 | if (cft->open) | |
3475 | return cft->open(of); | |
3476 | return 0; | |
3477 | } | |
3478 | ||
3479 | static void cgroup_file_release(struct kernfs_open_file *of) | |
3480 | { | |
3481 | struct cftype *cft = of->kn->priv; | |
3482 | ||
3483 | if (cft->release) | |
3484 | cft->release(of); | |
3485 | } | |
3486 | ||
2bd59d48 TH |
3487 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3488 | size_t nbytes, loff_t off) | |
355e0c48 | 3489 | { |
5136f636 | 3490 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; |
2bd59d48 TH |
3491 | struct cgroup *cgrp = of->kn->parent->priv; |
3492 | struct cftype *cft = of->kn->priv; | |
3493 | struct cgroup_subsys_state *css; | |
a742c59d | 3494 | int ret; |
355e0c48 | 3495 | |
5136f636 TH |
3496 | /* |
3497 | * If namespaces are delegation boundaries, disallow writes to | |
3498 | * files in an non-init namespace root from inside the namespace | |
3499 | * except for the files explicitly marked delegatable - | |
3500 | * cgroup.procs and cgroup.subtree_control. | |
3501 | */ | |
3502 | if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) && | |
3503 | !(cft->flags & CFTYPE_NS_DELEGATABLE) && | |
3504 | ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp) | |
3505 | return -EPERM; | |
3506 | ||
b4168640 TH |
3507 | if (cft->write) |
3508 | return cft->write(of, buf, nbytes, off); | |
3509 | ||
2bd59d48 TH |
3510 | /* |
3511 | * kernfs guarantees that a file isn't deleted with operations in | |
3512 | * flight, which means that the matching css is and stays alive and | |
3513 | * doesn't need to be pinned. The RCU locking is not necessary | |
3514 | * either. It's just for the convenience of using cgroup_css(). | |
3515 | */ | |
3516 | rcu_read_lock(); | |
3517 | css = cgroup_css(cgrp, cft->ss); | |
3518 | rcu_read_unlock(); | |
a742c59d | 3519 | |
451af504 | 3520 | if (cft->write_u64) { |
a742c59d TH |
3521 | unsigned long long v; |
3522 | ret = kstrtoull(buf, 0, &v); | |
3523 | if (!ret) | |
3524 | ret = cft->write_u64(css, cft, v); | |
3525 | } else if (cft->write_s64) { | |
3526 | long long v; | |
3527 | ret = kstrtoll(buf, 0, &v); | |
3528 | if (!ret) | |
3529 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3530 | } else { |
a742c59d | 3531 | ret = -EINVAL; |
e73d2c61 | 3532 | } |
2bd59d48 | 3533 | |
a742c59d | 3534 | return ret ?: nbytes; |
355e0c48 PM |
3535 | } |
3536 | ||
6612f05b | 3537 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3538 | { |
2bd59d48 | 3539 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3540 | } |
3541 | ||
6612f05b | 3542 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3543 | { |
2bd59d48 | 3544 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3545 | } |
3546 | ||
6612f05b | 3547 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3548 | { |
e90cbebc TH |
3549 | if (seq_cft(seq)->seq_stop) |
3550 | seq_cft(seq)->seq_stop(seq, v); | |
ddbcc7e8 PM |
3551 | } |
3552 | ||
91796569 | 3553 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3554 | { |
7da11279 TH |
3555 | struct cftype *cft = seq_cft(m); |
3556 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3557 | |
2da8ca82 TH |
3558 | if (cft->seq_show) |
3559 | return cft->seq_show(m, arg); | |
e73d2c61 | 3560 | |
f4c753b7 | 3561 | if (cft->read_u64) |
896f5199 TH |
3562 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3563 | else if (cft->read_s64) | |
3564 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3565 | else | |
3566 | return -EINVAL; | |
3567 | return 0; | |
91796569 PM |
3568 | } |
3569 | ||
2bd59d48 TH |
3570 | static struct kernfs_ops cgroup_kf_single_ops = { |
3571 | .atomic_write_len = PAGE_SIZE, | |
e90cbebc TH |
3572 | .open = cgroup_file_open, |
3573 | .release = cgroup_file_release, | |
2bd59d48 TH |
3574 | .write = cgroup_file_write, |
3575 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3576 | }; |
3577 | ||
2bd59d48 TH |
3578 | static struct kernfs_ops cgroup_kf_ops = { |
3579 | .atomic_write_len = PAGE_SIZE, | |
e90cbebc TH |
3580 | .open = cgroup_file_open, |
3581 | .release = cgroup_file_release, | |
2bd59d48 TH |
3582 | .write = cgroup_file_write, |
3583 | .seq_start = cgroup_seqfile_start, | |
3584 | .seq_next = cgroup_seqfile_next, | |
3585 | .seq_stop = cgroup_seqfile_stop, | |
3586 | .seq_show = cgroup_seqfile_show, | |
3587 | }; | |
ddbcc7e8 | 3588 | |
49957f8e TH |
3589 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3590 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3591 | { | |
3592 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3593 | .ia_uid = current_fsuid(), | |
3594 | .ia_gid = current_fsgid(), }; | |
3595 | ||
3596 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3597 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3598 | return 0; | |
3599 | ||
3600 | return kernfs_setattr(kn, &iattr); | |
3601 | } | |
3602 | ||
b12e3583 TH |
3603 | static void cgroup_file_notify_timer(struct timer_list *timer) |
3604 | { | |
3605 | cgroup_file_notify(container_of(timer, struct cgroup_file, | |
3606 | notify_timer)); | |
3607 | } | |
3608 | ||
4df8dc90 TH |
3609 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3610 | struct cftype *cft) | |
ddbcc7e8 | 3611 | { |
8d7e6fb0 | 3612 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3613 | struct kernfs_node *kn; |
3614 | struct lock_class_key *key = NULL; | |
49957f8e | 3615 | int ret; |
05ef1d7c | 3616 | |
2bd59d48 TH |
3617 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3618 | key = &cft->lockdep_key; | |
3619 | #endif | |
3620 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
488dee96 DT |
3621 | cgroup_file_mode(cft), |
3622 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, | |
3623 | 0, cft->kf_ops, cft, | |
dfeb0750 | 3624 | NULL, key); |
49957f8e TH |
3625 | if (IS_ERR(kn)) |
3626 | return PTR_ERR(kn); | |
3627 | ||
3628 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3629 | if (ret) { |
49957f8e | 3630 | kernfs_remove(kn); |
f8f22e53 TH |
3631 | return ret; |
3632 | } | |
3633 | ||
6f60eade TH |
3634 | if (cft->file_offset) { |
3635 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3636 | ||
b12e3583 TH |
3637 | timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0); |
3638 | ||
34c06254 | 3639 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3640 | cfile->kn = kn; |
34c06254 | 3641 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3642 | } |
3643 | ||
f8f22e53 | 3644 | return 0; |
ddbcc7e8 PM |
3645 | } |
3646 | ||
b1f28d31 TH |
3647 | /** |
3648 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3649 | * @css: the target css |
3650 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3651 | * @cfts: array of cftypes to be added |
3652 | * @is_add: whether to add or remove | |
3653 | * | |
3654 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3655 | * For removals, this function never fails. |
b1f28d31 | 3656 | */ |
4df8dc90 TH |
3657 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3658 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3659 | bool is_add) |
ddbcc7e8 | 3660 | { |
6732ed85 | 3661 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 3662 | int ret = 0; |
b1f28d31 | 3663 | |
01f6474c | 3664 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3665 | |
6732ed85 TH |
3666 | restart: |
3667 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3668 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3669 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3670 | continue; |
05ebb6e6 | 3671 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3672 | continue; |
d51f39b0 | 3673 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3674 | continue; |
d51f39b0 | 3675 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 | 3676 | continue; |
5cf8114d WL |
3677 | if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug) |
3678 | continue; | |
2739d3cc | 3679 | if (is_add) { |
4df8dc90 | 3680 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3681 | if (ret) { |
ed3d261b JP |
3682 | pr_warn("%s: failed to add %s, err=%d\n", |
3683 | __func__, cft->name, ret); | |
6732ed85 TH |
3684 | cft_end = cft; |
3685 | is_add = false; | |
3686 | goto restart; | |
b1f28d31 | 3687 | } |
2739d3cc LZ |
3688 | } else { |
3689 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3690 | } |
ddbcc7e8 | 3691 | } |
b598dde3 | 3692 | return ret; |
ddbcc7e8 PM |
3693 | } |
3694 | ||
21a2d343 | 3695 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 | 3696 | { |
2bb566cb | 3697 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3698 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3699 | struct cgroup_subsys_state *css; |
9ccece80 | 3700 | int ret = 0; |
8e3f6541 | 3701 | |
01f6474c | 3702 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3703 | |
e8c82d20 | 3704 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3705 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3706 | struct cgroup *cgrp = css->cgroup; |
3707 | ||
88cb04b9 | 3708 | if (!(css->flags & CSS_VISIBLE)) |
e8c82d20 LZ |
3709 | continue; |
3710 | ||
4df8dc90 | 3711 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3712 | if (ret) |
3713 | break; | |
8e3f6541 | 3714 | } |
21a2d343 TH |
3715 | |
3716 | if (is_add && !ret) | |
3717 | kernfs_activate(root->kn); | |
9ccece80 | 3718 | return ret; |
8e3f6541 TH |
3719 | } |
3720 | ||
2da440a2 | 3721 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3722 | { |
2bb566cb | 3723 | struct cftype *cft; |
8e3f6541 | 3724 | |
2bd59d48 TH |
3725 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3726 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3727 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3728 | kfree(cft->kf_ops); | |
3729 | cft->kf_ops = NULL; | |
2da440a2 | 3730 | cft->ss = NULL; |
a8ddc821 TH |
3731 | |
3732 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3733 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3734 | } |
2da440a2 TH |
3735 | } |
3736 | ||
2bd59d48 | 3737 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3738 | { |
3739 | struct cftype *cft; | |
3740 | ||
2bd59d48 TH |
3741 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3742 | struct kernfs_ops *kf_ops; | |
3743 | ||
0adb0704 TH |
3744 | WARN_ON(cft->ss || cft->kf_ops); |
3745 | ||
2bd59d48 TH |
3746 | if (cft->seq_start) |
3747 | kf_ops = &cgroup_kf_ops; | |
3748 | else | |
3749 | kf_ops = &cgroup_kf_single_ops; | |
3750 | ||
3751 | /* | |
3752 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3753 | * make a copy of kf_ops to set its atomic_write_len. | |
3754 | */ | |
3755 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3756 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3757 | if (!kf_ops) { | |
3758 | cgroup_exit_cftypes(cfts); | |
3759 | return -ENOMEM; | |
3760 | } | |
3761 | kf_ops->atomic_write_len = cft->max_write_len; | |
3762 | } | |
8e3f6541 | 3763 | |
2bd59d48 | 3764 | cft->kf_ops = kf_ops; |
2bb566cb | 3765 | cft->ss = ss; |
2bd59d48 | 3766 | } |
2bb566cb | 3767 | |
2bd59d48 | 3768 | return 0; |
2da440a2 TH |
3769 | } |
3770 | ||
21a2d343 TH |
3771 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3772 | { | |
01f6474c | 3773 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3774 | |
3775 | if (!cfts || !cfts[0].ss) | |
3776 | return -ENOENT; | |
3777 | ||
3778 | list_del(&cfts->node); | |
3779 | cgroup_apply_cftypes(cfts, false); | |
3780 | cgroup_exit_cftypes(cfts); | |
3781 | return 0; | |
8e3f6541 | 3782 | } |
8e3f6541 | 3783 | |
79578621 TH |
3784 | /** |
3785 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3786 | * @cfts: zero-length name terminated array of cftypes |
3787 | * | |
2bb566cb TH |
3788 | * Unregister @cfts. Files described by @cfts are removed from all |
3789 | * existing cgroups and all future cgroups won't have them either. This | |
3790 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3791 | * |
3792 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3793 | * registered. |
79578621 | 3794 | */ |
2bb566cb | 3795 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3796 | { |
21a2d343 | 3797 | int ret; |
79578621 | 3798 | |
01f6474c | 3799 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3800 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3801 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3802 | return ret; |
80b13586 TH |
3803 | } |
3804 | ||
8e3f6541 TH |
3805 | /** |
3806 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3807 | * @ss: target cgroup subsystem | |
3808 | * @cfts: zero-length name terminated array of cftypes | |
3809 | * | |
3810 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3811 | * existing cgroups to which @ss is attached and all future cgroups will | |
3812 | * have them too. This function can be called anytime whether @ss is | |
3813 | * attached or not. | |
3814 | * | |
3815 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3816 | * function currently returns 0 as long as @cfts registration is successful | |
3817 | * even if some file creation attempts on existing cgroups fail. | |
3818 | */ | |
2cf669a5 | 3819 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3820 | { |
9ccece80 | 3821 | int ret; |
8e3f6541 | 3822 | |
fc5ed1e9 | 3823 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3824 | return 0; |
3825 | ||
dc5736ed LZ |
3826 | if (!cfts || cfts[0].name[0] == '\0') |
3827 | return 0; | |
2bb566cb | 3828 | |
2bd59d48 TH |
3829 | ret = cgroup_init_cftypes(ss, cfts); |
3830 | if (ret) | |
3831 | return ret; | |
79578621 | 3832 | |
01f6474c | 3833 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3834 | |
0adb0704 | 3835 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3836 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3837 | if (ret) |
21a2d343 | 3838 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3839 | |
01f6474c | 3840 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3841 | return ret; |
79578621 TH |
3842 | } |
3843 | ||
a8ddc821 TH |
3844 | /** |
3845 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3846 | * @ss: target cgroup subsystem | |
3847 | * @cfts: zero-length name terminated array of cftypes | |
3848 | * | |
3849 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3850 | * the default hierarchy. | |
3851 | */ | |
3852 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3853 | { | |
3854 | struct cftype *cft; | |
3855 | ||
3856 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3857 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3858 | return cgroup_add_cftypes(ss, cfts); |
3859 | } | |
3860 | ||
3861 | /** | |
3862 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3863 | * @ss: target cgroup subsystem | |
3864 | * @cfts: zero-length name terminated array of cftypes | |
3865 | * | |
3866 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3867 | * the legacy hierarchies. | |
3868 | */ | |
2cf669a5 TH |
3869 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3870 | { | |
a8ddc821 TH |
3871 | struct cftype *cft; |
3872 | ||
e4b7037c TH |
3873 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3874 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3875 | return cgroup_add_cftypes(ss, cfts); |
3876 | } | |
3877 | ||
34c06254 TH |
3878 | /** |
3879 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3880 | * @cfile: target cgroup_file | |
3881 | * | |
3882 | * @cfile must have been obtained by setting cftype->file_offset. | |
3883 | */ | |
3884 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3885 | { | |
3886 | unsigned long flags; | |
3887 | ||
3888 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
b12e3583 TH |
3889 | if (cfile->kn) { |
3890 | unsigned long last = cfile->notified_at; | |
3891 | unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV; | |
3892 | ||
3893 | if (time_in_range(jiffies, last, next)) { | |
3894 | timer_reduce(&cfile->notify_timer, next); | |
3895 | } else { | |
3896 | kernfs_notify(cfile->kn); | |
3897 | cfile->notified_at = jiffies; | |
3898 | } | |
3899 | } | |
34c06254 TH |
3900 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); |
3901 | } | |
3902 | ||
53fa5261 | 3903 | /** |
492eb21b | 3904 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3905 | * @pos: the current position (%NULL to initiate traversal) |
3906 | * @parent: css whose children to walk | |
53fa5261 | 3907 | * |
c2931b70 | 3908 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3909 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3910 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3911 | * be returned regardless of their states. | |
3912 | * | |
3913 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3914 | * css which finished ->css_online() is guaranteed to be visible in the | |
3915 | * future iterations and will stay visible until the last reference is put. | |
3916 | * A css which hasn't finished ->css_online() or already finished | |
3917 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3918 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3919 | */ |
c2931b70 TH |
3920 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3921 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3922 | { |
c2931b70 | 3923 | struct cgroup_subsys_state *next; |
53fa5261 | 3924 | |
8353da1f | 3925 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3926 | |
3927 | /* | |
de3f0341 TH |
3928 | * @pos could already have been unlinked from the sibling list. |
3929 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3930 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3931 | * @pos is taken off list, at which time its next pointer is valid, | |
3932 | * and, as releases are serialized, the one pointed to by the next | |
3933 | * pointer is guaranteed to not have started release yet. This | |
3934 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3935 | * critical section, the one pointed to by its next pointer is | |
3936 | * guaranteed to not have finished its RCU grace period even if we | |
3937 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3938 | * |
de3f0341 TH |
3939 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3940 | * dereferenced; however, as each css is given a monotonically | |
3941 | * increasing unique serial number and always appended to the | |
3942 | * sibling list, the next one can be found by walking the parent's | |
3943 | * children until the first css with higher serial number than | |
3944 | * @pos's. While this path can be slower, it happens iff iteration | |
3945 | * races against release and the race window is very small. | |
53fa5261 | 3946 | */ |
3b287a50 | 3947 | if (!pos) { |
c2931b70 TH |
3948 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3949 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3950 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3951 | } else { |
c2931b70 | 3952 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3953 | if (next->serial_nr > pos->serial_nr) |
3954 | break; | |
53fa5261 TH |
3955 | } |
3956 | ||
3b281afb TH |
3957 | /* |
3958 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3959 | * the next sibling. |
3b281afb | 3960 | */ |
c2931b70 TH |
3961 | if (&next->sibling != &parent->children) |
3962 | return next; | |
3b281afb | 3963 | return NULL; |
53fa5261 | 3964 | } |
53fa5261 | 3965 | |
574bd9f7 | 3966 | /** |
492eb21b | 3967 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3968 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3969 | * @root: css whose descendants to walk |
574bd9f7 | 3970 | * |
492eb21b | 3971 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3972 | * to visit for pre-order traversal of @root's descendants. @root is |
3973 | * included in the iteration and the first node to be visited. | |
75501a6d | 3974 | * |
87fb54f1 TH |
3975 | * While this function requires cgroup_mutex or RCU read locking, it |
3976 | * doesn't require the whole traversal to be contained in a single critical | |
3977 | * section. This function will return the correct next descendant as long | |
3978 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3979 | * |
3980 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3981 | * css which finished ->css_online() is guaranteed to be visible in the | |
3982 | * future iterations and will stay visible until the last reference is put. | |
3983 | * A css which hasn't finished ->css_online() or already finished | |
3984 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3985 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3986 | */ |
492eb21b TH |
3987 | struct cgroup_subsys_state * |
3988 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3989 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3990 | { |
492eb21b | 3991 | struct cgroup_subsys_state *next; |
574bd9f7 | 3992 | |
8353da1f | 3993 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3994 | |
bd8815a6 | 3995 | /* if first iteration, visit @root */ |
7805d000 | 3996 | if (!pos) |
bd8815a6 | 3997 | return root; |
574bd9f7 TH |
3998 | |
3999 | /* visit the first child if exists */ | |
492eb21b | 4000 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
4001 | if (next) |
4002 | return next; | |
4003 | ||
4004 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 4005 | while (pos != root) { |
5c9d535b | 4006 | next = css_next_child(pos, pos->parent); |
75501a6d | 4007 | if (next) |
574bd9f7 | 4008 | return next; |
5c9d535b | 4009 | pos = pos->parent; |
7805d000 | 4010 | } |
574bd9f7 TH |
4011 | |
4012 | return NULL; | |
4013 | } | |
574bd9f7 | 4014 | |
12a9d2fe | 4015 | /** |
492eb21b TH |
4016 | * css_rightmost_descendant - return the rightmost descendant of a css |
4017 | * @pos: css of interest | |
12a9d2fe | 4018 | * |
492eb21b TH |
4019 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
4020 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 4021 | * subtree of @pos. |
75501a6d | 4022 | * |
87fb54f1 TH |
4023 | * While this function requires cgroup_mutex or RCU read locking, it |
4024 | * doesn't require the whole traversal to be contained in a single critical | |
4025 | * section. This function will return the correct rightmost descendant as | |
4026 | * long as @pos is accessible. | |
12a9d2fe | 4027 | */ |
492eb21b TH |
4028 | struct cgroup_subsys_state * |
4029 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 4030 | { |
492eb21b | 4031 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 4032 | |
8353da1f | 4033 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
4034 | |
4035 | do { | |
4036 | last = pos; | |
4037 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
4038 | pos = NULL; | |
492eb21b | 4039 | css_for_each_child(tmp, last) |
12a9d2fe TH |
4040 | pos = tmp; |
4041 | } while (pos); | |
4042 | ||
4043 | return last; | |
4044 | } | |
12a9d2fe | 4045 | |
492eb21b TH |
4046 | static struct cgroup_subsys_state * |
4047 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 4048 | { |
492eb21b | 4049 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
4050 | |
4051 | do { | |
4052 | last = pos; | |
492eb21b | 4053 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
4054 | } while (pos); |
4055 | ||
4056 | return last; | |
4057 | } | |
4058 | ||
4059 | /** | |
492eb21b | 4060 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 4061 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 4062 | * @root: css whose descendants to walk |
574bd9f7 | 4063 | * |
492eb21b | 4064 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
4065 | * to visit for post-order traversal of @root's descendants. @root is |
4066 | * included in the iteration and the last node to be visited. | |
75501a6d | 4067 | * |
87fb54f1 TH |
4068 | * While this function requires cgroup_mutex or RCU read locking, it |
4069 | * doesn't require the whole traversal to be contained in a single critical | |
4070 | * section. This function will return the correct next descendant as long | |
4071 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
4072 | * @cgroup. | |
c2931b70 TH |
4073 | * |
4074 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4075 | * css which finished ->css_online() is guaranteed to be visible in the | |
4076 | * future iterations and will stay visible until the last reference is put. | |
4077 | * A css which hasn't finished ->css_online() or already finished | |
4078 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4079 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4080 | */ |
492eb21b TH |
4081 | struct cgroup_subsys_state * |
4082 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
4083 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4084 | { |
492eb21b | 4085 | struct cgroup_subsys_state *next; |
574bd9f7 | 4086 | |
8353da1f | 4087 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4088 | |
58b79a91 TH |
4089 | /* if first iteration, visit leftmost descendant which may be @root */ |
4090 | if (!pos) | |
4091 | return css_leftmost_descendant(root); | |
574bd9f7 | 4092 | |
bd8815a6 TH |
4093 | /* if we visited @root, we're done */ |
4094 | if (pos == root) | |
4095 | return NULL; | |
4096 | ||
574bd9f7 | 4097 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 4098 | next = css_next_child(pos, pos->parent); |
75501a6d | 4099 | if (next) |
492eb21b | 4100 | return css_leftmost_descendant(next); |
574bd9f7 TH |
4101 | |
4102 | /* no sibling left, visit parent */ | |
5c9d535b | 4103 | return pos->parent; |
574bd9f7 | 4104 | } |
574bd9f7 | 4105 | |
f3d46500 TH |
4106 | /** |
4107 | * css_has_online_children - does a css have online children | |
4108 | * @css: the target css | |
4109 | * | |
4110 | * Returns %true if @css has any online children; otherwise, %false. This | |
4111 | * function can be called from any context but the caller is responsible | |
4112 | * for synchronizing against on/offlining as necessary. | |
4113 | */ | |
4114 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 4115 | { |
f3d46500 TH |
4116 | struct cgroup_subsys_state *child; |
4117 | bool ret = false; | |
cbc125ef TH |
4118 | |
4119 | rcu_read_lock(); | |
f3d46500 | 4120 | css_for_each_child(child, css) { |
99bae5f9 | 4121 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
4122 | ret = true; |
4123 | break; | |
cbc125ef TH |
4124 | } |
4125 | } | |
4126 | rcu_read_unlock(); | |
f3d46500 | 4127 | return ret; |
574bd9f7 | 4128 | } |
574bd9f7 | 4129 | |
450ee0c1 TH |
4130 | static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it) |
4131 | { | |
4132 | struct list_head *l; | |
4133 | struct cgrp_cset_link *link; | |
4134 | struct css_set *cset; | |
4135 | ||
4136 | lockdep_assert_held(&css_set_lock); | |
4137 | ||
4138 | /* find the next threaded cset */ | |
4139 | if (it->tcset_pos) { | |
4140 | l = it->tcset_pos->next; | |
4141 | ||
4142 | if (l != it->tcset_head) { | |
4143 | it->tcset_pos = l; | |
4144 | return container_of(l, struct css_set, | |
4145 | threaded_csets_node); | |
4146 | } | |
4147 | ||
4148 | it->tcset_pos = NULL; | |
4149 | } | |
4150 | ||
4151 | /* find the next cset */ | |
4152 | l = it->cset_pos; | |
4153 | l = l->next; | |
4154 | if (l == it->cset_head) { | |
4155 | it->cset_pos = NULL; | |
4156 | return NULL; | |
4157 | } | |
4158 | ||
4159 | if (it->ss) { | |
4160 | cset = container_of(l, struct css_set, e_cset_node[it->ss->id]); | |
4161 | } else { | |
4162 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
4163 | cset = link->cset; | |
4164 | } | |
4165 | ||
4166 | it->cset_pos = l; | |
4167 | ||
4168 | /* initialize threaded css_set walking */ | |
4169 | if (it->flags & CSS_TASK_ITER_THREADED) { | |
4170 | if (it->cur_dcset) | |
4171 | put_css_set_locked(it->cur_dcset); | |
4172 | it->cur_dcset = cset; | |
4173 | get_css_set(cset); | |
4174 | ||
4175 | it->tcset_head = &cset->threaded_csets; | |
4176 | it->tcset_pos = &cset->threaded_csets; | |
4177 | } | |
4178 | ||
4179 | return cset; | |
4180 | } | |
4181 | ||
0942eeee | 4182 | /** |
ecb9d535 | 4183 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
4184 | * @it: the iterator to advance |
4185 | * | |
4186 | * Advance @it to the next css_set to walk. | |
d515876e | 4187 | */ |
ecb9d535 | 4188 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 4189 | { |
d515876e TH |
4190 | struct css_set *cset; |
4191 | ||
f0d9a5f1 | 4192 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 4193 | |
d515876e TH |
4194 | /* Advance to the next non-empty css_set */ |
4195 | do { | |
450ee0c1 TH |
4196 | cset = css_task_iter_next_css_set(it); |
4197 | if (!cset) { | |
ecb9d535 | 4198 | it->task_pos = NULL; |
d515876e TH |
4199 | return; |
4200 | } | |
0de0942d | 4201 | } while (!css_set_populated(cset)); |
c7561128 | 4202 | |
c7561128 | 4203 | if (!list_empty(&cset->tasks)) |
0f0a2b4f | 4204 | it->task_pos = cset->tasks.next; |
c7561128 | 4205 | else |
0f0a2b4f TH |
4206 | it->task_pos = cset->mg_tasks.next; |
4207 | ||
4208 | it->tasks_head = &cset->tasks; | |
4209 | it->mg_tasks_head = &cset->mg_tasks; | |
ed27b9f7 TH |
4210 | |
4211 | /* | |
4212 | * We don't keep css_sets locked across iteration steps and thus | |
4213 | * need to take steps to ensure that iteration can be resumed after | |
4214 | * the lock is re-acquired. Iteration is performed at two levels - | |
4215 | * css_sets and tasks in them. | |
4216 | * | |
4217 | * Once created, a css_set never leaves its cgroup lists, so a | |
4218 | * pinned css_set is guaranteed to stay put and we can resume | |
4219 | * iteration afterwards. | |
4220 | * | |
4221 | * Tasks may leave @cset across iteration steps. This is resolved | |
4222 | * by registering each iterator with the css_set currently being | |
4223 | * walked and making css_set_move_task() advance iterators whose | |
4224 | * next task is leaving. | |
4225 | */ | |
4226 | if (it->cur_cset) { | |
4227 | list_del(&it->iters_node); | |
4228 | put_css_set_locked(it->cur_cset); | |
4229 | } | |
4230 | get_css_set(cset); | |
4231 | it->cur_cset = cset; | |
4232 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
4233 | } |
4234 | ||
ecb9d535 TH |
4235 | static void css_task_iter_advance(struct css_task_iter *it) |
4236 | { | |
74d0833c | 4237 | struct list_head *next; |
ecb9d535 | 4238 | |
f0d9a5f1 | 4239 | lockdep_assert_held(&css_set_lock); |
bc2fb7ed | 4240 | repeat: |
e9d81a1b TH |
4241 | if (it->task_pos) { |
4242 | /* | |
4243 | * Advance iterator to find next entry. cset->tasks is | |
4244 | * consumed first and then ->mg_tasks. After ->mg_tasks, | |
4245 | * we move onto the next cset. | |
4246 | */ | |
4247 | next = it->task_pos->next; | |
ecb9d535 | 4248 | |
e9d81a1b TH |
4249 | if (next == it->tasks_head) |
4250 | next = it->mg_tasks_head->next; | |
ecb9d535 | 4251 | |
e9d81a1b TH |
4252 | if (next == it->mg_tasks_head) |
4253 | css_task_iter_advance_css_set(it); | |
4254 | else | |
4255 | it->task_pos = next; | |
4256 | } else { | |
4257 | /* called from start, proceed to the first cset */ | |
ecb9d535 | 4258 | css_task_iter_advance_css_set(it); |
e9d81a1b | 4259 | } |
bc2fb7ed TH |
4260 | |
4261 | /* if PROCS, skip over tasks which aren't group leaders */ | |
4262 | if ((it->flags & CSS_TASK_ITER_PROCS) && it->task_pos && | |
4263 | !thread_group_leader(list_entry(it->task_pos, struct task_struct, | |
4264 | cg_list))) | |
4265 | goto repeat; | |
ecb9d535 TH |
4266 | } |
4267 | ||
0942eeee | 4268 | /** |
72ec7029 TH |
4269 | * css_task_iter_start - initiate task iteration |
4270 | * @css: the css to walk tasks of | |
bc2fb7ed | 4271 | * @flags: CSS_TASK_ITER_* flags |
0942eeee TH |
4272 | * @it: the task iterator to use |
4273 | * | |
72ec7029 TH |
4274 | * Initiate iteration through the tasks of @css. The caller can call |
4275 | * css_task_iter_next() to walk through the tasks until the function | |
4276 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
4277 | * called. | |
0942eeee | 4278 | */ |
bc2fb7ed | 4279 | void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags, |
72ec7029 | 4280 | struct css_task_iter *it) |
817929ec | 4281 | { |
56fde9e0 TH |
4282 | /* no one should try to iterate before mounting cgroups */ |
4283 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 4284 | |
ed27b9f7 TH |
4285 | memset(it, 0, sizeof(*it)); |
4286 | ||
82d6489d | 4287 | spin_lock_irq(&css_set_lock); |
c59cd3d8 | 4288 | |
3ebb2b6e | 4289 | it->ss = css->ss; |
bc2fb7ed | 4290 | it->flags = flags; |
3ebb2b6e TH |
4291 | |
4292 | if (it->ss) | |
4293 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
4294 | else | |
4295 | it->cset_pos = &css->cgroup->cset_links; | |
4296 | ||
0f0a2b4f | 4297 | it->cset_head = it->cset_pos; |
c59cd3d8 | 4298 | |
e9d81a1b | 4299 | css_task_iter_advance(it); |
ed27b9f7 | 4300 | |
82d6489d | 4301 | spin_unlock_irq(&css_set_lock); |
817929ec PM |
4302 | } |
4303 | ||
0942eeee | 4304 | /** |
72ec7029 | 4305 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
4306 | * @it: the task iterator being iterated |
4307 | * | |
4308 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
4309 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
4310 | * reaches the end. | |
0942eeee | 4311 | */ |
72ec7029 | 4312 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 4313 | { |
d5745675 | 4314 | if (it->cur_task) { |
ed27b9f7 | 4315 | put_task_struct(it->cur_task); |
d5745675 TH |
4316 | it->cur_task = NULL; |
4317 | } | |
ed27b9f7 | 4318 | |
82d6489d | 4319 | spin_lock_irq(&css_set_lock); |
ed27b9f7 | 4320 | |
d5745675 TH |
4321 | if (it->task_pos) { |
4322 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
4323 | cg_list); | |
4324 | get_task_struct(it->cur_task); | |
4325 | css_task_iter_advance(it); | |
4326 | } | |
ed27b9f7 | 4327 | |
82d6489d | 4328 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4329 | |
4330 | return it->cur_task; | |
817929ec PM |
4331 | } |
4332 | ||
0942eeee | 4333 | /** |
72ec7029 | 4334 | * css_task_iter_end - finish task iteration |
0942eeee TH |
4335 | * @it: the task iterator to finish |
4336 | * | |
72ec7029 | 4337 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 4338 | */ |
72ec7029 | 4339 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 4340 | { |
ed27b9f7 | 4341 | if (it->cur_cset) { |
82d6489d | 4342 | spin_lock_irq(&css_set_lock); |
ed27b9f7 TH |
4343 | list_del(&it->iters_node); |
4344 | put_css_set_locked(it->cur_cset); | |
82d6489d | 4345 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4346 | } |
4347 | ||
450ee0c1 TH |
4348 | if (it->cur_dcset) |
4349 | put_css_set(it->cur_dcset); | |
4350 | ||
ed27b9f7 TH |
4351 | if (it->cur_task) |
4352 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4353 | } |
4354 | ||
b4b90a8e | 4355 | static void cgroup_procs_release(struct kernfs_open_file *of) |
31a7df01 | 4356 | { |
b4b90a8e TH |
4357 | if (of->priv) { |
4358 | css_task_iter_end(of->priv); | |
4359 | kfree(of->priv); | |
4360 | } | |
4361 | } | |
6c694c88 | 4362 | |
b4b90a8e TH |
4363 | static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos) |
4364 | { | |
4365 | struct kernfs_open_file *of = s->private; | |
4366 | struct css_task_iter *it = of->priv; | |
31a7df01 | 4367 | |
bc2fb7ed | 4368 | return css_task_iter_next(it); |
b4b90a8e | 4369 | } |
31a7df01 | 4370 | |
8cfd8147 TH |
4371 | static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos, |
4372 | unsigned int iter_flags) | |
b4b90a8e TH |
4373 | { |
4374 | struct kernfs_open_file *of = s->private; | |
4375 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
4376 | struct css_task_iter *it = of->priv; | |
8cc99345 | 4377 | |
952aaa12 | 4378 | /* |
b4b90a8e TH |
4379 | * When a seq_file is seeked, it's always traversed sequentially |
4380 | * from position 0, so we can simply keep iterating on !0 *pos. | |
952aaa12 | 4381 | */ |
b4b90a8e TH |
4382 | if (!it) { |
4383 | if (WARN_ON_ONCE((*pos)++)) | |
4384 | return ERR_PTR(-EINVAL); | |
8cc99345 | 4385 | |
b4b90a8e TH |
4386 | it = kzalloc(sizeof(*it), GFP_KERNEL); |
4387 | if (!it) | |
4388 | return ERR_PTR(-ENOMEM); | |
4389 | of->priv = it; | |
450ee0c1 | 4390 | css_task_iter_start(&cgrp->self, iter_flags, it); |
b4b90a8e TH |
4391 | } else if (!(*pos)++) { |
4392 | css_task_iter_end(it); | |
450ee0c1 | 4393 | css_task_iter_start(&cgrp->self, iter_flags, it); |
b4b90a8e | 4394 | } |
bbcb81d0 | 4395 | |
b4b90a8e TH |
4396 | return cgroup_procs_next(s, NULL, NULL); |
4397 | } | |
24528255 | 4398 | |
8cfd8147 TH |
4399 | static void *cgroup_procs_start(struct seq_file *s, loff_t *pos) |
4400 | { | |
4401 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
4402 | ||
4403 | /* | |
4404 | * All processes of a threaded subtree belong to the domain cgroup | |
4405 | * of the subtree. Only threads can be distributed across the | |
4406 | * subtree. Reject reads on cgroup.procs in the subtree proper. | |
4407 | * They're always empty anyway. | |
4408 | */ | |
4409 | if (cgroup_is_threaded(cgrp)) | |
4410 | return ERR_PTR(-EOPNOTSUPP); | |
4411 | ||
4412 | return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS | | |
4413 | CSS_TASK_ITER_THREADED); | |
4414 | } | |
4415 | ||
b4b90a8e | 4416 | static int cgroup_procs_show(struct seq_file *s, void *v) |
bbcb81d0 | 4417 | { |
bc2fb7ed | 4418 | seq_printf(s, "%d\n", task_pid_vnr(v)); |
97978e6d DL |
4419 | return 0; |
4420 | } | |
4421 | ||
715c809d TH |
4422 | static int cgroup_procs_write_permission(struct cgroup *src_cgrp, |
4423 | struct cgroup *dst_cgrp, | |
4424 | struct super_block *sb) | |
4425 | { | |
4426 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; | |
4427 | struct cgroup *com_cgrp = src_cgrp; | |
4428 | struct inode *inode; | |
4429 | int ret; | |
4430 | ||
4431 | lockdep_assert_held(&cgroup_mutex); | |
4432 | ||
4433 | /* find the common ancestor */ | |
4434 | while (!cgroup_is_descendant(dst_cgrp, com_cgrp)) | |
4435 | com_cgrp = cgroup_parent(com_cgrp); | |
4436 | ||
4437 | /* %current should be authorized to migrate to the common ancestor */ | |
4438 | inode = kernfs_get_inode(sb, com_cgrp->procs_file.kn); | |
4439 | if (!inode) | |
4440 | return -ENOMEM; | |
4441 | ||
4442 | ret = inode_permission(inode, MAY_WRITE); | |
4443 | iput(inode); | |
4444 | if (ret) | |
4445 | return ret; | |
4446 | ||
4447 | /* | |
4448 | * If namespaces are delegation boundaries, %current must be able | |
4449 | * to see both source and destination cgroups from its namespace. | |
4450 | */ | |
4451 | if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) && | |
4452 | (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) || | |
4453 | !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp))) | |
4454 | return -ENOENT; | |
4455 | ||
4456 | return 0; | |
4457 | } | |
4458 | ||
4459 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, | |
4460 | char *buf, size_t nbytes, loff_t off) | |
4461 | { | |
4462 | struct cgroup *src_cgrp, *dst_cgrp; | |
4463 | struct task_struct *task; | |
4464 | ssize_t ret; | |
4465 | ||
4466 | dst_cgrp = cgroup_kn_lock_live(of->kn, false); | |
4467 | if (!dst_cgrp) | |
4468 | return -ENODEV; | |
4469 | ||
4470 | task = cgroup_procs_write_start(buf, true); | |
4471 | ret = PTR_ERR_OR_ZERO(task); | |
4472 | if (ret) | |
4473 | goto out_unlock; | |
4474 | ||
4475 | /* find the source cgroup */ | |
4476 | spin_lock_irq(&css_set_lock); | |
4477 | src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); | |
4478 | spin_unlock_irq(&css_set_lock); | |
4479 | ||
4480 | ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, | |
4481 | of->file->f_path.dentry->d_sb); | |
4482 | if (ret) | |
4483 | goto out_finish; | |
4484 | ||
4485 | ret = cgroup_attach_task(dst_cgrp, task, true); | |
4486 | ||
4487 | out_finish: | |
4488 | cgroup_procs_write_finish(task); | |
4489 | out_unlock: | |
4490 | cgroup_kn_unlock(of->kn); | |
4491 | ||
4492 | return ret ?: nbytes; | |
4493 | } | |
4494 | ||
8cfd8147 TH |
4495 | static void *cgroup_threads_start(struct seq_file *s, loff_t *pos) |
4496 | { | |
4497 | return __cgroup_procs_start(s, pos, 0); | |
4498 | } | |
4499 | ||
4500 | static ssize_t cgroup_threads_write(struct kernfs_open_file *of, | |
4501 | char *buf, size_t nbytes, loff_t off) | |
4502 | { | |
4503 | struct cgroup *src_cgrp, *dst_cgrp; | |
4504 | struct task_struct *task; | |
4505 | ssize_t ret; | |
4506 | ||
4507 | buf = strstrip(buf); | |
4508 | ||
4509 | dst_cgrp = cgroup_kn_lock_live(of->kn, false); | |
4510 | if (!dst_cgrp) | |
4511 | return -ENODEV; | |
4512 | ||
4513 | task = cgroup_procs_write_start(buf, false); | |
4514 | ret = PTR_ERR_OR_ZERO(task); | |
4515 | if (ret) | |
4516 | goto out_unlock; | |
4517 | ||
4518 | /* find the source cgroup */ | |
4519 | spin_lock_irq(&css_set_lock); | |
4520 | src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); | |
4521 | spin_unlock_irq(&css_set_lock); | |
4522 | ||
4523 | /* thread migrations follow the cgroup.procs delegation rule */ | |
4524 | ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, | |
4525 | of->file->f_path.dentry->d_sb); | |
4526 | if (ret) | |
4527 | goto out_finish; | |
4528 | ||
4529 | /* and must be contained in the same domain */ | |
4530 | ret = -EOPNOTSUPP; | |
4531 | if (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp) | |
4532 | goto out_finish; | |
4533 | ||
4534 | ret = cgroup_attach_task(dst_cgrp, task, false); | |
4535 | ||
4536 | out_finish: | |
4537 | cgroup_procs_write_finish(task); | |
4538 | out_unlock: | |
4539 | cgroup_kn_unlock(of->kn); | |
4540 | ||
4541 | return ret ?: nbytes; | |
4542 | } | |
4543 | ||
a14c6874 | 4544 | /* cgroup core interface files for the default hierarchy */ |
d62beb7f | 4545 | static struct cftype cgroup_base_files[] = { |
8cfd8147 TH |
4546 | { |
4547 | .name = "cgroup.type", | |
4548 | .flags = CFTYPE_NOT_ON_ROOT, | |
4549 | .seq_show = cgroup_type_show, | |
4550 | .write = cgroup_type_write, | |
4551 | }, | |
81a6a5cd | 4552 | { |
d5c56ced | 4553 | .name = "cgroup.procs", |
5136f636 | 4554 | .flags = CFTYPE_NS_DELEGATABLE, |
6f60eade | 4555 | .file_offset = offsetof(struct cgroup, procs_file), |
b4b90a8e TH |
4556 | .release = cgroup_procs_release, |
4557 | .seq_start = cgroup_procs_start, | |
4558 | .seq_next = cgroup_procs_next, | |
4559 | .seq_show = cgroup_procs_show, | |
acbef755 | 4560 | .write = cgroup_procs_write, |
102a775e | 4561 | }, |
8cfd8147 TH |
4562 | { |
4563 | .name = "cgroup.threads", | |
4f58424d | 4564 | .flags = CFTYPE_NS_DELEGATABLE, |
8cfd8147 TH |
4565 | .release = cgroup_procs_release, |
4566 | .seq_start = cgroup_threads_start, | |
4567 | .seq_next = cgroup_procs_next, | |
4568 | .seq_show = cgroup_procs_show, | |
4569 | .write = cgroup_threads_write, | |
4570 | }, | |
f8f22e53 TH |
4571 | { |
4572 | .name = "cgroup.controllers", | |
f8f22e53 TH |
4573 | .seq_show = cgroup_controllers_show, |
4574 | }, | |
4575 | { | |
4576 | .name = "cgroup.subtree_control", | |
5136f636 | 4577 | .flags = CFTYPE_NS_DELEGATABLE, |
f8f22e53 | 4578 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4579 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4580 | }, |
842b597e | 4581 | { |
4a07c222 | 4582 | .name = "cgroup.events", |
a14c6874 | 4583 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4584 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4585 | .seq_show = cgroup_events_show, |
842b597e | 4586 | }, |
1a926e0b RG |
4587 | { |
4588 | .name = "cgroup.max.descendants", | |
4589 | .seq_show = cgroup_max_descendants_show, | |
4590 | .write = cgroup_max_descendants_write, | |
4591 | }, | |
4592 | { | |
4593 | .name = "cgroup.max.depth", | |
4594 | .seq_show = cgroup_max_depth_show, | |
4595 | .write = cgroup_max_depth_write, | |
4596 | }, | |
ec39225c RG |
4597 | { |
4598 | .name = "cgroup.stat", | |
3e48930c | 4599 | .seq_show = cgroup_stat_show, |
ec39225c | 4600 | }, |
d41bf8c9 TH |
4601 | { |
4602 | .name = "cpu.stat", | |
4603 | .flags = CFTYPE_NOT_ON_ROOT, | |
4604 | .seq_show = cpu_stat_show, | |
4605 | }, | |
2ce7135a JW |
4606 | #ifdef CONFIG_PSI |
4607 | { | |
4608 | .name = "io.pressure", | |
4609 | .flags = CFTYPE_NOT_ON_ROOT, | |
4610 | .seq_show = cgroup_io_pressure_show, | |
4611 | }, | |
4612 | { | |
4613 | .name = "memory.pressure", | |
4614 | .flags = CFTYPE_NOT_ON_ROOT, | |
4615 | .seq_show = cgroup_memory_pressure_show, | |
4616 | }, | |
4617 | { | |
4618 | .name = "cpu.pressure", | |
4619 | .flags = CFTYPE_NOT_ON_ROOT, | |
4620 | .seq_show = cgroup_cpu_pressure_show, | |
4621 | }, | |
4622 | #endif | |
a14c6874 TH |
4623 | { } /* terminate */ |
4624 | }; | |
d5c56ced | 4625 | |
0c21ead1 TH |
4626 | /* |
4627 | * css destruction is four-stage process. | |
4628 | * | |
4629 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4630 | * Implemented in kill_css(). | |
4631 | * | |
4632 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4633 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4634 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4635 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4636 | * |
4637 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4638 | * accessors are inside RCU read sections. css_release() schedules the | |
4639 | * RCU callback. | |
4640 | * | |
4641 | * 4. After the grace period, the css can be freed. Implemented in | |
4642 | * css_free_work_fn(). | |
4643 | * | |
4644 | * It is actually hairier because both step 2 and 4 require process context | |
4645 | * and thus involve punting to css->destroy_work adding two additional | |
4646 | * steps to the already complex sequence. | |
4647 | */ | |
8f36aaec | 4648 | static void css_free_rwork_fn(struct work_struct *work) |
48ddbe19 | 4649 | { |
8f36aaec TH |
4650 | struct cgroup_subsys_state *css = container_of(to_rcu_work(work), |
4651 | struct cgroup_subsys_state, destroy_rwork); | |
01e58659 | 4652 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4653 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4654 | |
9a1049da TH |
4655 | percpu_ref_exit(&css->refcnt); |
4656 | ||
01e58659 | 4657 | if (ss) { |
9d755d33 | 4658 | /* css free path */ |
8bb5ef79 | 4659 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
4660 | int id = css->id; |
4661 | ||
01e58659 VD |
4662 | ss->css_free(css); |
4663 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 4664 | cgroup_put(cgrp); |
8bb5ef79 TH |
4665 | |
4666 | if (parent) | |
4667 | css_put(parent); | |
9d755d33 TH |
4668 | } else { |
4669 | /* cgroup free path */ | |
4670 | atomic_dec(&cgrp->root->nr_cgrps); | |
d62beb7f | 4671 | cgroup1_pidlist_destroy_all(cgrp); |
971ff493 | 4672 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4673 | |
d51f39b0 | 4674 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4675 | /* |
4676 | * We get a ref to the parent, and put the ref when | |
4677 | * this cgroup is being freed, so it's guaranteed | |
4678 | * that the parent won't be destroyed before its | |
4679 | * children. | |
4680 | */ | |
d51f39b0 | 4681 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 | 4682 | kernfs_put(cgrp->kn); |
2ce7135a | 4683 | psi_cgroup_free(cgrp); |
041cd640 | 4684 | if (cgroup_on_dfl(cgrp)) |
c58632b3 | 4685 | cgroup_rstat_exit(cgrp); |
9d755d33 TH |
4686 | kfree(cgrp); |
4687 | } else { | |
4688 | /* | |
4689 | * This is root cgroup's refcnt reaching zero, | |
4690 | * which indicates that the root should be | |
4691 | * released. | |
4692 | */ | |
4693 | cgroup_destroy_root(cgrp->root); | |
4694 | } | |
4695 | } | |
48ddbe19 TH |
4696 | } |
4697 | ||
25e15d83 | 4698 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4699 | { |
4700 | struct cgroup_subsys_state *css = | |
25e15d83 | 4701 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4702 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4703 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4704 | |
1fed1b2e TH |
4705 | mutex_lock(&cgroup_mutex); |
4706 | ||
de3f0341 | 4707 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4708 | list_del_rcu(&css->sibling); |
4709 | ||
9d755d33 TH |
4710 | if (ss) { |
4711 | /* css release path */ | |
8f53470b TH |
4712 | if (!list_empty(&css->rstat_css_node)) { |
4713 | cgroup_rstat_flush(cgrp); | |
4714 | list_del_rcu(&css->rstat_css_node); | |
4715 | } | |
4716 | ||
01e58659 | 4717 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
4718 | if (ss->css_released) |
4719 | ss->css_released(css); | |
9d755d33 | 4720 | } else { |
0679dee0 RG |
4721 | struct cgroup *tcgrp; |
4722 | ||
9d755d33 | 4723 | /* cgroup release path */ |
e4f8d81c | 4724 | TRACE_CGROUP_PATH(release, cgrp); |
ed1777de | 4725 | |
041cd640 | 4726 | if (cgroup_on_dfl(cgrp)) |
c58632b3 | 4727 | cgroup_rstat_flush(cgrp); |
041cd640 | 4728 | |
0679dee0 RG |
4729 | for (tcgrp = cgroup_parent(cgrp); tcgrp; |
4730 | tcgrp = cgroup_parent(tcgrp)) | |
4731 | tcgrp->nr_dying_descendants--; | |
4732 | ||
9d755d33 TH |
4733 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4734 | cgrp->id = -1; | |
a4189487 LZ |
4735 | |
4736 | /* | |
4737 | * There are two control paths which try to determine | |
4738 | * cgroup from dentry without going through kernfs - | |
4739 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4740 | * Those are supported by RCU protecting clearing of | |
4741 | * cgrp->kn->priv backpointer. | |
4742 | */ | |
6cd0f5bb TH |
4743 | if (cgrp->kn) |
4744 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
4745 | NULL); | |
30070984 DM |
4746 | |
4747 | cgroup_bpf_put(cgrp); | |
9d755d33 | 4748 | } |
d3daf28d | 4749 | |
1fed1b2e TH |
4750 | mutex_unlock(&cgroup_mutex); |
4751 | ||
8f36aaec TH |
4752 | INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn); |
4753 | queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork); | |
d3daf28d TH |
4754 | } |
4755 | ||
d3daf28d TH |
4756 | static void css_release(struct percpu_ref *ref) |
4757 | { | |
4758 | struct cgroup_subsys_state *css = | |
4759 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4760 | ||
25e15d83 TH |
4761 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
4762 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
4763 | } |
4764 | ||
ddfcadab TH |
4765 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4766 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4767 | { |
0cb51d71 TH |
4768 | lockdep_assert_held(&cgroup_mutex); |
4769 | ||
a590b90d | 4770 | cgroup_get_live(cgrp); |
ddfcadab | 4771 | |
d5c419b6 | 4772 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4773 | css->cgroup = cgrp; |
72c97e54 | 4774 | css->ss = ss; |
8fa3b8d6 | 4775 | css->id = -1; |
d5c419b6 TH |
4776 | INIT_LIST_HEAD(&css->sibling); |
4777 | INIT_LIST_HEAD(&css->children); | |
8f53470b | 4778 | INIT_LIST_HEAD(&css->rstat_css_node); |
0cb51d71 | 4779 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 4780 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 4781 | |
d51f39b0 TH |
4782 | if (cgroup_parent(cgrp)) { |
4783 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4784 | css_get(css->parent); |
ddfcadab | 4785 | } |
48ddbe19 | 4786 | |
8f53470b TH |
4787 | if (cgroup_on_dfl(cgrp) && ss->css_rstat_flush) |
4788 | list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list); | |
4789 | ||
ca8bdcaf | 4790 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4791 | } |
4792 | ||
2a4ac633 | 4793 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4794 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4795 | { |
623f926b | 4796 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4797 | int ret = 0; |
4798 | ||
a31f2d3f TH |
4799 | lockdep_assert_held(&cgroup_mutex); |
4800 | ||
92fb9748 | 4801 | if (ss->css_online) |
eb95419b | 4802 | ret = ss->css_online(css); |
ae7f164a | 4803 | if (!ret) { |
eb95419b | 4804 | css->flags |= CSS_ONLINE; |
aec25020 | 4805 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
4806 | |
4807 | atomic_inc(&css->online_cnt); | |
4808 | if (css->parent) | |
4809 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 4810 | } |
b1929db4 | 4811 | return ret; |
a31f2d3f TH |
4812 | } |
4813 | ||
2a4ac633 | 4814 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4815 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4816 | { |
623f926b | 4817 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4818 | |
4819 | lockdep_assert_held(&cgroup_mutex); | |
4820 | ||
4821 | if (!(css->flags & CSS_ONLINE)) | |
4822 | return; | |
4823 | ||
d7eeac19 | 4824 | if (ss->css_offline) |
eb95419b | 4825 | ss->css_offline(css); |
a31f2d3f | 4826 | |
eb95419b | 4827 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4828 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4829 | |
4830 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4831 | } |
4832 | ||
c81c925a | 4833 | /** |
6cd0f5bb | 4834 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
4835 | * @cgrp: the cgroup new css will be associated with |
4836 | * @ss: the subsys of new css | |
4837 | * | |
4838 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
4839 | * css is online and installed in @cgrp. This function doesn't create the |
4840 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 4841 | */ |
6cd0f5bb TH |
4842 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
4843 | struct cgroup_subsys *ss) | |
c81c925a | 4844 | { |
d51f39b0 | 4845 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4846 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4847 | struct cgroup_subsys_state *css; |
4848 | int err; | |
4849 | ||
c81c925a TH |
4850 | lockdep_assert_held(&cgroup_mutex); |
4851 | ||
1fed1b2e | 4852 | css = ss->css_alloc(parent_css); |
e7e15b87 TH |
4853 | if (!css) |
4854 | css = ERR_PTR(-ENOMEM); | |
c81c925a | 4855 | if (IS_ERR(css)) |
6cd0f5bb | 4856 | return css; |
c81c925a | 4857 | |
ddfcadab | 4858 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4859 | |
2aad2a86 | 4860 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 4861 | if (err) |
3eb59ec6 | 4862 | goto err_free_css; |
c81c925a | 4863 | |
cf780b7d | 4864 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 | 4865 | if (err < 0) |
b00c52da | 4866 | goto err_free_css; |
15a4c835 | 4867 | css->id = err; |
c81c925a | 4868 | |
15a4c835 | 4869 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 4870 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4871 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4872 | |
4873 | err = online_css(css); | |
4874 | if (err) | |
1fed1b2e | 4875 | goto err_list_del; |
94419627 | 4876 | |
c81c925a | 4877 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4878 | cgroup_parent(parent)) { |
ed3d261b | 4879 | 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 | 4880 | current->comm, current->pid, ss->name); |
c81c925a | 4881 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4882 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4883 | ss->warned_broken_hierarchy = true; |
4884 | } | |
4885 | ||
6cd0f5bb | 4886 | return css; |
c81c925a | 4887 | |
1fed1b2e TH |
4888 | err_list_del: |
4889 | list_del_rcu(&css->sibling); | |
3eb59ec6 | 4890 | err_free_css: |
8f53470b | 4891 | list_del_rcu(&css->rstat_css_node); |
8f36aaec TH |
4892 | INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn); |
4893 | queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork); | |
6cd0f5bb | 4894 | return ERR_PTR(err); |
c81c925a TH |
4895 | } |
4896 | ||
07cd1294 TH |
4897 | /* |
4898 | * The returned cgroup is fully initialized including its control mask, but | |
4899 | * it isn't associated with its kernfs_node and doesn't have the control | |
4900 | * mask applied. | |
4901 | */ | |
a5bca215 | 4902 | static struct cgroup *cgroup_create(struct cgroup *parent) |
ddbcc7e8 | 4903 | { |
a5bca215 | 4904 | struct cgroup_root *root = parent->root; |
a5bca215 TH |
4905 | struct cgroup *cgrp, *tcgrp; |
4906 | int level = parent->level + 1; | |
03970d3c | 4907 | int ret; |
ddbcc7e8 | 4908 | |
0a950f65 | 4909 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
acafe7e3 KC |
4910 | cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)), |
4911 | GFP_KERNEL); | |
a5bca215 TH |
4912 | if (!cgrp) |
4913 | return ERR_PTR(-ENOMEM); | |
0ab02ca8 | 4914 | |
2aad2a86 | 4915 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
4916 | if (ret) |
4917 | goto out_free_cgrp; | |
4918 | ||
041cd640 | 4919 | if (cgroup_on_dfl(parent)) { |
c58632b3 | 4920 | ret = cgroup_rstat_init(cgrp); |
041cd640 TH |
4921 | if (ret) |
4922 | goto out_cancel_ref; | |
4923 | } | |
4924 | ||
0ab02ca8 LZ |
4925 | /* |
4926 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4927 | * a half-baked cgroup. | |
4928 | */ | |
cf780b7d | 4929 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 4930 | if (cgrp->id < 0) { |
ba0f4d76 | 4931 | ret = -ENOMEM; |
041cd640 | 4932 | goto out_stat_exit; |
976c06bc TH |
4933 | } |
4934 | ||
cc31edce | 4935 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4936 | |
9d800df1 | 4937 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 4938 | cgrp->root = root; |
b11cfb58 | 4939 | cgrp->level = level; |
2ce7135a JW |
4940 | |
4941 | ret = psi_cgroup_alloc(cgrp); | |
324bda9e AS |
4942 | if (ret) |
4943 | goto out_idr_free; | |
b11cfb58 | 4944 | |
2ce7135a JW |
4945 | ret = cgroup_bpf_inherit(cgrp); |
4946 | if (ret) | |
4947 | goto out_psi_free; | |
4948 | ||
0679dee0 | 4949 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) { |
b11cfb58 | 4950 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; |
ddbcc7e8 | 4951 | |
0679dee0 RG |
4952 | if (tcgrp != cgrp) |
4953 | tcgrp->nr_descendants++; | |
4954 | } | |
4955 | ||
b6abdb0e LZ |
4956 | if (notify_on_release(parent)) |
4957 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4958 | ||
2260e7fc TH |
4959 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4960 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4961 | |
0cb51d71 | 4962 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 4963 | |
4e139afc | 4964 | /* allocation complete, commit to creation */ |
d5c419b6 | 4965 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 4966 | atomic_inc(&root->nr_cgrps); |
a590b90d | 4967 | cgroup_get_live(parent); |
415cf07a | 4968 | |
0d80255e TH |
4969 | /* |
4970 | * @cgrp is now fully operational. If something fails after this | |
4971 | * point, it'll be released via the normal destruction path. | |
4972 | */ | |
6fa4918d | 4973 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 4974 | |
bd53d617 TH |
4975 | /* |
4976 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 4977 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 4978 | */ |
03970d3c | 4979 | if (!cgroup_on_dfl(cgrp)) |
5531dc91 | 4980 | cgrp->subtree_control = cgroup_control(cgrp); |
03970d3c TH |
4981 | |
4982 | cgroup_propagate_control(cgrp); | |
4983 | ||
a5bca215 TH |
4984 | return cgrp; |
4985 | ||
2ce7135a JW |
4986 | out_psi_free: |
4987 | psi_cgroup_free(cgrp); | |
324bda9e AS |
4988 | out_idr_free: |
4989 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); | |
041cd640 TH |
4990 | out_stat_exit: |
4991 | if (cgroup_on_dfl(parent)) | |
c58632b3 | 4992 | cgroup_rstat_exit(cgrp); |
a5bca215 TH |
4993 | out_cancel_ref: |
4994 | percpu_ref_exit(&cgrp->self.refcnt); | |
4995 | out_free_cgrp: | |
4996 | kfree(cgrp); | |
4997 | return ERR_PTR(ret); | |
a5bca215 TH |
4998 | } |
4999 | ||
1a926e0b RG |
5000 | static bool cgroup_check_hierarchy_limits(struct cgroup *parent) |
5001 | { | |
5002 | struct cgroup *cgroup; | |
5003 | int ret = false; | |
5004 | int level = 1; | |
5005 | ||
5006 | lockdep_assert_held(&cgroup_mutex); | |
5007 | ||
5008 | for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) { | |
5009 | if (cgroup->nr_descendants >= cgroup->max_descendants) | |
5010 | goto fail; | |
5011 | ||
5012 | if (level > cgroup->max_depth) | |
5013 | goto fail; | |
5014 | ||
5015 | level++; | |
5016 | } | |
5017 | ||
5018 | ret = true; | |
5019 | fail: | |
5020 | return ret; | |
5021 | } | |
5022 | ||
1592c9b2 | 5023 | int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode) |
a5bca215 TH |
5024 | { |
5025 | struct cgroup *parent, *cgrp; | |
a5bca215 | 5026 | struct kernfs_node *kn; |
03970d3c | 5027 | int ret; |
a5bca215 TH |
5028 | |
5029 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ | |
5030 | if (strchr(name, '\n')) | |
5031 | return -EINVAL; | |
5032 | ||
945ba199 | 5033 | parent = cgroup_kn_lock_live(parent_kn, false); |
a5bca215 TH |
5034 | if (!parent) |
5035 | return -ENODEV; | |
5036 | ||
1a926e0b RG |
5037 | if (!cgroup_check_hierarchy_limits(parent)) { |
5038 | ret = -EAGAIN; | |
5039 | goto out_unlock; | |
5040 | } | |
5041 | ||
a5bca215 TH |
5042 | cgrp = cgroup_create(parent); |
5043 | if (IS_ERR(cgrp)) { | |
5044 | ret = PTR_ERR(cgrp); | |
5045 | goto out_unlock; | |
5046 | } | |
5047 | ||
195e9b6c TH |
5048 | /* create the directory */ |
5049 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
5050 | if (IS_ERR(kn)) { | |
5051 | ret = PTR_ERR(kn); | |
5052 | goto out_destroy; | |
5053 | } | |
5054 | cgrp->kn = kn; | |
5055 | ||
5056 | /* | |
5057 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
5058 | * that @cgrp->kn is always accessible. | |
5059 | */ | |
5060 | kernfs_get(kn); | |
5061 | ||
5062 | ret = cgroup_kn_set_ugid(kn); | |
5063 | if (ret) | |
5064 | goto out_destroy; | |
5065 | ||
334c3679 | 5066 | ret = css_populate_dir(&cgrp->self); |
195e9b6c TH |
5067 | if (ret) |
5068 | goto out_destroy; | |
5069 | ||
03970d3c TH |
5070 | ret = cgroup_apply_control_enable(cgrp); |
5071 | if (ret) | |
5072 | goto out_destroy; | |
195e9b6c | 5073 | |
e4f8d81c | 5074 | TRACE_CGROUP_PATH(mkdir, cgrp); |
ed1777de | 5075 | |
195e9b6c | 5076 | /* let's create and online css's */ |
2bd59d48 | 5077 | kernfs_activate(kn); |
ddbcc7e8 | 5078 | |
ba0f4d76 TH |
5079 | ret = 0; |
5080 | goto out_unlock; | |
ddbcc7e8 | 5081 | |
a5bca215 TH |
5082 | out_destroy: |
5083 | cgroup_destroy_locked(cgrp); | |
ba0f4d76 | 5084 | out_unlock: |
a9746d8d | 5085 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5086 | return ret; |
ddbcc7e8 PM |
5087 | } |
5088 | ||
223dbc38 TH |
5089 | /* |
5090 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
5091 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5092 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5093 | */ |
5094 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5095 | { |
223dbc38 TH |
5096 | struct cgroup_subsys_state *css = |
5097 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5098 | |
f20104de | 5099 | mutex_lock(&cgroup_mutex); |
09a503ea | 5100 | |
aa226ff4 TH |
5101 | do { |
5102 | offline_css(css); | |
5103 | css_put(css); | |
5104 | /* @css can't go away while we're holding cgroup_mutex */ | |
5105 | css = css->parent; | |
5106 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5107 | ||
5108 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5109 | } |
5110 | ||
223dbc38 TH |
5111 | /* css kill confirmation processing requires process context, bounce */ |
5112 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5113 | { |
5114 | struct cgroup_subsys_state *css = | |
5115 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5116 | ||
aa226ff4 TH |
5117 | if (atomic_dec_and_test(&css->online_cnt)) { |
5118 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5119 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5120 | } | |
d3daf28d TH |
5121 | } |
5122 | ||
f392e51c TH |
5123 | /** |
5124 | * kill_css - destroy a css | |
5125 | * @css: css to destroy | |
5126 | * | |
5127 | * This function initiates destruction of @css by removing cgroup interface | |
5128 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5129 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5130 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5131 | */ |
5132 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5133 | { |
01f6474c | 5134 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5135 | |
33c35aa4 WL |
5136 | if (css->flags & CSS_DYING) |
5137 | return; | |
5138 | ||
5139 | css->flags |= CSS_DYING; | |
5140 | ||
2bd59d48 TH |
5141 | /* |
5142 | * This must happen before css is disassociated with its cgroup. | |
5143 | * See seq_css() for details. | |
5144 | */ | |
334c3679 | 5145 | css_clear_dir(css); |
3c14f8b4 | 5146 | |
edae0c33 TH |
5147 | /* |
5148 | * Killing would put the base ref, but we need to keep it alive | |
5149 | * until after ->css_offline(). | |
5150 | */ | |
5151 | css_get(css); | |
5152 | ||
5153 | /* | |
5154 | * cgroup core guarantees that, by the time ->css_offline() is | |
5155 | * invoked, no new css reference will be given out via | |
ec903c0c | 5156 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5157 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5158 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5159 | * | |
5160 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5161 | * css is confirmed to be seen as killed on all CPUs. | |
5162 | */ | |
5163 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5164 | } |
5165 | ||
5166 | /** | |
5167 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5168 | * @cgrp: cgroup to be destroyed | |
5169 | * | |
5170 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5171 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5172 | * guarantee that css_tryget_online() won't succeed by the time |
5173 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5174 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5175 | * |
5176 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5177 | * userland visible parts and start killing the percpu refcnts of | |
5178 | * css's. Set up so that the next stage will be kicked off once all | |
5179 | * the percpu refcnts are confirmed to be killed. | |
5180 | * | |
5181 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5182 | * rest of destruction. Once all cgroup references are gone, the | |
5183 | * cgroup is RCU-freed. | |
5184 | * | |
5185 | * This function implements s1. After this step, @cgrp is gone as far as | |
5186 | * the userland is concerned and a new cgroup with the same name may be | |
5187 | * created. As cgroup doesn't care about the names internally, this | |
5188 | * doesn't cause any problem. | |
5189 | */ | |
42809dd4 TH |
5190 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5191 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5192 | { |
0679dee0 | 5193 | struct cgroup *tcgrp, *parent = cgroup_parent(cgrp); |
2bd59d48 | 5194 | struct cgroup_subsys_state *css; |
2b021cbf | 5195 | struct cgrp_cset_link *link; |
1c6727af | 5196 | int ssid; |
ddbcc7e8 | 5197 | |
42809dd4 TH |
5198 | lockdep_assert_held(&cgroup_mutex); |
5199 | ||
91486f61 TH |
5200 | /* |
5201 | * Only migration can raise populated from zero and we're already | |
5202 | * holding cgroup_mutex. | |
5203 | */ | |
5204 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5205 | return -EBUSY; |
a043e3b2 | 5206 | |
bb78a92f | 5207 | /* |
d5c419b6 TH |
5208 | * Make sure there's no live children. We can't test emptiness of |
5209 | * ->self.children as dead children linger on it while being | |
5210 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5211 | */ |
f3d46500 | 5212 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5213 | return -EBUSY; |
5214 | ||
455050d2 | 5215 | /* |
2b021cbf TH |
5216 | * Mark @cgrp and the associated csets dead. The former prevents |
5217 | * further task migration and child creation by disabling | |
5218 | * cgroup_lock_live_group(). The latter makes the csets ignored by | |
5219 | * the migration path. | |
455050d2 | 5220 | */ |
184faf32 | 5221 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5222 | |
82d6489d | 5223 | spin_lock_irq(&css_set_lock); |
2b021cbf TH |
5224 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
5225 | link->cset->dead = true; | |
82d6489d | 5226 | spin_unlock_irq(&css_set_lock); |
2b021cbf | 5227 | |
249f3468 | 5228 | /* initiate massacre of all css's */ |
1c6727af TH |
5229 | for_each_css(css, ssid, cgrp) |
5230 | kill_css(css); | |
455050d2 | 5231 | |
5faaf05f TH |
5232 | /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */ |
5233 | css_clear_dir(&cgrp->self); | |
01f6474c | 5234 | kernfs_remove(cgrp->kn); |
f20104de | 5235 | |
454000ad TH |
5236 | if (parent && cgroup_is_threaded(cgrp)) |
5237 | parent->nr_threaded_children--; | |
5238 | ||
0679dee0 RG |
5239 | for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) { |
5240 | tcgrp->nr_descendants--; | |
5241 | tcgrp->nr_dying_descendants++; | |
5242 | } | |
5243 | ||
5a621e6c | 5244 | cgroup1_check_for_release(parent); |
2bd59d48 | 5245 | |
249f3468 | 5246 | /* put the base reference */ |
9d755d33 | 5247 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5248 | |
ea15f8cc TH |
5249 | return 0; |
5250 | }; | |
5251 | ||
1592c9b2 | 5252 | int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5253 | { |
a9746d8d | 5254 | struct cgroup *cgrp; |
2bd59d48 | 5255 | int ret = 0; |
42809dd4 | 5256 | |
945ba199 | 5257 | cgrp = cgroup_kn_lock_live(kn, false); |
a9746d8d TH |
5258 | if (!cgrp) |
5259 | return 0; | |
42809dd4 | 5260 | |
a9746d8d | 5261 | ret = cgroup_destroy_locked(cgrp); |
ed1777de | 5262 | if (!ret) |
e4f8d81c | 5263 | TRACE_CGROUP_PATH(rmdir, cgrp); |
ed1777de | 5264 | |
a9746d8d | 5265 | cgroup_kn_unlock(kn); |
42809dd4 | 5266 | return ret; |
8e3f6541 TH |
5267 | } |
5268 | ||
2bd59d48 | 5269 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5136f636 | 5270 | .show_options = cgroup_show_options, |
2bd59d48 | 5271 | .remount_fs = cgroup_remount, |
2bd59d48 TH |
5272 | .mkdir = cgroup_mkdir, |
5273 | .rmdir = cgroup_rmdir, | |
4f41fc59 | 5274 | .show_path = cgroup_show_path, |
2bd59d48 TH |
5275 | }; |
5276 | ||
15a4c835 | 5277 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5278 | { |
ddbcc7e8 | 5279 | struct cgroup_subsys_state *css; |
cfe36bde | 5280 | |
a5ae9899 | 5281 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5282 | |
648bb56d TH |
5283 | mutex_lock(&cgroup_mutex); |
5284 | ||
15a4c835 | 5285 | idr_init(&ss->css_idr); |
0adb0704 | 5286 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5287 | |
3dd06ffa TH |
5288 | /* Create the root cgroup state for this subsystem */ |
5289 | ss->root = &cgrp_dfl_root; | |
5290 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5291 | /* We don't handle early failures gracefully */ |
5292 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5293 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5294 | |
5295 | /* | |
5296 | * Root csses are never destroyed and we can't initialize | |
5297 | * percpu_ref during early init. Disable refcnting. | |
5298 | */ | |
5299 | css->flags |= CSS_NO_REF; | |
5300 | ||
15a4c835 | 5301 | if (early) { |
9395a450 | 5302 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5303 | css->id = 1; |
5304 | } else { | |
5305 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5306 | BUG_ON(css->id < 0); | |
5307 | } | |
ddbcc7e8 | 5308 | |
e8d55fde | 5309 | /* Update the init_css_set to contain a subsys |
817929ec | 5310 | * pointer to this state - since the subsystem is |
e8d55fde | 5311 | * newly registered, all tasks and hence the |
3dd06ffa | 5312 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5313 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5314 | |
cb4a3167 AS |
5315 | have_fork_callback |= (bool)ss->fork << ss->id; |
5316 | have_exit_callback |= (bool)ss->exit << ss->id; | |
afcf6c8b | 5317 | have_free_callback |= (bool)ss->free << ss->id; |
7e47682e | 5318 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5319 | |
e8d55fde LZ |
5320 | /* At system boot, before all subsystems have been |
5321 | * registered, no tasks have been forked, so we don't | |
5322 | * need to invoke fork callbacks here. */ | |
5323 | BUG_ON(!list_empty(&init_task.tasks)); | |
5324 | ||
ae7f164a | 5325 | BUG_ON(online_css(css)); |
a8638030 | 5326 | |
cf5d5941 BB |
5327 | mutex_unlock(&cgroup_mutex); |
5328 | } | |
cf5d5941 | 5329 | |
ddbcc7e8 | 5330 | /** |
a043e3b2 LZ |
5331 | * cgroup_init_early - cgroup initialization at system boot |
5332 | * | |
5333 | * Initialize cgroups at system boot, and initialize any | |
5334 | * subsystems that request early init. | |
ddbcc7e8 PM |
5335 | */ |
5336 | int __init cgroup_init_early(void) | |
5337 | { | |
7b9a6ba5 | 5338 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5339 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5340 | int i; |
30159ec7 | 5341 | |
3dd06ffa | 5342 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5343 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5344 | ||
a4ea1cc9 | 5345 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5346 | |
3ed80a62 | 5347 | for_each_subsys(ss, i) { |
aec25020 | 5348 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 5349 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 5350 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 5351 | ss->id, ss->name); |
073219e9 TH |
5352 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5353 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5354 | ||
aec25020 | 5355 | ss->id = i; |
073219e9 | 5356 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5357 | if (!ss->legacy_name) |
5358 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5359 | |
5360 | if (ss->early_init) | |
15a4c835 | 5361 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5362 | } |
5363 | return 0; | |
5364 | } | |
5365 | ||
6e5c8307 | 5366 | static u16 cgroup_disable_mask __initdata; |
a3e72739 | 5367 | |
ddbcc7e8 | 5368 | /** |
a043e3b2 LZ |
5369 | * cgroup_init - cgroup initialization |
5370 | * | |
5371 | * Register cgroup filesystem and /proc file, and initialize | |
5372 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5373 | */ |
5374 | int __init cgroup_init(void) | |
5375 | { | |
30159ec7 | 5376 | struct cgroup_subsys *ss; |
035f4f51 | 5377 | int ssid; |
ddbcc7e8 | 5378 | |
6e5c8307 | 5379 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
1ed13287 | 5380 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
d62beb7f TH |
5381 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
5382 | BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files)); | |
ddbcc7e8 | 5383 | |
c58632b3 | 5384 | cgroup_rstat_boot(); |
041cd640 | 5385 | |
3942a9bd | 5386 | /* |
2af3024c | 5387 | * The latency of the synchronize_rcu() is too high for cgroups, |
3942a9bd PZ |
5388 | * avoid it at the cost of forcing all readers into the slow path. |
5389 | */ | |
5390 | rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss); | |
5391 | ||
a79a908f AK |
5392 | get_user_ns(init_cgroup_ns.user_ns); |
5393 | ||
54e7b4eb | 5394 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5395 | |
2378d8b8 TH |
5396 | /* |
5397 | * Add init_css_set to the hash table so that dfl_root can link to | |
5398 | * it during init. | |
5399 | */ | |
5400 | hash_add(css_set_table, &init_css_set.hlist, | |
5401 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 5402 | |
35ac1184 | 5403 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5404 | |
54e7b4eb TH |
5405 | mutex_unlock(&cgroup_mutex); |
5406 | ||
172a2c06 | 5407 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5408 | if (ss->early_init) { |
5409 | struct cgroup_subsys_state *css = | |
5410 | init_css_set.subsys[ss->id]; | |
5411 | ||
5412 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5413 | GFP_KERNEL); | |
5414 | BUG_ON(css->id < 0); | |
5415 | } else { | |
5416 | cgroup_init_subsys(ss, false); | |
5417 | } | |
172a2c06 | 5418 | |
2d8f243a TH |
5419 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5420 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5421 | |
5422 | /* | |
c731ae1d LZ |
5423 | * Setting dfl_root subsys_mask needs to consider the |
5424 | * disabled flag and cftype registration needs kmalloc, | |
5425 | * both of which aren't available during early_init. | |
172a2c06 | 5426 | */ |
a3e72739 TH |
5427 | if (cgroup_disable_mask & (1 << ssid)) { |
5428 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5429 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5430 | ss->name); | |
a8ddc821 | 5431 | continue; |
a3e72739 | 5432 | } |
a8ddc821 | 5433 | |
d62beb7f | 5434 | if (cgroup1_ssid_disabled(ssid)) |
223ffb29 JW |
5435 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", |
5436 | ss->name); | |
5437 | ||
a8ddc821 TH |
5438 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
5439 | ||
8cfd8147 TH |
5440 | /* implicit controllers must be threaded too */ |
5441 | WARN_ON(ss->implicit_on_dfl && !ss->threaded); | |
5442 | ||
f6d635ad TH |
5443 | if (ss->implicit_on_dfl) |
5444 | cgrp_dfl_implicit_ss_mask |= 1 << ss->id; | |
5445 | else if (!ss->dfl_cftypes) | |
a7165264 | 5446 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 5447 | |
8cfd8147 TH |
5448 | if (ss->threaded) |
5449 | cgrp_dfl_threaded_ss_mask |= 1 << ss->id; | |
5450 | ||
a8ddc821 TH |
5451 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5452 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5453 | } else { | |
5454 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5455 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5456 | } |
295458e6 VD |
5457 | |
5458 | if (ss->bind) | |
5459 | ss->bind(init_css_set.subsys[ssid]); | |
7af608e4 TH |
5460 | |
5461 | mutex_lock(&cgroup_mutex); | |
5462 | css_populate_dir(init_css_set.subsys[ssid]); | |
5463 | mutex_unlock(&cgroup_mutex); | |
676db4af GK |
5464 | } |
5465 | ||
2378d8b8 TH |
5466 | /* init_css_set.subsys[] has been updated, re-hash */ |
5467 | hash_del(&init_css_set.hlist); | |
5468 | hash_add(css_set_table, &init_css_set.hlist, | |
5469 | css_set_hash(init_css_set.subsys)); | |
5470 | ||
035f4f51 TH |
5471 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5472 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 5473 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
3f3942ac | 5474 | WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show)); |
ddbcc7e8 | 5475 | |
2bd59d48 | 5476 | return 0; |
ddbcc7e8 | 5477 | } |
b4f48b63 | 5478 | |
e5fca243 TH |
5479 | static int __init cgroup_wq_init(void) |
5480 | { | |
5481 | /* | |
5482 | * There isn't much point in executing destruction path in | |
5483 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5484 | * Use 1 for @max_active. |
e5fca243 TH |
5485 | * |
5486 | * We would prefer to do this in cgroup_init() above, but that | |
5487 | * is called before init_workqueues(): so leave this until after. | |
5488 | */ | |
1a11533f | 5489 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 TH |
5490 | BUG_ON(!cgroup_destroy_wq); |
5491 | return 0; | |
5492 | } | |
5493 | core_initcall(cgroup_wq_init); | |
5494 | ||
69fd5c39 SL |
5495 | void cgroup_path_from_kernfs_id(const union kernfs_node_id *id, |
5496 | char *buf, size_t buflen) | |
5497 | { | |
5498 | struct kernfs_node *kn; | |
5499 | ||
5500 | kn = kernfs_get_node_by_id(cgrp_dfl_root.kf_root, id); | |
5501 | if (!kn) | |
5502 | return; | |
5503 | kernfs_path(kn, buf, buflen); | |
5504 | kernfs_put(kn); | |
5505 | } | |
5506 | ||
a424316c PM |
5507 | /* |
5508 | * proc_cgroup_show() | |
5509 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5510 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5511 | */ |
006f4ac4 ZL |
5512 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5513 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5514 | { |
4c737b41 | 5515 | char *buf; |
a424316c | 5516 | int retval; |
3dd06ffa | 5517 | struct cgroup_root *root; |
a424316c PM |
5518 | |
5519 | retval = -ENOMEM; | |
e61734c5 | 5520 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5521 | if (!buf) |
5522 | goto out; | |
5523 | ||
a424316c | 5524 | mutex_lock(&cgroup_mutex); |
82d6489d | 5525 | spin_lock_irq(&css_set_lock); |
a424316c | 5526 | |
985ed670 | 5527 | for_each_root(root) { |
a424316c | 5528 | struct cgroup_subsys *ss; |
bd89aabc | 5529 | struct cgroup *cgrp; |
b85d2040 | 5530 | int ssid, count = 0; |
a424316c | 5531 | |
a7165264 | 5532 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
985ed670 TH |
5533 | continue; |
5534 | ||
2c6ab6d2 | 5535 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5536 | if (root != &cgrp_dfl_root) |
5537 | for_each_subsys(ss, ssid) | |
5538 | if (root->subsys_mask & (1 << ssid)) | |
5539 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5540 | ss->legacy_name); |
c6d57f33 PM |
5541 | if (strlen(root->name)) |
5542 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5543 | root->name); | |
a424316c | 5544 | seq_putc(m, ':'); |
2e91fa7f | 5545 | |
7717f7ba | 5546 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5547 | |
5548 | /* | |
5549 | * On traditional hierarchies, all zombie tasks show up as | |
5550 | * belonging to the root cgroup. On the default hierarchy, | |
5551 | * while a zombie doesn't show up in "cgroup.procs" and | |
5552 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5553 | * reporting the cgroup it belonged to before exiting. If | |
5554 | * the cgroup is removed before the zombie is reaped, | |
5555 | * " (deleted)" is appended to the cgroup path. | |
5556 | */ | |
5557 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
4c737b41 | 5558 | retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX, |
a79a908f | 5559 | current->nsproxy->cgroup_ns); |
e0223003 | 5560 | if (retval >= PATH_MAX) |
2e91fa7f | 5561 | retval = -ENAMETOOLONG; |
e0223003 | 5562 | if (retval < 0) |
2e91fa7f | 5563 | goto out_unlock; |
4c737b41 TH |
5564 | |
5565 | seq_puts(m, buf); | |
2e91fa7f | 5566 | } else { |
4c737b41 | 5567 | seq_puts(m, "/"); |
e61734c5 | 5568 | } |
2e91fa7f | 5569 | |
2e91fa7f TH |
5570 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) |
5571 | seq_puts(m, " (deleted)\n"); | |
5572 | else | |
5573 | seq_putc(m, '\n'); | |
a424316c PM |
5574 | } |
5575 | ||
006f4ac4 | 5576 | retval = 0; |
a424316c | 5577 | out_unlock: |
82d6489d | 5578 | spin_unlock_irq(&css_set_lock); |
a424316c | 5579 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5580 | kfree(buf); |
5581 | out: | |
5582 | return retval; | |
5583 | } | |
5584 | ||
b4f48b63 | 5585 | /** |
eaf797ab | 5586 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5587 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5588 | * |
eaf797ab TH |
5589 | * A task is associated with the init_css_set until cgroup_post_fork() |
5590 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5591 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5592 | */ |
5593 | void cgroup_fork(struct task_struct *child) | |
5594 | { | |
eaf797ab | 5595 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5596 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5597 | } |
5598 | ||
7e47682e AS |
5599 | /** |
5600 | * cgroup_can_fork - called on a new task before the process is exposed | |
5601 | * @child: the task in question. | |
5602 | * | |
5603 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5604 | * returns an error, the fork aborts with that error code. This allows for | |
5605 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5606 | */ | |
b53202e6 | 5607 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
5608 | { |
5609 | struct cgroup_subsys *ss; | |
5610 | int i, j, ret; | |
5611 | ||
b4e0eeaf | 5612 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
b53202e6 | 5613 | ret = ss->can_fork(child); |
7e47682e AS |
5614 | if (ret) |
5615 | goto out_revert; | |
b4e0eeaf | 5616 | } while_each_subsys_mask(); |
7e47682e AS |
5617 | |
5618 | return 0; | |
5619 | ||
5620 | out_revert: | |
5621 | for_each_subsys(ss, j) { | |
5622 | if (j >= i) | |
5623 | break; | |
5624 | if (ss->cancel_fork) | |
b53202e6 | 5625 | ss->cancel_fork(child); |
7e47682e AS |
5626 | } |
5627 | ||
5628 | return ret; | |
5629 | } | |
5630 | ||
5631 | /** | |
5632 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5633 | * @child: the task in question | |
5634 | * | |
5635 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5636 | * cgroup_can_fork() succeded. | |
5637 | */ | |
b53202e6 | 5638 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
5639 | { |
5640 | struct cgroup_subsys *ss; | |
5641 | int i; | |
5642 | ||
5643 | for_each_subsys(ss, i) | |
5644 | if (ss->cancel_fork) | |
b53202e6 | 5645 | ss->cancel_fork(child); |
7e47682e AS |
5646 | } |
5647 | ||
817929ec | 5648 | /** |
a043e3b2 LZ |
5649 | * cgroup_post_fork - called on a new task after adding it to the task list |
5650 | * @child: the task in question | |
5651 | * | |
5edee61e TH |
5652 | * Adds the task to the list running through its css_set if necessary and |
5653 | * call the subsystem fork() callbacks. Has to be after the task is | |
5654 | * visible on the task list in case we race with the first call to | |
0942eeee | 5655 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5656 | * list. |
a043e3b2 | 5657 | */ |
b53202e6 | 5658 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 5659 | { |
30159ec7 | 5660 | struct cgroup_subsys *ss; |
5edee61e TH |
5661 | int i; |
5662 | ||
3ce3230a | 5663 | /* |
251f8c03 | 5664 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5665 | * function sets use_task_css_set_links before grabbing |
5666 | * tasklist_lock and we just went through tasklist_lock to add | |
5667 | * @child, it's guaranteed that either we see the set | |
5668 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5669 | * @child during its iteration. | |
5670 | * | |
5671 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 5672 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
5673 | * association is stable, and, on completion of the parent's |
5674 | * migration, @child is visible in the source of migration or | |
5675 | * already in the destination cgroup. This guarantee is necessary | |
5676 | * when implementing operations which need to migrate all tasks of | |
5677 | * a cgroup to another. | |
5678 | * | |
251f8c03 | 5679 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5680 | * will remain in init_css_set. This is safe because all tasks are |
5681 | * in the init_css_set before cg_links is enabled and there's no | |
5682 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5683 | */ |
817929ec | 5684 | if (use_task_css_set_links) { |
eaf797ab TH |
5685 | struct css_set *cset; |
5686 | ||
82d6489d | 5687 | spin_lock_irq(&css_set_lock); |
0e1d768f | 5688 | cset = task_css_set(current); |
eaf797ab | 5689 | if (list_empty(&child->cg_list)) { |
eaf797ab | 5690 | get_css_set(cset); |
73a7242a | 5691 | cset->nr_tasks++; |
f6d7d049 | 5692 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 5693 | } |
82d6489d | 5694 | spin_unlock_irq(&css_set_lock); |
817929ec | 5695 | } |
5edee61e TH |
5696 | |
5697 | /* | |
5698 | * Call ss->fork(). This must happen after @child is linked on | |
5699 | * css_set; otherwise, @child might change state between ->fork() | |
5700 | * and addition to css_set. | |
5701 | */ | |
b4e0eeaf | 5702 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 5703 | ss->fork(child); |
b4e0eeaf | 5704 | } while_each_subsys_mask(); |
817929ec | 5705 | } |
5edee61e | 5706 | |
b4f48b63 PM |
5707 | /** |
5708 | * cgroup_exit - detach cgroup from exiting task | |
5709 | * @tsk: pointer to task_struct of exiting process | |
5710 | * | |
5711 | * Description: Detach cgroup from @tsk and release it. | |
5712 | * | |
5713 | * Note that cgroups marked notify_on_release force every task in | |
5714 | * them to take the global cgroup_mutex mutex when exiting. | |
5715 | * This could impact scaling on very large systems. Be reluctant to | |
5716 | * use notify_on_release cgroups where very high task exit scaling | |
5717 | * is required on large systems. | |
5718 | * | |
0e1d768f TH |
5719 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5720 | * call cgroup_exit() while the task is still competent to handle | |
5721 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5722 | * each hierarchy for the remainder of its exit. No need to bother with | |
5723 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5724 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5725 | */ |
1ec41830 | 5726 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5727 | { |
30159ec7 | 5728 | struct cgroup_subsys *ss; |
5abb8855 | 5729 | struct css_set *cset; |
d41d5a01 | 5730 | int i; |
817929ec PM |
5731 | |
5732 | /* | |
0e1d768f | 5733 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 5734 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 5735 | */ |
0de0942d TH |
5736 | cset = task_css_set(tsk); |
5737 | ||
817929ec | 5738 | if (!list_empty(&tsk->cg_list)) { |
82d6489d | 5739 | spin_lock_irq(&css_set_lock); |
f6d7d049 | 5740 | css_set_move_task(tsk, cset, NULL, false); |
73a7242a | 5741 | cset->nr_tasks--; |
82d6489d | 5742 | spin_unlock_irq(&css_set_lock); |
2e91fa7f TH |
5743 | } else { |
5744 | get_css_set(cset); | |
817929ec PM |
5745 | } |
5746 | ||
cb4a3167 | 5747 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 5748 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 5749 | ss->exit(tsk); |
b4e0eeaf | 5750 | } while_each_subsys_mask(); |
2e91fa7f | 5751 | } |
30159ec7 | 5752 | |
2e91fa7f TH |
5753 | void cgroup_free(struct task_struct *task) |
5754 | { | |
5755 | struct css_set *cset = task_css_set(task); | |
afcf6c8b TH |
5756 | struct cgroup_subsys *ss; |
5757 | int ssid; | |
5758 | ||
b4e0eeaf | 5759 | do_each_subsys_mask(ss, ssid, have_free_callback) { |
afcf6c8b | 5760 | ss->free(task); |
b4e0eeaf | 5761 | } while_each_subsys_mask(); |
d41d5a01 | 5762 | |
2e91fa7f | 5763 | put_css_set(cset); |
b4f48b63 | 5764 | } |
697f4161 | 5765 | |
8bab8dde PM |
5766 | static int __init cgroup_disable(char *str) |
5767 | { | |
30159ec7 | 5768 | struct cgroup_subsys *ss; |
8bab8dde | 5769 | char *token; |
30159ec7 | 5770 | int i; |
8bab8dde PM |
5771 | |
5772 | while ((token = strsep(&str, ",")) != NULL) { | |
5773 | if (!*token) | |
5774 | continue; | |
be45c900 | 5775 | |
3ed80a62 | 5776 | for_each_subsys(ss, i) { |
3e1d2eed TH |
5777 | if (strcmp(token, ss->name) && |
5778 | strcmp(token, ss->legacy_name)) | |
5779 | continue; | |
a3e72739 | 5780 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
5781 | } |
5782 | } | |
5783 | return 1; | |
5784 | } | |
5785 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5786 | |
5cf8114d WL |
5787 | void __init __weak enable_debug_cgroup(void) { } |
5788 | ||
5789 | static int __init enable_cgroup_debug(char *str) | |
5790 | { | |
5791 | cgroup_debug = true; | |
5792 | enable_debug_cgroup(); | |
5793 | return 1; | |
5794 | } | |
5795 | __setup("cgroup_debug", enable_cgroup_debug); | |
5796 | ||
b77d7b60 | 5797 | /** |
ec903c0c | 5798 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5799 | * @dentry: directory dentry of interest |
5800 | * @ss: subsystem of interest | |
b77d7b60 | 5801 | * |
5a17f543 TH |
5802 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5803 | * to get the corresponding css and return it. If such css doesn't exist | |
5804 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5805 | */ |
ec903c0c TH |
5806 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5807 | struct cgroup_subsys *ss) | |
e5d1367f | 5808 | { |
2bd59d48 | 5809 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 5810 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 5811 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 5812 | struct cgroup *cgrp; |
e5d1367f | 5813 | |
35cf0836 | 5814 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
5815 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
5816 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5817 | return ERR_PTR(-EBADF); |
5818 | ||
5a17f543 TH |
5819 | rcu_read_lock(); |
5820 | ||
2bd59d48 TH |
5821 | /* |
5822 | * This path doesn't originate from kernfs and @kn could already | |
5823 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 5824 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 | 5825 | */ |
e0aed7c7 | 5826 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); |
2bd59d48 TH |
5827 | if (cgrp) |
5828 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5829 | |
ec903c0c | 5830 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5831 | css = ERR_PTR(-ENOENT); |
5832 | ||
5833 | rcu_read_unlock(); | |
5834 | return css; | |
e5d1367f | 5835 | } |
e5d1367f | 5836 | |
1cb650b9 LZ |
5837 | /** |
5838 | * css_from_id - lookup css by id | |
5839 | * @id: the cgroup id | |
5840 | * @ss: cgroup subsys to be looked into | |
5841 | * | |
5842 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5843 | * Should be called under rcu_read_lock(). | |
5844 | */ | |
5845 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5846 | { | |
6fa4918d | 5847 | WARN_ON_ONCE(!rcu_read_lock_held()); |
d6ccc55e | 5848 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
5849 | } |
5850 | ||
16af4396 TH |
5851 | /** |
5852 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
5853 | * @path: path on the default hierarchy | |
5854 | * | |
5855 | * Find the cgroup at @path on the default hierarchy, increment its | |
5856 | * reference count and return it. Returns pointer to the found cgroup on | |
5857 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
5858 | * if @path points to a non-directory. | |
5859 | */ | |
5860 | struct cgroup *cgroup_get_from_path(const char *path) | |
5861 | { | |
5862 | struct kernfs_node *kn; | |
5863 | struct cgroup *cgrp; | |
5864 | ||
5865 | mutex_lock(&cgroup_mutex); | |
5866 | ||
5867 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
5868 | if (kn) { | |
5869 | if (kernfs_type(kn) == KERNFS_DIR) { | |
5870 | cgrp = kn->priv; | |
a590b90d | 5871 | cgroup_get_live(cgrp); |
16af4396 TH |
5872 | } else { |
5873 | cgrp = ERR_PTR(-ENOTDIR); | |
5874 | } | |
5875 | kernfs_put(kn); | |
5876 | } else { | |
5877 | cgrp = ERR_PTR(-ENOENT); | |
5878 | } | |
5879 | ||
5880 | mutex_unlock(&cgroup_mutex); | |
5881 | return cgrp; | |
5882 | } | |
5883 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
5884 | ||
1f3fe7eb MKL |
5885 | /** |
5886 | * cgroup_get_from_fd - get a cgroup pointer from a fd | |
5887 | * @fd: fd obtained by open(cgroup2_dir) | |
5888 | * | |
5889 | * Find the cgroup from a fd which should be obtained | |
5890 | * by opening a cgroup directory. Returns a pointer to the | |
5891 | * cgroup on success. ERR_PTR is returned if the cgroup | |
5892 | * cannot be found. | |
5893 | */ | |
5894 | struct cgroup *cgroup_get_from_fd(int fd) | |
5895 | { | |
5896 | struct cgroup_subsys_state *css; | |
5897 | struct cgroup *cgrp; | |
5898 | struct file *f; | |
5899 | ||
5900 | f = fget_raw(fd); | |
5901 | if (!f) | |
5902 | return ERR_PTR(-EBADF); | |
5903 | ||
5904 | css = css_tryget_online_from_dir(f->f_path.dentry, NULL); | |
5905 | fput(f); | |
5906 | if (IS_ERR(css)) | |
5907 | return ERR_CAST(css); | |
5908 | ||
5909 | cgrp = css->cgroup; | |
5910 | if (!cgroup_on_dfl(cgrp)) { | |
5911 | cgroup_put(cgrp); | |
5912 | return ERR_PTR(-EBADF); | |
5913 | } | |
5914 | ||
5915 | return cgrp; | |
5916 | } | |
5917 | EXPORT_SYMBOL_GPL(cgroup_get_from_fd); | |
5918 | ||
bd1060a1 TH |
5919 | /* |
5920 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
5921 | * definition in cgroup-defs.h. | |
5922 | */ | |
5923 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
5924 | ||
5925 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
5926 | ||
3fa4cc9c | 5927 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
5928 | static bool cgroup_sk_alloc_disabled __read_mostly; |
5929 | ||
5930 | void cgroup_sk_alloc_disable(void) | |
5931 | { | |
5932 | if (cgroup_sk_alloc_disabled) | |
5933 | return; | |
5934 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
5935 | cgroup_sk_alloc_disabled = true; | |
5936 | } | |
5937 | ||
5938 | #else | |
5939 | ||
5940 | #define cgroup_sk_alloc_disabled false | |
5941 | ||
5942 | #endif | |
5943 | ||
5944 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
5945 | { | |
5946 | if (cgroup_sk_alloc_disabled) | |
5947 | return; | |
5948 | ||
d979a39d JW |
5949 | /* Socket clone path */ |
5950 | if (skcd->val) { | |
a590b90d TH |
5951 | /* |
5952 | * We might be cloning a socket which is left in an empty | |
5953 | * cgroup and the cgroup might have already been rmdir'd. | |
5954 | * Don't use cgroup_get_live(). | |
5955 | */ | |
d979a39d JW |
5956 | cgroup_get(sock_cgroup_ptr(skcd)); |
5957 | return; | |
5958 | } | |
5959 | ||
bd1060a1 TH |
5960 | rcu_read_lock(); |
5961 | ||
5962 | while (true) { | |
5963 | struct css_set *cset; | |
5964 | ||
5965 | cset = task_css_set(current); | |
5966 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
5967 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
5968 | break; | |
5969 | } | |
5970 | cpu_relax(); | |
5971 | } | |
5972 | ||
5973 | rcu_read_unlock(); | |
5974 | } | |
5975 | ||
5976 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
5977 | { | |
5978 | cgroup_put(sock_cgroup_ptr(skcd)); | |
5979 | } | |
5980 | ||
5981 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
5982 | ||
30070984 | 5983 | #ifdef CONFIG_CGROUP_BPF |
324bda9e AS |
5984 | int cgroup_bpf_attach(struct cgroup *cgrp, struct bpf_prog *prog, |
5985 | enum bpf_attach_type type, u32 flags) | |
5986 | { | |
5987 | int ret; | |
5988 | ||
5989 | mutex_lock(&cgroup_mutex); | |
5990 | ret = __cgroup_bpf_attach(cgrp, prog, type, flags); | |
5991 | mutex_unlock(&cgroup_mutex); | |
5992 | return ret; | |
5993 | } | |
5994 | int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog, | |
5995 | enum bpf_attach_type type, u32 flags) | |
30070984 | 5996 | { |
7f677633 | 5997 | int ret; |
30070984 DM |
5998 | |
5999 | mutex_lock(&cgroup_mutex); | |
324bda9e | 6000 | ret = __cgroup_bpf_detach(cgrp, prog, type, flags); |
30070984 | 6001 | mutex_unlock(&cgroup_mutex); |
7f677633 | 6002 | return ret; |
30070984 | 6003 | } |
468e2f64 AS |
6004 | int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr, |
6005 | union bpf_attr __user *uattr) | |
6006 | { | |
6007 | int ret; | |
6008 | ||
6009 | mutex_lock(&cgroup_mutex); | |
6010 | ret = __cgroup_bpf_query(cgrp, attr, uattr); | |
6011 | mutex_unlock(&cgroup_mutex); | |
6012 | return ret; | |
6013 | } | |
30070984 | 6014 | #endif /* CONFIG_CGROUP_BPF */ |
01ee6cfb RG |
6015 | |
6016 | #ifdef CONFIG_SYSFS | |
6017 | static ssize_t show_delegatable_files(struct cftype *files, char *buf, | |
6018 | ssize_t size, const char *prefix) | |
6019 | { | |
6020 | struct cftype *cft; | |
6021 | ssize_t ret = 0; | |
6022 | ||
6023 | for (cft = files; cft && cft->name[0] != '\0'; cft++) { | |
6024 | if (!(cft->flags & CFTYPE_NS_DELEGATABLE)) | |
6025 | continue; | |
6026 | ||
6027 | if (prefix) | |
6028 | ret += snprintf(buf + ret, size - ret, "%s.", prefix); | |
6029 | ||
6030 | ret += snprintf(buf + ret, size - ret, "%s\n", cft->name); | |
6031 | ||
4d9ebbe2 | 6032 | if (WARN_ON(ret >= size)) |
01ee6cfb | 6033 | break; |
01ee6cfb RG |
6034 | } |
6035 | ||
6036 | return ret; | |
6037 | } | |
6038 | ||
6039 | static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr, | |
6040 | char *buf) | |
6041 | { | |
6042 | struct cgroup_subsys *ss; | |
6043 | int ssid; | |
6044 | ssize_t ret = 0; | |
6045 | ||
6046 | ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret, | |
6047 | NULL); | |
6048 | ||
6049 | for_each_subsys(ss, ssid) | |
6050 | ret += show_delegatable_files(ss->dfl_cftypes, buf + ret, | |
6051 | PAGE_SIZE - ret, | |
6052 | cgroup_subsys_name[ssid]); | |
6053 | ||
6054 | return ret; | |
6055 | } | |
6056 | static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate); | |
6057 | ||
5f2e6734 RG |
6058 | static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr, |
6059 | char *buf) | |
6060 | { | |
6061 | return snprintf(buf, PAGE_SIZE, "nsdelegate\n"); | |
6062 | } | |
6063 | static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features); | |
6064 | ||
01ee6cfb RG |
6065 | static struct attribute *cgroup_sysfs_attrs[] = { |
6066 | &cgroup_delegate_attr.attr, | |
5f2e6734 | 6067 | &cgroup_features_attr.attr, |
01ee6cfb RG |
6068 | NULL, |
6069 | }; | |
6070 | ||
6071 | static const struct attribute_group cgroup_sysfs_attr_group = { | |
6072 | .attrs = cgroup_sysfs_attrs, | |
6073 | .name = "cgroup", | |
6074 | }; | |
6075 | ||
6076 | static int __init cgroup_sysfs_init(void) | |
6077 | { | |
6078 | return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group); | |
6079 | } | |
6080 | subsys_initcall(cgroup_sysfs_init); | |
6081 | #endif /* CONFIG_SYSFS */ |